Handling Missing Outcomes in Time-to-Event Analyses: A Scoping Review of Multiple Imputation in Randomised Controlled Trials

Handling Missing Outcomes in Time-to-Event Analyses: A Scoping Review of Multiple Imputation in Randomised Controlled Trials | 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 Handling Missing Outcomes in Time-to-Event Analyses: A Scoping Review of Multiple Imputation in Randomised Controlled Trials Saravanaraj Karuppasamy, Prasanna Samuel Premkumar, Venkata Raghava Mohan This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6801188/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 28 Sep, 2025 Read the published version in BMC Medical Research Methodology → Version 1 posted 10 You are reading this latest preprint version Abstract Background Randomized Controlled Trials (RCTs) are the gold standard for evaluating treatment effects. However, several factors can threaten the validity of findings, including missing outcomes. Missing data pose a unique challenge in time-to-event analyses, where the event time may be censored rather than completely missing. Proper handling of missing event times is crucial to ensure unbiased and reliable conclusions in RCTs. This scoping review examines how missing outcomes in time-to-event studies have been addressed in high-impact medical journals and evaluates the implementation and reporting of multiple imputation (MI) techniques in RCTs. Method This scoping review assessed methods for handling missing time-to-event outcomes in RCTs published between 2019 and 2023 in three high-impact medical journals: The New England Journal of Medicine, The Lancet, and The Journal of the American Medical Association. Studies with time-to-event as the primary outcome were included. If missing outcomes were present, a full review was conducted to assess the methods used and how they were reported, including details on multiple imputation (MI). The review also explored theoretical approaches for imputing censored event times. Results A total of 694 articles were identified through a PubMed search. After screening, 321 RCTs underwent full-text review. Of these, 297 (92.2%) had no or < 10% missing outcomes without imputation. The remaining 17 (5.3%) addressed missing data using statistical methods: 10 used MI, 6 used best-/worst-case scenarios, and 1 applied a propensity score method. MI approaches varied, with some studies lacking detailed reporting. Conclusion In RCTs with survival outcomes, properly handling missing event times is essential. This scoping review reveals that, despite the availability of robust statistical methods, the treatment of missing time-to-event outcomes remains underutilized and often poorly documented. Many studies acknowledge censoring but fail to distinguish between informative and non-informative censoring. Additionally, the reporting of multiple imputation techniques is frequently insufficient. These findings highlight a critical gap in the handling and reporting of missing outcomes in survival analysis. Strengthening these practices will enhance the reliability and reproducibility of survival analyses in RCTs. Time-to-event data Survival Analysis Missing outcomes Multiple Imputation What is new? Key findings Randomised Controlled Trials with time-to-event outcomes often lack clear reporting on early dropout-related missing outcomes, particularly whether censoring is informative or non-informative, even in high-impact journals. Additionally, when multiple imputation is used for handling missing outcomes, details on its implementation in time-to-event analyses are frequently underreported. What this adds what is known? Previous literature reviews have given limited attention to missing outcomes in time-to-event analyses, primarily focusing on general reporting structures and the use of multiple imputation in RCTs without specifically addressing survival data. Furthermore, studies assessing the reporting of time-to-event outcomes in trial publications have largely overlooked how missing outcomes are handled. This review provides a comprehensive evaluation of the reporting and handling of missing outcomes in time-to-event analyses within high-impact journals. What are the implications and what should be changed? Researchers should explicitly assess whether early dropout, considered as a missing outcome, results in informative or non-informative censoring by examining its association with relevant covariates. If censoring is determined to be non-informative, this should be supported with appropriate data reporting. In cases where the censoring mechanism remains uncertain, it is advisable to assume informative censoring and conduct sensitivity analyses to assess the robustness of the findings. Furthermore, methodological experts should establish standardized guidelines for handling missing time-to-event outcome data in randomized controlled trials, incorporating software-specific recommendations and structured analytical workflows. 1. Background Randomized Controlled Trials (RCTs) are considered the gold standard for evaluating treatment effects in clinical research because RCTs minimize the bias and are closer to the true effects compared to the other research methods 1 . However, several challenges can lead to inconsistent results, including unmeasured confounders, inappropriate analysis plans, and missing data. Among these, missing data pose a particularly serious threat to the validity of findings 2 . Even when the analyses are conducted under appropriate assumptions, missing values in outcomes or covariates can introduce bias and reduce the reliability of results. In RCTs, outcomes should be tracked throughout the study. However, complete follow-up is often not possible in real-world settings. Missing data might have occurred in both continuous and categorical outcomes, but time-to-event outcomes pose unique issues. The major emphasis of time-to-event analysis is the time until an event of interest (i.e., death, relapse, failure, etc.,) occurs within a specified period. Some participants may not experience the event until the specific time point and are considered censored not missing; this situation is referred to as administrative censoring. In some cases, participants may decline to return to the follow-up for various reasons, but in RCTs, it should be considered important because the intervention involves to the participants and this is considered as missing event time also categorized as non-administrative censoring. If such non-administrative censoring is not well handled, it can significantly impair the validity of time-to-event analysis 3 . For instance, in a hypothetical study where the intervention group experiences twice the censoring rate of the placebo group—potentially due to treatment toxicity or lack of efficacy—the results may be unreliable. Since the exact event times remain unknown, the validity of the findings could be questioned. Kaplan-Meier (KM) survival estimates and Cox regression models are the most popular methods for time-to-event analysis (also known as survival analysis) which is based on the assumption that censoring is independent of the event time. Missing outcomes cannot always be regarded as independent in time-to-event analysis, necessitating procedures that account for uncertainty and include statistical models with sensitivity analysis. Rubin categorizes missing data mechanisms into three major types: Missing Completely at Random (MCAR), where the probability of missingness is unrelated to any observed data; Missing at Random (MAR), where the probability of missingness conditionally depends on the observed data; and Missing Not at Random (MNAR), where the probability of missingness is influenced by unobserved data. These classifications give a framework for selecting appropriate imputation algorithms 4 . Similarly, the censoring mechanisms can be categorized as Censoring Completely at Random (CCAR), Censored at Random (CAR), Censoring Not at Random (CNAR) 5 . In survival analysis, the validity of results depends on the assumption that censoring is not related to the event time. CCAR means that when censoring occurs for reasons unrelated to the event—such as a participant reaching the endpoint of the study period without experiencing the event also known as administrative censoring or withdrawing due to factors like relocation or seeking care elsewhere—the censoring is considered random and non-informative. In such cases, survival estimates remain unbiased, ensuring reliable conclusions about treatment effects in clinical trials. Typically, a predefined endpoint would be a good illustration of this kind of mechanism. The censoring mechanism is under CAR if, conditional on treatment exposure and other observed covariates in the survival model, the probability of censoring is independent of the event time. In contrast, CNAR is an informative censoring in which the probability of censoring depends on the event time which is closely related to non-administrative censoring, even after considering for covariates. Unlike CCAR and CAR, which are regarded as partially non-informative and can be safely ignored in survival analyses, CNAR requires additional attention even though the event times are imputed using rigorous statistical tools still sensitivity analysis is also needed to check the robustness of the results 6 . For instance, in an oncology trial, if patients with more severe disease are more likely to withdraw due to disease progression, their censoring times would be systematically related to their event times, making the censoring informative (CNAR) and this kind of situation we have to address the issue by considering departure from CAR assumptions. Multiple Imputation (MI) and Inverse Probability Weighting (IPW) both are flexible methods for handling missing outcomes in RCTs 7 . However, the primary recommended approach for addressing missing outcomes is MI under the MAR assumption. There are significant contributions to the developing MI methods for continuous and binary variables and comparatively less attention given to the time-to-event data. Nevertheless, over the past decades, a few researchers have contributed to the development of MI methodologies for time-to-event outcomes. Notably, Taylor et al. 8 proposed a nonparametric multiple imputation approach for imputing missing event times in censored data, employing techniques such as risk set imputation and KM imputation. In risk set imputation, censored times are replaced by random draws from observed event times among individuals still at risk after the censoring time. In contrast, KM imputation draws from the estimated distribution of event times for those at risk post-censoring. These methods aim to reproduce the KM estimator, thereby facilitating survival estimation and hypothesis testing through multiple imputed datasets. Building on this, Zhao et al. 9 proposed a multiple imputation method for sensitivity analyses of time-to-event data with possibly informative censoring. Their approach imputes missing event times based on the failure time distribution conditional on the follow-up discontinuation time. This method incorporates different assumptions regarding post-discontinuation event occurrences through a hazard ratio parameter, enabling the analysis of multiple imputed datasets using standard survival analysis techniques, with results combined via Rubin’s rules. Jackson et al. 10 extended the field by proposing a bootstrap-based multiple imputation approach for handling non-independent censoring. This method imputes censored observations by randomly selecting one of the observed event times using a step function. However, as this approach does not provide full estimation and relies on the distribution of observed events, it may not be suitable for datasets with very few events. Lipkovich et al. 6 developed a sensitivity analysis framework for non-ignorable missing data, applying a tipping point approach under a censored-at-random assumption within a multiple imputation framework. Their methodology incorporates a covariate-adjusted piecewise exponential model and compares parametric, semi-parametric, and non-parametric imputation models via simulations. Their findings suggest that parametric imputation models generally exhibit lower bias compared to other methods, particularly when the piecewise exponential-based multiple imputation approach is applied. Atkinson et al. 11 introduced a reference-based sensitivity analysis approach for time-to-event data by integrating multiple imputation with a Weibull proportional hazards model. This method enables the assessment of treatment effects under various assumptions about the missing data mechanism, offering a structured framework for evaluating potential biases. More recently, Wang et al. 3 proposed a retrieved-dropout-based multiple imputation approach for time-to-event data. This method constructs imputation models using data from participants who discontinued study treatment but remained in follow-up, addressing the treatment policy estimand and providing practical implementation guidelines. The aforementioned approaches are based on non-parametric, semi-parametric, and parametric multiple imputation methods for time-to-event data, incorporating both simulation studies and real-world datasets. Additionally, sensitivity analyses were conducted to assess the impact of departures from CAR to CNAR scenarios. Despite these advances in methodological development in missing outcomes handling in time-to-event data, prior literature studies have found that missing outcomes are frequently handled incorrectly, with Complete Case Analysis being the most prevalent technique 12 in RCTs. O. U. Corrol et al. 13 did a systematic analysis on time-to-event missing variables and discovered that 53% of the 148 evaluated publications employed Complete Case Analysis, whereas 22% of the missing outcomes were reported unclearly in oncology research. This review recommended that the Substantive Model Compatible Fully Conditional Specification (SMC-FCS) as the gold standard for MI in time-to-event data, however, this recommendation was limited to oncology research for the time-to-event covariates. More recently, Marius et al. 14 reviewed the reporting and handling of time-to-event outcomes in trials included in meta-analyses of systematic review. The study examined 235 trials, finding that half of them did not adequately report missing outcomes. Additionally, they reported that 57% of the trials had missing outcomes, which was most frequently handled by exclusion from the analysis. Given the advancements and existing gaps in the field, a thorough evaluation of MI methodologies employed in RCTs for survival outcomes—regardless of illness or treatment—is required. Additionally, while several literature reviews have been published, most have focused on missing outcomes in continuous and categorical data, with limited attention given to the methodologies used in published RCTs for time-to-event analyses or, in some cases, not addressing them at all 13 , 15 , 16 . This scoping review aims to systematically examine how missing outcomes in time-to-event studies have been addressed in the top three high-impact medical journals—The NEJM, The Lancet, and The Journal of the American Medical Association (JAMA)—over the past five years (2019–2023). If MI procedures were used to assess time-to-event outcomes, the review would look at how they were implemented and reported. Furthermore, this article investigates the theoretical discussion of existing approaches for imputing censored event timings under various censoring assumptions using Multiple Imputation technique for both parametric and non-parametric approach. 2. Methods 2.1. Multiple Imputation (MI) MI fits a statistical model to the observed data to generate estimates that account for uncertainty. It is a more robust strategy under the MAR assumption since it accommodates uncertainty. To incorporate uncertainty, MI imputes missing values multiple times, generating several completed datasets and estimating parameters from each. Generally, let \(\:X\) be the data matrix, where \(\:{X}_{O}\) represents observed values and \(\:{X}_{M}\) denotes missing values. The multiple imputation procedure is 5 : For \(\:k=\text{1,2},3,\dots\:,\:K\) impute the missing values from the posterior predictive distribution \(\:f\left({X}_{M}|{X}_{O}\right)\) given the observed values, generating K completed datasets. Fit the substantive model to each imputed dataset, and compute the K parameter estimates and their variances. Combine these estimates using Rubin’s rules to obtain pooled parameter estimates and their associated variances. While MI is typically implemented under the MAR assumption, it can also be extended to MNAR scenarios. If the missing data assumption is met correctly and the imputation model is specified correctly then, MI produces asymptotically efficient parameter estimates with nominal coverage 17 . Among various MI methods, multiple imputation by chained equations (MICE) has been widely used for handling missing data in RCTs 18 . However, its suitability for time-to-event outcomes remains still questionable. To address this, Carpenter et al., 5 discussed both parametric and non-parametric MI approaches for survival outcomes. 2.2. Impute the censored event time In time-to-event analysis, censored times are often treated as missing outcomes. When censoring is non-informative, the CCAR assumption, combined with appropriate statistical methods, provides reliable estimates. However, if censoring is informative, it may still be assumed to follow the CAR assumption. In such cases, MI can be a valuable tool for imputing censored event times, given that other covariates are available. The imputation process should be based on either the marginal survival distribution or a survival regression model that incorporates relevant covariates, though not necessarily all observed variables 5 . Two multiple imputation methods have been proposed for handling censored event times in survival outcomes under the CAR assumption: a parametric approach and a non-parametric approach 5 . Both methods involve generating multiple imputed datasets and combining the resulting estimates to account for uncertainty. Let’s consider the survivor function \(\:\widehat{S}\left({c}_{i}\right|\widehat{\beta\:},\:{X}_{1},\:{X}_{2},\dots\:,\:{X}_{p})\) for the \(\:{i}^{th}\) subject and \(\:p\) covariate values observed at and \(\:\widehat{\beta\:}\) is a vector of model parameters sampled from a Bayesian posterior distribution estimated from the observed data. The imputed event time \(\:{t}_{i}^{\text{*}}\) for the \(\:{i}^{th}\) subject is then computed as follows: Compute \(\:{p}_{i}=1-\widehat{S}\left({c}_{i}\right|\widehat{\beta\:},\:{X}_{1},\:{X}_{2},\dots\:,\:{X}_{p})\) Draw a uniform random value \(\:{u}_{i}\sim[{p}_{i},1]\) Impute the event time \(\:{t}_{i}^{\text{*}}\) as the solution of \(\:{u}_{i}=1-\widehat{S}\left({t}_{i}\right|{x}_{i},\widehat{\beta\:})\) . This will ensure that the event time is greater than the censoring time. Therefore, \(\:{u}_{i}=\widehat{F}\left({t}_{i}\right|{x}_{i},\widehat{\beta\:})\) and \(\:{t}_{i}^{\text{*}}={\widehat{F}}^{-1}\left({u}_{i}\right|{x}_{i},\widehat{\beta\:})\) . Thus, imputed time can be found from the inverse of the failure function. This approach is then repeated \(\:K\) times to generate multiple completed datasets. Each completed dataset is analyzed using appropriate statistical methods (most often Cox regression), and the resulting parameter estimates are combined using Rubin’s rules to account for variability across imputations. This ensures valid statistical inference by incorporating both within-imputation and between-imputation uncertainty. Using this approach, Lipkovich et al. 6 developed a distribution-based imputation method that incorporates a stepwise exponential distribution under the CAR assumption. They also conducted sensitivity analyses to assess the robustness of imputation results compared to existing methods using the stress test tipping point analysis. The non-parametric approach is based on the Bayesian bootstrap method, which involves forming a subgroup of the fully observed dataset as a donor pool. Missing values are then imputed by drawing replacements from this donor pool with equal probability. This process is repeated \(\:K\) times to generate multiple imputed datasets, which are subsequently analyzed and pooled to obtain final estimates by using Rubin’s rules. This method was proposed by Rubin and Schenker (1986). Unlike parametric methods, which rely on statistical models to estimate and impute missing data from observed values, the non-parametric approach does not assume an underlying distribution. This makes it particularly useful for large datasets, such as census data, where parametric models can be complex and computationally intensive. Another approach, developed by Hsu, Taylor, and Murray, imputes event times based on auxiliary variables under the CAR assumption. Their method involves creating a bootstrap dataset, identifying the nearest neighbors based on distance metrics, and imputing the missing event times using either the Kaplan-Meier estimator or a proportional hazards model. While these approaches are robust in large samples, they may introduce bias in the parametric approach due to neighborhood censoring and the potential dependence of event times 5 . 2.3. Search strategy To assess how missing outcomes in survival analysis are reported in RCTs and how the MI methods are used to handle them, we focused on clinical trials published in high-impact journals. Specifically, we selected The Lancet, The New England Journal of Medicine (NEJM), and JAMA as our primary sources. Relevant articles were identified through PubMed using a predefined set of search terms. The search strategy and data extraction procedures were outlined in the study protocol and registered in the open science framework 19 . Our analysis included RCTs published between 2019 and 2023, ensuring a comprehensive review of recent practices in handling missing survival outcomes. 2.4. Inclusion criteria Studies were included if they: Employed RCT designs (cluster or parallel) with no blinding, single blinding, or double blinding. Conducted survival analysis using Kaplan-Meier estimation or Cox proportional hazards regression as part of the primary outcome assessment. Explicitly reported methods for handling loss to follow-up (early discontinuation) in the Methods section or supplementary. Involved either primary data collection or secondary data analysis. 2.5. Exclusion Criteria Studies were excluded if they: Were study protocols. Focused on primary outcomes other than survival analysis. Were meta-analyses. Involved in non-clinical trial designs (e.g., cohort studies, retrospective studies, cost-effectiveness analyses). 2.6. Study selection The study selection process began with extracting research articles from PubMed using predefined search terms provided in the protocol, with the results downloaded as a CSV file. A prevalidated questionnaire was prepared in an Excel sheet, and each abstract underwent an initial review to determine whether the study was an RCT and whether survival analysis was included as a primary outcome. The first stage of reviewing studies meets the above mentioned inclusion criteria and then proceeds to a second-stage review to assess the presence of missing outcomes. If no missing outcomes were identified, the review moved on to the next paper. If missing outcomes were present, the paper underwent a detailed investigation to quantify the percentage of dropouts or missing data and to evaluate how the data were handled. For studies that utilized multiple imputation, supplementary or supporting documents were reviewed, and reference papers detailing the imputation methods were documented. 2.7. Statistical methods We reported the frequency and percentage of the study characteristics, with results presented in tables. If multiple imputation was used, the methodology for handling missing data was described in detail. All results were recorded in Microsoft Excel, and tabulation and analysis were performed using R version 4.3.1 20 . The findings were summarized using frequency tables and percentages, and the identified methods were described in detail. 3. Results 3.1. Study selection A total of 694 articles were initially identified through a PubMed search. During the first screening, titles and abstracts were reviewed, and 321 articles were selected for further evaluation. A total of 373 articles were excluded due to study design, type, or lack of a primary focus on survival outcomes. These excluded articles were based on cohort studies, cost-effectiveness analyses, population surveys, systematic reviews, meta-analyses, and methodology papers (329 articles). An additional 44 articles were excluded because survival analysis was not the primary outcome; these articles primarily included studies that considered competing risks, used binary outcomes or treated survival outcomes as secondary (Figure-1). The 321 included articles were RCTs, consisting of parallel or cluster designs, as well as pooled RCT analyses. In these studies, survival analysis—defined as time-to-event analysis—was the primary focus, examining key events such as death, progression-free survival, relapse, or clinical failure, etc., Among the 321 studies, 297 (92.2%) reported no missing outcomes, indicating no dropouts or loss to follow-up, with some cases classified as censored. Studies that reported censoring had missing outcomes due to discontinuations, dropouts, or loss to follow-up, with rates below 10%, considering these cases as censored. Seven studies (2.2%) excluded missing outcomes from their time-to-event analyses but used complete data for primary outcome analyses. The remaining 17 studies (5.3%) underwent a detailed review, as they applied additional methods such as multiple imputations or other approaches for handling missing outcomes in time-to-event analyses (Table 1). 3.2. Study characteristics The 17 articles included in a detailed review of methodology: The NEJM (7 articles), The Lancet (4 articles), and JAMA (6 articles). Notably, five articles each were published in 2020 and 2021, while no survival-related studies discussing missing outcomes were identified in 2022. The majority of the articles focused on the following conditions: cardiovascular diseases (5 articles), cancer (3 articles), infectious diseases (2 articles), pediatric conditions (2 articles), and other conditions (5 articles). Most studies were drug-based RCTs addressing survival outcomes (14 articles), while the remaining focused on health service strategies aimed at improving patient outcomes. The primary outcome in 4 articles was mortality over a prespecified time period, while cardiovascular outcomes were the primary endpoint in another 4 studies. Efficacy, safety, and clinical outcomes were the primary endpoints for the remaining 9 articles. 3.3. Reporting and extent of missing data analysis Among the 17 articles, the majority of RCTs included sample sizes between 100 and 499 participants (7 articles), while three studies enrolled more than 1,000 participants per arm. Regarding missing outcomes, 10 articles reported less than 10% missing data related to time-to-event outcomes, such as dropouts or treatment discontinuations. Six articles reported dropout rates exceeding 20% and applied missing data analysis approaches. In total, 10 articles discussed the use of MI-based methods to address missing outcomes. Six of these 17 conducted best-case and worst-case sensitivity analyses to assess the robustness of their findings, and one study applied the IPW method (Table − 2). 3.4. Missing handling procedure using MI Among the ten studies that applied MI to address missing outcomes given in the appendix, three did not provide a clear explanation of how the imputation was conducted (S7-S9). Four studies explicitly mentioned using the MICE method (S1,S2,S6 and S10); however, they did not specify how the imputation was performed for event status or event time, leaving key methodological details unclear. There are three articles used survival models (S3-S5). Articles with MI based survival models Study-I The study conducted by Pereira et al. 21 aimed to evaluate the impact of a genotype-guided oral P2Y12 inhibitor strategy on ischemic outcomes among CYP2C19 loss-of-function (LOF) allele carriers following percutaneous coronary intervention (PCI). A total of 5,302 participants were enrolled, with 2,641 allocated to the genotype-guided therapy group and 2,635 to the conventional therapy group. The primary outcome was a composite of cardiovascular death, myocardial infarction, stroke, definite or probable stent thrombosis, and severe recurrent ischemia assessed at 12 months. Of the total population, 269 participants (5%) either withdrew prior to 12 months or were lost to follow-up. The primary analysis, which excluded participants with missing outcome data, yielded a hazard ratio (HR) of 0.66 (95% CI: 0.43–1.02) with a p-value of 0.06, indicating no statistically significant difference between the groups. As part of the sensitivity analyses, multiple imputation was performed to address missing laboratory-based genotyping data. Imputation was conducted using a logistic regression model incorporating race/ethnicity, stratification factors, treatment arm, and Spartan genotyping results. Ten imputed datasets were generated, and the primary outcome analysis was repeated across these datasets. The resulting hazard ratio was 0.68 (95% CI: 0.45–1.04), with the estimates combined using Rubin’s rules. However, the study did not specify whether the imputation approach extended to missing event times. Study-II Milstone et al., 22 conducted a study to evaluate whether treating parents with intranasal mupirocin and topical chlorhexidine, as compared to placebo, could reduce the transmission of Staphylococcus aureus (S. aureus) from parents to neonates. The primary outcome was defined as neonatal acquisition of an S. aureus strain concordant with the parental strain identified during prerandomization screening (baseline colonization). A total of 236 neonates were randomized: 117 to the active treatment group (chlorhexidine-impregnated cloths and intranasal mupirocin) and 119 to the placebo group (non-medicated soap cloths and petrolatum ointment). The unadjusted analysis showed a hazard ratio (HR) of 0.43 (95% CI: 0.16–0.79) for concordant S. aureus colonization, and an HR of 0.57 (95% CI: 0.31–0.88) for colonization with any S. aureus strain. The proportional hazards assumption was supported by the data (P = .79). A total of 18 neonates (7.6%) were lost to follow-up, including 12 from the intervention group and 6 from the control group. To address the potential bias introduced by missing outcome data, the investigators conducted sensitivity analyses, including a multiple imputation approach. The primary analysis corrected for missing outcomes yielded a similar hazard ratio of 0.45 (95% CI: 0.18–0.87), supporting the robustness of the findings. Two approaches were used to handle missing colonization outcomes. In the first, all neonates lost to follow-up were assumed not to have acquired concordant colonization. In the second, a more detailed multiple imputation strategy was implemented. This approach involved generating matched sets for each neonate with missing colonization status based on key baseline covariates, including treatment group, birthweight (within 1000 grams), gestational age (within 4 weeks), singleton versus non-singleton birth, and mode of delivery (vaginal vs. other). For each matched neonate, colonization status was imputed using time-at-risk information: if the matched neonate's time at risk was equal to or less than the observed time at risk, the colonization status was carried forward; if the matched time at risk was more than one week longer than the observed time, the status was imputed as no colonization; if the time difference was within one week, the matched colonization status was used. This imputation strategy was applied across 10 datasets, and hazard ratios were estimated from each and combined using Rubin’s rules. This structured two-step imputation process—combining covariate matching with time-at-risk adjustment—provided a transparent and reproducible approach for addressing missing outcomes in the primary analysis. Study-III Yamamura et al. 23 conducted a randomized, placebo-controlled trial evaluating the effect of an intervention on time to worsening of neurologic symptoms. The primary endpoint was time to protocol-defined relapse, assessed using a time-to-event analysis. A total of 83 participants were enrolled, with 41 assigned to the intervention group and 42 to the placebo group. The primary analysis yielded a hazard ratio (HR) of 0.38 (95% CI: 0.16–0.88), suggesting a reduced risk of relapse in the intervention group. Censoring due to discontinuation occurred in 3 participants (7%) in the intervention arm and 10 participants (24%) in the placebo arm. These participants were treated as censored in the primary analysis. Recognizing the potential influence of informative censoring, the authors conducted a series of sensitivity analyses using multiple imputation methods designed for time-to-event data. Four imputation models were implemented, each using 100 imputed datasets, and results were pooled using Rubin’s rules. The models differed based on the statistical method and reference approach used: Model I: Kaplan–Meier-based multiple imputation following the approach of Hsu and Taylor, yielding an HR of 0.34 (95% CI: 0.14–0.78). Model II: Cox proportional hazards model-based imputation using the method described by Jackson et al., with an HR of 0.37 (95% CI: 0.16–0.86). Model III: Kaplan–Meier-based imputation following Lipkovich et al., resulting in an HR of 0.44 (95% CI: 0.20–0.95). Model IV: Cox model-based multiple imputation using the framework of Lipkovich et al., with an HR of 0.35 (95% CI: 0.15–0.81). Each of these imputation strategies was implemented without variation in the number of imputations (n = 100), and the analyses consistently supported the treatment effect observed in the primary analysis. This study illustrates the application of multiple imputation tailored to survival outcomes, leveraging both Kaplan–Meier and Cox model-based approaches. The consistency of results across imputation methods underscores the robustness of the treatment effect and highlights the importance of sensitivity analyses in trials with differential dropout rates. Overall, despite variations in methodological transparency, all ten studies reported robust findings, suggesting that the use of multiple imputations did not introduce significant bias. However, best-case worst-case scenario was conducted as the sensitivity analysis for the trials. Trials utilizing MI methods did not incorporate a tipping point analysis to assess the robustness of imputed results. While tipping point approaches are commonly applied to continuous and categorical outcomes to evaluate the impact of missing data, their use in survival analysis remains limited. However, such methods can serve as a stress test for imputation, assessing the potential impact of worsening conditions on treatment outcomes. Sterne et al., 24 proposed six key recommendations for handling missing outcomes in RCTs using multiple imputation (MI). These include providing a complete description of the MI method, specifying the variables included in the imputation model, reporting the number of imputations, detailing the software package or procedure used, and presenting the pooled effect estimates using Rubin’s rule along with confidence intervals and p-values. However, seven out of ten studies fully adhered to these guidelines. Additionally, among the ten studies reviewed, only three assessed the proportional hazards assumption, while the remaining studies did not report sufficient information, raising concerns about their validity. 4. Discussions This scoping review evaluates how missing outcomes are handled and reported in time-to-event analyses. Over the past five years (2020–2023), only 10 studies applied MI techniques to address missing data in survival analyses. Among these, MICE was the most commonly used approach (4 studies). While MICE can be extended to survival outcomes, it does not inherently account for censoring time distributions, making its direct application to time-to-event data methodologically challenging. Notably, only one study employed MI based on the event time distribution, while 4 out of 10 used the MICE package for imputing missing outcomes. Given that event time distributions are typically skewed, MICE methods may not be appropriate. Parametric approaches based on lifetime distributions offer a more reliable alternative, as they not only better account for the distributional characteristics of event times but also incorporate uncertainty into the imputation process for the RCTs. Many research articles that employed MI for missing outcomes in time-to-event analyses did not mention the software used for imputations. In the PubMed search, we aimed to retrieve all relevant articles from high-impact journals published over the past five years. Additionally, for verification, we cross-checked methodological papers along with their cited references. No additional RCT articles were identified beyond those retrieved through the initial search. 4.1. Key findings & challenges Several interesting patterns were found in this review. Many studies aimed to minimize missing data to preserve statistical power and reduce bias. However, 90 studies did not provide clear information on the nature of censoring—whether it was informative, independent, or random. While drug-based clinical trials frequently applied multiple imputations to test the robustness of their results, the sensitivity of these analyses remains questionable due to the lack of explicit methodological details. In many cases, studies acknowledged missing outcomes and treated dropouts or discontinuations as censored observations. Some trials attempted to maintain statistical power by designing studies with higher power (90%) and ensuring the missing data proportion remained minimal. However, these articles did not explicitly state whether censoring assumptions—non-informative, independent, or random—were upheld. Given the increasing reliance on multiple imputations in clinical research, more rigorous methodological frameworks are needed to ensure accurate and reliable effect estimates. This review further highlights that multiple imputation procedures are still underutilized or insufficiently documented in survival analysis. While best-case and worst-case scenario analyses are commonly used to assess the impact of missing data, they do not provide actual effect estimates. To improve methodological transparency, clear documentation of multiple imputation techniques is essential, including details on the specific MI methods used, the covariates included in the imputation model, and the handling of missing data. Our findings align with those of previous reviews on missing data in survival analysis 12 – 15 , 18 , reinforcing the need for standardized reporting practices. 4.2. Limitations This review has several limitations. First, it was restricted to a PubMed search, as this database includes articles from top-tier medical journals, ensuring comprehensive coverage of high-impact research. Although some additional journals were randomly reviewed, many did not apply appropriate MI methods for handling missing survival outcomes. Therefore, the PubMed search was considered sufficient for this study. Second, this review focused exclusively on primary survival outcomes, without examining secondary endpoints, which may limit the generalizability of findings. Future research should extend beyond RCTs to explore how missing survival outcomes are handled in other study designs, such as observational studies and registry-based analyses. Lastly, while this review provides an overview of how missing data are managed in time-to-event analyses, further investigations are needed to develop structured guidelines for addressing missing survival outcomes in clinical research. 4.3. Future recommendations Researchers need to give an explicit and detailed explanation of censoring data. If missing outcomes are dealt with by imputation, specialized methods for time-to-event data are required. For example, the event time distribution needs to be properly included in the imputation model for reliable effect estimation. In addition, the proportional hazards assumption needs to be investigated and reported in RCTs, especially when missing outcomes exist, for the validation of results. In applying MI for time-to-event missing outcomes, researchers should follow the similar procedures by Tan et al., 15 to time-to-event such as reporting the censoring mechanism, model variables that were used in imputation, number of imputations, Rubin's pooled estimates, confidence intervals, and p-values. Sensitivity analyses should be conducted to evaluate MI results under varying censoring assumptions, that is, both CAR and CNAR. To improve consistency and transparency in managing missing outcomes, a universal checklist for the reporting of the imputation process should be employed. Methodological researchers should develop comprehensive practical guidelines on implementing MI for missing outcomes in time-to-event analyses across various statistical software. Providing clear documentation and software-specific workflows would facilitate the adoption of appropriate MI procedures, enabling researchers to effectively integrate these methods into their analyses. Future manuscripts on research studies should prioritize methodological description clarity and accuracy, employ standardized statistical and clinical methods, organize information in a logical sequence, and cite seminal literature to justify methodological decisions. In-depth sensitivity analyses are recommended to be performed, considering varying missing data assumptions and explaining the chosen imputation method. Finally, reporting transparency should be given precedence by adhering to guidelines to make survival analysis and missing outcomes management reproducible and reliable. 5. Conclusion In RCTs with survival outcomes, proper handling of missing event times is crucial, particularly in the presence of censoring. This scoping review highlights that despite the availability of rigorous statistical methods, the handling of missing event times remains underutilized and often inadequately reported. Many studies acknowledge censoring but fail to specify whether it is informative or non-informative, and the reporting of multiple imputation techniques is often insufficiently documented. These findings underscore a critical gap in survival analysis reporting, emphasizing the need for structured guidelines on the transparent handling and reporting of missing outcomes in time-to-event studies. Addressing these gaps will enhance the reliability and reproducibility of survival analyses in RCTs. Declarations Ethics approval and consent to participate Not applicable Consent for publication Not applicable Availability of data and materials The data supporting the findings of this study are available from the corresponding author upon reasonable request. Competing interests The authors declare no competing interests Funding This research received no specific grant from any funding agency Author’s contributions S.K. and P.S. conceptualized and designed the study. S.K. performed the data extraction, and P.S. conducted a pilot review of the extracted data and provided feedback. S.K. drafted the initial version of the manuscript. V.R. and P.S. critically reviewed and revised the draft. All authors read and approved the final manuscript. Acknowledgements Not applicable References Akobeng AK. Understanding randomised controlled trials. Arch Dis Child . 2005;90(8):840-844. doi:10.1136/ADC.2004.058222 O’Neill RT, Temple R. The prevention and treatment of missing data in clinical trials: an FDA perspective on the importance of dealing with it. Clin Pharmacol Ther . 2012;91(3):550-554. doi:10.1038/CLPT.2011.340 Wang S, Frederich R, Mancuso JP. Imputation of Missing Data for Time-to-Event Endpoints Using Retrieved Dropouts. Ther Innov Regul Sci . 2024;58(1):114-126. doi:10.1007/S43441-023-00575-5/TABLES/3 Rubin DB. Inference and missing data. Biometrika . 1976;63(3):581-592. doi:10.1093/BIOMET/63.3.581 Carpenter JR., Bartlett Johnathon, Morris Tim, Wood Angela, Quartagno Matteo, Kenward MG. Multiple imputation and its application. Published online 2023:440. Lipkovich I, Ratitch B, O’Kelly M. Sensitivity to censored-at-random assumption in the analysis of time-to-event endpoints. Pharm Stat . 2016;15(3):216-229. doi:10.1002/PST.1738 Groenwold RHH, Moons KGM, Vandenbroucke JP. Randomized trials with missing outcome data: how to analyze and what to report. CMAJ : Canadian Medical Association Journal . 2014;186(15):1153. doi:10.1503/CMAJ.131353 Taylor JMG, Murray S, Hsu CH. Survival estimation and testing via multiple imputation. Stat Probab Lett . 2002;58:221-232. Zhao Y, Herring AH, Zhou H, Ali MW, Koch GG. A multiple imputation method for sensitivity analyses of time-to-event data with possibly informative censoring. J Biopharm Stat . 2014;24(2):229-253. doi:10.1080/10543406.2013.860769 Jackson D, White IR, Seaman S, Evans H, Baisley K, Carpenter J. Relaxing the independent censoring assumption in the Cox proportional hazards model using multiple imputation. Stat Med . 2014;33(27):4681-4694. doi:10.1002/SIM.6274 Atkinson A, Kenward MG, Clayton T, Carpenter JR. Reference-based sensitivity analysis for time-to-event data. Pharm Stat . 2019;18(6):645-658. doi:10.1002/PST.1954 Bell ML, Fiero M, Horton NJ, Hsu CH. Handling missing data in RCTs; A review of the top medical journals. BMC Med Res Methodol . 2014;14(1):1-8. doi:10.1186/1471-2288-14-118/TABLES/4 Carroll OU, Morris TP, Keogh RH. How are missing data in covariates handled in observational time-to-event studies in oncology? A systematic review. BMC Med Res Methodol . 2020;20(1):1-15. doi:10.1186/S12874-020-01018-7/TABLES/4 Goldkuhle M, Hirsch C, Iannizzi C, et al. Meta-epidemiological review identified variable reporting and handling of time-to-event analyses in publications of trials included in meta-analyses of systematic reviews. J Clin Epidemiol . 2023;159:174-189. doi:10.1016/J.JCLINEPI.2023.05.023 Tan PT, Cro S, Van Vogt E, Szigeti M, Cornelius VR. A review of the use of controlled multiple imputation in randomised controlled trials with missing outcome data. BMC Med Res Methodol . 2021;21(1). doi:10.1186/S12874-021-01261-6 Goldkuhle M, Hirsch C, Iannizzi C, et al. Meta-epidemiological review identified variable reporting and handling of time-to-event analyses in publications of trials included in meta-analyses of systematic reviews. J Clin Epidemiol . 2023;159:174-189. doi:10.1016/J.JCLINEPI.2023.05.023 Rubin DB. Multiple Imputation for Nonresponse in Surveys. Published online June 9, 1987. doi:10.1002/9780470316696 Sullivan TR, Yelland LN, Lee KJ, Ryan P, Salter AB. Treatment of missing data in follow-up studies of randomised controlled trials: A systematic review of the literature. https://doi.org/101177/1740774517703319 . 2017;14(4):387-395. doi:10.1177/1740774517703319 Karuppasamy S, Samuel P. A Scoping Review Protocol for Handling Missing Outcomes in Time-to-Event Analyses in Randomised Controlled Trials. osf.io/6xz95. Published April 9, 2025. R Core Team (2023). R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. Pereira NL, Farkouh ME, So D, et al. Effect of Genotype-Guided Oral P2Y12 Inhibitor Selection vs Conventional Clopidogrel Therapy on Ischemic Outcomes After Percutaneous Coronary Intervention: The TAILOR-PCI Randomized Clinical Trial. JAMA . 2020;324(8):761-771. doi:10.1001/JAMA.2020.12443 Milstone AM, Voskertchian A, Koontz DW, et al. Effect of Treating Parents Colonized With Staphylococcus aureus on Transmission to Neonates in the Intensive Care Unit: A Randomized Clinical Trial. JAMA . 2020;323(4):319-328. doi:10.1001/JAMA.2019.20785 Yamamura T, Kleiter I, Fujihara K, et al. Trial of Satralizumab in Neuromyelitis Optica Spectrum Disorder. New England Journal of Medicine . 2019;381(22):2114-2124. doi:10.1056/nejmoa1901747 Sterne JAC, White IR, Carlin JB, et al. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ . 2009;338(7713):157-160. doi:10.1136/BMJ.B2393 Additional Declarations No competing interests reported. Supplementary Files Appendix.docx Cite Share Download PDF Status: Published Journal Publication published 28 Sep, 2025 Read the published version in BMC Medical Research Methodology → Version 1 posted Editorial decision: Revision requested 14 Jul, 2025 Reviews received at journal 09 Jul, 2025 Reviews received at journal 01 Jul, 2025 Reviewers agreed at journal 20 Jun, 2025 Reviewers agreed at journal 11 Jun, 2025 Reviewers invited by journal 11 Jun, 2025 Editor assigned by journal 11 Jun, 2025 Editor invited by journal 09 Jun, 2025 Submission checks completed at journal 09 Jun, 2025 First submitted to journal 09 Jun, 2025 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-6801188","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":470184387,"identity":"ff7a2576-e339-49fd-b38b-d77a3921d53b","order_by":0,"name":"Saravanaraj Karuppasamy","email":"","orcid":"","institution":"Christian Medical College","correspondingAuthor":false,"prefix":"","firstName":"Saravanaraj","middleName":"","lastName":"Karuppasamy","suffix":""},{"id":470184388,"identity":"638e5b77-baac-4c01-a6ab-541f6cfb635a","order_by":1,"name":"Prasanna Samuel Premkumar","email":"data:image/png;base64,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","orcid":"","institution":"Christian Medical College","correspondingAuthor":true,"prefix":"","firstName":"Prasanna","middleName":"Samuel","lastName":"Premkumar","suffix":""},{"id":470184389,"identity":"c18371dc-b87b-4954-9092-848264026aa2","order_by":2,"name":"Venkata Raghava Mohan","email":"","orcid":"","institution":"Christian Medical College","correspondingAuthor":false,"prefix":"","firstName":"Venkata","middleName":"Raghava","lastName":"Mohan","suffix":""}],"badges":[],"createdAt":"2025-06-02 10:23:25","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6801188/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6801188/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12874-025-02676-1","type":"published","date":"2025-09-29T00:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":92601237,"identity":"2044f2e0-a0bb-446d-86da-d418d1dcb9ef","added_by":"auto","created_at":"2025-10-01 14:36:35","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":755692,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6801188/v1/928638f2-52d7-428d-b5aa-9dacbb1dd1ba.pdf"},{"id":84567164,"identity":"6308b25e-321b-407a-a9a7-5fcb82056d36","added_by":"auto","created_at":"2025-06-13 14:30:56","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":196182,"visible":true,"origin":"","legend":"","description":"","filename":"Appendix.docx","url":"https://assets-eu.researchsquare.com/files/rs-6801188/v1/6e6343cda56a5a2b2c06bd62.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Handling Missing Outcomes in Time-to-Event Analyses: A Scoping Review of Multiple Imputation in Randomised Controlled Trials","fulltext":[{"header":"What is new?","content":"\u003cp\u003e\u003cstrong\u003eKey findings\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRandomised Controlled Trials with time-to-event outcomes often lack clear reporting on early dropout-related missing outcomes, particularly whether censoring is informative or non-informative, even in high-impact journals. Additionally, when multiple imputation is used for handling missing outcomes, details on its implementation in time-to-event analyses are frequently underreported.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eWhat this adds what is known?\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePrevious literature reviews have given limited attention to missing outcomes in time-to-event analyses, primarily focusing on general reporting structures and the use of multiple imputation in RCTs without specifically addressing survival data. Furthermore, studies assessing the reporting of time-to-event outcomes in trial publications have largely overlooked how missing outcomes are handled. This review provides a comprehensive evaluation of the reporting and handling of missing outcomes in time-to-event analyses within high-impact journals.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eWhat are the implications and what should be changed?\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eResearchers should explicitly assess whether early dropout, considered as a missing outcome, results in informative or non-informative censoring by examining its association with relevant covariates. If censoring is determined to be non-informative, this should be supported with appropriate data reporting. In cases where the censoring mechanism remains uncertain, it is advisable to assume informative censoring and conduct sensitivity analyses to assess the robustness of the findings. Furthermore, methodological experts should establish standardized guidelines for handling missing time-to-event outcome data in randomized controlled trials, incorporating software-specific recommendations and structured analytical workflows.\u003c/p\u003e"},{"header":"1. Background","content":"\u003cp\u003eRandomized Controlled Trials (RCTs) are considered the gold standard for evaluating treatment effects in clinical research because RCTs minimize the bias and are closer to the true effects compared to the other research methods \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. However, several challenges can lead to inconsistent results, including unmeasured confounders, inappropriate analysis plans, and missing data. Among these, missing data pose a particularly serious threat to the validity of findings \u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. Even when the analyses are conducted under appropriate assumptions, missing values in outcomes or covariates can introduce bias and reduce the reliability of results.\u003c/p\u003e \u003cp\u003eIn RCTs, outcomes should be tracked throughout the study. However, complete follow-up is often not possible in real-world settings. Missing data might have occurred in both continuous and categorical outcomes, but time-to-event outcomes pose unique issues. The major emphasis of time-to-event analysis is the time until an event of interest (i.e., death, relapse, failure, etc.,) occurs within a specified period. Some participants may not experience the event until the specific time point and are considered censored not missing; this situation is referred to as administrative censoring. In some cases, participants may decline to return to the follow-up for various reasons, but in RCTs, it should be considered important because the intervention involves to the participants and this is considered as missing event time also categorized as non-administrative censoring. If such non-administrative censoring is not well handled, it can significantly impair the validity of time-to-event analysis \u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. For instance, in a hypothetical study where the intervention group experiences twice the censoring rate of the placebo group\u0026mdash;potentially due to treatment toxicity or lack of efficacy\u0026mdash;the results may be unreliable. Since the exact event times remain unknown, the validity of the findings could be questioned.\u003c/p\u003e \u003cp\u003eKaplan-Meier (KM) survival estimates and Cox regression models are the most popular methods for time-to-event analysis (also known as survival analysis) which is based on the assumption that censoring is independent of the event time. Missing outcomes cannot always be regarded as independent in time-to-event analysis, necessitating procedures that account for uncertainty and include statistical models with sensitivity analysis.\u003c/p\u003e \u003cp\u003eRubin categorizes missing data mechanisms into three major types: Missing Completely at Random (MCAR), where the probability of missingness is unrelated to any observed data; Missing at Random (MAR), where the probability of missingness conditionally depends on the observed data; and Missing Not at Random (MNAR), where the probability of missingness is influenced by unobserved data. These classifications give a framework for selecting appropriate imputation algorithms \u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. Similarly, the censoring mechanisms can be categorized as Censoring Completely at Random (CCAR), Censored at Random (CAR), Censoring Not at Random (CNAR) \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. In survival analysis, the validity of results depends on the assumption that censoring is not related to the event time.\u003c/p\u003e \u003cp\u003eCCAR means that when censoring occurs for reasons unrelated to the event\u0026mdash;such as a participant reaching the endpoint of the study period without experiencing the event also known as administrative censoring or withdrawing due to factors like relocation or seeking care elsewhere\u0026mdash;the censoring is considered random and non-informative. In such cases, survival estimates remain unbiased, ensuring reliable conclusions about treatment effects in clinical trials. Typically, a predefined endpoint would be a good illustration of this kind of mechanism.\u003c/p\u003e \u003cp\u003eThe censoring mechanism is under CAR if, conditional on treatment exposure and other observed covariates in the survival model, the probability of censoring is independent of the event time. In contrast, CNAR is an informative censoring in which the probability of censoring depends on the event time which is closely related to non-administrative censoring, even after considering for covariates. Unlike CCAR and CAR, which are regarded as partially non-informative and can be safely ignored in survival analyses, CNAR requires additional attention even though the event times are imputed using rigorous statistical tools still sensitivity analysis is also needed to check the robustness of the results \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. For instance, in an oncology trial, if patients with more severe disease are more likely to withdraw due to disease progression, their censoring times would be systematically related to their event times, making the censoring informative (CNAR) and this kind of situation we have to address the issue by considering departure from CAR assumptions.\u003c/p\u003e \u003cp\u003eMultiple Imputation (MI) and Inverse Probability Weighting (IPW) both are flexible methods for handling missing outcomes in RCTs \u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. However, the primary recommended approach for addressing missing outcomes is MI under the MAR assumption. There are significant contributions to the developing MI methods for continuous and binary variables and comparatively less attention given to the time-to-event data. Nevertheless, over the past decades, a few researchers have contributed to the development of MI methodologies for time-to-event outcomes. Notably, Taylor et al. \u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e proposed a nonparametric multiple imputation approach for imputing missing event times in censored data, employing techniques such as risk set imputation and KM imputation. In risk set imputation, censored times are replaced by random draws from observed event times among individuals still at risk after the censoring time. In contrast, KM imputation draws from the estimated distribution of event times for those at risk post-censoring. These methods aim to reproduce the KM estimator, thereby facilitating survival estimation and hypothesis testing through multiple imputed datasets.\u003c/p\u003e \u003cp\u003eBuilding on this, Zhao et al. \u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e proposed a multiple imputation method for sensitivity analyses of time-to-event data with possibly informative censoring. Their approach imputes missing event times based on the failure time distribution conditional on the follow-up discontinuation time. This method incorporates different assumptions regarding post-discontinuation event occurrences through a hazard ratio parameter, enabling the analysis of multiple imputed datasets using standard survival analysis techniques, with results combined via Rubin\u0026rsquo;s rules.\u003c/p\u003e \u003cp\u003eJackson et al. \u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e extended the field by proposing a bootstrap-based multiple imputation approach for handling non-independent censoring. This method imputes censored observations by randomly selecting one of the observed event times using a step function. However, as this approach does not provide full estimation and relies on the distribution of observed events, it may not be suitable for datasets with very few events.\u003c/p\u003e \u003cp\u003eLipkovich et al. \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e developed a sensitivity analysis framework for non-ignorable missing data, applying a tipping point approach under a censored-at-random assumption within a multiple imputation framework. Their methodology incorporates a covariate-adjusted piecewise exponential model and compares parametric, semi-parametric, and non-parametric imputation models via simulations. Their findings suggest that parametric imputation models generally exhibit lower bias compared to other methods, particularly when the piecewise exponential-based multiple imputation approach is applied.\u003c/p\u003e \u003cp\u003eAtkinson et al. \u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e introduced a reference-based sensitivity analysis approach for time-to-event data by integrating multiple imputation with a Weibull proportional hazards model. This method enables the assessment of treatment effects under various assumptions about the missing data mechanism, offering a structured framework for evaluating potential biases.\u003c/p\u003e \u003cp\u003eMore recently, Wang et al. \u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e proposed a retrieved-dropout-based multiple imputation approach for time-to-event data. This method constructs imputation models using data from participants who discontinued study treatment but remained in follow-up, addressing the treatment policy estimand and providing practical implementation guidelines.\u003c/p\u003e \u003cp\u003eThe aforementioned approaches are based on non-parametric, semi-parametric, and parametric multiple imputation methods for time-to-event data, incorporating both simulation studies and real-world datasets. Additionally, sensitivity analyses were conducted to assess the impact of departures from CAR to CNAR scenarios.\u003c/p\u003e \u003cp\u003eDespite these advances in methodological development in missing outcomes handling in time-to-event data, prior literature studies have found that missing outcomes are frequently handled incorrectly, with Complete Case Analysis being the most prevalent technique \u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e in RCTs. O. U. Corrol et al. \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e did a systematic analysis on time-to-event missing variables and discovered that 53% of the 148 evaluated publications employed Complete Case Analysis, whereas 22% of the missing outcomes were reported unclearly in oncology research. This review recommended that the Substantive Model Compatible Fully Conditional Specification (SMC-FCS) as the gold standard for MI in time-to-event data, however, this recommendation was limited to oncology research for the time-to-event covariates. More recently, Marius et al.\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e reviewed the reporting and handling of time-to-event outcomes in trials included in meta-analyses of systematic review. The study examined 235 trials, finding that half of them did not adequately report missing outcomes. Additionally, they reported that 57% of the trials had missing outcomes, which was most frequently handled by exclusion from the analysis.\u003c/p\u003e \u003cp\u003eGiven the advancements and existing gaps in the field, a thorough evaluation of MI methodologies employed in RCTs for survival outcomes\u0026mdash;regardless of illness or treatment\u0026mdash;is required. Additionally, while several literature reviews have been published, most have focused on missing outcomes in continuous and categorical data, with limited attention given to the methodologies used in published RCTs for time-to-event analyses or, in some cases, not addressing them at all \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. This scoping review aims to systematically examine how missing outcomes in time-to-event studies have been addressed in the top three high-impact medical journals\u0026mdash;The NEJM, The Lancet, and The Journal of the American Medical Association (JAMA)\u0026mdash;over the past five years (2019\u0026ndash;2023). If MI procedures were used to assess time-to-event outcomes, the review would look at how they were implemented and reported. Furthermore, this article investigates the theoretical discussion of existing approaches for imputing censored event timings under various censoring assumptions using Multiple Imputation technique for both parametric and non-parametric approach.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Multiple Imputation (MI)\u003c/h2\u003e \u003cp\u003eMI fits a statistical model to the observed data to generate estimates that account for uncertainty. It is a more robust strategy under the MAR assumption since it accommodates uncertainty. To incorporate uncertainty, MI imputes missing values multiple times, generating several completed datasets and estimating parameters from each.\u003c/p\u003e \u003cp\u003eGenerally, let \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:X\\)\u003c/span\u003e\u003c/span\u003e be the data matrix, where \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{X}_{O}\\)\u003c/span\u003e\u003c/span\u003e represents observed values and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{X}_{M}\\)\u003c/span\u003e\u003c/span\u003e denotes missing values. The multiple imputation procedure is \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eFor \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:k=\\text{1,2},3,\\dots\\:,\\:K\\)\u003c/span\u003e\u003c/span\u003e impute the missing values from the posterior predictive distribution \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:f\\left({X}_{M}|{X}_{O}\\right)\\)\u003c/span\u003e\u003c/span\u003e given the observed values, generating K completed datasets.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eFit the substantive model to each imputed dataset, and compute the K parameter estimates and their variances.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eCombine these estimates using Rubin\u0026rsquo;s rules to obtain pooled parameter estimates and their associated variances.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eWhile MI is typically implemented under the MAR assumption, it can also be extended to MNAR scenarios. If the missing data assumption is met correctly and the imputation model is specified correctly then, MI produces asymptotically efficient parameter estimates with nominal coverage \u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. Among various MI methods, multiple imputation by chained equations (MICE) has been widely used for handling missing data in RCTs \u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. However, its suitability for time-to-event outcomes remains still questionable. To address this, Carpenter et al.,\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e discussed both parametric and non-parametric MI approaches for survival outcomes.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. Impute the censored event time\u003c/h2\u003e \u003cp\u003eIn time-to-event analysis, censored times are often treated as missing outcomes. When censoring is non-informative, the CCAR assumption, combined with appropriate statistical methods, provides reliable estimates. However, if censoring is informative, it may still be assumed to follow the CAR assumption. In such cases, MI can be a valuable tool for imputing censored event times, given that other covariates are available. The imputation process should be based on either the marginal survival distribution or a survival regression model that incorporates relevant covariates, though not necessarily all observed variables \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eTwo multiple imputation methods have been proposed for handling censored event times in survival outcomes under the CAR assumption: a parametric approach and a non-parametric approach \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. Both methods involve generating multiple imputed datasets and combining the resulting estimates to account for uncertainty.\u003c/p\u003e \u003cp\u003eLet\u0026rsquo;s consider the survivor function \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\widehat{S}\\left({c}_{i}\\right|\\widehat{\\beta\\:},\\:{X}_{1},\\:{X}_{2},\\dots\\:,\\:{X}_{p})\\)\u003c/span\u003e\u003c/span\u003e for the \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{i}^{th}\\)\u003c/span\u003e\u003c/span\u003e subject and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\)\u003c/span\u003e\u003c/span\u003e covariate values observed at and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\widehat{\\beta\\:}\\)\u003c/span\u003e\u003c/span\u003e is a vector of model parameters sampled from a Bayesian posterior distribution estimated from the observed data. The imputed event time \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{t}_{i}^{\\text{*}}\\)\u003c/span\u003e\u003c/span\u003e for the \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{i}^{th}\\)\u003c/span\u003e\u003c/span\u003e subject is then computed as follows:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eCompute \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{p}_{i}=1-\\widehat{S}\\left({c}_{i}\\right|\\widehat{\\beta\\:},\\:{X}_{1},\\:{X}_{2},\\dots\\:,\\:{X}_{p})\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eDraw a uniform random value \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{u}_{i}\\sim[{p}_{i},1]\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eImpute the event time \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{t}_{i}^{\\text{*}}\\)\u003c/span\u003e\u003c/span\u003eas the solution of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{u}_{i}=1-\\widehat{S}\\left({t}_{i}\\right|{x}_{i},\\widehat{\\beta\\:})\\)\u003c/span\u003e\u003c/span\u003e. This will ensure that the event time is greater than the censoring time.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eTherefore, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{u}_{i}=\\widehat{F}\\left({t}_{i}\\right|{x}_{i},\\widehat{\\beta\\:})\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{t}_{i}^{\\text{*}}={\\widehat{F}}^{-1}\\left({u}_{i}\\right|{x}_{i},\\widehat{\\beta\\:})\\)\u003c/span\u003e\u003c/span\u003e. Thus, imputed time can be found from the inverse of the failure function.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eThis approach is then repeated \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:K\\)\u003c/span\u003e\u003c/span\u003e times to generate multiple completed datasets. Each completed dataset is analyzed using appropriate statistical methods (most often Cox regression), and the resulting parameter estimates are combined using Rubin\u0026rsquo;s rules to account for variability across imputations. This ensures valid statistical inference by incorporating both within-imputation and between-imputation uncertainty.\u003c/p\u003e \u003cp\u003eUsing this approach, Lipkovich et al. \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e developed a distribution-based imputation method that incorporates a stepwise exponential distribution under the CAR assumption. They also conducted sensitivity analyses to assess the robustness of imputation results compared to existing methods using the stress test tipping point analysis.\u003c/p\u003e \u003cp\u003eThe non-parametric approach is based on the Bayesian bootstrap method, which involves forming a subgroup of the fully observed dataset as a donor pool. Missing values are then imputed by drawing replacements from this donor pool with equal probability. This process is repeated \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:K\\)\u003c/span\u003e\u003c/span\u003e times to generate multiple imputed datasets, which are subsequently analyzed and pooled to obtain final estimates by using Rubin\u0026rsquo;s rules. This method was proposed by Rubin and Schenker (1986).\u003c/p\u003e \u003cp\u003eUnlike parametric methods, which rely on statistical models to estimate and impute missing data from observed values, the non-parametric approach does not assume an underlying distribution. This makes it particularly useful for large datasets, such as census data, where parametric models can be complex and computationally intensive.\u003c/p\u003e \u003cp\u003eAnother approach, developed by Hsu, Taylor, and Murray, imputes event times based on auxiliary variables under the CAR assumption. Their method involves creating a bootstrap dataset, identifying the nearest neighbors based on distance metrics, and imputing the missing event times using either the Kaplan-Meier estimator or a proportional hazards model. While these approaches are robust in large samples, they may introduce bias in the parametric approach due to neighborhood censoring and the potential dependence of event times \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3. Search strategy\u003c/h2\u003e \u003cp\u003eTo assess how missing outcomes in survival analysis are reported in RCTs and how the MI methods are used to handle them, we focused on clinical trials published in high-impact journals. Specifically, we selected The Lancet, The New England Journal of Medicine (NEJM), and JAMA as our primary sources.\u003c/p\u003e \u003cp\u003eRelevant articles were identified through PubMed using a predefined set of search terms. The search strategy and data extraction procedures were outlined in the study protocol and registered in the open science framework\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. Our analysis included RCTs published between 2019 and 2023, ensuring a comprehensive review of recent practices in handling missing survival outcomes.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4. Inclusion criteria\u003c/h2\u003e \u003cp\u003eStudies were included if they:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eEmployed RCT designs (cluster or parallel) with no blinding, single blinding, or double blinding.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eConducted survival analysis using Kaplan-Meier estimation or Cox proportional hazards regression as part of the primary outcome assessment.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eExplicitly reported methods for handling loss to follow-up (early discontinuation) in the Methods section or supplementary.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eInvolved either primary data collection or secondary data analysis.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5. Exclusion Criteria\u003c/h2\u003e \u003cp\u003eStudies were excluded if they:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eWere study protocols.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eFocused on primary outcomes other than survival analysis.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eWere meta-analyses.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eInvolved in non-clinical trial designs (e.g., cohort studies, retrospective studies, cost-effectiveness analyses).\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6. Study selection\u003c/h2\u003e \u003cp\u003eThe study selection process began with extracting research articles from PubMed using predefined search terms provided in the protocol, with the results downloaded as a CSV file. A prevalidated questionnaire was prepared in an Excel sheet, and each abstract underwent an initial review to determine whether the study was an RCT and whether survival analysis was included as a primary outcome.\u003c/p\u003e \u003cp\u003eThe first stage of reviewing studies meets the above mentioned inclusion criteria and then proceeds to a second-stage review to assess the presence of missing outcomes. If no missing outcomes were identified, the review moved on to the next paper. If missing outcomes were present, the paper underwent a detailed investigation to quantify the percentage of dropouts or missing data and to evaluate how the data were handled.\u003c/p\u003e \u003cp\u003eFor studies that utilized multiple imputation, supplementary or supporting documents were reviewed, and reference papers detailing the imputation methods were documented.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.7. Statistical methods\u003c/h2\u003e \u003cp\u003eWe reported the frequency and percentage of the study characteristics, with results presented in tables. If multiple imputation was used, the methodology for handling missing data was described in detail. All results were recorded in Microsoft Excel, and tabulation and analysis were performed using R version 4.3.1\u003csup\u003e20\u003c/sup\u003e. The findings were summarized using frequency tables and percentages, and the identified methods were described in detail.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.1. Study selection\u003c/h2\u003e \u003cp\u003eA total of 694 articles were initially identified through a PubMed search. During the first screening, titles and abstracts were reviewed, and 321 articles were selected for further evaluation. A total of 373 articles were excluded due to study design, type, or lack of a primary focus on survival outcomes. These excluded articles were based on cohort studies, cost-effectiveness analyses, population surveys, systematic reviews, meta-analyses, and methodology papers (329 articles). An additional 44 articles were excluded because survival analysis was not the primary outcome; these articles primarily included studies that considered competing risks, used binary outcomes or treated survival outcomes as secondary (Figure-1).\u003c/p\u003e \u003cp\u003eThe 321 included articles were RCTs, consisting of parallel or cluster designs, as well as pooled RCT analyses. In these studies, survival analysis\u0026mdash;defined as time-to-event analysis\u0026mdash;was the primary focus, examining key events such as death, progression-free survival, relapse, or clinical failure, etc.,\u003c/p\u003e \u003cp\u003eAmong the 321 studies, 297 (92.2%) reported no missing outcomes, indicating no dropouts or loss to follow-up, with some cases classified as censored. Studies that reported censoring had missing outcomes due to discontinuations, dropouts, or loss to follow-up, with rates below 10%, considering these cases as censored. Seven studies (2.2%) excluded missing outcomes from their time-to-event analyses but used complete data for primary outcome analyses. The remaining 17 studies (5.3%) underwent a detailed review, as they applied additional methods such as multiple imputations or other approaches for handling missing outcomes in time-to-event analyses (Table\u0026nbsp;1).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.2. Study characteristics\u003c/h2\u003e \u003cp\u003eThe 17 articles included in a detailed review of methodology: The NEJM (7 articles), The Lancet (4 articles), and JAMA (6 articles). Notably, five articles each were published in 2020 and 2021, while no survival-related studies discussing missing outcomes were identified in 2022.\u003c/p\u003e \u003cp\u003eThe majority of the articles focused on the following conditions: cardiovascular diseases (5 articles), cancer (3 articles), infectious diseases (2 articles), pediatric conditions (2 articles), and other conditions (5 articles). Most studies were drug-based RCTs addressing survival outcomes (14 articles), while the remaining focused on health service strategies aimed at improving patient outcomes.\u003c/p\u003e \u003cp\u003eThe primary outcome in 4 articles was mortality over a prespecified time period, while cardiovascular outcomes were the primary endpoint in another 4 studies. Efficacy, safety, and clinical outcomes were the primary endpoints for the remaining 9 articles.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.3. Reporting and extent of missing data analysis\u003c/h2\u003e \u003cp\u003eAmong the 17 articles, the majority of RCTs included sample sizes between 100 and 499 participants (7 articles), while three studies enrolled more than 1,000 participants per arm. Regarding missing outcomes, 10 articles reported less than 10% missing data related to time-to-event outcomes, such as dropouts or treatment discontinuations. Six articles reported dropout rates exceeding 20% and applied missing data analysis approaches.\u003c/p\u003e \u003cp\u003eIn total, 10 articles discussed the use of MI-based methods to address missing outcomes. Six of these 17 conducted best-case and worst-case sensitivity analyses to assess the robustness of their findings, and one study applied the IPW method (Table \u0026minus;\u0026thinsp;2).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.4. Missing handling procedure using MI\u003c/h2\u003e \u003cp\u003eAmong the ten studies that applied MI to address missing outcomes given in the appendix, three did not provide a clear explanation of how the imputation was conducted (S7-S9). Four studies explicitly mentioned using the MICE method (S1,S2,S6 and S10); however, they did not specify how the imputation was performed for event status or event time, leaving key methodological details unclear. There are three articles used survival models (S3-S5).\u003c/p\u003e \u003cp\u003e \u003cb\u003eArticles with MI based survival models\u003c/b\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eStudy-I\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThe study conducted by Pereira et al.\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e aimed to evaluate the impact of a genotype-guided oral P2Y12 inhibitor strategy on ischemic outcomes among CYP2C19 loss-of-function (LOF) allele carriers following percutaneous coronary intervention (PCI). A total of 5,302 participants were enrolled, with 2,641 allocated to the genotype-guided therapy group and 2,635 to the conventional therapy group. The primary outcome was a composite of cardiovascular death, myocardial infarction, stroke, definite or probable stent thrombosis, and severe recurrent ischemia assessed at 12 months. Of the total population, 269 participants (5%) either withdrew prior to 12 months or were lost to follow-up. The primary analysis, which excluded participants with missing outcome data, yielded a hazard ratio (HR) of 0.66 (95% CI: 0.43\u0026ndash;1.02) with a p-value of 0.06, indicating no statistically significant difference between the groups.\u003c/p\u003e \u003cp\u003eAs part of the sensitivity analyses, multiple imputation was performed to address missing laboratory-based genotyping data. Imputation was conducted using a logistic regression model incorporating race/ethnicity, stratification factors, treatment arm, and Spartan genotyping results. Ten imputed datasets were generated, and the primary outcome analysis was repeated across these datasets. The resulting hazard ratio was 0.68 (95% CI: 0.45\u0026ndash;1.04), with the estimates combined using Rubin\u0026rsquo;s rules. However, the study did not specify whether the imputation approach extended to missing event times.\u003c/p\u003e \u003cp\u003e \u003cb\u003eStudy-II\u003c/b\u003e \u003c/p\u003e \u003cp\u003eMilstone et al.,\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e conducted a study to evaluate whether treating parents with intranasal mupirocin and topical chlorhexidine, as compared to placebo, could reduce the transmission of Staphylococcus aureus (S. aureus) from parents to neonates. The primary outcome was defined as neonatal acquisition of an S. aureus strain concordant with the parental strain identified during prerandomization screening (baseline colonization).\u003c/p\u003e \u003cp\u003eA total of 236 neonates were randomized: 117 to the active treatment group (chlorhexidine-impregnated cloths and intranasal mupirocin) and 119 to the placebo group (non-medicated soap cloths and petrolatum ointment). The unadjusted analysis showed a hazard ratio (HR) of 0.43 (95% CI: 0.16\u0026ndash;0.79) for concordant S. aureus colonization, and an HR of 0.57 (95% CI: 0.31\u0026ndash;0.88) for colonization with any S. aureus strain. The proportional hazards assumption was supported by the data (P\u0026thinsp;=\u0026thinsp;.79). A total of 18 neonates (7.6%) were lost to follow-up, including 12 from the intervention group and 6 from the control group.\u003c/p\u003e \u003cp\u003eTo address the potential bias introduced by missing outcome data, the investigators conducted sensitivity analyses, including a multiple imputation approach. The primary analysis corrected for missing outcomes yielded a similar hazard ratio of 0.45 (95% CI: 0.18\u0026ndash;0.87), supporting the robustness of the findings.\u003c/p\u003e \u003cp\u003eTwo approaches were used to handle missing colonization outcomes. In the first, all neonates lost to follow-up were assumed not to have acquired concordant colonization. In the second, a more detailed multiple imputation strategy was implemented. This approach involved generating matched sets for each neonate with missing colonization status based on key baseline covariates, including treatment group, birthweight (within 1000 grams), gestational age (within 4 weeks), singleton versus non-singleton birth, and mode of delivery (vaginal vs. other). For each matched neonate, colonization status was imputed using time-at-risk information: if the matched neonate's time at risk was equal to or less than the observed time at risk, the colonization status was carried forward; if the matched time at risk was more than one week longer than the observed time, the status was imputed as no colonization; if the time difference was within one week, the matched colonization status was used.\u003c/p\u003e \u003cp\u003eThis imputation strategy was applied across 10 datasets, and hazard ratios were estimated from each and combined using Rubin\u0026rsquo;s rules. This structured two-step imputation process\u0026mdash;combining covariate matching with time-at-risk adjustment\u0026mdash;provided a transparent and reproducible approach for addressing missing outcomes in the primary analysis.\u003c/p\u003e \u003cp\u003e \u003cb\u003eStudy-III\u003c/b\u003e \u003c/p\u003e \u003cp\u003eYamamura et al.\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e conducted a randomized, placebo-controlled trial evaluating the effect of an intervention on time to worsening of neurologic symptoms. The primary endpoint was time to protocol-defined relapse, assessed using a time-to-event analysis. A total of 83 participants were enrolled, with 41 assigned to the intervention group and 42 to the placebo group. The primary analysis yielded a hazard ratio (HR) of 0.38 (95% CI: 0.16\u0026ndash;0.88), suggesting a reduced risk of relapse in the intervention group.\u003c/p\u003e \u003cp\u003eCensoring due to discontinuation occurred in 3 participants (7%) in the intervention arm and 10 participants (24%) in the placebo arm. These participants were treated as censored in the primary analysis. Recognizing the potential influence of informative censoring, the authors conducted a series of sensitivity analyses using multiple imputation methods designed for time-to-event data.\u003c/p\u003e \u003cp\u003eFour imputation models were implemented, each using 100 imputed datasets, and results were pooled using Rubin\u0026rsquo;s rules. The models differed based on the statistical method and reference approach used: Model I: Kaplan\u0026ndash;Meier-based multiple imputation following the approach of Hsu and Taylor, yielding an HR of 0.34 (95% CI: 0.14\u0026ndash;0.78). Model II: Cox proportional hazards model-based imputation using the method described by Jackson et al., with an HR of 0.37 (95% CI: 0.16\u0026ndash;0.86). Model III: Kaplan\u0026ndash;Meier-based imputation following Lipkovich et al., resulting in an HR of 0.44 (95% CI: 0.20\u0026ndash;0.95). Model IV: Cox model-based multiple imputation using the framework of Lipkovich et al., with an HR of 0.35 (95% CI: 0.15\u0026ndash;0.81).\u003c/p\u003e \u003cp\u003eEach of these imputation strategies was implemented without variation in the number of imputations (n\u0026thinsp;=\u0026thinsp;100), and the analyses consistently supported the treatment effect observed in the primary analysis. This study illustrates the application of multiple imputation tailored to survival outcomes, leveraging both Kaplan\u0026ndash;Meier and Cox model-based approaches. The consistency of results across imputation methods underscores the robustness of the treatment effect and highlights the importance of sensitivity analyses in trials with differential dropout rates. Overall, despite variations in methodological transparency, all ten studies reported robust findings, suggesting that the use of multiple imputations did not introduce significant bias. However, best-case worst-case scenario was conducted as the sensitivity analysis for the trials.\u003c/p\u003e \u003cp\u003eTrials utilizing MI methods did not incorporate a tipping point analysis to assess the robustness of imputed results. While tipping point approaches are commonly applied to continuous and categorical outcomes to evaluate the impact of missing data, their use in survival analysis remains limited. However, such methods can serve as a stress test for imputation, assessing the potential impact of worsening conditions on treatment outcomes.\u003c/p\u003e \u003cp\u003eSterne et al.,\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e proposed six key recommendations for handling missing outcomes in RCTs using multiple imputation (MI). These include providing a complete description of the MI method, specifying the variables included in the imputation model, reporting the number of imputations, detailing the software package or procedure used, and presenting the pooled effect estimates using Rubin\u0026rsquo;s rule along with confidence intervals and p-values. However, seven out of ten studies fully adhered to these guidelines.\u003c/p\u003e \u003cp\u003eAdditionally, among the ten studies reviewed, only three assessed the proportional hazards assumption, while the remaining studies did not report sufficient information, raising concerns about their validity.\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussions","content":"\u003cp\u003eThis scoping review evaluates how missing outcomes are handled and reported in time-to-event analyses. Over the past five years (2020\u0026ndash;2023), only 10 studies applied MI techniques to address missing data in survival analyses. Among these, MICE was the most commonly used approach (4 studies). While MICE can be extended to survival outcomes, it does not inherently account for censoring time distributions, making its direct application to time-to-event data methodologically challenging. Notably, only one study employed MI based on the event time distribution, while 4 out of 10 used the MICE package for imputing missing outcomes. Given that event time distributions are typically skewed, MICE methods may not be appropriate. Parametric approaches based on lifetime distributions offer a more reliable alternative, as they not only better account for the distributional characteristics of event times but also incorporate uncertainty into the imputation process for the RCTs.\u003c/p\u003e \u003cp\u003eMany research articles that employed MI for missing outcomes in time-to-event analyses did not mention the software used for imputations.\u003c/p\u003e \u003cp\u003eIn the PubMed search, we aimed to retrieve all relevant articles from high-impact journals published over the past five years. Additionally, for verification, we cross-checked methodological papers along with their cited references. No additional RCT articles were identified beyond those retrieved through the initial search.\u003c/p\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e4.1. Key findings \u0026amp; challenges\u003c/h2\u003e \u003cp\u003eSeveral interesting patterns were found in this review. Many studies aimed to minimize missing data to preserve statistical power and reduce bias. However, 90 studies did not provide clear information on the nature of censoring\u0026mdash;whether it was informative, independent, or random. While drug-based clinical trials frequently applied multiple imputations to test the robustness of their results, the sensitivity of these analyses remains questionable due to the lack of explicit methodological details.\u003c/p\u003e \u003cp\u003eIn many cases, studies acknowledged missing outcomes and treated dropouts or discontinuations as censored observations. Some trials attempted to maintain statistical power by designing studies with higher power (90%) and ensuring the missing data proportion remained minimal. However, these articles did not explicitly state whether censoring assumptions\u0026mdash;non-informative, independent, or random\u0026mdash;were upheld. Given the increasing reliance on multiple imputations in clinical research, more rigorous methodological frameworks are needed to ensure accurate and reliable effect estimates.\u003c/p\u003e \u003cp\u003eThis review further highlights that multiple imputation procedures are still underutilized or insufficiently documented in survival analysis. While best-case and worst-case scenario analyses are commonly used to assess the impact of missing data, they do not provide actual effect estimates. To improve methodological transparency, clear documentation of multiple imputation techniques is essential, including details on the specific MI methods used, the covariates included in the imputation model, and the handling of missing data.\u003c/p\u003e \u003cp\u003eOur findings align with those of previous reviews on missing data in survival analysis\u003csup\u003e\u003cspan additionalcitationids=\"CR13 CR14\" citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e, reinforcing the need for standardized reporting practices.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e4.2. Limitations\u003c/h2\u003e \u003cp\u003eThis review has several limitations. First, it was restricted to a PubMed search, as this database includes articles from top-tier medical journals, ensuring comprehensive coverage of high-impact research. Although some additional journals were randomly reviewed, many did not apply appropriate MI methods for handling missing survival outcomes. Therefore, the PubMed search was considered sufficient for this study. Second, this review focused exclusively on primary survival outcomes, without examining secondary endpoints, which may limit the generalizability of findings. Future research should extend beyond RCTs to explore how missing survival outcomes are handled in other study designs, such as observational studies and registry-based analyses. Lastly, while this review provides an overview of how missing data are managed in time-to-event analyses, further investigations are needed to develop structured guidelines for addressing missing survival outcomes in clinical research.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e4.3. Future recommendations\u003c/h2\u003e \u003cp\u003eResearchers need to give an explicit and detailed explanation of censoring data. If missing outcomes are dealt with by imputation, specialized methods for time-to-event data are required. For example, the event time distribution needs to be properly included in the imputation model for reliable effect estimation. In addition, the proportional hazards assumption needs to be investigated and reported in RCTs, especially when missing outcomes exist, for the validation of results. In applying MI for time-to-event missing outcomes, researchers should follow the similar procedures by Tan et al., \u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e to time-to-event such as reporting the censoring mechanism, model variables that were used in imputation, number of imputations, Rubin's pooled estimates, confidence intervals, and p-values. Sensitivity analyses should be conducted to evaluate MI results under varying censoring assumptions, that is, both CAR and CNAR. To improve consistency and transparency in managing missing outcomes, a universal checklist for the reporting of the imputation process should be employed. Methodological researchers should develop comprehensive practical guidelines on implementing MI for missing outcomes in time-to-event analyses across various statistical software. Providing clear documentation and software-specific workflows would facilitate the adoption of appropriate MI procedures, enabling researchers to effectively integrate these methods into their analyses. Future manuscripts on research studies should prioritize methodological description clarity and accuracy, employ standardized statistical and clinical methods, organize information in a logical sequence, and cite seminal literature to justify methodological decisions. In-depth sensitivity analyses are recommended to be performed, considering varying missing data assumptions and explaining the chosen imputation method. Finally, reporting transparency should be given precedence by adhering to guidelines to make survival analysis and missing outcomes management reproducible and reliable.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eIn RCTs with survival outcomes, proper handling of missing event times is crucial, particularly in the presence of censoring. This scoping review highlights that despite the availability of rigorous statistical methods, the handling of missing event times remains underutilized and often inadequately reported. Many studies acknowledge censoring but fail to specify whether it is informative or non-informative, and the reporting of multiple imputation techniques is often insufficiently documented. These findings underscore a critical gap in survival analysis reporting, emphasizing the need for structured guidelines on the transparent handling and reporting of missing outcomes in time-to-event studies. Addressing these gaps will enhance the reliability and reproducibility of survival analyses in RCTs.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data supporting the findings of this study are available from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research received no specific grant from any funding agency\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor\u0026rsquo;s contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eS.K. and P.S. conceptualized and designed the study. S.K. performed the data extraction, and P.S. conducted a pilot review of the extracted data and provided feedback. S.K. drafted the initial version of the manuscript. V.R. and P.S. critically reviewed and revised the draft. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAkobeng AK. Understanding randomised controlled trials. \u003cem\u003eArch Dis Child\u003c/em\u003e. 2005;90(8):840-844. doi:10.1136/ADC.2004.058222\u003c/li\u003e\n\u003cli\u003eO\u0026rsquo;Neill RT, Temple R. The prevention and treatment of missing data in clinical trials: an FDA perspective on the importance of dealing with it. \u003cem\u003eClin Pharmacol Ther\u003c/em\u003e. 2012;91(3):550-554. doi:10.1038/CLPT.2011.340\u003c/li\u003e\n\u003cli\u003eWang S, Frederich R, Mancuso JP. Imputation of Missing Data for Time-to-Event Endpoints Using Retrieved Dropouts. \u003cem\u003eTher Innov Regul Sci\u003c/em\u003e. 2024;58(1):114-126. doi:10.1007/S43441-023-00575-5/TABLES/3 \u003c/li\u003e\n\u003cli\u003eRubin DB. Inference and missing data. \u003cem\u003eBiometrika\u003c/em\u003e. 1976;63(3):581-592. doi:10.1093/BIOMET/63.3.581\u003c/li\u003e\n\u003cli\u003eCarpenter JR., Bartlett Johnathon, Morris Tim, Wood Angela, Quartagno Matteo, Kenward MG. Multiple imputation and its application. Published online 2023:440.\u003c/li\u003e\n\u003cli\u003eLipkovich I, Ratitch B, O\u0026rsquo;Kelly M. Sensitivity to censored-at-random assumption in the analysis of time-to-event endpoints. \u003cem\u003ePharm Stat\u003c/em\u003e. 2016;15(3):216-229. doi:10.1002/PST.1738\u003c/li\u003e\n\u003cli\u003eGroenwold RHH, Moons KGM, Vandenbroucke JP. Randomized trials with missing outcome data: how to analyze and what to report. \u003cem\u003eCMAJ : Canadian Medical Association Journal\u003c/em\u003e. 2014;186(15):1153. doi:10.1503/CMAJ.131353\u003c/li\u003e\n\u003cli\u003eTaylor JMG, Murray S, Hsu CH. Survival estimation and testing via multiple imputation. \u003cem\u003eStat Probab Lett\u003c/em\u003e. 2002;58:221-232.\u003c/li\u003e\n\u003cli\u003eZhao Y, Herring AH, Zhou H, Ali MW, Koch GG. A multiple imputation method for sensitivity analyses of time-to-event data with possibly informative censoring. \u003cem\u003eJ Biopharm Stat\u003c/em\u003e. 2014;24(2):229-253. doi:10.1080/10543406.2013.860769\u003c/li\u003e\n\u003cli\u003eJackson D, White IR, Seaman S, Evans H, Baisley K, Carpenter J. Relaxing the independent censoring assumption in the Cox proportional hazards model using multiple imputation. \u003cem\u003eStat Med\u003c/em\u003e. 2014;33(27):4681-4694. doi:10.1002/SIM.6274\u003c/li\u003e\n\u003cli\u003eAtkinson A, Kenward MG, Clayton T, Carpenter JR. Reference-based sensitivity analysis for time-to-event data. \u003cem\u003ePharm Stat\u003c/em\u003e. 2019;18(6):645-658. doi:10.1002/PST.1954\u003c/li\u003e\n\u003cli\u003eBell ML, Fiero M, Horton NJ, Hsu CH. Handling missing data in RCTs; A review of the top medical journals. \u003cem\u003eBMC Med Res Methodol\u003c/em\u003e. 2014;14(1):1-8. doi:10.1186/1471-2288-14-118/TABLES/4\u003c/li\u003e\n\u003cli\u003eCarroll OU, Morris TP, Keogh RH. How are missing data in covariates handled in observational time-to-event studies in oncology? A systematic review. \u003cem\u003eBMC Med Res Methodol\u003c/em\u003e. 2020;20(1):1-15. doi:10.1186/S12874-020-01018-7/TABLES/4\u003c/li\u003e\n\u003cli\u003eGoldkuhle M, Hirsch C, Iannizzi C, et al. Meta-epidemiological review identified variable reporting and handling of time-to-event analyses in publications of trials included in meta-analyses of systematic reviews. \u003cem\u003eJ Clin Epidemiol\u003c/em\u003e. 2023;159:174-189. doi:10.1016/J.JCLINEPI.2023.05.023\u003c/li\u003e\n\u003cli\u003eTan PT, Cro S, Van Vogt E, Szigeti M, Cornelius VR. A review of the use of controlled multiple imputation in randomised controlled trials with missing outcome data. \u003cem\u003eBMC Med Res Methodol\u003c/em\u003e. 2021;21(1). doi:10.1186/S12874-021-01261-6\u003c/li\u003e\n\u003cli\u003eGoldkuhle M, Hirsch C, Iannizzi C, et al. Meta-epidemiological review identified variable reporting and handling of time-to-event analyses in publications of trials included in meta-analyses of systematic reviews. \u003cem\u003eJ Clin Epidemiol\u003c/em\u003e. 2023;159:174-189. doi:10.1016/J.JCLINEPI.2023.05.023\u003c/li\u003e\n\u003cli\u003eRubin DB. Multiple Imputation for Nonresponse in Surveys. Published online June 9, 1987. doi:10.1002/9780470316696\u003c/li\u003e\n\u003cli\u003eSullivan TR, Yelland LN, Lee KJ, Ryan P, Salter AB. Treatment of missing data in follow-up studies of randomised controlled trials: A systematic review of the literature. \u003cem\u003ehttps://doi.org/101177/1740774517703319\u003c/em\u003e. 2017;14(4):387-395. doi:10.1177/1740774517703319\u003c/li\u003e\n\u003cli\u003eKaruppasamy S, Samuel P. A Scoping Review Protocol for Handling Missing Outcomes in Time-to-Event Analyses in Randomised Controlled Trials. \u003cem\u003eosf.io/6xz95. \u003c/em\u003ePublished April 9, 2025.\u003c/li\u003e\n\u003cli\u003eR Core Team (2023). R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria.\u003c/li\u003e\n\u003cli\u003ePereira NL, Farkouh ME, So D, et al. Effect of Genotype-Guided Oral P2Y12 Inhibitor Selection vs Conventional Clopidogrel Therapy on Ischemic Outcomes After Percutaneous Coronary Intervention: The TAILOR-PCI Randomized Clinical Trial. \u003cem\u003eJAMA\u003c/em\u003e. 2020;324(8):761-771. doi:10.1001/JAMA.2020.12443\u003c/li\u003e\n\u003cli\u003eMilstone AM, Voskertchian A, Koontz DW, et al. Effect of Treating Parents Colonized With Staphylococcus aureus on Transmission to Neonates in the Intensive Care Unit: A Randomized Clinical Trial. \u003cem\u003eJAMA\u003c/em\u003e. 2020;323(4):319-328. doi:10.1001/JAMA.2019.20785\u003c/li\u003e\n\u003cli\u003eYamamura T, Kleiter I, Fujihara K, et al. Trial of Satralizumab in Neuromyelitis Optica Spectrum Disorder. \u003cem\u003eNew England Journal of Medicine\u003c/em\u003e. 2019;381(22):2114-2124. doi:10.1056/nejmoa1901747\u003c/li\u003e\n\u003cli\u003eSterne JAC, White IR, Carlin JB, et al. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. \u003cem\u003eBMJ\u003c/em\u003e. 2009;338(7713):157-160. doi:10.1136/BMJ.B2393\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"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":"Time-to-event data, Survival Analysis, Missing outcomes, Multiple Imputation","lastPublishedDoi":"10.21203/rs.3.rs-6801188/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6801188/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eRandomized Controlled Trials (RCTs) are the gold standard for evaluating treatment effects. However, several factors can threaten the validity of findings, including missing outcomes. Missing data pose a unique challenge in time-to-event analyses, where the event time may be censored rather than completely missing. Proper handling of missing event times is crucial to ensure unbiased and reliable conclusions in RCTs. This scoping review examines how missing outcomes in time-to-event studies have been addressed in high-impact medical journals and evaluates the implementation and reporting of multiple imputation (MI) techniques in RCTs.\u003c/p\u003e\u003ch2\u003eMethod\u003c/h2\u003e \u003cp\u003eThis scoping review assessed methods for handling missing time-to-event outcomes in RCTs published between 2019 and 2023 in three high-impact medical journals: The New England Journal of Medicine, The Lancet, and The Journal of the American Medical Association. Studies with time-to-event as the primary outcome were included. If missing outcomes were present, a full review was conducted to assess the methods used and how they were reported, including details on multiple imputation (MI). The review also explored theoretical approaches for imputing censored event times.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eA total of 694 articles were identified through a PubMed search. After screening, 321 RCTs underwent full-text review. Of these, 297 (92.2%) had no or \u0026lt;\u0026thinsp;10% missing outcomes without imputation. The remaining 17 (5.3%) addressed missing data using statistical methods: 10 used MI, 6 used best-/worst-case scenarios, and 1 applied a propensity score method. MI approaches varied, with some studies lacking detailed reporting.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eIn RCTs with survival outcomes, properly handling missing event times is essential. This scoping review reveals that, despite the availability of robust statistical methods, the treatment of missing time-to-event outcomes remains underutilized and often poorly documented. Many studies acknowledge censoring but fail to distinguish between informative and non-informative censoring. Additionally, the reporting of multiple imputation techniques is frequently insufficient. These findings highlight a critical gap in the handling and reporting of missing outcomes in survival analysis. Strengthening these practices will enhance the reliability and reproducibility of survival analyses in RCTs.\u003c/p\u003e","manuscriptTitle":"Handling Missing Outcomes in Time-to-Event Analyses: A Scoping Review of Multiple Imputation in Randomised Controlled Trials","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-13 14:30:51","doi":"10.21203/rs.3.rs-6801188/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-07-14T18:44:00+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-09T16:04:00+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-01T07:52:57+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"106051466010804581756581128499019145272","date":"2025-06-20T14:07:25+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"77455228386551657098705628134077730860","date":"2025-06-11T07:43:20+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-06-11T06:18:05+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-06-11T06:15:52+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-06-09T14:15:31+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-06-09T12:00:36+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Research Methodology","date":"2025-06-09T11:57:24+00:00","index":"","fulltext":""}],"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":"4502c8d9-c949-454d-89a1-bd3dc1550027","owner":[],"postedDate":"June 13th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-10-01T13:56:54+00:00","versionOfRecord":{"articleIdentity":"rs-6801188","link":"https://doi.org/10.1186/s12874-025-02676-1","journal":{"identity":"bmc-medical-research-methodology","isVorOnly":false,"title":"BMC Medical Research Methodology"},"publishedOn":"2025-09-29 00:00:00","publishedOnDateReadable":"September 29th, 2025"},"versionCreatedAt":"2025-06-13 14:30:51","video":"","vorDoi":"10.1186/s12874-025-02676-1","vorDoiUrl":"https://doi.org/10.1186/s12874-025-02676-1","workflowStages":[]},"version":"v1","identity":"rs-6801188","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6801188","identity":"rs-6801188","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}