Functional attrition and Quality of Life in Advanced Cancer Trial: modeling patients’ trajectories and evaluating the impact of missing outcome data handling on quality-of-life predictive model performance

Functional attrition and Quality of Life in Advanced Cancer Trial: modeling patients’ trajectories and evaluating the impact of missing outcome data handling on quality-of-life predictive model performance | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Functional attrition and Quality of Life in Advanced Cancer Trial: modeling patients’ trajectories and evaluating the impact of missing outcome data handling on quality-of-life predictive model performance Selassi Komi Gayi, Prerna Priyam, Sangenis Ayao Assogba, Samadou Tchakondo, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8407124/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract Background: Functional attrition (non-response) of Patient-Reported Outcomes (PROs) and Quality of Life (QoL) deterioration are common and non-random in oncology trials. While functional attrition is informative, it is often treated as simple missing data, which biases the interpretation of QoL trajectories and prediction models. Objectives and Methods: This study aimed (1) To examine prognostic factors of transitions between active participation, functional attrition, and death using multinomial models and (2) To evaluate how different methods of dealing with missing data affect the reliability of QoL deterioration prediction models. A discrete-state modeling framework was used to analyze transitions between active participation, functional attrition, and death in advanced cancer patients. QoL decline was then predicted using LASSO, Random Forest, and XGBoost algorithms, while Inverse Probability of Completion Weighting (IPCW) under a conditional Missing At Random (MAR) assumption was systematically compared against Multiple Imputation by Chained Equations (MICE). At the end, we examined robustness to deviations from MAR through delta-adjusted Missing Not At Random (MNAR) sensitivity analyses. Results: Functional attrition accumulated more rapidly than death within the first 12 weeks, confirming its high prevalence and distinct determinants. Lower educational attainment was associated with functional attrition, whereas poor performance status and depressive symptoms primarily predicted mortality. Among predictive models, LASSO combined with MICE achieved the best performance (median AUC = 0.671 [IQR 0.136]), reflecting the modest predictability of dynamic PRO outcomes when relying on baseline covariates. Methodological comparison demonstrated superior discrimination and substantially improved calibration for MICE relative to IPCW, with IPCW models showing marked miscalibration. These findings remained stable across MNAR sensitivity scenarios. Conclusion: By modeling the functional attrition as a distinct state, we revealed its specific socio-demographic determinants and clarified how covariate profiles shape patient trajectories . For QoL prediction under conditional MAR, MICE yields more stable and better-calibrated models than IPCW. Robustness across MNAR scenarios supports the methodological validity of these conclusions and highlights the importance of principled handling of informative non-response in longitudinal PRO research. Functional attrition quality of life patient-reported outcomes advanced cancer Figures Figure 1 Figure 2 Figure 3 Figure 4 INTRODUCTION Patient Reported Outcomes (PROs) such as quality of life (QoL) measures have become increasingly important endpoints in Oncology Clinical Trials [ 1 , 2 ]. PROs such as FACT (Functional Assessment of Cancer Therapy) measure validated patient-centered assessment of a patient's physical, emotional and social well-being [ 3 ]. Inclusion of QoL endpoints was strongly recommended by both the CONSORT-PRO extension [ 4 ] and the ISOQOL (International Society for Quality-of-Life Research) [ 5 ]. However, QoL data are commonly impacted by "functional attrition" (patients alive and formally enrolled in the study, however they do not complete QoL assessments at follow-up). Functional attrition is different than formal withdraw or death and occurs most commonly in advanced cancer settings due to fatigue, low education level, lack of social support and/or psychological distress [ 6 , 7 ]. Conventional approaches usually treat functional attrition as "missing at random", and attempt to recover the missing data through multiple imputation, etc. [ 8 ]; however, this assumption is rarely met because functional attrition is typically informative and represents socio- demographic and/psychosocial characteristics that are also predictive of QoL outcomes; ignoring functional attrition can introduce bias and mask clinically significant trends [ 9 ]. Recent methodologic research has demonstrated the importance of accounting for the type of missing data in PRO analyses; intercurrent events [ 10 ]; and regulatory frameworks such as the ICH E9(R3) addendum on estimands, which emphasize the need to develop methods to address death, dropout and treatment discontinuation [ 11 ]. While these primarily address survival and treatment-related events, the same reasoning can be extended to PROs, functional attrition is not just a nuisance but an informative state that should be explicitly modelled. At the same time, there is increasing interest in the development of predictive models of QoL deterioration, aiming to identify patients at high risk for early decline and target supportive interventions [ 12 , 13 ]. Modern statistical learning methods, including penalized regression, random forests, and gradient boosting, have been applied to PRO data, most studies have not explicitly addressed functional attrition [ 14 , 15 ]. Importantly, since only a subset of patients remains evaluable at follow-up, the performance and robustness of these models can be heavily influenced by how outcome missing data are handled. The choice between different methods such as multiple imputation and Inverse Probability of Completion Weighting (IPCW) is a vital methodological question, as each approach relies on different assumptions and can lead to different conclusions. Following this background, we propose a dual analytical strategy applied to data from the randomized trial NCT02349412. First, we model patient trajectories across three states: Active (completed QoL), Functional attrition (alive but missing QoL), and Death using a discrete-time multinomial regression framework. This method enables us to characterize baseline predictors of attrition and death and to estimate predicted state probabilities at 12 and 24 weeks, including the marginal effects of key covariates on the likelihood of being in each state. Second, we evaluate the impact of missing outcome data handling on prognostic prediction by developing predictive models of QoL deterioration at 24 weeks among patients active at baseline, comparing penalized regression and machine learning approaches and using two distinct methodologies to handle missing outcome data: multiple imputation and IPCW. As the missingness of quality of life data may not always satisfy the missing at random (MAR) assumption, we also conducted sensitivity analyses under missing not at random (MNAR) mechanisms to examine the robustness of the findings. By integrating these perspectives, we aim to clarify the role of functional attrition, offer significant insights into the methodological challenges of longitudinal QoL studies and to guide future research in the development of robust predictive models. METHODOLOGY Data Source and Study Population : The data analysed come from a multicenter randomized controlled trial (ClinicalTrials.gov identifier: NCT02349412), registered on 23 January 2015, which assessed the impact of early palliative care on quality of life, psychological outcomes, and health care use in adult patients with advanced cancer. Eligibility criteria consisted of adults (≥ 18 years) with an advanced malignancy, ECOG performance status of 0–2, and expected survival ≥ 6 months who were randomized to receive either standard oncology care or integrated early palliative care. Socio-demographic, clinical, vital status, and patient-reported outcomes (FACT-G and HADS) were measured at baseline and at 12- and 24-weeks post-baseline. In this secondary analysis, only de-identified patients alive and with complete baseline QoL data were included. Outcome and Predictors Patient states at 12- and 24-weeks post-baseline (Active (A), Functional Attrition (F), Deceased (D)) and QoL deterioration at 24 weeks (FACT-G score decrease ≥ minimal clinically important differences of 0, -5, -7, -10) were primary outcomes of interest. Predictors of these outcomes included socio-demographic characteristics (age, sex, education level, whether partnered, race/ethnicity), clinical characteristics (ECOG, tumor type), and baseline FACT-G and HADS scores. Data Processing After completing the process of cleaning and harmonizing the data, functional attrition was explicitly defined as living patients without QoL data available at follow-up time points; thereby allowing for the creation of three discrete states (A/F/D) and further enabling modeling of transitions to state F, and comparative analyses for descriptive, multinomial and predictive analyses. Listwise deletion was applied to handle missing covariates when they were less than 5%, because the loss of data would not significantly affect the total statistical power of the study. Statistical Analysis Transitions between states were modelled using penalized multinomial logistic regression (LASSO/ridge) for variable selection, refitted for ORs and 95% CIs. For enhanced clinical interpretation, Average Marginal Effects (AMEs) were calculated, quantifying predicted probability changes for each covariate. Example clinical profiles combining age, ECOG, anxiety, caregiver presence, tumor type and treatment arm were constructed, with associated transition probabilities summarized to illustrate risk stratification. Predictive models for QoL decline at 24 weeks were developed using LASSO, Random Forest, and XGBoost, evaluating missingness approaches (MICE, IPCW) and conducting MNAR sensitivity analyses. Model validation used repeated cross-validation (AUC, calibration slope, Brier score, median/IQR), See Fig. 1. All statistical analyses were conducted using R version 4.3.2 with appropriate packages (dplyr, glmnet, mice, caret, pROC, ranger, xgboost). RESULTS Study Population From 405 patients enrolled, 352 (87%) completed QoL assessment at baseline and were therefore included (Table 1). Mean age was 65.3 years and 56% were men. The median magnitude of ECOG performance was 0.94. Baseline QoL (FACT-G) was 73.8 and anxiety/depression symptoms moderate in level. Lung cancer was the most prevalent diagnosis (61% of patients), followed by hepatobiliary (32%), and esophageal/GI cancers (8%). Over half had a college education and 70.5% reported caregiver spouse or partner. Table 1. Baseline Characteristics of the Study Cohort (N = 352) Characteristic Overall (N = 352) Age, years – mean (SD) 65.3 (9.9) Sex, male – n (%) 196 (55.7) ECOG performance status – mean (SD) 0.94 (0.62) Baseline FACT score – mean (SD) 73.8 (16.4) HADS anxiety score – mean (SD) 7.2 (3.5) HADS depression score – mean (SD) 5.6 (4.2) Tumor type – n (%) Esophageal/GEJ/Gastric 28 (8.0) Hepatic/Biliary/Pancreatic/Unknown primary 111 (31.5) Lung 213 (60.5) Trial arm 2 – n (%) 183 (52.0) Education – n (%) College+ 177 (52.2) High school 128 (37.8) Low 34 (10.0) Race/ethnicity – n (%) Asian 14 (4.2) Black 42 (12.5) Minority (other) 6 (1.8) White 274 (81.5) Not married – n (%) 130 (37.7) Number of prior visits – mean (SD) 1.81 (2.38) Caregiver partner present – n (%) 248 (70.5) Attrition and State Transitions At week 12, 55% remained active but 29% were functionally attrited and 16.5% deceased. Of those active at week 12, 67% remained active at week 24, 21% were attrited, and 12% were deceased. Among those functionally attrited at week 12, 20% had returned to active status by week 24, 52% remained attrited, and 28% were deceased. (Figure 1. Patient state transitions). Figure 2. Patient state transitions from baseline (T0) to week 12 and week 24. A = active; F = functional attrition; D = death. Baseline Characteristics by Attrition Status Baseline profiles (Supplementary Table S1) showed that patients who died by week 12 were older on average, had generally poorer ECOG, lower QoL, and higher depression scores, compared to patients who were active. Those who had functionally attrited at week 12 had intermediate characteristics including lower baseline quality of life scores and more anxiety than active patients, though less severe than those who died. These trends were observed again at week 24. Other sociodemographic variables such as sex, race, education, caregiver present showed no clinically meaningful differences between patient state. All comparisons were exploratory (alpha = 0.05, two-tailed, no adjustment). Multinomial Modeling of State Transitions In multinomial logistic regression (Table 2), lower education significantly predictive of functional attrition at T0 →T12 (high school completed vs college OR=2.07; low vs college =3.26), the presence of a caregiver reduced the likelihood of functional attrition, although this association was not statistically significant; Age, ECOG, and depression are all predictive of mortality. The Average Marginal Effects (AME) show that each unit increase in ECOG related to increase of ~8% risk of functional attrition and ~12% increase in risk of mortality from T0→T12 and the Presence of a caregiver raised the likelihood of being active by 11%. At T12→T24, however these marginal changes were not supported by statistically significant multinomial coefficients for the attrition state and should therefore be interpreted descriptively rather than as evidence of a robust association. The type of tumor appears particularly influential with respect to risk of dying (Supplementary Table S2, Figure 3). Table 2 . Multinomial Logistic Regression Results for State Transitions (Odds ratios [OR] and 95% confidence intervals relative to remaining active [A]) Transition Outcome Variable OR 95% CI p-value T0 → T12 Attrition(F) Age (per year) 1.00 0.97 – 1.03 0.99 Male (vs female) 0.84 0.49 – 1.46 0.54 ECOG (per unit) 1.00 0.62 – 1.60 0.98 FACT baseline (per point) 1.00 0.98 – 1.03 0.87 Anxiety (HADS-A, per point) 1.07 0.97 – 1.19 0.17 Depression (HADS-D, per point) 0.99 0.89 – 1.10 0.82 Tumor: Hepatobiliary vs Esophageal/GEJ 1.04 0.36 – 3.02 0.95 Tumor: Lung vs Esophageal/GEJ 0.65 0.23 – 1.79 0.40 Education: High school vs College+ 2.07 1.16 – 3.69 0.014 Education: Low vs College+ 3.26 1.31 – 8.12 0.011 Caregiver present 0.57 0.32 – 1.01 0.053 Death (D) Age (per year) 1.05 1.01 – 1.09 0.010 Male (vs female) 0.73 0.37 – 1.44 0.36 ECOG (per unit) 2.37 1.31 – 4.29 0.004 FACT baseline (per point) 0.98 0.95 – 1.01 0.13 Anxiety (HADS-A, per point) 0.89 0.79 – 1.01 0.075 Depression (HADS-D, per point) 1.14 1.02 – 1.28 0.027 Tumor: Hepatobiliary vs Esophageal/GEJ 1.39 0.32 – 5.94 0.66 Tumor: Lung vs Esophageal/GEJ 0.60 0.14 – 2.50 0.48 Education: High school vs College+ 1.55 0.75 – 3.22 0.24 Education: Low vs College+ 0.84 0.23 – 3.03 0.79 Caregiver present 0.73 0.35 – 1.54 0.41 T12 → T24 Attrition(F) Age (per year) 0.99 0.96 – 1.03 0.68 Male (vs female) 1.14 0.53 – 2.44 0.73 ECOG (per unit) 1.55 0.80 – 3.00 0.19 FACT baseline (per point) 0.99 0.97 – 1.02 0.66 Anxiety (HADS-A, per point) 0.91 0.79 – 1.05 0.19 Tumor: Hepatobiliary vs Esophageal/GEJ 0.51 0.10 – 2.49 0.40 Tumor: Lung vs Esophageal/GEJ 0.62 0.14 – 2.78 0.53 Arm 2 vs Arm 1 0.73 0.34 – 1.54 0.41 Death (D) Age (per year) 1.00 0.95 – 1.05 0.91 Male (vs female) 0.86 0.31 – 2.35 0.76 ECOG (per unit) 1.57 0.64 – 3.86 0.33 FACT baseline (per point) 0.98 0.94 – 1.02 0.34 Anxiety (HADS-A, per point) 0.80 0.66 – 0.98 0.029 Tumor: Hepatobiliary vs Esophageal/GEJ 0.08 0.01 – 0.55 0.011 Tumor: Lung vs Esophageal/GEJ 0.24 0.05 – 1.07 0.062 Arm 2 vs Arm 1 0.63 0.23 – 1.73 0.37 Figure 3 . Average marginal effects of key covariates on transition probabilities. Predicted Probabilities by Clinical Profile Patient profiles combining age, ECOG, anxiety, presence of caregiver, tumor, and treatment status were evaluated to characterize clinical risk: younger fitter patient profiles had 65–75% chance of active status at T0→T12; older frail patients reflected a greater degree of death risk (~34-40%). Anxiety and no caregivers raised the risk of functional attrition (~48%). At T12→T24, among those who are active, most remain likewise, particularly if they are younger, and less anxious (>80%) (Supplementary Table S3). QoL prediction Modeling and Comparison of Missing Data Handling Strategies Among the 319 patients scheduled for 24-week follow-up visit, only 137 (43%) provided complete QoL data. Of those with complete data, 43.8% experienced QoL deterioration (≥0 points). By using more traditional definitions, the rates of QoL decline fell to 23.4%, 19.7% and 16.8% for rates of decline ≥5, ≥7, and ≥10 FACT-G points, respectively (see Supplementary Table S4). The substantial degree of informative missingness justified the use of advanced methods, including Inverse Probability of Completion Weighting (IPCW) and Multiple Imputation by Chained Equations (MICE). The IPCW weights we calculated using logistic regression were well-stabilized (mean ESS 127–132; see Figure 4 and upplementary table S4), which suggested that they were satisfactory for statistical inference. Figure 4 . Distribution of IPCW Weights For all analytic thresholds of interest, MICE-derived models outperformed IPCW-derived models. For example, the highest AUC was achieved using LASSO-MICE (AUC = 0.671, IQR = 0.136) at the -5 FACT-G threshold, with Random Forest and XGBoost using MICE also exhibiting superior performance to model using IPCW. MICE produced lower Brier scores (up to 0.10 lower than IPCW) which suggests greater accuracy in the predicted probabilities of observing actual risk especially for nonlinear models. Importantly, calibration analysis further favored MICE since slopes were closer to the ideal of 1 (e.g., slope = 1.005 for LASSO-MICE at –10), while IPCW often produced suboptimal (<0.5) or negative slopes, indicating systematic risk misestimation. MICE also returned lower variability (smaller IQRs) confirming more stable performance. See Table 3 and Supplementary Figure S1 for full model performance results. Table 3 . Comparison of Model Discrimination , Overall Accuracy (Brier Score), Calibration (Slope) by Missing Data Method Threshold Model IPCW ( M edian [IQR]) MICE ( Median [IQR]) Comment Model Discrimination (AUC) by Missing Data Method (median [IQR]) -10 glmnet 0.577 [0.130] 0.664 [0.158] Higher AUC with MICE rf 0.604 [0.095] 0.644 [0.169] Slight ↑ with MICE xgb 0.623 [0.110] 0.663 [0.155] Higher AUC with MICE -7 glmnet 0.613 [0.098] 0.629 [0.195] Stable, larger IQR MICE rf 0.643 [0.158] 0.608 [0.170] Slight ↓ with MICE xgb 0.590 [0.100] 0.625 [0.176] Higher AUC with MICE -5 glmnet 0.560 [0.113] 0.671 [0.136] Clear improvement rf 0.579 [0.103] 0.622 [0.169] Higher AUC with MICE xgb 0.579 [0.130] 0.661 [0.165] Higher AUC with MICE 0 glmnet 0.612 [0.141] 0.670 [0.082] Higher AUC with MICE rf 0.586 [0.073] 0.645 [0.097] Higher AUC with MICE xgb 0.575 [0.066] 0.652 [0.089] Higher AUC with MICE Model Overall Accuracy (Brier Score) by Missing Data Method (median [IQR]) -10 glmnet 0.136 [0.073] 0.145 [0.044] Similar rf 0.147 [0.047] 0.150 [0.056] Similar xgb 0.176 [0.067] 0.146 [0.055] Lower with MICE (better) -7 glmnet 0.169 [0.066] 0.161 [0.038] Lower with MICE (better) rf 0.179 [0.052] 0.169 [0.047] Lower with MICE (better) xgb 0.215 [0.058] 0.172 [0.049] Lower with MICE (better) -5 glmnet 0.175 [0.036] 0.177 [0.032] Similar rf 0.187 [0.054] 0.184 [0.037] Similar xgb 0.215 [0.084] 0.184 [0.042] Lower with MICE (better) 0 glmnet 0.239 [0.023] 0.221 [0.023] Lower with MICE (better) rf 0.269 [0.065] 0.227 [0.027] Lower with MICE (better) xgb 0.329 [0.082] 0.228 [0.029] Lower with MICE (better) Model Calibration (Slope) by Missing Data Method (median [IQR]) -10 glmnet 0.775 [2.166] 1.005 [1.199] Closer to 1 with MICE rf 0.199 [0.691] 0.656 [0.817] Marked improvement with MICE xgb 0.124 [0.381] 0.658 [0.606] Marked improvement with MICE -7 glmnet 0.532 [0.858] 0.840 [0.773] Closer to 1 with MICE rf 0.366 [0.590] 0.497 [0.624] Slight improvement with MICE xgb 0.124 [0.201] 0.456 [0.610] Improvement with MICE -5 glmnet 0.522 [0.796] 0.858 [0.727] Closer to 1 with MICE rf 0.092 [0.648] 0.666 [0.668] Marked improvement with MICE xgb 0.066 [0.309] 0.571 [0.571] Marked improvement with MICE 0 glmnet 1.681 [3.263] 0.882 [0.519] Corrected towards 1 with MICE rf 0.132 [0.671] 0.728 [0.502] Marked improvement with MICE xgb -0.029 [0.302] 0.619 [0.380] Corrected (better with MICE) Sensitivity Analysis under MNAR Scenarios Robustness analyses with delta shifts (Δ=5,7,10) showed minor reductions in AUC and Brier scores but maintained model rankings and stable calibration slopes near 1, indicating consistent predictive performance despite departures from MAR (Table 4). Variable Importance Baseline QoL, age, and anxiety and depression were consistently in the top ranks of important predictors across models and substrategies for missing data methods overall. Tumor site and treatment arm varied by method. ECOG less consistently predictive. This reflected the overall greater consistency between the MICE and IPCW (Figure 5). Table 4 . Sensitivity of AUC under MNAR assumptions Threshold Model IPCW (median [IQR]) MICE (median [IQR]) MNAR Δ5 MNAR Δ7 MNAR Δ10 Sensitivity interpretation Sensitivity of AUC -10 glmnet 0.577 [0.130] 0.664 [0.158] 0.639 [0.119] 0.626 [0.116] 0.630 [0.118] Slightly lower than MICE but within MAR range → stable rf 0.604 [0.095] 0.644 [0.169] 0.614 [0.127] 0.619 [0.164] 0.582 [0.118] Minor sensitivity at Δ10, otherwise stable xgb 0.623 [0.110] 0.663 [0.155] 0.623 [0.147] 0.614 [0.143] 0.625 [0.113] Very stable across MNAR settings -7 glmnet 0.613 [0.098] 0.629 [0.195] 0.616 [0.121] 0.632 [0.120] 0.603 [0.115] Consistent discrimination under MNAR rf 0.643 [0.158] 0.608 [0.170] 0.622 [0.127] 0.617 [0.117] 0.567 [0.084] Stable to Δ7, minor drop at Δ10 xgb 0.590 [0.100] 0.625 [0.176] 0.627 [0.123] 0.609 [0.105] 0.590 [0.084] Stable under MNAR, slight decrease at Δ10 -5 glmnet 0.560 [0.113] 0.671 [0.136] 0.643 [0.119] 0.635 [0.084] 0.632 [0.112] Minor decrease vs MICE, overall robust rf 0.579 [0.103] 0.622 [0.169] 0.622 [0.125] 0.614 [0.099] 0.590 [0.090] Stable up to Δ7, degradation under Δ10 xgb 0.579 [0.130] 0.661 [0.165] 0.615 [0.116] 0.621 [0.071] 0.598 [0.094] Slight drop under MNAR, within tolerance Sensitivity of Brier Score -10 glmnet 0.136 [0.073] 0.145 [0.044] 0.168 [0.041] 0.184 [0.035] 0.204 [0.038] Progressive increase → slightly worse under MNAR rf 0.147 [0.047] 0.150 [0.056] 0.172 [0.045] 0.191 [0.036] 0.215 [0.036] Gradual degradation with higher Δ xgb 0.176 [0.067] 0.146 [0.055] 0.168 [0.048] 0.192 [0.036] 0.211 [0.036] Clear loss in accuracy at Δ≥7 -7 glmnet 0.169 [0.066] 0.161 [0.038] 0.195 [0.041] 0.208 [0.030] 0.224 [0.034] Gradual deterioration under MNAR rf 0.179 [0.052] 0.169 [0.047] 0.202 [0.041] 0.215 [0.033] 0.230 [0.037] Similar pattern, accuracy slightly worse xgb 0.215 [0.058] 0.172 [0.049] 0.206 [0.046] 0.221 [0.036] 0.231 [0.035] Stable but slightly higher error under Δ10 -5 glmnet 0.175 [0.036] 0.177 [0.032] 0.214 [0.031] 0.219 [0.027] 0.230 [0.030] Slight degradation, consistent trend rf 0.187 [0.054] 0.184 [0.037] 0.216 [0.035] 0.223 [0.033] 0.239 [0.029] Gradual increase → less accurate xgb 0.215 [0.084] 0.184 [0.042] 0.217 [0.039] 0.223 [0.037] 0.240 [0.035] Accuracy declines modestly with Δ Sensitivity of Calibration Slope -10 glmnet 0.136 [0.073] 0.145 [0.044] 0.168 [0.041] 0.184 [0.035] 0.204 [0.038] Progressive increase → slightly worse under MNAR rf 0.147 [0.047] 0.150 [0.056] 0.172 [0.045] 0.191 [0.036] 0.215 [0.036] Gradual degradation with higher Δ xgb 0.176 [0.067] 0.146 [0.055] 0.168 [0.048] 0.192 [0.036] 0.211 [0.036] Clear loss in accuracy at Δ≥7 -7 glmnet 0.169 [0.066] 0.161 [0.038] 0.195 [0.041] 0.208 [0.030] 0.224 [0.034] Gradual deterioration under MNAR rf 0.179 [0.052] 0.169 [0.047] 0.202 [0.041] 0.215 [0.033] 0.230 [0.037] Similar pattern, accuracy slightly worse xgb 0.215 [0.058] 0.172 [0.049] 0.206 [0.046] 0.221 [0.036] 0.231 [0.035] Stable but slightly higher error under Δ10 -5 glmnet 0.175 [0.036] 0.177 [0.032] 0.214 [0.031] 0.219 [0.027] 0.230 [0.030] Slight degradation, consistent trend rf 0.187 [0.054] 0.184 [0.037] 0.216 [0.035] 0.223 [0.033] 0.239 [0.029] Gradual increase → less accurate xgb 0.215 [0.084] 0.184 [0.042] 0.217 [0.039] 0.223 [0.037] 0.240 [0.035] Accuracy declines modestly with Δ DISCUSSION Our study evaluated the attrition dynamics and predictors of 24-week quality-of-life (QoL) deterioration in advanced cancer patients. Approximately one-third of the patients experienced functional attrition at 24 weeks, and similarly, approximately one-third died, indicating selective and informative attrition. Lower education, higher distress, and poorer clinical performance at baseline were associated with higher functional attrition. Multinomial regression and predictive modelling showed that lower education increased the risk of functional attrition, while older age, poor ECOG status and depression symptoms increased the risk of dying; early engagement of caregivers was protective and MICE gave more stable and better calibrated estimates than IPCW with moderate discriminative power (AUC ≈ 0.67). The MNAR sensitivity analysis confirmed that the major conclusions of the study remained valid. These results are consistent with prior oncology studies reporting that patient-reported outcome (PRO) dropout is informative and related to declining functional status and prognosis rather than randomness. For example, Efficace et al [ 17 ] found that performance status and symptom burden predicted PRO failure in cancer patients, while Groenvold et al demonstrated the link between advanced disease and PRO dropout in lung cancer trials [ 18 ]. Cuer et al showed similar patterns of PRO attrition in esophageal cancer [ 19 ], patients who drop out early tend to have substantially worse HRQoL trajectories than those who remain. Other studies, PMC8059010 [ 20 ] and PMC5834396 [ 21 ], have associated lower baseline health status, greater symptom burden with increased dropout. These observations in our cohort reinforce previous findings depression is a strong predictor of attrition and mortality [ 22 ] but moreover, the association between lower educational attainment and a higher likelihood of functional attrition adds nuance to prior work, suggesting that social determinants contribute not only to survival disparities but also to engagement with longitudinal QoL assessments. ECOG performance status did not significantly predict functional attrition in multivariable modeling. A small marginal increase in predicted attrition probability observed in AME analysis should therefore be interpreted with caution and likely reflects statistical noise rather than a true underlying effect. Additionally, our study provides an empirical comparison of IPCW and MICE for modeling QoL deterioration and our findings both support and challenge established methodologic norms in the use of informative missing data. Whereas some cancer prognosis studies utilizing molecular or clinical predictors have achieved very high AUCs (> 0.90) with advanced machine learning [ 23 ], our results show that the discriminative ability for PRO deterioration is notably lower, reflecting the inherent complexity of these outcomes compared to biomarker- or imaging-based endpoints. Although the mean AUC values (approximately 0.67) for the top performing model MICE-LASSO, with an IQR of approximately 0.136 indicated relatively stable and moderate prediction performance for all imputed models across all folds, the results in our study are consistent with studies on PRO research and predictive modeling in general. Examples include studies by Lian et al [ 14 ], who showed that traditional baseline models for predicting changes in quality-of-life had average AUC values of approximately 0.60 whereas only approaches leveraging deep learning in very large, feature-rich datasets have reported AUC values exceeding 0.75 [ 14 ] [ 24 ], exceptions rather than the rule; Additionally, studies by Wahl et al [ 15 ] also showed that even after applying internal validation and multiple imputation techniques, the predictive performance of health data remains modest when dealing with incomplete health data. Although IPCW theoretically has the ability to reduce selection bias [ 25 ], the improved discrimination (AUC) and accuracy (Brier Score) of MICE models, compared to IPCW, reflect the practical limitations of IPCW in real-world data, especially given its sensitivity to model specification and overlap between cases and controls. While MICE carry a theoretical risk of optimism bias under the MAR assumption [ 26 ], our analysis shows substantial advantages in variance reduction, stability, and calibration, making MICE preferable in most practical scenarios and reinforces the importance of empirical validation over theoretical preferences. Additionally, the relatively small variability of our models further indicates that while accuracy is moderate, the estimates are robust and reproducible. One of the most important findings was the severe miscalibration of IPCW models, with slopes often near zero, especially in nonlinear models. The issue of miscalibration due to extreme weighting is a recognized, but often understated problem [ 27 ]. Accurate calibration is essential for clinical utility, determining whether predicted risk matches observed outcomes [ 28 ]; The inability of IPCW to maintain a stable calibration slope, especially in nonlinear models (RF, XGBoost), represents a significant limitation of this approach in developing clinical decision support systems. Our finding that MICE significantly improved the calibration slope, often from near zero (IPCW) to near 1.0 (MICE) demonstrates a critical empirical contribution to the growing consensus that predictive model stability and robust calibration are often more important than theoretical bias correction when choosing a methodology for the development of clinical decision support systems [29]. The identification of Baseline QoL, age, and baseline psychological distress (HADS scores) as the three most influential predictors, is in alignment with prior literature [30] [31]. The consistency of the top-ranking predictors across the two different missing data methodologies increases the evidence that these PROs are the dominant factors in determining the patient outcome within this cohort. Result from the MNAR sensitivity analysis support the notion that the observed superior calibration of MICE, and equivalent discrimination compared to IPCW, is not a result of the MAR assumption. Methodological and clinical contribution One of the primary contributions of this study from a methodological perspective is the explicit consideration of functional attrition as an informative state and the comparative evaluation of IPCW and MICE methods to handle missing data. In addition, through the evaluation of model performance under MNAR conditions, we present a methodologically strong and empirically grounded framework for the analysis of longitudinal patient reported outcomes (PROs), however, the moderately AUC ceiling (≈ 0.67) underscore the need for time-updated longitudinal predictors to capture the complex, dynamic trajectories of QoL decline beyond baseline variables, which alone yield only moderate predictive accuracy (AUC ≈ 0.67). From a clinical perspective, baseline quality of life, patient age, and psychological distress were the main determinants of QoL decline, whereas poor ECOG status and depressive symptoms predicted mortality with lower educational attainment increasing the risk of functional attrition and the quantified marginal effects support early identification of at-risk patients and tailored interventions. Strengths and Limitations Strengths of our study include explicit modeling of functional attrition as an informative state, the evaluation of the marginal effects and predicted probabilities enhancing clinical interpretability, a robust cross-validation of classifiers, and transparent handling of missing data showing a combination of both methodological rigor and practical clinical relevance. Limitations include modest sample size and restriction to baseline predictors, reducing power for transition analyses and individual-level predictions. The MNAR sensitivity analyses demonstrated reassuring robustness but may not fully account for real world non-ignorable missingness. We did not fit continuous-time Markov or semi-Markov multistate models. Instead, we used discrete-time multinomial regression to model state transitions, which was appropriate given the fixed follow-up schedule (12 and 24 weeks) and the study’s objective to characterize baseline determinants rather than transition intensities. Implications and future research Clinically, treating functional attrition as an explicit process rather than as random missingness can improve the interpretation of QoL trajectories. Therefore, identify vulnerable patients early and implementing strategies to retain them in the study should be a priority, and future work should explore targeted strategies such as caregiver training, reduced complexity of questionnaires, enhanced communication, more frequent contacts or adapting follow-up to enhance retention, provide timely supportive care, and assist in palliative planning with potentially minimal effect to allow for stratification and individualized interventions for each patient. Regarding the methodological aspect, future research should include time-updated predictors (e.g. repeated assessments of QoL, symptoms, treatment response, biomarkers) and utilize advanced frameworks (e.g. Joint Modeling, Landmark Analysis, Recurrent Event Models) to better identify relationships among attrition, declines in QoL, and mortality over time. Additionally, validating the study’s findings across various cancer types is necessary to determine the study's generalizability and facilitate real-time risk stratification to inform individualized monitoring and enhance the delivery of supportive care and the reliability of QoL assessment. CONCLUSION Our study provides new insights into how quality of life (QoL) changes in advanced cancer by directly modeling functional attrition as an informative state and not treating it as missing data that can be ignored. State-transition analyses demonstrated that the transition to functional attrition or death were systemically related to the baseline characteristics of the participants: Poorer ECOG Performance Status and greater levels of depression at baseline resulted in earlier mortality, while lower educational attainment resulted in a greater risk of functional attrition, indicating clinical and sociodemographic determinants. The predictive modeling of QoL based solely on baseline data produced modest discriminative ability (AUCmax ≈ 0.67), consistent with limited signal provided by complex Patient Reported Outcomes (PRO). Most importantly, Multiple Imputation by Chained Equations (MICE) empirically performed better than Inverse Probability of Completion Weighting (IPCW), providing more stable and better-calibrated models as evidenced through MNAR Sensitivity Analyses. Although MICE outperformed IPCW in this study, the findings should be interpreted in context and validated against additional datasets and disease settings prior to generalization. Overall, our study highlights the clinical and methodological importance of incorporating functional attrition into QoL research, validates the role of baseline patient reported outcomes as prognostic indicators, and emphasizes the necessity of developing time-updated and multimodal predictors to improve the predictive accuracy of future studies. Furthermore, the study demonstrates the importance of integrating functional attrition and mortality within a unified framework to allow for more accurate interpretation of PROs and support the development of targeted and data-driven supportive care strategies in advanced cancer. Declarations Individual participant-level data can be made available, but only through a controlled access mechanism via the NCI/NIH official repository, as required by federal regulations. My Data Use Agreement (DUA) with the National Cancer Institute strictly prohibits releasing or distributing data to any entity (Section 4) and specifically forbids posting it on any public site (Section 9). Consequently, uploading the raw file directly to Research Square would constitute a breach of this federal contract. However, the data is accessible to any researcher who requests it through the official NCI portal, which satisfies standard scientific transparency requirements. Human Ethics and Consent to Participate This study is a secondary analysis of anonymized data from the clinical trial NCT02349412 registered on 23 January 2015, which evaluated early palliative care versus standard care in patients with advanced cancer. The original trial was conducted in accordance with the Declaration of Helsinki and ICH-GCP guidelines and received all required ethical approvals, with written informed consent obtained from all participants. As the present analysis used only fully de-identified data, no additional ethical approval or consent was required. Competing interests The authors declare that they have no known conflicting financial interests or personal relationships that might appear to affect the work presented in this manuscript. Funding This research did not receive any funding from any funding agencies in the public, private, or non-profit sectors. Author Contribution K.G. conceptualized the study, extracted and analyzed the data, interpreted the results, and wrote the manuscript. All authors reviewed and approved the final manuscript. P.P., A.A., T.S., K.Y., and A.J. prepared the figures. Acknowledgement The authors gratefully acknowledge the support of the Government of Togo for providing the scholarship that enabled the pursuit of graduate studies. We thank the faculty members of the SRM School of Public Health for their academic environment and institutional support.We also acknowledge the guidance and feedback provided by Dr. Jennifer Gladius (Department of Biostatistics, SRM School of Public Health), Dr. Prakash (School of Public Health, SRMIST), and Dr. Dhivyia (School of Public Health, SRMIST), which contributed to the development of this work. Data Availability The datasets supporting this analysis are available from the [NCTN/NCORP](https:/nctn-data-archive.nci.nih.gov/node/1534) Data Archive in accordance with NCI and NIH data-sharing policies. Due to privacy restrictions and limitations of the original informed consent, individual participant-level data cannot be made publicly available. Aggregated and de-identified data used in this study may be requested from the study investigators or through the Data Archive portal, subject to submission of a brief proposal and relevant institutional approvals. Analytical scripts are available from the corresponding author upon reasonable request or through the corresponding author’s [GitHub](https:/github.com/jack-junior/MSc_project) References Basch E, Deal AM, Kris MG, Scher HI, Hudis CA, Sabbatini P, et al. Symptom monitoring with patient-reported outcomes during routine cancer treatment: a randomized controlled trial. JAMA Oncol. 2016;2(6):902–9. 10.1001/jamaoncol.2016.1990 . U.S. Food and Drug Administration. Patient-Reported Outcome Measures: Use in Medical Product Development to Support Labelling Claims. Silver Spring: FDA; 2009 (updated 2019). Cella DF, Tulsky DS, Gray G, Sarafian B, Linn E, Bonomi A, et al. The Functional Assessment of Cancer Therapy scale: development and validation of the general measure. J Clin Oncol. 1993;11(3):570–9. 10.1200/JCO.1993.11.3.570 . Calvert M, Blazeby J, Altman DG, Revicki DA, Moher D, Brundage MD, et al. Reporting of patient-reported outcomes in randomized trials: the CONSORT PRO extension. JAMA. 2013;309(8):814–22. 10.1001/jama.2013.879 . Snyder CF, Aaronson NK, Choucair AK, Elliott TE, Greenhalgh J, Halyard MY, et al. Implementing patient-reported outcomes assessment in clinical practice: a review of the options and considerations. Qual Life Res. 2014;23(5):1415–33. 10.1007/s11136-013-0396-9 . Fairclough DL. Design and Analysis of Quality of Life Studies in Clinical Trials. 2nd ed. Boca Raton: Chapman & Hall/CRC; 2010. Hamel JF, Tuen Mun HS, Bryant AL, Ikwuazom D, Schrag D. Do patients and oncologists discuss the cost of cancer treatment? An observational study of clinical interactions. Support Care Cancer. 2017;25(10):3275–82. 10.1007/s00520-017-3731-6 . Little RJA, Rubin DB. Statistical Analysis with Missing Data. 3rd ed. Hoboken: Wiley; 2019. Fayers PM, Machin D. Quality of Life: The Assessment, Analysis and Interpretation of Patient-Reported Outcomes. 2nd ed. Chichester: Wiley; 2016. Bell ML, Fiero M, Horton NJ, Hsu CH. Handling missing data in RCTs in palliative care: a simulation study. BMJ Support Palliat Care. 2016;6(3):295–301. 10.1136/bmjspcare-2015-000889 . International Council for Harmonisation. ICH E9(R. Addendum on Estimands and Sensitivity Analysis in Clinical Trials. Geneva: ICH; 2019. 1. Van Cutsem E, Cervantes A, Adam R, Sobrero A, Van Krieken JH, Aderka D, et al. ESMO consensus guidelines for the management of patients with metastatic colorectal cancer. Ann Oncol. 2022;33(7):659–701. 10.1016/j.annonc.2022.04.003 . Warrington L, Absolom K, Velikova G. Prediction of patient-reported outcomes in oncology: current status and future directions. J Clin Epidemiol. 2021;137:175–88. 10.1016/j.jclinepi.2021.04.009 . Lian R, Hsiao V, Hwang J, Ou Y, Robbins SE, Connor NP, et al. Predicting HRQOL change using natural language processing in thyroid cancer. iScience Biomed Inf. 2023;100097. 10.1016/j.isci.2023.107123 . Wahl S, Löwe B, Bjorner JB, Rose M, Brähler E, Walter OB. Assessment of predictive performance in incomplete data by combining internal validation and multiple imputation. BMC Med Res Methodol. 2016;16:144. 10.1186/s12874-016-0228-0 . Andersen PK, Keiding N. Multi-state models for event history analysis. Stat Methods Med Res. 2002;11(2):91–115. 10.1191/0962280202SM276ra . Efficace F, Bottomley A, Osoba D, Gotay C, Flechtner H, D’haese S, et al. Beyond the development of HRQOL measures: a checklist for evaluating HRQOL outcomes in cancer clinical trials. J Clin Oncol. 2006;24(25):4232–9. 10.1200/JCO.2006.06.7239 . Groenvold M, Petersen MA, Aaronson NK, Arraras JI, Blazeby JM, Bottomley A, et al. The impact of dropout on the assessment of HRQOL in cancer clinical trials. Qual Life Res. 2010;19(7):1087–94. 10.1007/s11136-010-9660-5 . Cuer B, Mollevi C, Anota A, Bouché O, Bedenne L, Conroy T, et al. Handling informative dropout in longitudinal analysis of health-related quality of life: application of three approaches to data from the esophageal cancer clinical trial PRODIGE 5/ACCORD 17. BMC Med Res Methodol. 2020;20:223. 10.1186/s12874-020-01104-w . Gebert P, Schindel D, Frick J, Schenk L, Grittner U. Characteristics and patient-reported outcomes associated with dropout in severely affected oncological patients: an exploratory study. Support Care Cancer. 2021;29(6):3311–20. 10.1007/s00520-020-05893-3 . Perez-Cruz PE, Shamieh O, Paiva CE, Kwon JH, Muckaden MA, Bruera E, et al. Factors associated with attrition in a multicenter longitudinal observational study of patients with advanced cancer. J Pain Symptom Manage. 2018;55(3):938–45. 10.1016/j.jpainsymman.2017.11.019 . Quinten C, Coens C, Mauer M, Comte S, Sprangers MAG, Cleeland C, et al. Baseline quality of life as a prognostic indicator of survival: a meta-analysis of individual patient data from EORTC clinical trials. Lancet Oncol. 2009;10(9):865–71. 10.1016/S1470-2045(09)70200-1 . Kourou K, Exarchos TP, Exarchos KP, Karamouzis MV, Fotiadis DI. Machine learning applications in cancer prognosis and prediction. Comput Struct Biotechnol J. 2015;13:8–17. PMID:25750696; PMCID:PMC4348437. Abegaz TM, Ahmed M, Ali AA, Bhagavathula AS. Predicting HRQOL using social determinants of health: a machine learning approach with the All of Us cohort. Risk Manag Healthc Policy. 2024;17:1665–78. 10.2147/RMHP.S436294 . Robins JM, Finkelstein DM. Correcting for noncompliance and dependent censoring in an AIDS clinical trial with IPCW log-rank tests. Biometrics. 2000;56(3):779–88. 10.1111/j.0006-341X.2000.00779.x . White IR, Royston P, Wood AM. Multiple imputation using chained equations: issues and guidance for practice. Int J Epidemiol. 2011;40(2):338–48. 10.1093/ije/dyr080 . Seaman SR, White IR. Review of inverse probability weighting for dealing with missing data. Int J Epidemiol. 2013;42(4):1091–9. 10.1093/ije/dyt110 . Van Calster B, McLernon DJ, van Smeden M, Wynants L, Steyerberg EW. Calibration: the Achilles heel of predictive analytics. Eur J Epidemiol. 2019;34(8):749–61. 10.1007/s10654-019-00546-7 . Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterial.docx Cite Share Download PDF Status: Under Review Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8407124","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":574042630,"identity":"6d8d1220-ac24-43eb-a828-4d41ad01a573","order_by":0,"name":"Selassi Komi Gayi","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA7ElEQVRIie2RsQrCMBCGrxQ6FVyvBH0GoaAOoq9iEOorOAqFuIhuUsHBR/ARTgq6iH2ALorgHHBx05MuTomjYL4h3MH/kcsFwOH4QRB8Iv3oculNoDqrwqAE8rKaJVWYvlPCOA6DvOoIbHkmWmxaURYW/fYyT+8auvUt+epsUgSeE9SdUq5LqZAgibfkTZsmpYG051vKAQqpeLBcsqLQrOyUCINTn5VUEzztiqilPj+fvEzICQ9GdiXKAo+XPJSsKDw2h/EqtyhYFJq/sseDja56PO7V54fpzagADj6796p8Y56pkS3hcDgcf88LPvFRBh/8GXUAAAAASUVORK5CYII=","orcid":"","institution":"SRM Institute of Science and Technology","correspondingAuthor":true,"prefix":"","firstName":"Selassi","middleName":"Komi","lastName":"Gayi","suffix":""},{"id":574042631,"identity":"772f1e92-2c6c-4f71-96a9-a94932fd178b","order_by":1,"name":"Prerna Priyam","email":"","orcid":"","institution":"SRM Institute of Science and Technology","correspondingAuthor":false,"prefix":"","firstName":"Prerna","middleName":"","lastName":"Priyam","suffix":""},{"id":574042632,"identity":"f8654308-b913-4a5b-b0a2-554e3bdbc46c","order_by":2,"name":"Sangenis Ayao Assogba","email":"","orcid":"","institution":"SRM Institute of Science and Technology","correspondingAuthor":false,"prefix":"","firstName":"Sangenis","middleName":"Ayao","lastName":"Assogba","suffix":""},{"id":574042633,"identity":"865bdd43-257f-4907-be4e-56ec5de4b7e8","order_by":3,"name":"Samadou Tchakondo","email":"","orcid":"","institution":"SRM Institute of Science and Technology","correspondingAuthor":false,"prefix":"","firstName":"Samadou","middleName":"","lastName":"Tchakondo","suffix":""},{"id":574042634,"identity":"8c40c108-b668-4c45-8a5e-021c0a445476","order_by":4,"name":"Yendouname Kandjoni","email":"","orcid":"","institution":"SRM Institute of Science and Technology","correspondingAuthor":false,"prefix":"","firstName":"Yendouname","middleName":"","lastName":"Kandjoni","suffix":""},{"id":574042635,"identity":"08284376-9f41-47bf-ade1-b96e39944101","order_by":5,"name":"Richard Tugbeh","email":"","orcid":"","institution":"SRM Institute of Science and Technology","correspondingAuthor":false,"prefix":"","firstName":"Richard","middleName":"","lastName":"Tugbeh","suffix":""},{"id":574042636,"identity":"4cd66000-e030-4bba-999e-5d14da988e63","order_by":6,"name":"Alex Joseph","email":"","orcid":"","institution":"SRM Institute of Science and Technology","correspondingAuthor":false,"prefix":"","firstName":"Alex","middleName":"","lastName":"Joseph","suffix":""}],"badges":[],"createdAt":"2025-12-19 17:26:37","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8407124/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8407124/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":100599614,"identity":"e9af61c7-eff3-47d0-89af-5551736cba52","added_by":"auto","created_at":"2026-01-19 14:44:18","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1867679,"visible":true,"origin":"","legend":"","description":"","filename":"GayietalBMCMRMManuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/3406e978bd96e0667cb7bd8f.docx"},{"id":100599694,"identity":"d75131e1-ab6d-4f6c-859d-c71d657ba210","added_by":"auto","created_at":"2026-01-19 14:44:41","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10246,"visible":true,"origin":"","legend":"","description":"","filename":"17fd6ad2944f451e84a638346090c5fa.json","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/81ace9bcccb15b5bec0542d6.json"},{"id":100599363,"identity":"7d295dc8-5e3b-4e84-ba0c-74dcba77c97e","added_by":"auto","created_at":"2026-01-19 14:43:06","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":163684,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/286fe68ce37347b43184d60b.docx"},{"id":100599494,"identity":"438e268d-a30b-4551-8598-2ab3d3875626","added_by":"auto","created_at":"2026-01-19 14:43:30","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":151427,"visible":true,"origin":"","legend":"","description":"","filename":"17fd6ad2944f451e84a638346090c5fa1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/882424f300a7dda9e46821e2.xml"},{"id":100599437,"identity":"93520509-ff6b-4ddb-b588-84c0ce47b04f","added_by":"auto","created_at":"2026-01-19 14:43:16","extension":"eps","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":455,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage2.eps","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/6bcd4a355cf42ee0e49d96db.eps"},{"id":100600055,"identity":"fc7768ad-3cdb-4e0e-9800-f0c54834afa1","added_by":"auto","created_at":"2026-01-19 14:46:06","extension":"jpeg","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":75660,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/a78743ead0b01f9913556b82.jpeg"},{"id":100599570,"identity":"69573da6-e52e-4a30-8f4c-19e94b57b752","added_by":"auto","created_at":"2026-01-19 14:44:04","extension":"jpeg","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":129753,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/6d74480864e685500cc36e70.jpeg"},{"id":100599595,"identity":"7fa272c7-70d5-4393-ada3-0d3162e75058","added_by":"auto","created_at":"2026-01-19 14:44:09","extension":"jpeg","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":639392,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/e07c4a995e4ca7b38bd7783a.jpeg"},{"id":100600066,"identity":"2b761c60-bbf6-4a11-a7a2-9963a4a44c29","added_by":"auto","created_at":"2026-01-19 14:46:09","extension":"jpeg","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1074,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/cef1d2818dfb958e306f3566.jpeg"},{"id":100599136,"identity":"4453c755-0d8d-4220-81e6-8278eb5c1eb2","added_by":"auto","created_at":"2026-01-19 14:42:02","extension":"jpeg","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":325881,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/a623d756169bfcfdb37a0362.jpeg"},{"id":100599486,"identity":"3ee3bbba-3096-42b3-8c7a-09e9e1bc869e","added_by":"auto","created_at":"2026-01-19 14:43:28","extension":"jpeg","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":143604,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/bfdcd6c6a8f18b56f64ecd35.jpeg"},{"id":100599384,"identity":"ebaa0cf7-aa7e-4bfa-b9db-93ce851057a4","added_by":"auto","created_at":"2026-01-19 14:43:12","extension":"jpeg","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":194381,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/0dd547a6737aafd04b5a72ce.jpeg"},{"id":100599591,"identity":"8f3f28f4-7f52-4eb0-912a-822ed5915d32","added_by":"auto","created_at":"2026-01-19 14:44:07","extension":"jpeg","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1074,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/20486d0201253683526ffb8a.jpeg"},{"id":100599841,"identity":"b6298c7b-f783-4c69-afcc-39bdf4f5657c","added_by":"auto","created_at":"2026-01-19 14:45:22","extension":"jpeg","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":237851,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage9.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/cfc524afb1fc6b76f4718a06.jpeg"},{"id":100599443,"identity":"ec4e22fd-289d-4443-9f0a-8fcfc077245f","added_by":"auto","created_at":"2026-01-19 14:43:19","extension":"jpeg","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1700,"visible":true,"origin":"","legend":"","description":"","filename":"groupimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/a055e0a4b850c460b5d68c07.jpeg"},{"id":100600062,"identity":"09065bf4-3cd9-4089-942e-b65f2f06d66e","added_by":"auto","created_at":"2026-01-19 14:46:08","extension":"jpeg","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":74317,"visible":true,"origin":"","legend":"","description":"","filename":"groupimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/b10b85f8bb81ee1844adbb96.jpeg"},{"id":100599750,"identity":"22cee953-906f-4971-b86c-b78ff7a5588e","added_by":"auto","created_at":"2026-01-19 14:44:53","extension":"jpeg","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":181433,"visible":true,"origin":"","legend":"","description":"","filename":"groupimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/9fa2cc0e1f84edf9e1f4c813.jpeg"},{"id":100599612,"identity":"006e361a-8dc0-4795-a55e-01d7c14a2a8e","added_by":"auto","created_at":"2026-01-19 14:44:16","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":25290,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/1208e92161dc9ca09114a7a5.png"},{"id":100599681,"identity":"a35d25a4-d144-432e-98c6-1ee84a09f9fe","added_by":"auto","created_at":"2026-01-19 14:44:40","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":21838,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/6ea73845fc9d9d3969454e02.png"},{"id":100599499,"identity":"b34bc81a-cda4-42f6-9d16-60bbae9f27fe","added_by":"auto","created_at":"2026-01-19 14:43:32","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":132818,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/0fed6b796c8500b986a86995.png"},{"id":100599804,"identity":"40890a72-fbe1-442b-804c-af854884cb76","added_by":"auto","created_at":"2026-01-19 14:45:05","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":935,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/7d9786498d6efa0cb3c215a2.png"},{"id":100599504,"identity":"2b5614d8-5abb-4a25-b422-c159cb9f18c5","added_by":"auto","created_at":"2026-01-19 14:43:33","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":90770,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/3e1c7fad753724c2958af690.png"},{"id":100599188,"identity":"71ad53d9-3dbf-4517-9610-9063326b8261","added_by":"auto","created_at":"2026-01-19 14:42:15","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":31711,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/d7f620520a799ab470cd6308.png"},{"id":100599785,"identity":"45491acb-bff5-4cc8-be33-d6df5deb0fff","added_by":"auto","created_at":"2026-01-19 14:45:01","extension":"png","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":29000,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/11dd7ba97b98045fb098a617.png"},{"id":100600045,"identity":"08a65429-c927-4f37-8601-6a483c0f2c01","added_by":"auto","created_at":"2026-01-19 14:46:03","extension":"png","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":935,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/9ad851ad9c9dbbc9dac58697.png"},{"id":100599874,"identity":"66a2c22b-c329-4c5a-a76a-8ac5b498e82e","added_by":"auto","created_at":"2026-01-19 14:45:26","extension":"png","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":35314,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/c533699e8367dcae26cf4a57.png"},{"id":100599473,"identity":"21d99826-1552-4f6b-9462-38268fb846d4","added_by":"auto","created_at":"2026-01-19 14:43:25","extension":"png","order_by":27,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":620,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinegroupimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/a203ee1d9f85329cd9eee2f4.png"},{"id":100599493,"identity":"f57c0c2e-cc2d-4180-8fca-fce7c22efa0d","added_by":"auto","created_at":"2026-01-19 14:43:30","extension":"png","order_by":28,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":26468,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinegroupimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/4b1972980d8f7bc80ab9242a.png"},{"id":100599810,"identity":"95dfa031-4a7f-4777-ae35-08e3bc320547","added_by":"auto","created_at":"2026-01-19 14:45:06","extension":"png","order_by":29,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":47904,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinegroupimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/a0c8670595970a347546b8ef.png"},{"id":100599269,"identity":"1fe6dacf-1da9-4c69-aa99-921d88e561e2","added_by":"auto","created_at":"2026-01-19 14:42:33","extension":"xml","order_by":30,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":150362,"visible":true,"origin":"","legend":"","description":"","filename":"17fd6ad2944f451e84a638346090c5fa1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/8c00ad222f9cca223fb14166.xml"},{"id":100599597,"identity":"38deab30-507e-4631-a8cc-323a8e62cca4","added_by":"auto","created_at":"2026-01-19 14:44:10","extension":"html","order_by":31,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":166385,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/41eddf30dc4aec41739f4065.html"},{"id":100599667,"identity":"5bde405d-7820-4263-9f21-1c6699f7604b","added_by":"auto","created_at":"2026-01-19 14:44:38","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":155417,"visible":true,"origin":"","legend":"\u003cp\u003eSee image above for figure legend.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/284c8eec0e0909e48f954f89.png"},{"id":100599243,"identity":"abb1a307-4f51-4428-bb74-f6c9e1d068be","added_by":"auto","created_at":"2026-01-19 14:42:28","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":45908,"visible":true,"origin":"","legend":"\u003cp\u003ePatient state transitions from baseline (T0) to week 12 and week 24. A = active; F = functional attrition; D = death.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/8822ebb024329921e9509d36.png"},{"id":100599844,"identity":"cd8cf608-9c73-4633-987e-8506457b1ec1","added_by":"auto","created_at":"2026-01-19 14:45:23","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":40614,"visible":true,"origin":"","legend":"\u003cp\u003eAverage marginal effects of key covariates on transition probabilities.\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/4e6bbd2681b4d537afd72a3d.jpg"},{"id":100599669,"identity":"e10c95c7-093d-4f77-ac5b-d849d3fcc528","added_by":"auto","created_at":"2026-01-19 14:44:40","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":23462,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of IPCW Weights\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/fc58550f77845f3893c326dc.jpg"},{"id":100602592,"identity":"047653f2-7a71-4b30-9e82-3a17e0f12c0d","added_by":"auto","created_at":"2026-01-19 15:16:50","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1726296,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/b9a70e9e-d0c9-4a69-83dc-b7d14d0ad2ed.pdf"},{"id":100599630,"identity":"ec562558-ce7d-41c5-bed2-986ca10f6b70","added_by":"auto","created_at":"2026-01-19 14:44:29","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":163684,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-8407124/v1/421cf424c054c9d4f2d624b6.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Functional attrition and Quality of Life in Advanced Cancer Trial: modeling patients’ trajectories and evaluating the impact of missing outcome data handling on quality-of-life predictive model performance","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003ePatient Reported Outcomes (PROs) such as quality of life (QoL) measures have become increasingly important endpoints in Oncology Clinical Trials [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. PROs such as FACT (Functional Assessment of Cancer Therapy) measure validated patient-centered assessment of a patient's physical, emotional and social well-being [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Inclusion of QoL endpoints was strongly recommended by both the CONSORT-PRO extension [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e] and the ISOQOL (International Society for Quality-of-Life Research) [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. However, QoL data are commonly impacted by \"functional attrition\" (patients alive and formally enrolled in the study, however they do not complete QoL assessments at follow-up). Functional attrition is different than formal withdraw or death and occurs most commonly in advanced cancer settings due to fatigue, low education level, lack of social support and/or psychological distress [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Conventional approaches usually treat functional attrition as \"missing at random\", and attempt to recover the missing data through multiple imputation, etc. [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]; however, this assumption is rarely met because functional attrition is typically informative and represents socio- demographic and/psychosocial characteristics that are also predictive of QoL outcomes; ignoring functional attrition can introduce bias and mask clinically significant trends [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eRecent methodologic research has demonstrated the importance of accounting for the type of missing data in PRO analyses; intercurrent events [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]; and regulatory frameworks such as the ICH E9(R3) addendum on estimands, which emphasize the need to develop methods to address death, dropout and treatment discontinuation [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. While these primarily address survival and treatment-related events, the same reasoning can be extended to PROs, functional attrition is not just a nuisance but an informative state that should be explicitly modelled.\u003c/p\u003e \u003cp\u003eAt the same time, there is increasing interest in the development of predictive models of QoL deterioration, aiming to identify patients at high risk for early decline and target supportive interventions [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Modern statistical learning methods, including penalized regression, random forests, and gradient boosting, have been applied to PRO data, most studies have not explicitly addressed functional attrition [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Importantly, since only a subset of patients remains evaluable at follow-up, the performance and robustness of these models can be heavily influenced by how outcome missing data are handled. The choice between different methods such as multiple imputation and Inverse Probability of Completion Weighting (IPCW) is a vital methodological question, as each approach relies on different assumptions and can lead to different conclusions.\u003c/p\u003e \u003cp\u003eFollowing this background, we propose a dual analytical strategy applied to data from the randomized trial NCT02349412. First, we model patient trajectories across three states: Active (completed QoL), Functional attrition (alive but missing QoL), and Death using a discrete-time multinomial regression framework. This method enables us to characterize baseline predictors of attrition and death and to estimate predicted state probabilities at 12 and 24 weeks, including the marginal effects of key covariates on the likelihood of being in each state. Second, we evaluate the impact of missing outcome data handling on prognostic prediction by developing predictive models of QoL deterioration at 24 weeks among patients active at baseline, comparing penalized regression and machine learning approaches and using two distinct methodologies to handle missing outcome data: multiple imputation and IPCW. As the missingness of quality of life data may not always satisfy the missing at random (MAR) assumption, we also conducted sensitivity analyses under missing not at random (MNAR) mechanisms to examine the robustness of the findings.\u003c/p\u003e \u003cp\u003eBy integrating these perspectives, we aim to clarify the role of functional attrition, offer significant insights into the methodological challenges of longitudinal QoL studies and to guide future research in the development of robust predictive models.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"METHODOLOGY","content":"\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003e\u003cb\u003eData Source and Study Population\u003c/b\u003e: The data analysed come from a multicenter randomized controlled trial (ClinicalTrials.gov identifier: NCT02349412), registered on 23 January 2015, which assessed the impact of early palliative care on quality of life, psychological outcomes, and health care use in adult patients with advanced cancer. Eligibility criteria consisted of adults (\u0026ge;\u0026thinsp;18 years) with an advanced malignancy, ECOG performance status of 0\u0026ndash;2, and expected survival\u0026thinsp;\u0026ge;\u0026thinsp;6 months who were randomized to receive either standard oncology care or integrated early palliative care. Socio-demographic, clinical, vital status, and patient-reported outcomes (FACT-G and HADS) were measured at baseline and at 12- and 24-weeks post-baseline. In this secondary analysis, only de-identified patients alive and with complete baseline QoL data were included.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eOutcome and Predictors\u003c/strong\u003e \u003cp\u003ePatient states at 12- and 24-weeks post-baseline (Active (A), Functional Attrition (F), Deceased (D)) and QoL deterioration at 24 weeks (FACT-G score decrease\u0026thinsp;\u0026ge;\u0026thinsp;minimal clinically important differences of 0, -5, -7, -10) were primary outcomes of interest. Predictors of these outcomes included socio-demographic characteristics (age, sex, education level, whether partnered, race/ethnicity), clinical characteristics (ECOG, tumor type), and baseline FACT-G and HADS scores.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eData Processing\u003c/strong\u003e \u003cp\u003eAfter completing the process of cleaning and harmonizing the data, functional attrition was explicitly defined as living patients without QoL data available at follow-up time points; thereby allowing for the creation of three discrete states (A/F/D) and further enabling modeling of transitions to state F, and comparative analyses for descriptive, multinomial and predictive analyses. Listwise deletion was applied to handle missing covariates when they were less than 5%, because the loss of data would not significantly affect the total statistical power of the study.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eStatistical Analysis\u003c/strong\u003e \u003cp\u003eTransitions between states were modelled using penalized multinomial logistic regression (LASSO/ridge) for variable selection, refitted for ORs and 95% CIs. For enhanced clinical interpretation, Average Marginal Effects (AMEs) were calculated, quantifying predicted probability changes for each covariate. Example clinical profiles combining age, ECOG, anxiety, caregiver presence, tumor type and treatment arm were constructed, with associated transition probabilities summarized to illustrate risk stratification. Predictive models for QoL decline at 24 weeks were developed using LASSO, Random Forest, and XGBoost, evaluating missingness approaches (MICE, IPCW) and conducting MNAR sensitivity analyses. Model validation used repeated cross-validation (AUC, calibration slope, Brier score, median/IQR), See Fig.\u0026nbsp;1. All statistical analyses were conducted using R version\u003c/p\u003e \u003c/p\u003e \u003cp\u003e\u003cdiv\u003e\u003cp\u003e 4.3.2 with appropriate packages (dplyr, glmnet, mice, caret, pROC, ranger, xgboost).\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e"},{"header":"RESULTS","content":"\u003ch2\u003eStudy Population\u003c/h2\u003e\n\u003cp\u003eFrom 405 patients enrolled, 352 (87%) completed QoL assessment at baseline and were therefore included (Table 1). Mean age was 65.3 years and 56% were men. The median magnitude of ECOG performance was 0.94. Baseline QoL (FACT-G) was 73.8 and anxiety/depression\u0026nbsp;symptoms\u0026nbsp;moderate\u0026nbsp;in\u0026nbsp;level.\u0026nbsp;Lung\u0026nbsp;cancer\u0026nbsp;was\u0026nbsp;the\u0026nbsp;most\u0026nbsp;prevalent\u0026nbsp;diagnosis (61%\u0026nbsp;of\u0026nbsp;patients),\u0026nbsp;followed\u0026nbsp;by\u0026nbsp;hepatobiliary\u0026nbsp;(32%),\u0026nbsp;and\u0026nbsp;esophageal/GI\u0026nbsp;cancers\u0026nbsp;(8%).\u0026nbsp;Over\u0026nbsp;half had a college education and 70.5% reported caregiver spouse or partner.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1.\u0026nbsp;\u003c/strong\u003eBaseline Characteristics of the Study Cohort (N = 352)\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cem\u003eCharacteristic\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e\u003cem\u003eOverall\u0026nbsp;(N\u0026nbsp;= 352)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eAge, years\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e\u0026ndash;\u0026nbsp;mean (SD)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e65.3 (9.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eSex, male\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e\u0026ndash; n (%)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e196 (55.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eECOG performance status\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e\u0026ndash;\u0026nbsp;mean (SD)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e0.94 (0.62)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eBaseline FACT score\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e\u0026ndash;\u0026nbsp;mean (SD)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e73.8 (16.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eHADS anxiety score\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e\u0026ndash; mean (SD)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e7.2 (3.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eHADS depression score\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e\u0026ndash;\u0026nbsp;mean (SD)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e5.6 (4.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eTumor type\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e\u0026ndash; n (%)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cem\u003eEsophageal/GEJ/Gastric\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e28 (8.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cem\u003eHepatic/Biliary/Pancreatic/Unknown primary\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e111 (31.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cem\u003eLung\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e213 (60.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eTrial arm 2\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e\u0026ndash; n (%)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e183 (52.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eEducation\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e\u0026ndash; n (%)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cem\u003eCollege+\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e177 (52.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cem\u003eHigh\u0026nbsp;school\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e128 (37.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cem\u003eLow\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e34\u0026nbsp;(10.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eRace/ethnicity\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e\u0026ndash;\u0026nbsp;n (%)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cem\u003eAsian\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e14 (4.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cem\u003eBlack\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e42 (12.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cem\u003eMinority (other)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e6 (1.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cem\u003eWhite\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e274 (81.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eNot married\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e\u0026ndash; n (%)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e130 (37.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eNumber of prior visits\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e\u0026ndash;\u0026nbsp;mean (SD)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e1.81 (2.38)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 423px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eCaregiver partner present\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e\u0026ndash;\u0026nbsp;n (%)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003e248 (70.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003ch2\u003eAttrition\u0026nbsp;and\u0026nbsp;State\u0026nbsp;Transitions\u003c/h2\u003e\n\u003cp\u003eAt week 12, 55% remained active but 29% were functionally attrited and 16.5% deceased. Of those active at week 12, 67% remained active at week 24, 21% were attrited, and 12% were deceased. Among those functionally attrited at week 12, 20% had returned to active status by week 24, 52% remained attrited, and 28% were deceased. (Figure 1. Patient state transitions).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 2.\u0026nbsp;\u003c/strong\u003ePatient state transitions from baseline (T0) to week 12 and week 24. A = active; F = functional attrition; D = death.\u003c/p\u003e\n\u003ch2\u003eBaseline\u0026nbsp;Characteristics\u0026nbsp;by\u0026nbsp;Attrition\u0026nbsp;Status\u003c/h2\u003e\n\u003cp\u003eBaseline profiles (Supplementary Table S1) showed that patients who died by week 12 were older on average, had generally poorer ECOG, lower QoL, and higher depression scores, compared to patients who were active. Those who had functionally attrited at week 12 had intermediate characteristics including lower baseline quality of life scores and more anxiety than\u0026nbsp;active\u0026nbsp;patients,\u0026nbsp;though\u0026nbsp;less\u0026nbsp;severe\u0026nbsp;than\u0026nbsp;those\u0026nbsp;who\u0026nbsp;died.\u0026nbsp;These\u0026nbsp;trends\u0026nbsp;were\u0026nbsp;observed\u0026nbsp;again at week 24. Other sociodemographic variables such as sex, race, education, caregiver present showed no clinically meaningful differences between patient state. All comparisons were exploratory (alpha = 0.05, two-tailed, no adjustment).\u003c/p\u003e\n\u003ch2\u003eMultinomial\u0026nbsp;Modeling\u0026nbsp;of\u0026nbsp;State Transitions\u003c/h2\u003e\n\u003cp\u003eIn multinomial logistic regression (Table 2), lower education significantly predictive of functional attrition at T0 \u0026rarr;T12 (high school completed vs college OR=2.07; low vs college =3.26), the presence of a caregiver reduced the likelihood of functional attrition, although this\u003c/p\u003e\n\u003cp\u003eassociation was not statistically significant; Age, ECOG, and depression are all predictive of mortality. The Average Marginal Effects (AME) show that each unit increase in ECOG related to increase of ~8% risk of functional attrition and ~12% increase in risk of mortality from T0\u0026rarr;T12 and the Presence of a caregiver raised the likelihood of being active by 11%. At T12\u0026rarr;T24, however these marginal changes were not supported by statistically significant multinomial coefficients for the attrition state and should therefore be interpreted descriptively rather than as evidence of a robust association. The type of tumor appears particularly influential with respect to risk of dying (Supplementary Table S2, Figure 3).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable\u0026nbsp;2\u003c/strong\u003e. Multinomial Logistic Regression Results for State Transitions (Odds ratios [OR] and 95% confidence intervals relative to remaining active [A])\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 109px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTransition\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOutcome\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eVariable\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOR\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e95% CI\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"22\" valign=\"top\" style=\"width: 109px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eT0 \u0026rarr; T12\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"11\" valign=\"top\" style=\"width: 92px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eAttrition(F)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eAge\u0026nbsp;(per year)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.97 \u0026ndash; 1.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.99\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eMale\u0026nbsp;(vs\u0026nbsp;female)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.49 \u0026ndash; 1.46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.54\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eECOG\u0026nbsp;(per unit)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.62 \u0026ndash; 1.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eFACT\u0026nbsp;baseline\u0026nbsp;(per\u0026nbsp;point)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.98 \u0026ndash; 1.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eAnxiety\u0026nbsp;(HADS-A,\u0026nbsp;per point)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e1.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.97 \u0026ndash; 1.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.17\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eDepression\u0026nbsp;(HADS-D,\u0026nbsp;per point)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e0.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.89 \u0026ndash; 1.10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.82\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eTumor:\u0026nbsp;Hepatobiliary\u0026nbsp;vs\u003c/p\u003e\n \u003cp\u003eEsophageal/GEJ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e1.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.36 \u0026ndash; 3.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eTumor:\u0026nbsp;Lung\u0026nbsp;vs Esophageal/GEJ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e0.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.23 \u0026ndash; 1.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.40\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eEducation:\u0026nbsp;High\u0026nbsp;school\u0026nbsp;vs College+\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e2.07\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e1.16 \u0026ndash; 3.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.014\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eEducation:\u0026nbsp;Low\u0026nbsp;vs College+\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e3.26\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e1.31 \u0026ndash; 8.12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.011\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eCaregiver present\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e0.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.32 \u0026ndash; 1.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.053\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"11\" valign=\"top\" style=\"width: 92px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eDeath (D)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eAge\u0026nbsp;(per year)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1.05\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e1.01 \u0026ndash; 1.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.010\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eMale\u0026nbsp;(vs\u0026nbsp;female)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e0.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.37 \u0026ndash; 1.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.36\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eECOG\u0026nbsp;(per unit)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e2.37\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e1.31 \u0026ndash; 4.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.004\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eFACT\u0026nbsp;baseline\u0026nbsp;(per\u0026nbsp;point)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.95 \u0026ndash; 1.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.13\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eAnxiety\u0026nbsp;(HADS-A,\u0026nbsp;per point)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.79 \u0026ndash; 1.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.075\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eDepression\u0026nbsp;(HADS-D,\u0026nbsp;per point)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1.14\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e1.02 \u0026ndash; 1.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.027\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eTumor:\u0026nbsp;Hepatobiliary\u0026nbsp;vs\u003c/p\u003e\n \u003cp\u003eEsophageal/GEJ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e1.39\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.32 \u0026ndash; 5.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eTumor:\u0026nbsp;Lung\u0026nbsp;vs Esophageal/GEJ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e0.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.14 \u0026ndash; 2.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.48\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eEducation:\u0026nbsp;High\u0026nbsp;school\u0026nbsp;vs College+\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e1.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.75 \u0026ndash; 3.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.24\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eEducation:\u0026nbsp;Low\u0026nbsp;vs College+\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.23 \u0026ndash; 3.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eCaregiver present\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 49px;\"\u003e\n \u003cp\u003e0.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.35 \u0026ndash; 1.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.41\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"16\" valign=\"top\" style=\"width: 109px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eT12 \u0026rarr; T24\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"8\" valign=\"top\" style=\"width: 92px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eAttrition(F)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eAge\u0026nbsp;(per year)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e0.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.96 \u0026ndash; 1.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.68\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eMale\u0026nbsp;(vs\u0026nbsp;female)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.53 \u0026ndash; 2.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.73\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eECOG\u0026nbsp;(per unit)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.80 \u0026ndash; 3.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.19\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eFACT\u0026nbsp;baseline\u0026nbsp;(per\u0026nbsp;point)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e0.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.97 \u0026ndash; 1.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eAnxiety\u0026nbsp;(HADS-A,\u0026nbsp;per point)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.79 \u0026ndash; 1.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.19\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eTumor:\u0026nbsp;Hepatobiliary\u0026nbsp;vs\u003c/p\u003e\n \u003cp\u003eEsophageal/GEJ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e0.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.10 \u0026ndash; 2.49\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.40\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eTumor:\u0026nbsp;Lung\u0026nbsp;vs Esophageal/GEJ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.14 \u0026ndash; 2.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.53\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eArm\u0026nbsp;2\u0026nbsp;vs\u0026nbsp;Arm 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e0.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.34 \u0026ndash; 1.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.41\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"8\" valign=\"top\" style=\"width: 92px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eDeath (D)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eAge\u0026nbsp;(per year)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.95 \u0026ndash; 1.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eMale\u0026nbsp;(vs\u0026nbsp;female)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.31 \u0026ndash; 2.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.76\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eECOG\u0026nbsp;(per unit)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.64 \u0026ndash; 3.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.33\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eFACT\u0026nbsp;baseline\u0026nbsp;(per\u0026nbsp;point)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.94 \u0026ndash; 1.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.34\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eAnxiety\u0026nbsp;(HADS-A,\u0026nbsp;per point)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.80\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.66 \u0026ndash; 0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.029\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eTumor:\u0026nbsp;Hepatobiliary\u0026nbsp;vs\u003c/p\u003e\n \u003cp\u003eEsophageal/GEJ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.08\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.01 \u0026ndash; 0.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.011\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eTumor:\u0026nbsp;Lung\u0026nbsp;vs Esophageal/GEJ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e0.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.05 \u0026ndash; 1.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.062\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 224px;\"\u003e\n \u003cp\u003eArm\u0026nbsp;2\u0026nbsp;vs\u0026nbsp;Arm 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e0.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.23 \u0026ndash; 1.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 75px;\"\u003e\n \u003cp\u003e0.37\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eFigure\u0026nbsp;3\u003c/strong\u003e. Average marginal effects of key covariates on transition probabilities.\u003c/p\u003e\n\u003ch2\u003ePredicted\u0026nbsp;Probabilities\u0026nbsp;by\u0026nbsp;Clinical\u0026nbsp;Profile\u003c/h2\u003e\n\u003cp\u003ePatient profiles combining age, ECOG, anxiety, presence of caregiver, tumor, and treatment status were evaluated to characterize clinical risk: younger fitter patient profiles had 65\u0026ndash;75% chance of active status at T0\u0026rarr;T12; older frail patients reflected a greater degree of death risk (~34-40%). Anxiety and no caregivers raised the risk of functional attrition (~48%). At T12\u0026rarr;T24, among those who are active, most remain likewise, particularly if they are younger, and less anxious (\u0026gt;80%) (Supplementary Table S3).\u003c/p\u003e\n\u003ch2\u003eQoL\u0026nbsp;prediction Modeling\u0026nbsp;and\u0026nbsp;Comparison of\u0026nbsp;Missing\u0026nbsp;Data\u0026nbsp;Handling Strategies\u003c/h2\u003e\n\u003cp\u003eAmong the 319 patients scheduled for 24-week follow-up visit, only 137 (43%) provided complete QoL data. Of those with complete data, 43.8% experienced QoL deterioration (\u0026ge;0 points). By using more traditional definitions, the rates of QoL decline fell to 23.4%, 19.7% and 16.8% for rates of decline \u0026ge;5, \u0026ge;7, and \u0026ge;10 FACT-G points, respectively (see Supplementary Table S4). The substantial degree of informative missingness justified the use of advanced methods, including Inverse Probability of Completion Weighting (IPCW) and Multiple Imputation by Chained Equations (MICE). The IPCW weights we calculated using logistic regression were well-stabilized (mean ESS 127\u0026ndash;132; see Figure 4 and upplementary table S4), which suggested that they were satisfactory for statistical inference.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure\u0026nbsp;4\u003c/strong\u003e. Distribution of IPCW Weights\u003c/p\u003e\n\u003cp\u003eFor all analytic thresholds of interest, MICE-derived models outperformed IPCW-derived models.\u0026nbsp;For\u0026nbsp;example,\u0026nbsp;the\u0026nbsp;highest\u0026nbsp;AUC\u0026nbsp;was\u0026nbsp;achieved\u0026nbsp;using\u0026nbsp;LASSO-MICE\u0026nbsp;(AUC\u0026nbsp;=\u0026nbsp;0.671, IQR\u003c/p\u003e\n\u003cp\u003e= 0.136) at the -5 FACT-G threshold, with Random Forest and XGBoost using MICE also exhibiting superior performance to model using IPCW. MICE produced lower Brier scores (up to 0.10 lower than IPCW) which suggests greater accuracy in the predicted probabilities of observing actual risk especially for nonlinear models. Importantly, calibration analysis further favored MICE since slopes were closer to the ideal of 1 (e.g., slope = 1.005 for LASSO-MICE at \u0026ndash;10), while IPCW often produced suboptimal (\u0026lt;0.5) or negative slopes, indicating systematic risk misestimation. MICE also returned lower variability (smaller IQRs) confirming more stable performance. See Table 3 and Supplementary Figure S1 for full model performance results.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3\u003c/strong\u003e. Comparison of Model Discrimination\u003cstrong\u003e\u003cem\u003e,\u0026nbsp;\u003c/em\u003e\u003c/strong\u003eOverall Accuracy (Brier Score), Calibration (Slope) by Missing Data Method\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eThreshold\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 140px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eIPCW\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(\u003cem\u003eM\u003c/em\u003eedian [IQR])\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 167px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMICE\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(\u003cem\u003eMedian\u0026nbsp;\u003c/em\u003e[IQR])\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eComment\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"6\" valign=\"top\" style=\"width: 708px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u0026nbsp;Discrimination\u0026nbsp;(AUC)\u0026nbsp;by\u0026nbsp;Missing\u0026nbsp;Data\u0026nbsp;Method\u0026nbsp;(median [IQR])\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.577 [0.130]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.664 [0.158]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eHigher\u0026nbsp;AUC\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.604 [0.095]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.644 [0.169]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eSlight\u0026nbsp;\u0026uarr; with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.623 [0.110]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.663 [0.155]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eHigher\u0026nbsp;AUC\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.613 [0.098]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.629 [0.195]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eStable,\u0026nbsp;larger\u0026nbsp;IQR MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.643 [0.158]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.608 [0.170]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eSlight\u0026nbsp;\u0026darr; with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.590 [0.100]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.625 [0.176]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eHigher\u0026nbsp;AUC\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.560 [0.113]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.671 [0.136]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eClear\u0026nbsp;improvement\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.579 [0.103]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.622 [0.169]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eHigher\u0026nbsp;AUC\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.579 [0.130]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.661 [0.165]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eHigher\u0026nbsp;AUC\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.612 [0.141]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.670 [0.082]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eHigher\u0026nbsp;AUC\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.586 [0.073]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.645 [0.097]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eHigher\u0026nbsp;AUC\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.575 [0.066]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.652 [0.089]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eHigher\u0026nbsp;AUC\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"6\" valign=\"top\" style=\"width: 708px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u0026nbsp;Overall\u0026nbsp;Accuracy\u0026nbsp;(Brier\u0026nbsp;Score)\u0026nbsp;by\u0026nbsp;Missing\u0026nbsp;Data\u0026nbsp;Method\u0026nbsp;(median [IQR])\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.136 [0.073]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.145 [0.044]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eSimilar\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.147 [0.047]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.150 [0.056]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eSimilar\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.176 [0.067]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.146 [0.055]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eLower\u0026nbsp;with\u0026nbsp;MICE\u0026nbsp;(better)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.169 [0.066]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.161 [0.038]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eLower\u0026nbsp;with\u0026nbsp;MICE\u0026nbsp;(better)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.179 [0.052]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.169 [0.047]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eLower\u0026nbsp;with\u0026nbsp;MICE\u0026nbsp;(better)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.215 [0.058]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.172 [0.049]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eLower\u0026nbsp;with\u0026nbsp;MICE\u0026nbsp;(better)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.175 [0.036]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.177 [0.032]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eSimilar\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.187 [0.054]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.184 [0.037]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eSimilar\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.215 [0.084]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.184 [0.042]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eLower\u0026nbsp;with\u0026nbsp;MICE\u0026nbsp;(better)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.239 [0.023]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.221 [0.023]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eLower\u0026nbsp;with\u0026nbsp;MICE\u0026nbsp;(better)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.269 [0.065]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.227 [0.027]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eLower\u0026nbsp;with\u0026nbsp;MICE\u0026nbsp;(better)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.329 [0.082]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.228 [0.029]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eLower\u0026nbsp;with\u0026nbsp;MICE\u0026nbsp;(better)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"5\" valign=\"top\" style=\"width: 708px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003cstrong\u003eModel\u0026nbsp;Calibration\u0026nbsp;(Slope)\u0026nbsp;by\u0026nbsp;Missing\u0026nbsp;Data\u0026nbsp;Method\u0026nbsp;(median [IQR])\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.775 [2.166]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e1.005 [1.199]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eCloser\u0026nbsp;to 1 with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.199 [0.691]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.656 [0.817]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eMarked\u0026nbsp;improvement\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.124 [0.381]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.658 [0.606]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eMarked\u0026nbsp;improvement\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.532 [0.858]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.840 [0.773]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eCloser\u0026nbsp;to 1 with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.366 [0.590]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.497 [0.624]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eSlight\u0026nbsp;improvement\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.124 [0.201]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.456 [0.610]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eImprovement\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.522 [0.796]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.858 [0.727]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eCloser\u0026nbsp;to 1 with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.092 [0.648]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.666 [0.668]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eMarked\u0026nbsp;improvement\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.066 [0.309]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.571 [0.571]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eMarked\u0026nbsp;improvement\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e1.681 [3.263]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.882 [0.519]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eCorrected\u0026nbsp;towards\u0026nbsp;1\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e0.132 [0.671]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.728 [0.502]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eMarked\u0026nbsp;improvement\u0026nbsp;with MICE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 83px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e-0.029 [0.302]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e0.619 [0.380]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 233px;\"\u003e\n \u003cp\u003eCorrected\u0026nbsp;(better\u0026nbsp;with\u0026nbsp;MICE)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003ch2\u003eSensitivity\u0026nbsp;Analysis\u0026nbsp;under\u0026nbsp;MNAR\u0026nbsp;Scenarios\u003c/h2\u003e\n\u003cp\u003eRobustness analyses with delta shifts (\u0026Delta;=5,7,10) showed minor reductions in AUC and Brier scores\u0026nbsp;but\u0026nbsp;maintained\u0026nbsp;model\u0026nbsp;rankings\u0026nbsp;and\u0026nbsp;stable\u0026nbsp;calibration\u0026nbsp;slopes\u0026nbsp;near\u0026nbsp;1,\u0026nbsp;indicating\u0026nbsp;consistent predictive performance despite departures from MAR (Table 4).\u003c/p\u003e\n\u003ch2\u003eVariable Importance\u003c/h2\u003e\n\u003cp\u003eBaseline QoL, age, and anxiety and depression were consistently in the top ranks of important predictors across models and substrategies for missing data methods overall. Tumor site and treatment arm varied by method. ECOG less consistently predictive. This reflected the overall greater consistency between the MICE and IPCW (Figure 5).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable\u0026nbsp;4\u003c/strong\u003e. Sensitivity of AUC under MNAR assumptions\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003eThreshold\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003eIPCW\u003c/p\u003e\n \u003cp\u003e(median [IQR])\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eMICE\u003c/p\u003e\n \u003cp\u003e(median [IQR])\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eMNAR\u0026nbsp;\u0026Delta;5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003eMNAR\u0026nbsp;\u0026Delta;7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003eMNAR \u0026Delta;10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eSensitivity interpretation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"8\" valign=\"top\" style=\"width: 715px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSensitivity\u0026nbsp;of AUC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.577 [0.130]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.664 [0.158]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.639 [0.119]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.626\u003c/p\u003e\n \u003cp\u003e[0.116]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.630\u003c/p\u003e\n \u003cp\u003e[0.118]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eSlightly\u0026nbsp;lower than\u003c/p\u003e\n \u003cp\u003eMICE\u0026nbsp;but\u0026nbsp;within\u0026nbsp;MAR range \u0026rarr; stable\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.604 [0.095]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.644 [0.169]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.614 [0.127]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.619\u003c/p\u003e\n \u003cp\u003e[0.164]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.582\u003c/p\u003e\n \u003cp\u003e[0.118]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eMinor\u0026nbsp;sensitivity\u0026nbsp;at\u003c/p\u003e\n \u003cp\u003e\u0026Delta;10,\u0026nbsp;otherwise stable\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.623 [0.110]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.663 [0.155]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.623 [0.147]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.614\u003c/p\u003e\n \u003cp\u003e[0.143]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.625\u003c/p\u003e\n \u003cp\u003e[0.113]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eVery\u0026nbsp;stable\u0026nbsp;across\u003c/p\u003e\n \u003cp\u003eMNAR settings\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.613 [0.098]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.629 [0.195]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.616 [0.121]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.632\u003c/p\u003e\n \u003cp\u003e[0.120]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.603\u003c/p\u003e\n \u003cp\u003e[0.115]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eConsistent\u003c/p\u003e\n \u003cp\u003ediscrimination\u0026nbsp;under MNAR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.643 [0.158]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.608 [0.170]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.622 [0.127]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.617\u003c/p\u003e\n \u003cp\u003e[0.117]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.567\u003c/p\u003e\n \u003cp\u003e[0.084]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eStable\u0026nbsp;to\u0026nbsp;\u0026Delta;7, minor\u003c/p\u003e\n \u003cp\u003edrop\u0026nbsp;at \u0026Delta;10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.590 [0.100]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.625 [0.176]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.627 [0.123]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.609\u003c/p\u003e\n \u003cp\u003e[0.105]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.590\u003c/p\u003e\n \u003cp\u003e[0.084]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eStable\u0026nbsp;under MNAR,\u003c/p\u003e\n \u003cp\u003eslight\u0026nbsp;decrease\u0026nbsp;at \u0026Delta;10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.560 [0.113]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.671 [0.136]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.643 [0.119]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.635\u003c/p\u003e\n \u003cp\u003e[0.084]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.632\u003c/p\u003e\n \u003cp\u003e[0.112]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eMinor\u0026nbsp;decrease vs\u003c/p\u003e\n \u003cp\u003eMICE,\u0026nbsp;overall robust\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.579 [0.103]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.622 [0.169]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.622 [0.125]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.614\u003c/p\u003e\n \u003cp\u003e[0.099]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.590\u003c/p\u003e\n \u003cp\u003e[0.090]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eStable\u0026nbsp;up\u0026nbsp;to \u0026Delta;7,\u003c/p\u003e\n \u003cp\u003edegradation\u0026nbsp;under \u0026Delta;10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.579 [0.130]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.661 [0.165]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.615 [0.116]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.621\u003c/p\u003e\n \u003cp\u003e[0.071]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.598\u003c/p\u003e\n \u003cp\u003e[0.094]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eSlight\u0026nbsp;drop\u0026nbsp;under MNAR, within\u003c/p\u003e\n \u003cp\u003etolerance\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"8\" valign=\"top\" style=\"width: 715px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSensitivity\u0026nbsp;of\u0026nbsp;Brier Score\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.136 [0.073]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.145 [0.044]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.168 [0.041]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.184\u003c/p\u003e\n \u003cp\u003e[0.035]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.204\u003c/p\u003e\n \u003cp\u003e[0.038]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eProgressive\u0026nbsp;increase\u0026nbsp;\u0026rarr; slightly worse under\u003c/p\u003e\n \u003cp\u003eMNAR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.147 [0.047]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.150 [0.056]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.172 [0.045]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.191\u003c/p\u003e\n \u003cp\u003e[0.036]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.215\u003c/p\u003e\n \u003cp\u003e[0.036]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eGradual degradation\u003c/p\u003e\n \u003cp\u003ewith\u0026nbsp;higher \u0026Delta;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.176 [0.067]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.146 [0.055]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.168 [0.048]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.192\u003c/p\u003e\n \u003cp\u003e[0.036]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.211\u003c/p\u003e\n \u003cp\u003e[0.036]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eClear\u0026nbsp;loss\u0026nbsp;in accuracy\u003c/p\u003e\n \u003cp\u003eat \u0026Delta;\u0026ge;7\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.169 [0.066]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.161 [0.038]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.195 [0.041]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.208\u003c/p\u003e\n \u003cp\u003e[0.030]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.224\u003c/p\u003e\n \u003cp\u003e[0.034]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eGradual deterioration\u003c/p\u003e\n \u003cp\u003eunder\u0026nbsp;MNAR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.179 [0.052]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.169 [0.047]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.202 [0.041]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.215\u003c/p\u003e\n \u003cp\u003e[0.033]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.230\u003c/p\u003e\n \u003cp\u003e[0.037]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eSimilar\u0026nbsp;pattern,\u003c/p\u003e\n \u003cp\u003eaccuracy\u0026nbsp;slightly worse\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.215 [0.058]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.172 [0.049]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.206 [0.046]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.221\u003c/p\u003e\n \u003cp\u003e[0.036]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.231\u003c/p\u003e\n \u003cp\u003e[0.035]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eStable\u0026nbsp;but slightly\u003c/p\u003e\n \u003cp\u003ehigher\u0026nbsp;error\u0026nbsp;under \u0026Delta;10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.175 [0.036]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.177 [0.032]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.214 [0.031]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.219\u003c/p\u003e\n \u003cp\u003e[0.027]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.230\u003c/p\u003e\n \u003cp\u003e[0.030]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eSlight degradation,\u003c/p\u003e\n \u003cp\u003econsistent trend\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.187 [0.054]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.184 [0.037]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.216 [0.035]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.223\u003c/p\u003e\n \u003cp\u003e[0.033]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.239\u003c/p\u003e\n \u003cp\u003e[0.029]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eGradual\u0026nbsp;increase \u0026rarr;\u003c/p\u003e\n \u003cp\u003eless accurate\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.215 [0.084]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.184 [0.042]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.217 [0.039]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.223\u003c/p\u003e\n \u003cp\u003e[0.037]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.240\u003c/p\u003e\n \u003cp\u003e[0.035]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 160px;\"\u003e\n \u003cp\u003eAccuracy\u0026nbsp;declines modestly with \u0026Delta;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"8\" valign=\"top\" style=\"width: 716px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSensitivity\u0026nbsp;of\u0026nbsp;Calibration Slope\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.136 [0.073]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.145 [0.044]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.168 [0.041]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.184\u003c/p\u003e\n \u003cp\u003e[0.035]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.204\u003c/p\u003e\n \u003cp\u003e[0.038]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 161px;\"\u003e\n \u003cp\u003eProgressive\u0026nbsp;increase \u0026rarr;\u003c/p\u003e\n \u003cp\u003eslightly\u0026nbsp;worse\u0026nbsp;under MNAR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.147 [0.047]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.150 [0.056]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.172 [0.045]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.191\u003c/p\u003e\n \u003cp\u003e[0.036]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.215\u003c/p\u003e\n \u003cp\u003e[0.036]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 161px;\"\u003e\n \u003cp\u003eGradual degradation\u003c/p\u003e\n \u003cp\u003ewith\u0026nbsp;higher \u0026Delta;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.176 [0.067]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.146 [0.055]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.168 [0.048]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.192\u003c/p\u003e\n \u003cp\u003e[0.036]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.211\u003c/p\u003e\n \u003cp\u003e[0.036]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 161px;\"\u003e\n \u003cp\u003eClear\u0026nbsp;loss\u0026nbsp;in accuracy\u003c/p\u003e\n \u003cp\u003eat \u0026Delta;\u0026ge;7\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.169 [0.066]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.161 [0.038]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.195 [0.041]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.208\u003c/p\u003e\n \u003cp\u003e[0.030]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.224\u003c/p\u003e\n \u003cp\u003e[0.034]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 161px;\"\u003e\n \u003cp\u003eGradual deterioration\u003c/p\u003e\n \u003cp\u003eunder\u0026nbsp;MNAR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.179 [0.052]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.169 [0.047]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.202 [0.041]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.215\u003c/p\u003e\n \u003cp\u003e[0.033]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.230\u003c/p\u003e\n \u003cp\u003e[0.037]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 161px;\"\u003e\n \u003cp\u003eSimilar\u0026nbsp;pattern,\u003c/p\u003e\n \u003cp\u003eaccuracy\u0026nbsp;slightly worse\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.215 [0.058]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.172 [0.049]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.206 [0.046]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.221\u003c/p\u003e\n \u003cp\u003e[0.036]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.231\u003c/p\u003e\n \u003cp\u003e[0.035]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 161px;\"\u003e\n \u003cp\u003eStable\u0026nbsp;but slightly\u003c/p\u003e\n \u003cp\u003ehigher\u0026nbsp;error\u0026nbsp;under \u0026Delta;10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e-5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eglmnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.175 [0.036]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.177 [0.032]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.214 [0.031]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.219\u003c/p\u003e\n \u003cp\u003e[0.027]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.230\u003c/p\u003e\n \u003cp\u003e[0.030]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 161px;\"\u003e\n \u003cp\u003eSlight degradation,\u003c/p\u003e\n \u003cp\u003econsistent trend\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003erf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.187 [0.054]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.184 [0.037]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.216 [0.035]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.223\u003c/p\u003e\n \u003cp\u003e[0.033]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.239\u003c/p\u003e\n \u003cp\u003e[0.029]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 161px;\"\u003e\n \u003cp\u003eGradual\u0026nbsp;increase \u0026rarr;\u003c/p\u003e\n \u003cp\u003eless accurate\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003exgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e0.215 [0.084]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.184 [0.042]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e0.217 [0.039]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.223\u003c/p\u003e\n \u003cp\u003e[0.037]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.240\u003c/p\u003e\n \u003cp\u003e[0.035]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 161px;\"\u003e\n \u003cp\u003eAccuracy declines\u003c/p\u003e\n \u003cp\u003emodestly\u0026nbsp;with \u0026Delta;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eOur study evaluated the attrition dynamics and predictors of 24-week quality-of-life (QoL) deterioration in advanced cancer patients. Approximately one-third of the patients experienced functional attrition at 24 weeks, and similarly, approximately one-third died, indicating selective and informative attrition. Lower education, higher distress, and poorer clinical performance at baseline were associated with higher functional attrition.\u003c/p\u003e \u003cp\u003eMultinomial regression and predictive modelling showed that lower education increased the risk of functional attrition, while older age, poor ECOG status and depression symptoms increased the risk of dying; early engagement of caregivers was protective and MICE gave more stable and better calibrated estimates than IPCW with moderate discriminative power (AUC\u0026thinsp;\u0026asymp;\u0026thinsp;0.67). The MNAR sensitivity analysis confirmed that the major conclusions of the study remained valid.\u003c/p\u003e \u003cp\u003eThese results are consistent with prior oncology studies reporting that patient-reported outcome (PRO) dropout is informative and related to declining functional status and prognosis rather than randomness. For example, Efficace et al [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] found that performance status and symptom burden predicted PRO failure in cancer patients, while Groenvold et al demonstrated the link between advanced disease and PRO dropout in lung cancer trials [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Cuer et al showed similar patterns of PRO attrition in esophageal cancer [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e], patients who drop out early tend to have substantially worse HRQoL trajectories than those who remain. Other studies, PMC8059010\u003c/p\u003e \u003cp\u003e[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] and PMC5834396 [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], have associated lower baseline health status, greater symptom burden with increased dropout. These observations in our cohort reinforce previous findings depression is a strong predictor of attrition and mortality [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] but moreover, the association between lower educational attainment and a higher likelihood of functional attrition adds nuance to prior work, suggesting that social determinants contribute not only to survival disparities but also to engagement with longitudinal QoL assessments. ECOG performance status did not significantly predict functional attrition in multivariable modeling. A small marginal increase in predicted attrition probability observed in AME analysis should therefore be interpreted with caution and likely reflects statistical noise rather than a true underlying effect.\u003c/p\u003e \u003cp\u003eAdditionally, our study provides an empirical comparison of IPCW and MICE for modeling QoL deterioration and our findings both support and challenge established methodologic norms in the use of informative missing data. Whereas some cancer prognosis studies utilizing molecular or clinical predictors have achieved very high AUCs (\u0026gt;\u0026thinsp;0.90) with advanced machine\u003c/p\u003e \u003cp\u003elearning [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e], our results show that the discriminative ability for PRO deterioration is notably lower, reflecting the inherent complexity of these outcomes compared to biomarker- or imaging-based endpoints. Although the mean AUC values (approximately 0.67) for the top performing model MICE-LASSO, with an IQR of approximately 0.136 indicated relatively stable and moderate prediction performance for all imputed models across all folds, the results in our study are consistent with studies on PRO research and predictive modeling in general. Examples include studies by Lian et al [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e], who showed that traditional baseline models for predicting changes in quality-of-life had average AUC values of approximately 0.60 whereas only approaches leveraging deep learning in very large, feature-rich datasets have reported AUC values exceeding 0.75 [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e] [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e], exceptions rather than the rule; Additionally, studies by Wahl et al [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] also showed that even after applying internal validation and multiple imputation techniques, the predictive performance of health data remains modest when dealing with incomplete health data. Although IPCW theoretically has the ability to reduce selection bias [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e], the improved discrimination (AUC) and accuracy (Brier Score) of MICE models, compared to IPCW, reflect the practical limitations of IPCW in real-world data, especially given its sensitivity to model specification and overlap between cases and controls. While MICE carry a theoretical risk of optimism bias under the MAR assumption [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], our analysis shows substantial advantages in variance reduction, stability, and calibration, making MICE preferable in most practical scenarios and reinforces the importance of empirical validation over theoretical preferences. Additionally, the relatively small variability of our models further indicates that while accuracy is moderate, the estimates are robust and reproducible.\u003c/p\u003e \u003cp\u003eOne of the most important findings was the severe miscalibration of IPCW models, with slopes often near zero, especially in nonlinear models. The issue of miscalibration due to extreme weighting is a recognized, but often understated problem [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Accurate calibration is essential for clinical utility, determining whether predicted risk matches observed outcomes [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]; The inability of IPCW to maintain a stable calibration slope, especially in nonlinear models (RF, XGBoost), represents a significant limitation of this approach in developing clinical decision support systems.\u003c/p\u003e \u003cp\u003eOur finding that MICE significantly improved the calibration slope, often from near zero (IPCW) to near 1.0 (MICE) demonstrates a critical empirical contribution to the growing consensus that predictive model stability and robust calibration are often more important than theoretical bias correction when choosing a methodology for the development of clinical decision support systems [29]. The identification of Baseline QoL, age, and baseline psychological distress (HADS scores) as the three most influential predictors, is in alignment\u003c/p\u003e \u003cp\u003ewith prior literature [30] [31]. The consistency of the top-ranking predictors across the two different missing data methodologies increases the evidence that these PROs are the dominant factors in determining the patient outcome within this cohort.\u003c/p\u003e \u003cp\u003eResult from the MNAR sensitivity analysis support the notion that the observed superior calibration of MICE, and equivalent discrimination compared to IPCW, is not a result of the MAR assumption.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eMethodological and clinical contribution\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eOne of the primary contributions of this study from a methodological perspective is the explicit consideration of functional attrition as an informative state and the comparative evaluation of IPCW and MICE methods to handle missing data. In addition, through the evaluation of model performance under MNAR conditions, we present a methodologically strong and empirically grounded framework for the analysis of longitudinal patient reported outcomes (PROs), however, the moderately AUC ceiling (\u0026asymp;\u0026thinsp;0.67) underscore the need for time-updated longitudinal predictors to capture the complex, dynamic trajectories of QoL decline beyond baseline variables, which alone yield only moderate predictive accuracy (AUC\u0026thinsp;\u0026asymp;\u0026thinsp;0.67).\u003c/p\u003e \u003cp\u003eFrom a clinical perspective, baseline quality of life, patient age, and psychological distress were the main determinants of QoL decline, whereas poor ECOG status and depressive symptoms predicted mortality with lower educational attainment increasing the risk of functional attrition and the quantified marginal effects support early identification of at-risk patients and tailored interventions.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eStrengths and Limitations\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eStrengths of our study include explicit modeling of functional attrition as an informative state, the evaluation of the marginal effects and predicted probabilities enhancing clinical interpretability, a robust cross-validation of classifiers, and transparent handling of missing data showing a combination of both methodological rigor and practical clinical relevance. Limitations include modest sample size and restriction to baseline predictors, reducing power for transition analyses and individual-level predictions. The MNAR sensitivity analyses demonstrated reassuring robustness but may not fully account for real world non-ignorable missingness. We did not fit continuous-time Markov or semi-Markov multistate models. Instead, we used discrete-time multinomial regression to model state transitions, which was appropriate given the fixed follow-up schedule (12 and 24 weeks) and the study\u0026rsquo;s objective to characterize baseline determinants rather than transition intensities.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eImplications and future research\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eClinically, treating functional attrition as an explicit process rather than as random missingness can improve the interpretation of QoL trajectories. Therefore, identify vulnerable patients early and implementing strategies to retain them in the study should be a priority, and future work should explore targeted strategies such as caregiver training, reduced complexity of questionnaires, enhanced communication, more frequent contacts or adapting follow-up to enhance retention, provide timely supportive care, and assist in palliative planning with potentially minimal effect to allow for stratification and individualized interventions for each patient. Regarding the methodological aspect, future research should include time-updated predictors (e.g. repeated assessments of QoL, symptoms, treatment response, biomarkers) and utilize advanced frameworks (e.g. Joint Modeling, Landmark Analysis, Recurrent Event Models) to better identify relationships among attrition, declines in QoL, and mortality over time. Additionally, validating the study\u0026rsquo;s findings across various cancer types is necessary to determine the study's generalizability and facilitate real-time risk stratification to inform individualized monitoring and enhance the delivery of supportive care and the reliability of QoL assessment.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"CONCLUSION","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eOur study provides new insights into how quality of life (QoL) changes in advanced cancer by directly modeling functional attrition as an informative state and not treating it as missing data that can be ignored. State-transition analyses demonstrated that the transition to functional attrition or death were systemically related to the baseline characteristics of the participants: Poorer ECOG Performance Status and greater levels of depression at baseline resulted in earlier mortality, while lower educational attainment resulted in a greater risk of functional attrition, indicating clinical and sociodemographic determinants.\u003c/p\u003e \u003cp\u003eThe predictive modeling of QoL based solely on baseline data produced modest discriminative ability (AUCmax\u0026thinsp;\u0026asymp;\u0026thinsp;0.67), consistent with limited signal provided by complex Patient Reported Outcomes (PRO). Most importantly, Multiple Imputation by Chained Equations (MICE) empirically performed better than Inverse Probability of Completion Weighting (IPCW), providing more stable and better-calibrated models as evidenced through MNAR Sensitivity Analyses. Although MICE outperformed IPCW in this study, the findings should be interpreted in context and validated against additional datasets and disease settings prior to generalization. Overall, our study highlights the clinical and methodological importance of incorporating functional attrition into QoL research, validates the role of baseline patient reported outcomes as prognostic indicators, and emphasizes the necessity of developing time-updated and multimodal predictors to improve the predictive accuracy of future studies. Furthermore, the study demonstrates the importance of integrating functional attrition and mortality within a unified framework to allow for more accurate interpretation of PROs and support the development of targeted and data-driven supportive care strategies in advanced cancer.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eIndividual participant-level data can be made available, but only through a controlled access mechanism via the NCI/NIH official repository, as required by federal regulations. My Data Use Agreement (DUA) with the National Cancer Institute strictly prohibits releasing or distributing data to any entity (Section 4) and specifically forbids posting it on any public site (Section 9). Consequently, uploading the raw file directly to Research Square would constitute a breach of this federal contract. However, the data is accessible to any researcher who requests it through the official NCI portal, which satisfies standard scientific transparency requirements.\u003c/p\u003e\n \u003cp\u003e \u003cb\u003eHuman Ethics and Consent to Participate\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThis study is a secondary analysis of anonymized data from the clinical trial NCT02349412 registered on 23 January 2015, which evaluated early palliative care versus standard care in patients with advanced cancer. The original trial was conducted in accordance with the Declaration of Helsinki and ICH-GCP guidelines and received all required ethical approvals, with written informed consent obtained from all participants. As the present analysis used only fully de-identified data, no additional ethical approval or consent was required.\u003c/p\u003e\u003cp\u003e \u003ch2\u003eCompeting interests\u003c/h2\u003e \u003cp\u003eThe authors declare that they have no known conflicting financial interests or personal relationships that might appear to affect the work presented in this manuscript.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis research did not receive any funding from any funding agencies in the public, private, or non-profit sectors.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eK.G. conceptualized the study, extracted and analyzed the data, interpreted the results, and wrote the manuscript. All authors reviewed and approved the final manuscript. P.P., A.A., T.S., K.Y., and A.J. prepared the figures.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThe authors gratefully acknowledge the support of the Government of Togo for providing the scholarship that enabled the pursuit of graduate studies. We thank the faculty members of the SRM School of Public Health for their academic environment and institutional support.We also acknowledge the guidance and feedback provided by Dr. Jennifer Gladius (Department of Biostatistics, SRM School of Public Health), Dr. Prakash (School of Public Health, SRMIST), and Dr. Dhivyia (School of Public Health, SRMIST), which contributed to the development of this work.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets supporting this analysis are available from the [NCTN/NCORP](https:/nctn-data-archive.nci.nih.gov/node/1534) Data Archive in accordance with NCI and NIH data-sharing policies. Due to privacy restrictions and limitations of the original informed consent, individual participant-level data cannot be made publicly available. Aggregated and de-identified data used in this study may be requested from the study investigators or through the Data Archive portal, subject to submission of a brief proposal and relevant institutional approvals. Analytical scripts are available from the corresponding author upon reasonable request or through the corresponding author\u0026rsquo;s [GitHub](https:/github.com/jack-junior/MSc_project)\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBasch E, Deal AM, Kris MG, Scher HI, Hudis CA, Sabbatini P, et al. Symptom monitoring with patient-reported outcomes during routine cancer treatment: a randomized controlled trial. JAMA Oncol. 2016;2(6):902\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/jamaoncol.2016.1990\u003c/span\u003e\u003cspan address=\"10.1001/jamaoncol.2016.1990\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eU.S. Food and Drug Administration. Patient-Reported Outcome Measures: Use in Medical Product Development to Support Labelling Claims. Silver Spring: FDA; 2009 (updated 2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCella DF, Tulsky DS, Gray G, Sarafian B, Linn E, Bonomi A, et al. The Functional Assessment of Cancer Therapy scale: development and validation of the general measure. J Clin Oncol. 1993;11(3):570\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1200/JCO.1993.11.3.570\u003c/span\u003e\u003cspan address=\"10.1200/JCO.1993.11.3.570\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCalvert M, Blazeby J, Altman DG, Revicki DA, Moher D, Brundage MD, et al. Reporting of patient-reported outcomes in randomized trials: the CONSORT PRO extension. JAMA. 2013;309(8):814\u0026ndash;22. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/jama.2013.879\u003c/span\u003e\u003cspan address=\"10.1001/jama.2013.879\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSnyder CF, Aaronson NK, Choucair AK, Elliott TE, Greenhalgh J, Halyard MY, et al. Implementing patient-reported outcomes assessment in clinical practice: a review of the options and considerations. Qual Life Res. 2014;23(5):1415\u0026ndash;33. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s11136-013-0396-9\u003c/span\u003e\u003cspan address=\"10.1007/s11136-013-0396-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFairclough DL. Design and Analysis of Quality of Life Studies in Clinical Trials. 2nd ed. Boca Raton: Chapman \u0026amp; Hall/CRC; 2010.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHamel JF, Tuen Mun HS, Bryant AL, Ikwuazom D, Schrag D. Do patients and oncologists discuss the cost of cancer treatment? An observational study of clinical interactions. Support Care Cancer. 2017;25(10):3275\u0026ndash;82. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00520-017-3731-6\u003c/span\u003e\u003cspan address=\"10.1007/s00520-017-3731-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLittle RJA, Rubin DB. Statistical Analysis with Missing Data. 3rd ed. Hoboken: Wiley; 2019.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFayers PM, Machin D. Quality of Life: The Assessment, Analysis and Interpretation of Patient-Reported Outcomes. 2nd ed. Chichester: Wiley; 2016.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBell ML, Fiero M, Horton NJ, Hsu CH. Handling missing data in RCTs in palliative care: a simulation study. BMJ Support Palliat Care. 2016;6(3):295\u0026ndash;301. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/bmjspcare-2015-000889\u003c/span\u003e\u003cspan address=\"10.1136/bmjspcare-2015-000889\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eInternational Council for Harmonisation. ICH E9(R. Addendum on Estimands and Sensitivity Analysis in Clinical Trials. Geneva: ICH; 2019. 1.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVan Cutsem E, Cervantes A, Adam R, Sobrero A, Van Krieken JH, Aderka D, et al. ESMO consensus guidelines for the management of patients with metastatic colorectal cancer. Ann Oncol. 2022;33(7):659\u0026ndash;701. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.annonc.2022.04.003\u003c/span\u003e\u003cspan address=\"10.1016/j.annonc.2022.04.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWarrington L, Absolom K, Velikova G. Prediction of patient-reported outcomes in oncology: current status and future directions. J Clin Epidemiol. 2021;137:175\u0026ndash;88. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jclinepi.2021.04.009\u003c/span\u003e\u003cspan address=\"10.1016/j.jclinepi.2021.04.009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLian R, Hsiao V, Hwang J, Ou Y, Robbins SE, Connor NP, et al. Predicting HRQOL change using natural language processing in thyroid cancer. iScience Biomed Inf. 2023;100097. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.isci.2023.107123\u003c/span\u003e\u003cspan address=\"10.1016/j.isci.2023.107123\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWahl S, L\u0026ouml;we B, Bjorner JB, Rose M, Br\u0026auml;hler E, Walter OB. Assessment of predictive performance in incomplete data by combining internal validation and multiple imputation. BMC Med Res Methodol. 2016;16:144. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12874-016-0228-0\u003c/span\u003e\u003cspan address=\"10.1186/s12874-016-0228-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAndersen PK, Keiding N. Multi-state models for event history analysis. Stat Methods Med Res. 2002;11(2):91\u0026ndash;115. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1191/0962280202SM276ra\u003c/span\u003e\u003cspan address=\"10.1191/0962280202SM276ra\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEfficace F, Bottomley A, Osoba D, Gotay C, Flechtner H, D\u0026rsquo;haese S, et al. Beyond the development of HRQOL measures: a checklist for evaluating HRQOL outcomes in cancer clinical trials. J Clin Oncol. 2006;24(25):4232\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1200/JCO.2006.06.7239\u003c/span\u003e\u003cspan address=\"10.1200/JCO.2006.06.7239\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGroenvold M, Petersen MA, Aaronson NK, Arraras JI, Blazeby JM, Bottomley A, et al. The impact of dropout on the assessment of HRQOL in cancer clinical trials. Qual Life Res. 2010;19(7):1087\u0026ndash;94. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s11136-010-9660-5\u003c/span\u003e\u003cspan address=\"10.1007/s11136-010-9660-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCuer B, Mollevi C, Anota A, Bouch\u0026eacute; O, Bedenne L, Conroy T, et al. Handling informative dropout in longitudinal analysis of health-related quality of life: application of three approaches to data from the esophageal cancer clinical trial PRODIGE 5/ACCORD 17. BMC Med Res Methodol. 2020;20:223. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12874-020-01104-w\u003c/span\u003e\u003cspan address=\"10.1186/s12874-020-01104-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGebert P, Schindel D, Frick J, Schenk L, Grittner U. Characteristics and patient-reported outcomes associated with dropout in severely affected oncological patients: an exploratory study. Support Care Cancer. 2021;29(6):3311\u0026ndash;20. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00520-020-05893-3\u003c/span\u003e\u003cspan address=\"10.1007/s00520-020-05893-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePerez-Cruz PE, Shamieh O, Paiva CE, Kwon JH, Muckaden MA, Bruera E, et al. Factors associated with attrition in a multicenter longitudinal observational study of patients with advanced cancer. J Pain Symptom Manage. 2018;55(3):938\u0026ndash;45. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jpainsymman.2017.11.019\u003c/span\u003e\u003cspan address=\"10.1016/j.jpainsymman.2017.11.019\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQuinten C, Coens C, Mauer M, Comte S, Sprangers MAG, Cleeland C, et al. Baseline quality of life as a prognostic indicator of survival: a meta-analysis of individual patient data from EORTC clinical trials. Lancet Oncol. 2009;10(9):865\u0026ndash;71. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S1470-2045(09)70200-1\u003c/span\u003e\u003cspan address=\"10.1016/S1470-2045(09)70200-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKourou K, Exarchos TP, Exarchos KP, Karamouzis MV, Fotiadis DI. Machine learning applications in cancer prognosis and prediction. Comput Struct Biotechnol J. 2015;13:8\u0026ndash;17. PMID:25750696; PMCID:PMC4348437.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbegaz TM, Ahmed M, Ali AA, Bhagavathula AS. Predicting HRQOL using social determinants of health: a machine learning approach with the All of Us cohort. Risk Manag Healthc Policy. 2024;17:1665\u0026ndash;78. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.2147/RMHP.S436294\u003c/span\u003e\u003cspan address=\"10.2147/RMHP.S436294\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRobins JM, Finkelstein DM. Correcting for noncompliance and dependent censoring in an AIDS clinical trial with IPCW log-rank tests. Biometrics. 2000;56(3):779\u0026ndash;88. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/j.0006-341X.2000.00779.x\u003c/span\u003e\u003cspan address=\"10.1111/j.0006-341X.2000.00779.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWhite IR, Royston P, Wood AM. Multiple imputation using chained equations: issues and guidance for practice. Int J Epidemiol. 2011;40(2):338\u0026ndash;48. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/ije/dyr080\u003c/span\u003e\u003cspan address=\"10.1093/ije/dyr080\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSeaman SR, White IR. Review of inverse probability weighting for dealing with missing data. Int J Epidemiol. 2013;42(4):1091\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/ije/dyt110\u003c/span\u003e\u003cspan address=\"10.1093/ije/dyt110\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVan Calster B, McLernon DJ, van Smeden M, Wynants L, Steyerberg EW. Calibration: the Achilles heel of predictive analytics. Eur J Epidemiol. 2019;34(8):749\u0026ndash;61. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10654-019-00546-7\u003c/span\u003e\u003cspan address=\"10.1007/s10654-019-00546-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-research-methodology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmrm","sideBox":"Learn more about [BMC Medical Research Methodology](http://bmcmedresmethodol.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmrm/default.aspx","title":"BMC Medical Research Methodology","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Functional attrition, quality of life, patient-reported outcomes, advanced cancer","lastPublishedDoi":"10.21203/rs.3.rs-8407124/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8407124/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground: \u003c/strong\u003eFunctional attrition (non-response) of Patient-Reported Outcomes (PROs) and Quality of Life (QoL) deterioration are common and non-random in oncology trials. While functional attrition is informative, it is often treated as simple missing data, which biases the interpretation of QoL trajectories and prediction models.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eObjectives and Methods: \u003c/strong\u003eThis study aimed (1) To examine prognostic factors of transitions between active participation, functional attrition, and death using multinomial models and (2) To evaluate how different methods of dealing with missing data affect the reliability of QoL deterioration prediction models. A discrete-state modeling framework was used to analyze transitions between active participation, functional attrition, and death in advanced cancer patients. QoL decline was then predicted using LASSO, Random Forest, and XGBoost algorithms, while Inverse Probability of Completion Weighting (IPCW) under a conditional Missing At Random (MAR) assumption was systematically compared against Multiple Imputation by Chained Equations (MICE). At the end, we examined robustness to deviations from MAR through delta-adjusted Missing Not At Random (MNAR) sensitivity analyses.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults: \u003c/strong\u003eFunctional attrition accumulated more rapidly than death within the first 12 weeks, confirming its high prevalence and distinct determinants. Lower educational attainment was associated with functional attrition, whereas poor performance status and depressive symptoms primarily predicted mortality. Among predictive models, LASSO combined with MICE achieved the best performance (median AUC = 0.671 [IQR 0.136]), reflecting the modest predictability of dynamic PRO outcomes when relying on baseline covariates. Methodological comparison demonstrated superior discrimination and substantially improved calibration for MICE relative to IPCW, with IPCW models showing marked miscalibration. These findings remained stable across MNAR sensitivity scenarios.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion: \u003c/strong\u003eBy modeling the functional attrition as a distinct state, we revealed its specific socio-demographic determinants and clarified how covariate profiles shape patient trajectories\u003cstrong\u003e. \u003c/strong\u003eFor QoL prediction under conditional MAR, MICE yields more stable and better-calibrated models than IPCW. Robustness across MNAR scenarios supports the methodological validity of these conclusions and highlights the importance of principled handling of informative non-response in longitudinal PRO research.\u003c/p\u003e","manuscriptTitle":"Functional attrition and Quality of Life in Advanced Cancer Trial: modeling patients’ trajectories and evaluating the impact of missing outcome data handling on quality-of-life predictive model performance","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-19 13:34:38","doi":"10.21203/rs.3.rs-8407124/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-research-methodology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmrm","sideBox":"Learn more about [BMC Medical Research Methodology](http://bmcmedresmethodol.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmrm/default.aspx","title":"BMC Medical Research Methodology","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"bb14510a-06fa-46a3-a324-18f7dbd37561","owner":[],"postedDate":"January 19th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-01-20T12:37:38+00:00","versionOfRecord":[],"versionCreatedAt":"2026-01-19 13:34:38","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8407124","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8407124","identity":"rs-8407124","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}