Multimodal Prediction of Future Depressive Symptoms in Adolescents

Multimodal Prediction of Future Depressive Symptoms in Adolescents | 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 Multimodal Prediction of Future Depressive Symptoms in Adolescents Linshanshan Wang, Ningxuan Zhou, Nigel M. Jaffe, Kristina Pidvirny, and 7 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6585192/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 03 Dec, 2025 Read the published version in BMC Psychiatry → Version 1 posted 9 You are reading this latest preprint version Abstract Background Depression rates surge during adolescence. Early identification of youth at increased risk for depression is crucial for timely intervention and, ideally, prevention. This study aims to improve the prediction of future depressive symptoms in adolescents by using a multimodal approach that integrates relevant clinical, demographic, behavioral, and neural characteristics. Methods 103 adolescents (ages 12–18; 72.8% female) underwent a baseline assessment including self-report questionnaires, ecological momentary assessment, a clinical interview, and behavioral and neural measures of reward responsiveness. We used nested cross-validation to compare machine learning approaches as well as conventional linear regression in predicting depressive symptoms (Center for Epidemiological Studies Depression Scale [CES-D] and the Mood and Feelings Questionnaire [MFQ]) at a 3-month follow-up. Results For the prediction of CES-D depression scores, the best performing model was a multivariable linear regression using as predictors five principal component scores from a principal component analysis of baseline variables (RMSE = 6.501, R 2 = 0.688). For the MFQ, the best performing model was a univariable linear regression with baseline MFQ scores as the sole predictor (RMSE = 8.054, R 2 = 0.671). A factor analysis revealed that items assessing melancholic features were most predictive of future depressive symptoms. Conclusion More complex machine learning approaches did not outperform regression in predicting future depression. The integration of relevant multimodal predictors reveals which adolescent characteristics (e.g., melancholic features and physical anxiety) have a larger contribution to predicting short-term future depression. Future studies are needed with larger sample sizes and longer follow-up periods to provide a more comprehensive test of such models. Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Rates of major depressive disorder (MDD) among adolescents and young adults have risen substantially over the last 15 years [ 1 ]. More recently, evidence points to further increases in child and adolescent depression symptoms during the COVID pandemic [ 2 ]. This notable increase underscores the critical need for early identification of adolescents at risk for depression, which could inform early intervention and preventative efforts. Despite widespread awareness of a mental health epidemic afflicting youth, researchers have not yet succeeded in developing statistical models that can reliably forecast depressive symptoms in adolescents. Comparable systems already exist for certain physical health problems. For example, the Predicting Risk of Cardiovascular Disease Events (PREVENT) calculator is an algorithm incorporating several relevant predictors (e.g., age, sex, cholesterol, current smoking, and body mass index) as inputs to provide personalized estimates of a patient’s risk of developing cardiovascular disease outcomes such as coronary heart disease or heart failure [ 3 ]. A similar predictive model for depressive symptoms would represent an important step forward for the field, as more reliable identification of early warning signs of future depression (i.e., “red flags”) may mitigate MDD’s negative consequences by allowing the timely delivery of preventive interventions. In addition, a fuller understanding of the extent to which certain factors (e.g., demographic and clinical characteristics, environmental/social variables or neural features) each contribute to depression risk might allow for the development of algorithms that offer personalized prediction of future symptoms, with the potential for tailoring preventative interventions to an individual’s particular risk profile (e.g., intervention recommendations may differ if neuroticism vs. rumination vs. anhedonia were especially elevated in a given individual) [ 4 ]. In an effort to better understand and predict the onset of depression during adolescence, previous studies have identified a range of risk factors that prospectively predict depression (for a recent review see [ 5 ]). Studies have typically only examined one or a few risk factors in isolation. For instance, previous studies have identified a range of predictors of depression, including demographic variables such as age and gender [ 6 , 7 ], clinical characteristics like depressive symptoms, anhedonia, anxiety symptoms, and rumination [ 8 – 11 ], socioeconomic status [ 12 ], pubertal status [ 13 ], as well as neural markers such as decreased striatal activation to reward [ 14 , 15 ]. Considering each variable in isolation may account for only limited variance in future depression outcomes. Since depression is a heterogeneous disorder emerging from a complex interplay of environmental/social, psychological, and biological factors, it is difficult to predict the onset of depression based on any single variable [ 16 ]. Therefore, there is a need for a more comprehensive, data-driven approach that considers multiple risk markers simultaneously to improve predictive accuracy and to better reflect the reality of depression’s multifaceted etiology. Recently, machine learning techniques have gained traction in predicting depression. These approaches are well-suited to handling a large number of predictor variables (including multicollinearity among variables) and modeling complex relations such as interactions and other non-linear associations that traditional statistical methods might overlook (e.g., in ordinary least squares (OLS) regression if relevant interactions and non-linear relations are not explicitly specified). Prior studies using machine learning have shown promising results. For instance, a recent large study achieved area under the receiver operating characteristic curve (AUROC) values of 0.72–0.74 in identifying depression in 6,310 children and adolescents aged 4–17 [ 17 ]. However, while these studies incorporated several predictors, they typically collected data within one modality (e.g., all self-reports) anddid not combine multi-modal data from several relevant sources [ 18 ], for an exception see [ 19 ]. An important next step involves determining if combining measures of depression-relevant characteristics assessed from various modalities – such as conventional self-report questionnaires (e.g., assessing clinical, demographic, and personality characteristics), ecological momentary assessments (EMAs; e.g., affect dynamics in daily life), behavioral tasks (e.g., reward learning and cognitive control abilities), and neuroimaging (e.g., reward-related brain activation) – can enhance our ability to predict depression in adolescents. In addition, most recent studies employing machine learning to examine multiple predictors of depression tested associations with current depression. Longitudinal studies are required to study the predictive value of these factors for the onset of future depressive symptoms. One exception is a recent study that used machine learning to combine various modalities such as clinical data, cognitive assessments, environmental factors, and structural magnetic resonance imaging (MRI) to predict depression onset at a 2-year and 5-year follow-up [ 19 ]. The study used penalized logistic regression and found that baseline depressive symptoms, female sex, neuroticism, stressful life events, and the surface area of the supramarginal gyrus were the strongest predictors. The predictive model achieved an AUROC of 0.68–0.72. However, penalized logistic regression, while effective in handling multicollinearity and overfitting, does not account for complex interactions and nonlinear associations (unless they are explicitly modeled). Other algorithms, such as decision-tree based models (e.g., random forest) are well-designed to do so and may yield better predictive performance, highlighting the value of comparing a range of machine learning approaches. The current study aims to address the abovementioned limitations by combining multimodal data, including depression-relevant self-reports/EMA (e.g., assessing neuroticism, rumination, stress, and affect dynamics), behavioral assessments (reward learning), and neural measures (neural reward sensitivity), to prospectively predict future depressive symptoms in youth. The integration of different data sources may also help in identifying novel predictors and interactions that may be important for early detection and intervention. In addition, the current study will compare various machine learning approaches, as well as conventional regression, to identify which best predicts depressive symptoms in adolescents. Methods and Materials Participants Participants were 103 English-speaking adolescents aged 12–18 (75 female, 28 male; m age = 16.0, SD = 1.9) recruited from the greater Boston area for two larger studies that recruited typically developing (non-anhedonic) adolescents (n = 68 included in the current study), as well as adolescents with elevated levels of anhedonia (n = 35 included in the current study). Exclusion criteria included history or current diagnosis of any of the following DSM-5 psychiatric illnesses: major depressive disorder, schizophrenia spectrum or other psychotic disorder, bipolar disorder, substance or alcohol use disorder within the past 12 months or lifetime severe substance or alcohol use disorder, and current diagnosis of anorexia nervosa or bulimia nervosa. Psychotropic medications were exclusionary, with the exception of stable-dose (at least one month) selective serotonin reuptake inhibitors (SSRIs) (n = 2; see Table 1 ). See Supplement for additional exclusion criteria. Procedure All procedures were approved by the Mass General Brigham IRB. Written informed consent was provided by participants who were 18 years of age, as well as from parents of participants who were under 18 years of age, along with the participant’s written informed assent. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. At the baseline session, either in-person or over Zoom, participants were administered a semi-structured clinical interview, the Kiddie Schedule for Affective Disorders and Schizophrenia (K-SADS; [ 20 ]), and completed self-report measures. Following the baseline session, participants completed an MRI scan session including a Reward Task [ 21 ] functional MRI (fMRI) probing neural response to the anticipation and receipt of, loss of, or no change in monetary reward vs. losses. After the MRI session, participants performed a Probabilistic Reward Task (PRT; [ 22 ]) assessing reward learning. See Supplement for further details on MRI acquisition, task design, and MRI data processing. At the MRI scan session, participants installed the MetricWire App on their smartphones to complete EMAs asking about current positive affect (PA) and negative affect (NA). As described in Murray et al. [ 23 ] , following the fMRI scan session EMA surveys were delivered 2–3 times per day whereby participants were randomly signaled during two timeslots (4pm to 6:30pm and 6:30pm to 9pm) during a 5-day period (Thursday - Monday). A third survey was sent on weekends (11am-4pm). Three months after the baseline session, participants completed the baseline self-report measures again, including depression measures (see below). Outcome Measures : Two self-report depression outcome measures were assessed at the 3-month follow-up: Center for Epidemiological Studies Depression Scale (CES-D; [ 24 ]) total score, assessing past-week symptoms, and the Mood and Feelings Questionnaire – Child Self-Report – Long Version (MFQ; [ 25 ]) total score, assessing symptoms over the past two weeks. Both measures are widely used but differ on several dimensions, including content (e.g., the MFQ has greater coverage of DSM-5 symptoms), sensitivity (the CESD has been shown to be relatively sensitive in detecting depressive symptoms at the lower severity range), and timeframe (assessing symptoms over the past week vs. past 2 weeks) [ 26 – 28 ]. Baseline Predictor Variables Clinical self-report scales Depression (CES-D and MFQ), anhedonia (Snaith-Hamilton Pleasure Scale [SHAPS; [ 29 ]] total score), rumination (Children’s Response Styles Questionnaire [CRSQ; [ 30 ]] rumination subscale score), perceived stress (Perceived Stress Scale [PSS; [ 31 ]] total score), physical anxiety (Multidimensional Anxiety Scale for Children [MASC; [ 32 ]] subscale score), social anxiety (MASC social subscore), separation anxiety (MASC separation subscore), harm avoidance (MASC harm subscore), negative life events (Adolescent Life Event Questionnaire-Revised [ALEQ-R; [ 33 ]] total score), extraversion (NEO Five-Factor Inventory-3 [NEO-FFI-3; [ 34 ]] extraversion factor), and neuroticism (NEO-FFI-3 neuroticism factor). Ecological Momentary Assessment : Participants rated emotions on a 5-point Likert scale, with mean PA and NA calculated from respective emotions. See Supplement for full details. Several affect dynamic measures previously linked to depression were also included: the variability in PA and NA were measured using the standard deviation (SD) and mean square successive difference (MSSD; [ 35 – 38 ]), and we also included temporal dependency (autocorrelation; [ 39 ]) of PA and NA as a measure of emotional inertia. Clinical Interview Participants were administered the K-SADS. As noted above, we included participants from a no anhedonia vs. elevated anhedonia group (binary variable). Elevated anhedonia was defined as having a K-SADS anhedonia item (from the MDD module) score greater than one. Neuroimaging Participants completed an fMRI monetary reward task assessing neural responses during reward anticipation and outcome [ 21 , 23 , 40 ]. See Supplement for task details. Behavioral : A reward learning task (Probabilistic Reward Task [PRT]) previously validated in adolescents [ 41 – 43 ] was used. We included two variables as predictors: reward learning (change in response bias over the course of the task computed as Response Bias in block 2 minus Response Bias in block 1) and mean response bias (averaging the response bias score over the two blocks). For task details, see [ 40 ]. Demographic Self-reported age and sex. Pubertal Pubertal status (Tanner Staging Questionnaire [TSQ; [ 44 ]] total score). Statistical Analysis Concurrent correlations with depressive symptoms at baseline Prior to testing the relation between baseline predictors and future (3 month) depressive (CESD and MFQ) symptoms, we first examined the concurrent correlations between depression scores and other predictor variables at baseline (see Fig. 1 for correlation coefficients and p -values). Predicting future (3 months) depressive symptoms See Fig. 2 for a schematic of the data processing and modeling pipeline. Data Preprocessing We standardized all variables to have means of 0 and standard deviations of 1 (Data type: Raw ). To reduce the potential instability in model training due to the high correlations between the baseline predictors, we re-ran analyses applying two separate approaches to data pre-processing: (1) To remove the confounding influence of baseline depression severity from our other predictors, for each variable other than the baseline depression scores (MFQ or CESD), we regressed the variable against baseline depression and took the residuals as the feature value. We trained the models using the residualized features along with baseline depression (Data type: Residuals ). (2) To reduce the number of predictors, we applied principal component analysis (PCA) to all baseline predictors (across all modalities), then evaluated the first 5, 10, or 20 principal component scores as features (Data type: PC5, PC10, PC20 ). See “data type” column in Table 2 . See Supplementary Fig. 3 for the cumulative variance explained by the top principal components. Model Training and Validation We fit five different predictive models with all baseline predictors: (1) conventional linear regression, (2) linear regression with elastic net penalty [ 45 ], (3) random forest [ 46 ], (4) XGBoost [ 47 ], and (5) Support vector machine (SVM, [ 48 ]). We further trained an ensemble learning model [ 49 ] which combines predictions from the base models (linear regression, elastic net, random forest, XGBoost and SVM) in an effort to maximize predictive performance. We also compared these models to a univariable model with baseline depression (CESD or MFQ) total score as the only predictor. We fit a linear regression to predict the depression outcome at 3 months with the corresponding baseline measure as the sole predictor variable. To perform model training with hyperparameter optimization while robustly evaluating the model performance, we performed nested cross-validation (CV; see Supplement for details). Preprocessing steps (standardization, residualization, and PCA) were performed within the training folds during nested cross-validation. This ensures that no information from the test folds was used to inform any transformation. See Table 2 (also see Fig. 3 ) for results. Feature Importance To assess the most influential features in the multivariable model for the prediction of depression, SHapley Additive exPlanations (SHAP) were implemented, and the top ten most influential features were reported for the random forest model, which achieved the highest performance using standardized predictors (data type: Raw ). See Supplement for additional details. Factor Analysis We conducted two exploratory factor analyses (EFA) to examine factors underlying the depression questionnaire (MFQ and CESD) items at baseline, in an effort to determine which depression features, if any, were most predictive of future depressive symptoms. See Supplement for details. Results Participant characteristics Participants were predominantly White ( n = 64, 62.1%), non-Hispanic ( n = 92, 89.3%) teens, most of whom were assigned female at birth ( n = 75, 72.8%). Baseline anhedonia (SHAPS: M = 23.0, SD = 7.4) and depression symptoms (CES-D: M = 12.3, SD = 11.1) were relatively mild in severity. We report additional demographic and clinical characteristics in Table 1 . At the 3-month timepoint, CES-D scores were slightly higher ( M = 13.3, SD = 13.1) than at the baseline assessment, but below the conventional cutoff score of 16 [ 24 ], with substantial variability at both timepoints. Baseline correlations We report Spearman correlation coefficient between depression symptoms and other predictive variables at baseline, along with the p-values in Fig. 1 . Both depression measures (CESD and MFQ) exhibit statistically significant correlations with a range of self-report measures: anhedonia (SHAPS), anxiety facets (MASC physical, social, and separation anxiety subscores), rumination (CRSQ) perceived stress (PSS), negative life events (ALEQR), and neuroticism (NEO), ranging from r = 0.4 to 0.75 . Strong negative correlations were found between both baseline depression measures and extraversion (NEO) (CESD: r = -0.58, p = 2e-10; MFQ: r = -0.58, p = 1e-10). Among the EMA measures, lower mean PA and higher mean and variability in NA (both SD and MSSD) were significantly associated with higher depression. Among the neuroimaging variables, only blunted striatal (left caudate) response to the anticipation of rewards was significantly correlated with greater depressive symptoms on both measures. Our behavioral measure of reward learning (PRT) was not correlated with depression (nor anhedonia). Principal Component Analysis The PCA decomposition (see Supplementary Fig. 3 ) showed that the first 5, 10, and 20 components explained 55.6%, 74.8%, and 91.3% of the total variance, respectively, and loadings of the first 5 PCs are summarized in Supplementary Fig. 4 . Model Performance For the model predicting future (3-month) depressive symptoms as measured by the CESD, the best performing model based on RMSE and R2 was the multivariable linear regression model with the first 5 PCs as features (RMSE = 6.501, R2 = 0.688, Table 2 and Fig. 3 ). It achieved better performance compared with the univariable model using baseline CESD as the only predictor (RMSE = 6.750, R2 = 0.645). Across all models using raw features as input, the random forest model and ensemble approach had the best prediction accuracy, achieving R2 of 0.614 and 0.616, and RMSE of 6.723 and 6.738, respectively. For the MFQ model, the best performing model based on RMSE and R2 was the simplest model: the univariable linear regression model with baseline MFQ total scores as the sole predictor (RMSE = 8.054, R2 = 0.671). The best multivariable model was also the linear regression model with the first 5 PCs as features (RMSE = 8.128, R2 = 0.647), but the performance was worse than the univariable model. Using raw features as input, the random forest model achieved the best prediction accuracy, with R2 of 0.562 and RMSE of 8.616. Feature Importance To evaluate feature importance of each predictor variable, we fit a random forest model using all raw features on the full dataset (i.e. outside of the nested cross-validation framework). We focused on the random forest model since that raw feature model performed the best across the two depression outcome measures. For the model predicting CESD scores at 3 months, depressive (MFQ) symptom severity at baseline appears as the most influential feature, with a mean SHAP value five times greater than the second most influential feature, the baseline CESD score (Fig. 4 a). Other features such as greater physical anxiety (MASC), anhedonia (SHAPS), and blunted striatal (right caudate) response to the anticipation of rewards also demonstrated predictive influence. For the model predicting MFQ scores at 3 months, greater baseline depressive (MFQ) symptoms again emerged as the most important contributor to the predicted outcomes (Fig. 4 b). The set of top influential features shows some overlap with those important for predicting MFQ scores, including greater physical anxiety (MASC), neuroticism (NEO), and blunted striatal (right caudate and putamen) response to the anticipation of rewards. Depression Symptom Factors Given that baseline depression was the most robust predictor of future depression we sought to understand if there were particular factors or subsets of depression symptoms/items that were most predictive. For the CESD questionnaire, the first factor described depressed mood. The second factor was related to depressogenic social cognitions (e.g., “ I felt that people disliked me ”). The third factor described anhedonia, while the last factor was related to sleep issues and fear ( Supplementary Fig. 6a ). Within the MFQ questionnaire, the first factor reflected items related to melancholia, characterized by psychomotor changes (restlessness and psychomotor retardation), sleep difficulties, and difficulty concentrating. The second factor was related to low self-esteem and self-deprecation. The third factor points to items related to suicidal ideation, while the fourth factor included anhedonia symptoms and hopelessness ( Supplementary Fig. 6b ). We explored whether certain factors were more strongly related to future depression. For the CESD model, only Factor 2 scores (depressogenic social cognitions) significantly correlated with the total CESD scores at 3 months (r 2 = 0.16, p = 0.0014, Supplementary Fig. 6a ). In the MFQ model, only factor 1 (melancholia) baseline scores significantly correlated with total MFQ scores at three months (r 2 = 0.33, p = 1.3e-6, Supplementary Fig. 6b ). Discussion Early identification of depression risk is crucial for the timely implementation of prevention strategies, particularly during adolescence, a developmental period marked by heightened vulnerability to depressive symptoms. A key challenge in predicting future depression is its complex, multidimensional nature. Depression is influenced by a wide array of factors spanning psychological, behavioral, biological, and environmental domains, making it important to integrate data from multiple modalities when developing predictive models of future symptoms. In this study, we explored machine learning approaches to predict future depressive symptoms in youth by integrating a wide range of theoretically-relevant predictors, including from conventional self-report scales (e.g., neuroticism and anhedonia), ecological momentary assessments (EMA; e.g., mean and variance in negative and positive affect), behavioral tasks (e.g., reward responsiveness), neuroimaging data (e.g., reward circuitry response), demographic characteristics (e.g., age and sex), and developmental measures (e.g., puberty status). Additionally, we employed machine learning (ML) techniques, combined with dimensionality reduction of baseline predictors, to enhance model accuracy and interpretability. In an effort to further improve predictive performance, this approach not only integrated numerous predictors, but also captured potential non-linear effects and interactions that studies using fewer predictors might overlook. Our use of a multivariable approach, combined with an analysis of feature importance, enabled a detailed assessment of which predictors exerted the most significant influence on depression outcomes. For the CES-D, the multivariable linear regression model that used the first five principal components as predictors achieved the best performance, with an RMSE of 6.501 and an R² of 0.688. This model outperformed the univariable approach, which only included baseline CES-D scores as the sole predictor. Previous research on predicting depression outcomes in treatment-seeking populations has similarly shown that multivariable models outperform univariable ones [ 50 ]. Our findings contribute to this body of evidence by extending these results to an adolescent sample recruited from the community, using a wide range of predictors. In contrast, for the MFQ, the best-performing model was a simple univariable linear regression that used baseline MFQ scores as the sole predictor, achieving an RMSE of 8.054 and an R² of 0.671. In line with this, feature importance analysis in the multivariable models indicated that baseline depressive symptoms, as measured by the MFQ, was the most robust predictor of future depressive symptoms. This finding aligns with previous research [ 17 ] and highlights the critical role that the severity of initial symptoms plays in predicting future depression outcomes (at least within the relatively short timeframe of 3 months). This emphasizes the importance of early symptom assessment as a key factor in forecasting long-term mental health trajectories. Interestingly, feature importance showed that baseline MFQ was the best predictor of both CES-D and MFQ outcomes, even exceeding the predictive value of baseline CES-D for future CES-D scores. One explanation is that the MFQ is specifically designed for adolescents and covers a broader range of symptoms highly relevant to this age group. In addition, because the MFQ assesses symptoms over a two-week period rather than one week, it may provide a more stable index of severity. Together, these features may capture more trait-like aspects of adolescent depression vulnerability, whereas the CES-D may be somewhat more sensitive to short-term, state-related changes in symptoms. While the univariable model using baseline depressive symptoms as the sole predictor performed best for MFQ, a random forest model incorporating all raw features also showed relatively strong performance in predicting both CES-D and MFQ depression scores. Feature importance revealed that the variables contributing most strongly to depression predictions were from conventional self-report measures (e.g., baseline depression severity, neuroticism and physical anxiety), EMA (mean positive affect) and neuroimaging (striatal response to the anticipation of rewards). These findings reinforce the multifaceted nature of depression, where a combination of psychological, biological, and behavioral factors contribute to its trajectory. The observed importance of physical anxiety and neuroticism emphasizes the need to incorporate psychological characteristics into predictive models, which are relatively easily assessed via low-cost self-report measures, supporting the idea that depression risk is influenced by a constellation of internal and external factors [ 16 ]. These results align with previous research showing that depression is prospectively predicted by psychological variables, particularly neuroticism [ 51 ]. Moreover, findings also align with recent work suggesting that a blunted striatal response to reward is a hallmark of maladaptive reward processing in major depressive disorder [ 52 ], which may precede and predict future depression in youth [ 53 ]. However, when comparing across modalities, the bulk of the top predictors were self-report measures (CESD: 6/10 conventional self-report questionnaires and 1/10 EMA; MFQ: 6/10 conventional self-report questionnaires and 2/10 EMA). The relatively poor predictive performance of neural and behavioral measures may be due at least in part to fundamental differences in measurement reliability and error characteristics across modalities, as self-report measures have been shown to have higher reliability than single-session behavioral [ 54 – 57 ] or neuroimaging tasks [ 58 ]. Lower reliability in these tasks is in part due to low between-person variance, which can attenuate predictor-outcome associations. This reflects a fundamental design feature of experimental tasks, which are optimized to detect robust group-level condition effects at the expense of minimizing individual differences (which ultimately hampers reliability). Also, laboratory-based neural and behavioral measures may not capture real-world functioning as effectively as self-reported symptoms and traits. These findings highlight both the robust predictive value of carefully selected self-report measures and the need for improved approaches to capturing meaningful individual differences in neural and behavioral functioning. Given the robust role of baseline depressive symptom severity in predicting future symptoms, with the univariable model performing the best for predicting MFQ scores, and baseline depression being the strongest predictor of CES-D scores in that multivariable model, we wanted to more closely examine which elements of depression were driving these effects. Therefore, we conducted a factor analysis of baseline MFQ and CES-D scores. Our analysis identified several distinct dimensions of depressive symptoms, including melancholia, low self-esteem, suicidal ideation, and anhedonia. These factors resemble those uncovered in previous research using the MFQ, namely vegetative symptoms, suicidality, cognitive symptoms, and agitated distress [ 59 , 60 ]. The strong correlation of the melancholia MFQ factor with future MFQ total scores, and the depressogenic social cognition CES-D factor with future CES-D total scores, suggest that it may be important to emphasize targeting these specific symptom dimensions in preventive interventions. Our results showed that complex ML approaches did not outperform regression-based techniques, aligning with prior work [ 50 ]. However, tree-based machine learning models such as random forest performed relatively well and provided complementary insights via SHAP-based feature importance analyses. These models may still offer meaningful value, particularly in larger samples or when interactions and nonlinear relationships are more prominent. More research is needed to determine in which contexts ML approaches are likely to outperform simpler statistical models (e.g., see [ 61 , 62 ] for examples of ML models incorporating passive smartphone sensor data which outperformed simpler statistical approaches in predicting negative affective states). Future studies could benefit from assessing other depression-relevant baseline individual characteristics which were not assessed in this study that may improve predictive performance (e.g., emotion regulation abilities, sleep habits, home and school environment, early adversity, or executive function). Furthermore, there are of course factors in the intervening period between baseline and follow-up assessments that were not captured in our study which are known to be robust predictors of depression (e.g., stressors and negative life events, especially in the interpersonal domain). The current study had several limitations. One notable limitation of our study is the relatively small sample size, in particular for the multivariable ML models. Although this risk was somewhat mitigated through dimensionality reduction of baseline predictors and the use of penalized ML models, a larger sample size is needed to more reliably estimate these models [ 63 , 64 ]. Relatedly, while we used a nested cross-validation approach to mitigate overfitting, external validation in a larger, independent sample will be important to further evaluate the generalizability of these findings. Second, although our current sample is relatively diverse (38% non-White), it may not fully represent the broader adolescent population, which could limit generalizability. Another limitation is the short follow-up period, with only a single follow-up assessment after three months. This design restricts our ability to capture longer-term depression symptoms [ 65 ] and assess the stability of the predictive factors over time. Additionally, our study used two different depression outcome measures, CES-D and MFQ, analyzing them separately. This approach may complicate clinical interpretation and the integration of findings. Future research should explore methods for combining or integrating different outcome measures (including clinical interviews) to enhance the clinical applicability of predictive models and provide a clearer understanding of depression dynamics. These limitations notwithstanding, the present study demonstrated the utility of models that integrate diverse predictors from psychological, behavioral, developmental, and neural assessments to characterize risk for future depression and their relative predictive importance. Further research is needed to test the generalizability of these models across multiple time scales and populations. Table 1 Demographic and Clinical Characteristics of the Sample Sample Characteristics N % Biological Sex Female 75 72.8 Male 28 27.2 Race American Indian or Alaska Native 0 0.0 Asian 13 12.6 Black or African American 11 10.7 Native Hawaiian or Other Pacific Islander 1 1.0 White 64 62.1 Other 3 2.9 More than one race 10 9.7 Unknown race 1 1.0 Ethnicity Hispanic or Latino 10 9.7 Not Hispanic or Latino 92 89.3 Unknown Ethnicity 1 1.0 Current Diagnoses (DSM-V) Major Depressive Episode 0 0.0 Generalized Anxiety Disorder 5 4.9 Social Anxiety Disorder 3 2.9 Panic Disorder 1 1.0 Specific Phobia 0 0.0 Attention-Deficit / Hyperactivity Disorder 3 2.9 Obsessive Compulsive Disorder 2 1.9 Medication SSRI 2 1.9 M SD Age (in years) 16.0 1.9 Family Income (dollars) 165,844.5 92,242.4 Baseline SHAPS Score 23.0 7.4 Baseline CES-D Score 12.3 11.1 Note . The SHAPS is scored on a 1–4 scale where higher scores indicate greater anhedonia; the possible range is 14–56. The CES-D is scored on a 0–3 scale where higher scores indicate greater depression; the possible range is 0–60. Table 2 Model performances for prediction of depressive symptoms (CESD and MFQ) at 3 months follow-up. Results are summarized over 30 repetitions. RMSE: root mean squared error, R2: R squared, MAE: mean absolute error. Univariable: model with respective baseline depression as the sole predictor, Raw: all normalized baseline variables as predictors, Residuals: residualized baseline variables as predictors, PC5: first 5 principal component scores, PC10: first 10 principal component scores, PC20: first 20 principal component scores. The best performance (highest R2, lowest RMSE and MAE) across all data pre-processing procedures are bolded, and the best performance with “Raw” features are underlined. Data type Model CESD MFQ R2 RMSE MAE R2 RMSE MAE Univariable Linear Regression 0.645 6.750 4.908 0.671 8.054 5.769 Multivariable- Raw Linear Regression 0.502 8.589 6.631 0.469 10.812 8.210 Elastic Net 0.570 7.098 5.217 0.521 9.019 6.002 Random Forest 0.614 6.723 4.808 0.562 8.616 5.764 XGBoost 0.545 7.476 5.209 0.543 9.039 5.966 SVM 0.510 8.134 5.937 0.459 9.985 7.101 Ensemble 0.616 6.738 4.819 0.549 8.755 5.697 Multivariable-Residuals Linear Regression 0.323 8.891 6.574 0.268 11.11 7.941 Elastic Net 0.540 7.327 5.451 0.524 8.969 6.117 Random Forest 0.503 7.621 5.447 0.558 8.635 5.893 XGBoost 0.332 8.833 6.359 0.464 9.510 6.459 SVM 0.396 8.395 6.215 0.378 10.249 7.331 Ensemble 0.544 7.301 5.302 0.541 8.805 5.849 Multivariable-PC 5 Linear Regression 0.688 6.501 4.873 0.647 8.128 5.931 Elastic Net 0.591 6.912 4.919 0.511 9.093 5.937 Random Forest 0.606 6.785 4.932 0.535 8.866 5.961 XGBoost 0.463 7.891 5.651 0.343 10.523 6.753 SVM 0.602 6.823 4.806 0.509 9.106 5.817 Ensemble 0.600 6.836 4.846 0.528 8.933 5.931 Multivariable-PC 10 Linear Regression 0.588 7.204 4.823 0.521 9.356 5.812 Elastic Net 0.584 6.976 5.014 0.534 8.869 5.792 Random Forest 0.604 6.807 5.022 0.541 8.801 5.883 XGBoost 0.453 7.982 5.672 0.426 9.792 6.404 SVM 0.616 6.705 4.708 0.557 8.649 5.450 Ensemble 0.623 6.638 4.685 0.547 8.752 5.683 Multivariable-PC 20 Linear Regression 0.546 7.468 5.078 0.477 9.248 6.289 Elastic Net 0.562 7.156 5.312 0.529 8.922 5.811 Random Forest 0.580 7.004 5.254 0.506 9.132 6.234 XGBoost 0.425 8.190 5.911 0.364 10.341 6.779 SVM 0.589 6.929 5.007 0.550 8.722 5.662 Ensemble 0.575 7.046 5.096 0.522 8.982 5.959 Declarations Ethics approval and consent to participate All procedures were approved by the Mass General Brigham IRB. Written informed consent was provided by participants who were 18 years of age, as well as from parents of participants who were under 18 years of age, along with the participant’s written informed assent. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. Consent for publication: Not applicable Competing Interests Over the past three years, Dr. Pizzagalli has received consulting fees from Boehringer Ingelheim, Compass Pathways, Engrail Therapeutics, Neumora Therapeutics (formerly BlackThorn Therapeutics), Neurocrine Biosciences, Neuroscience Software, Sage Therapeutics, Sama Therapeutics, and Takeda; he has received honoraria from the American Psychological Association, Psychonomic Society and Springer (for editorial work) and from Alkermes; he has received research funding from the BIRD Foundation, Brain and Behavior Research Foundation, Dana Foundation, DARPA, Millennium Pharmaceuticals, NIMH and Wellcome Leap MCPsych; he has received stock options from Compass Pathways, Engrail Therapeutics, Neumora Therapeutics, and Neuroscience Software. Dr. Pizzagalli has a financial interest in Neumora Therapeutics, which has licensed the copyright to the probabilistic reward task through Harvard University. Dr. Webb has received consulting fees from King & Spalding law firm. Dr. Pizzagalli’s and Dr. Webb’s interests were reviewed and are managed by McLean Hospital and Mass General Brigham in accordance with their conflict of interest policies. No funding from these entities was used to support the current work, and all views expressed are solely those of the authors. The other authors declare no competing financial interests. Funding This research was supported by NIMH K23MH108752, the Tommy Fuss Fund, a NARSAD Young Investigator Grant from the Brain & Behavior Research Foundation and the Klingenstein Third Generation Foundation (CAW). CAW was partially supported by NIMH R01MH116969 and NCCIH R01AT011002. DAP was partially supported by the National Institute of Mental Health grants P50MH119467 and R014R37MH068376-17. Author Contribution CW acquired funding for the project. CW, LW, NZ, DP, and EF conceptualized the study. CW, NJ, KP, HF, and AT acquired and processed the data. CW, LW and NZ processed and analyzed the data. All authors contributed to interpretation of findings and drafting the manuscript. All authors approved the final version of the manuscript. Since completion of this study, DP has moved to the University of California, Irvine. Acknowledgements: None Clinical trial number : not applicable Data Availability Mass General Brigham requires IRB approval and a signed Data Use Agreement for data sharing. Please contact the first author. References Twenge JM, Cooper AB, Joiner TE, Duffy ME, Binau SG. Age, period, and cohort trends in mood disorder indicators and suicide-related outcomes in a nationally representative dataset, 2005–2017. J Abnorm Psychol. 2019;128:185–99. Madigan S, Racine N, Vaillancourt T, Korczak DJ, Hewitt JMA, Pador P, et al. Changes in Depression and Anxiety Among Children and Adolescents From Before to During the COVID-19 Pandemic: A Systematic Review and Meta-analysis. JAMA Pediatr. 2023;177:567–81. Larkin H. New Cardiovascular Disease Risk Calculator Could Eliminate the Need for Statins for Millions. JAMA. 2024;332:186–7. Webb CA, Dillon DG, Pechtel P, Goer FK, Murray L, Huys QJ, et al. Neural Correlates of Three Promising Endophenotypes of Depression: Evidence from the EMBARC Study. Neuropsychopharmacology. 2016;41:454–63. Hankin BL, Griffith JM. What Do We Know About Depression Among Youth and How Can We Make Progress Toward Improved Understanding and Reducing Distress? A New Hope. Clin Child Fam Psychol Rev. 2023;26:919–42. Dürrbaum T, Sattler FA. Minority stress and mental health in lesbian, gay male, and bisexual youths: A meta-analysis. J LGBT Youth. 2020;17:298–314. Pellicane MJ, Ciesla JA. Associations between minority stress, depression, and suicidal ideation and attempts in transgender and gender diverse (TGD) individuals: Systematic review and meta-analysis. Clin Psychol Rev. 2022;91:102113. Abela JRZ, Hankin BL. Rumination as a vulnerability factor to depression during the transition from early to middle adolescence: A multiwave longitudinal study. J Abnorm Psychol. 2011;120:259–71. Bennik EC, Nederhof E, Ormel J, Oldehinkel AJ. Anhedonia and depressed mood in adolescence: course, stability, and reciprocal relation in the TRAILS study. Eur Child Adolesc Psychiatry. 2014;23:579–86. Keenan K, Feng X, Hipwell A, Klostermann S. Depression begets depression: Comparing the predictive utility of depression and anxiety symptoms to later depression. J Child Psychol Psychiatry. 2009;50:1167–75. Stein MB, Fuetsch M, Müller N, Höfler M, Lieb R, Wittchen HU. Social anxiety disorder and the risk of depression: a prospective community study of adolescents and young adults. Arch Gen Psychiatry. 2001;58:251–6. Reiss F. Socioeconomic inequalities and mental health problems in children and adolescents: A systematic review. Soc Sci Med. 2013;90:24–31. Hamlat EJ, McCormick KC, Young JF, Hankin BL. Early pubertal timing predicts onset and recurrence of depressive episodes in boys and girls. J Child Psychol Psychiatry. 2020;61:1266–74. Nelson SK, Layous K, Cole SW, Lyubomirsky S. Do unto others or treat yourself? The effects of prosocial and self-focused behavior on psychological flourishing. Emot Wash DC. 2016;16:850–61. Toenders YJ, van Velzen LS, Heideman IZ, Harrison BJ, Davey CG, Schmaal L. Neuroimaging predictors of onset and course of depression in childhood and adolescence: A systematic review of longitudinal studies. Dev Cogn Neurosci. 2019;39:100700. Fried EI. The 52 symptoms of major depression: Lack of content overlap among seven common depression scales. J Affect Disord. 2017;208:191–7. Haque UM, Kabir E, Khanam R. Detection of child depression using machine learning methods. PLoS ONE. 2021;16:e0261131. Highland D, Zhou G. A review of detection techniques for depression and bipolar disorder. Smart Health. 2022;24:100282. Toenders YJ, Kottaram A, Dinga R, Davey CG, Banaschewski T, Bokde ALW, et al. Predicting Depression Onset in Young People Based on Clinical, Cognitive, Environmental, and Neurobiological Data. Biol Psychiatry Cogn Neurosci Neuroimaging. 2022;7:376–84. Kaufman J, Birmaher B, Brent D, Rao U, Flynn C, Moreci P, et al. Schedule for Affective Disorders and Schizophrenia for School-Age Children-Present and Lifetime Version (K-SADS-PL): initial reliability and validity data. J Am Acad Child Adolesc Psychiatry. 1997;36:980–8. Forbes EE, Hariri AR, Martin SL, Silk JS, Moyles DL, Fisher PM, et al. Altered striatal activation predicting real-world positive affect in adolescent major depressive disorder. Am J Psychiatry. 2009;166:64–73. Pizzagalli DA, Jahn AL, O’Shea JP. Toward an objective characterization of an anhedonic phenotype: a signal-detection approach. Biol Psychiatry. 2005;57:319–27. Murray L, Israel ES, Balkind EG, Pastro B, Lovell-Smith N, Lukas SE et al. Multi-modal assessment of reward functioning in adolescent anhedonia. Psychol Med. 2022:1–10. Radloff LS, The CES-D, Scale. A Self-Report Depression Scale for Research in the General Population. Appl Psychol Meas. 1977;1:385–401. Costello EJ, Angold A. Scales to assess child and adolescent depression: Checklists, screens, and nets. J Am Acad Child Adolesc Psychiatry. 1988;27:726–37. Olino TM, Yu L, Klein DN, Rohde P, Seeley JR, Pilkonis PA, et al. Measuring depression using item response theory: An examination of three measures of depressive symptomatology. Int J Methods Psychiatr Res. 2012;21:76–85. Olino TM, Yu L, McMakin DL, Forbes EE, Seeley JR, Lewinsohn PM, et al. Comparisons across depression assessment instruments in adolescence and young adulthood: An Item Response Theory study using two linking methods. J Abnorm Child Psychol. 2013;41:1267–77. Vilar A, Sánchez-Martínez N, Blasco MJ, Álvarez-Salazar S, Batlle Vila S, Forero G. Content agreement of depressive symptomatology in children and adolescents: a review of eighteen self-report questionnaires. Eur Child Adolesc Psychiatry. 2024;33:2019–33. Snaith RP, Hamilton M, Morley S, Humayan A, Hargreaves D, Trigwell P. A scale for the assessment of hedonic tone the Snaith-Hamilton Pleasure Scale. Br J Psychiatry J Ment Sci. 1995;167:99–103. Abela JRZ, Verroneau M, Taylor G, Bryan C. The development of a ruminative response style in early adolescents. Unpublished questionnaire. 2002. 2002. Cohen S, Kamarck T, Mermelstein R. A Global Measure of Perceived Stress. J Health Soc Behav. 1983;24:385–96. March JS, Parker JD, Sullivan K, Stallings P, Conners CK. The Multidimensional Anxiety Scale for Children (MASC): factor structure, reliability, and validity. J Am Acad Child Adolesc Psychiatry. 1997;36:554–65. Hankin BL, Abramson LY. Measuring cognitive vulnerability to depression in adolescence: reliability, validity, and gender differences. J Clin Child Adolesc Psychol Off J Soc Clin Child Adolesc Psychol Am Psychol Assoc Div. 2002;53:31:491–504. Costa PT Jr., McCrae RR. Bridging the gap with the five-factor model. Personal Disord Theory Res Treat. 2010;1:127–30. Smit A, Helmich MA, Bringmann L, Oldehinkel A, Wichers M, Snippe E. Critical slowing down in momentary affect as early warning signal of impending transitions in depression. 2022. van de Leemput IA, Wichers M, Cramer AOJ, Borsboom D, Tuerlinckx F, Kuppens P, et al. Critical slowing down as early warning for the onset and termination of depression. Proc Natl Acad Sci U S A. 2014;111:87–92. Wichers M, Groot PC, Psychosystems, EG EWS Group. Critical Slowing Down as a Personalized Early Warning Signal for Depression. Psychother Psychosom. 2016;85:114–6. Wichers M, Smit AC, Snippe E. Early Warning Signals Based on Momentary Affect Dynamics can Expose Nearby Transitions in Depression: A Confirmatory Single-Subject Time-Series Study. J Pers-Oriented Res. 2020;6:1–15. Kuppens P, Allen NB, Sheeber L. Emotional inertia and psychological maladjustment. Psychol Sci. 2010;21:984–91. Webb CA, Murray L, Tierney AO, Forbes EE, Pizzagalli DA. Reward-related predictors of symptom change in behavioral activation therapy for anhedonic adolescents: a multimodal approach. Neuropsychopharmacology. 2023;48:623–32. Belleau E, Kremens R, Ang Y-S, Pisoni A, Bondy E, Durham K, et al. Behavioral and Neural Markers of Reward Processing Deficits in Adolescents at High Familial Risk for Depressive Disorders. Biol Psychiatry. 2021;89:S27–8. Boger KD, Auerbach RP, Pechtel P, Busch AB, Greenfield SF, Pizzagalli DA. Co-Occurring Depressive and Substance Use Disorders in Adolescents: An Examination of Reward Responsiveness During Treatment. J Psychother Integr. 2014;24:109–21. Luking KR, Pagliaccio D, Luby JL, Barch DM. Child Gain Approach and Loss Avoidance Behavior: Relationships With Depression Risk, Negative Mood, and Anhedonia. J Am Acad Child Adolesc Psychiatry. 2015;54:643–51. Tanner J. Growth at adolescence. 2nd ed. Oxford: Blackwell; 1962. Zou H, Hastie T. Regularization and Variable Selection Via the Elastic Net. J R Stat Soc Ser B Stat Methodol. 2005;67:301–20. Breiman L. Random Forests. Mach Learn. 2001;45:5–32. Chen T, Guestrin C, XGBoost:. A Scalable Tree Boosting System. Proc. 22nd ACM SIGKDD Int. Conf. Knowl. Discov. Data Min., New York, NY, USA: Association for Computing Machinery; 2016. pp. 785–794. Steinwart I, Christmann A. Support Vector Machines. Springer Science & Business Media; 2008. Zhou Z-H. Ensemble Learning. In: Zhou Z-H, editor. Mach. Learn. Singapore: Springer; 2021. pp. 181–210. Webb CA, Cohen ZD, Beard C, Forgeard M, Peckham A, Björgvinsson T. Personalized Prognostic Prediction of Treatment Outcome for Depressed Patients in a Naturalistic Psychiatric Hospital Setting: A Comparison of Machine Learning Approaches. J Consult Clin Psychol. 2020;88:25–38. Fu Z, Brouwer M, Kennis M, Williams A, Cuijpers P, Bockting C. Psychological factors for the onset of depression: a meta-analysis of prospective studies. BMJ Open. 2021;11:e050129. Yang X, Su Y, Yang F, Song Y, Yan J, Luo Y, et al. Neurofunctional mapping of reward anticipation and outcome for major depressive disorder: a voxel-based meta-analysis. Psychol Med. 2022;52:3309–22. Mackin DM, Goldstein BL, Mumper E, Kujawa A, Kessel EM, Olino TM, et al. Longitudinal Associations Between Reward Responsiveness and Depression Across Adolescence. J Am Acad Child Adolesc Psychiatry. 2023;62:816–28. Enkavi AZ, Eisenberg IW, Bissett PG, Mazza GL, MacKinnon DP, Marsch LA, et al. Large-scale analysis of test–retest reliabilities of self-regulation measures. Proc Natl Acad Sci. 2019;116:5472–7. Enkavi AZ, Poldrack RA. Implications of the Lacking Relationship Between Cognitive Task and Self-report Measures for Psychiatry. Biol Psychiatry Cogn Neurosci Neuroimaging. 2021;6:670–2. Hedge C, Powell G, Sumner P. The reliability paradox: Why robust cognitive tasks do not produce reliable individual differences. Behav Res Methods. 2018;50:1166–86. Rouder JN, Haaf JM. A psychometrics of individual differences in experimental tasks. Psychon Bull Rev. 2019;26:452–67. Elliott ML, Knodt AR, Ireland D, Morris ML, Poulton R, Ramrakha S, et al. What Is the Test-Retest Reliability of Common Task-Functional MRI Measures? New Empirical Evidence and a Meta-Analysis. Psychol Sci. 2020;31:792–806. Jeffreys M, Rozenman M, Gonzalez A, Warnick EM, Dauser C, Scahill L, et al. Factor Structure of the Parent-Report Mood and Feelings Questionnaire (MFQ) in an Outpatient Mental Health Sample. J Abnorm Child Psychol. 2016;44:1111–20. Supreeyaporn N, Watson R, Damrongpanit S, Orchard F, Reynolds S, Jenkins PE. Factor structure and measurement invariance of the Mood and Feelings Questionnaire: a cross-cultural study among Thai and British adolescents. Eur Child Adolesc Psychiatry. 2024;33:1563–72. Ren B, Balkind EG, Pastro B, Israel ES, Pizzagalli DA, Rahimi-Eichi H, et al. Predicting states of elevated negative affect in adolescents from smartphone sensors: a novel personalized machine learning approach. Psychol Med. 2023;53:5146–54. Webb CA, Ren B, Rahimi-Eichi H, Gillis BW, Chung Y, Baker JT. Personalized prediction of negative affect in individuals with serious mental illness followed using long-term multimodal mobile phenotyping. Transl Psychiatry. 2025;15:174. Huang AA, Huang SY. Exploring Depression and Nutritional Covariates Amongst US Adults using Shapely Additive Explanations. Health Sci Rep. 2023;6:e1635. Luedtke A, Sadikova E, Kessler RC. Sample Size Requirements for Multivariate Models to Predict Between-Patient Differences in Best Treatments of Major Depressive Disorder. Clin Psychol Sci. 2019;7:445–61. Gijzen MWM, Creemers DHM, Rasing SPA, Smit F, Engels RCME. Evaluation of a multimodal school-based depression and suicide prevention program among Dutch adolescents: design of a cluster-randomized controlled trial. BMC Psychiatry. 2018;18:124. Footnotes Murray et al (2022) examined concurrent, baseline (not prospective) associations between reward-related variables in a subset of the current sample. Additional Declarations Competing interest reported. Over the past three years, Dr. Pizzagalli has received consulting fees from Boehringer Ingelheim, Compass Pathways, Engrail Therapeutics, Neumora Therapeutics (formerly BlackThorn Therapeutics), Neurocrine Biosciences, Neuroscience Software, Sage Therapeutics, Sama Therapeutics, and Takeda; he has received honoraria from the American Psychological Association, Psychonomic Society and Springer (for editorial work) and from Alkermes; he has received research funding from the BIRD Foundation, Brain and Behavior Research Foundation, Dana Foundation, DARPA, Millennium Pharmaceuticals, NIMH and Wellcome Leap MCPsych; he has received stock options from Compass Pathways, Engrail Therapeutics, Neumora Therapeutics, and Neuroscience Software. Dr. Pizzagalli has a financial interest in Neumora Therapeutics, which has licensed the copyright to the probabilistic reward task through Harvard University. Dr. Webb has received consulting fees from King & Spalding law firm. Dr. Pizzagalli’s and Dr. Webb’s interests were reviewed and are managed by McLean Hospital and Mass General Brigham in accordance with their conflict of interest policies. No funding from these entities was used to support the current work, and all views expressed are solely those of the authors. The other authors declare no competing financial interests. Supplementary Files WangsupplementR1clean.docx Cite Share Download PDF Status: Published Journal Publication published 03 Dec, 2025 Read the published version in BMC Psychiatry → Version 1 posted Editorial decision: Accepted 25 Nov, 2025 Reviews received at journal 18 Nov, 2025 Reviews received at journal 11 Nov, 2025 Reviewers agreed at journal 08 Nov, 2025 Reviewers agreed at journal 03 Nov, 2025 Reviewers agreed at journal 30 Oct, 2025 Reviewers invited by journal 30 Oct, 2025 Submission checks completed at journal 30 Oct, 2025 First submitted to journal 29 Oct, 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-6585192","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":537655423,"identity":"8a687dbe-f36b-4b3b-9826-83c9b542c15b","order_by":0,"name":"Linshanshan Wang","email":"","orcid":"","institution":"Harvard T.H. Chan School of Public Health","correspondingAuthor":false,"prefix":"","firstName":"Linshanshan","middleName":"","lastName":"Wang","suffix":""},{"id":537655424,"identity":"942db384-037d-4ac8-a8c5-ee13230f2123","order_by":1,"name":"Ningxuan Zhou","email":"","orcid":"","institution":"Harvard T.H. Chan School of Public Health","correspondingAuthor":false,"prefix":"","firstName":"Ningxuan","middleName":"","lastName":"Zhou","suffix":""},{"id":537655425,"identity":"a66cc406-1389-4f79-98dc-8cb3f28d2e37","order_by":2,"name":"Nigel M. Jaffe","email":"","orcid":"","institution":"McLean Hospital","correspondingAuthor":false,"prefix":"","firstName":"Nigel","middleName":"M.","lastName":"Jaffe","suffix":""},{"id":537655428,"identity":"9dfb51ee-5f6e-4ce4-907a-3a665a3c5b75","order_by":3,"name":"Kristina Pidvirny","email":"","orcid":"","institution":"McLean Hospital","correspondingAuthor":false,"prefix":"","firstName":"Kristina","middleName":"","lastName":"Pidvirny","suffix":""},{"id":537655429,"identity":"26e14c23-345a-4948-a1e3-119fc3a2d1fa","order_by":4,"name":"Anna O. Tierney","email":"","orcid":"","institution":"McLean Hospital","correspondingAuthor":false,"prefix":"","firstName":"Anna","middleName":"O.","lastName":"Tierney","suffix":""},{"id":537655438,"identity":"4d449812-961c-419f-b5b5-5e8d47b29dd7","order_by":5,"name":"Hadar B. Fisher","email":"","orcid":"","institution":"McLean Hospital","correspondingAuthor":false,"prefix":"","firstName":"Hadar","middleName":"B.","lastName":"Fisher","suffix":""},{"id":537655440,"identity":"7a21b7f8-5928-46af-b640-442635f6e61b","order_by":6,"name":"Francesca Morfini","email":"","orcid":"","institution":"McLean Hospital","correspondingAuthor":false,"prefix":"","firstName":"Francesca","middleName":"","lastName":"Morfini","suffix":""},{"id":537655442,"identity":"5f7742ce-fa18-47ed-9516-13af7eb7b243","order_by":7,"name":"Erika E. Forbes","email":"","orcid":"","institution":"University of Pittsburgh","correspondingAuthor":false,"prefix":"","firstName":"Erika","middleName":"E.","lastName":"Forbes","suffix":""},{"id":537655446,"identity":"40151bf5-4713-42ad-80ee-8875e2021dd5","order_by":8,"name":"Diego A. Pizzagalli","email":"","orcid":"","institution":"McLean Hospital","correspondingAuthor":false,"prefix":"","firstName":"Diego","middleName":"A.","lastName":"Pizzagalli","suffix":""},{"id":537655448,"identity":"c7afe38b-f050-4e0e-a97a-9f85c458f050","order_by":9,"name":"Tianxi Cai","email":"","orcid":"","institution":"Harvard T.H. Chan School of Public Health","correspondingAuthor":false,"prefix":"","firstName":"Tianxi","middleName":"","lastName":"Cai","suffix":""},{"id":537655451,"identity":"4e4060de-f349-4ddb-8ff3-16eb1a6dc966","order_by":10,"name":"Christian A. Webb","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA90lEQVRIiWNgGAWjYHACNmaGCgYGPhCTsQEiJMHAwExAyxkgSZoWxjZStPDPSD72uHDeNnk29uZnDxh32OUbHD/78AZDhXViAw4tEjfS0o1nbrtt2MZzzNyA8Uyy5YYz6cYWDGfScWoxkM4xk+bddjuBTSLBTIKxjdlAsiGNDcg4jEdL/jdp3jlALfLPvwFV1htI9j8DavmHT0sOmzRvA8gWHpAthw34JUC2NODWInH/mbnxjGMgv+SUSSS2HQdqecZskXAs3RiXFv6ew88eF9TcludnP75N4mNbtQEbfxrjjQ811rK4tKCCBAzGKBgFo2AUjAKyAADIrlBnAjWX0AAAAABJRU5ErkJggg==","orcid":"","institution":"McLean Hospital","correspondingAuthor":true,"prefix":"","firstName":"Christian","middleName":"A.","lastName":"Webb","suffix":""}],"badges":[],"createdAt":"2025-05-03 16:53:16","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6585192/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6585192/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12888-025-07665-8","type":"published","date":"2025-12-03T15:57:57+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":94887656,"identity":"8a8cb625-806c-49ce-8c5c-66cfd47bee3b","added_by":"auto","created_at":"2025-10-31 19:04:21","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6939559,"visible":true,"origin":"","legend":"","description":"","filename":"WangmanuscriptR1clean.docx","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/d73ebf38c69ac86735e6fc88.docx"},{"id":94887650,"identity":"64350973-1f36-4e0b-bf8b-ec2d7cedff6d","added_by":"auto","created_at":"2025-10-31 19:04:21","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":13138,"visible":true,"origin":"","legend":"","description":"","filename":"8af98da653ee444e83448f434f38ab9b.json","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/c03d7319475a08af6ec759db.json"},{"id":94887663,"identity":"11b919ba-84aa-4bad-8d0e-641811325c5f","added_by":"auto","created_at":"2025-10-31 19:04:21","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2512952,"visible":true,"origin":"","legend":"","description":"","filename":"WangsupplementR1clean.docx","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/e4261cd8545b0bf56c0a514d.docx"},{"id":94987302,"identity":"28fc99fe-cf6a-4559-8e40-9031d4afbefe","added_by":"auto","created_at":"2025-11-03 07:01:41","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":166089,"visible":true,"origin":"","legend":"","description":"","filename":"8af98da653ee444e83448f434f38ab9b1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/b570deff53c6780b37de2e42.xml"},{"id":94987191,"identity":"e8c150a3-d5df-4f85-8aea-a7b81d6bbb43","added_by":"auto","created_at":"2025-11-03 07:01:27","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":152721,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/8a8c769c0ed32984397b11e8.png"},{"id":94887654,"identity":"39115ac7-12a8-4ae8-9331-fd31a4119d49","added_by":"auto","created_at":"2025-10-31 19:04:21","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":77913,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/18859c5440638dd279c1df85.png"},{"id":94986318,"identity":"e76e450b-d616-4349-821f-d35b808d1b9d","added_by":"auto","created_at":"2025-11-03 07:00:12","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":71643,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/6f6a04d42378d9185481251b.png"},{"id":94987321,"identity":"840ccb3f-66b7-4865-a3ac-301a5021283e","added_by":"auto","created_at":"2025-11-03 07:01:42","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":58264,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/6ae42c12afc55b8ce88f4bb7.png"},{"id":94987158,"identity":"f64b1924-b4ef-4a6d-8d58-a610d1ac8370","added_by":"auto","created_at":"2025-11-03 07:01:23","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":55883,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/ef30e76af96934a75d04b938.png"},{"id":94986100,"identity":"78b70c79-80ef-4098-b74a-9164368d86db","added_by":"auto","created_at":"2025-11-03 06:59:47","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":54711,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/66e75f4bab97e2a56d7630ec.png"},{"id":94887668,"identity":"9d044908-2457-466a-b24d-7f10e6ef945e","added_by":"auto","created_at":"2025-10-31 19:04:22","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":58335,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/cfd66ca9279a84e232accc1c.png"},{"id":94986197,"identity":"5ee44fc9-2362-4963-9b68-0e9c683129db","added_by":"auto","created_at":"2025-11-03 07:00:03","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":54431,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/9d2843b1282c0f6c52b814dc.png"},{"id":94887670,"identity":"15d156bb-293e-4e29-b4e3-f2652f5f0b8c","added_by":"auto","created_at":"2025-10-31 19:04:22","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":58516,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/2d0a85a419b38027ded50cd3.png"},{"id":94887665,"identity":"74b132ef-5960-4474-9f36-e81508372fde","added_by":"auto","created_at":"2025-10-31 19:04:21","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":32604,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/da7c1e6876ce6fe923745c4c.png"},{"id":94887673,"identity":"512ff579-a611-4938-b0e3-80f977917bb7","added_by":"auto","created_at":"2025-10-31 19:04:22","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":25141,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/b07b9d8155175332c9e7d6f5.png"},{"id":94887672,"identity":"52501e13-0370-4316-b1fd-c1aa6165c6d8","added_by":"auto","created_at":"2025-10-31 19:04:22","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":24780,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/eb241fb4789108b2161367b7.png"},{"id":94987242,"identity":"210bb36a-8ae8-45ed-91af-7ad9005054e9","added_by":"auto","created_at":"2025-11-03 07:01:33","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16918,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/ecc345e1eaa68a0ed63c7e01.png"},{"id":94887661,"identity":"1d83646d-753f-4e89-8b82-bd8c79368733","added_by":"auto","created_at":"2025-10-31 19:04:21","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":17723,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/e59907cbad05d3bdd28b2b48.png"},{"id":94887676,"identity":"024493bc-2693-422c-8ddb-1e560665e345","added_by":"auto","created_at":"2025-10-31 19:04:22","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":17287,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/3fb7b58d1e558715a5c94491.png"},{"id":94887666,"identity":"0d8064a0-963c-4b98-af21-e51c59747ab3","added_by":"auto","created_at":"2025-10-31 19:04:21","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16690,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/c8ed7b3ba3438b3915370342.png"},{"id":94887675,"identity":"0d577481-9fd7-48ea-8b59-b6f50339a3d3","added_by":"auto","created_at":"2025-10-31 19:04:22","extension":"xml","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":165144,"visible":true,"origin":"","legend":"","description":"","filename":"8af98da653ee444e83448f434f38ab9b1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/ae96fa0b831a126f0192a49f.xml"},{"id":94887674,"identity":"368b165f-3444-4718-a35c-c6aa0f782d75","added_by":"auto","created_at":"2025-10-31 19:04:22","extension":"html","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":175290,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/8c6bd85a56c5bcd84234bf07.html"},{"id":94887653,"identity":"6498eb86-e751-4cb6-9149-ba4d9da10806","added_by":"auto","created_at":"2025-10-31 19:04:21","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":181901,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConcurrent correlation between depressive symptoms and other predictors at baseline. \u003c/strong\u003eSignificance level is indicated by *.\u003c/p\u003e\n\u003cp\u003e* \u0026lt;0.05, ** \u0026lt;0.01, ***\u0026lt;0.001\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/e825fbe7cc80236ab939b93f.png"},{"id":94887652,"identity":"9907b092-861b-4cc3-a266-9123e0eaefb9","added_by":"auto","created_at":"2025-10-31 19:04:21","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":316580,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSchematic of data processing and modeling pipeline.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/9cc139c6a67b88da4fb0365a.jpg"},{"id":94887651,"identity":"8b7bd1e4-2864-4ab6-838e-ea4862d5d497","added_by":"auto","created_at":"2025-10-31 19:04:21","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":202087,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePerformance of models for predicting depressive symptoms at 3 months follow-up, as measured by (a) R2: R squared, (b) RMSE: root mean squared error, \u0026nbsp;and (c) MAE: mean absolute error. \u003c/strong\u003eResults are summarized over 30 repetitions. Univariate: model with respective baseline depression measures as the sole predictor, Raw: all normalized baseline variables as predictors, Residuals: residualized baseline variables as predictors, PC5: first 5 principal component scores, PC10: first 10 principal component scores, PC20: first 20 principal component scores.\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/a428c2a93bb7948fad32b001.jpg"},{"id":94987246,"identity":"c18165b0-5e7d-4921-b236-968703611d15","added_by":"auto","created_at":"2025-11-03 07:01:34","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":281335,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP value-based feature importance for prediction of depressive symptoms at 3 months follow-up based on Random Forest model trained on raw features, with (a) CESD or (b) MFQ as the outcome measure.\u003c/strong\u003e (Left) The barplot summarizes the overall impact of each feature on model output, ranked from the highest. (Right) The individual dot color corresponds to the value of the variable, and location on the plot’s \u003cem\u003ex\u003c/em\u003e axis corresponds to that point’s relative impact on the model output. A high-feature value (red) with a corresponding high \u003cem\u003ex\u003c/em\u003e axis value (SHAP value) represents a point that strongly, positively influences the model’s prediction of depression outcome.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/06824486f70f669c9c199b5a.png"},{"id":97724097,"identity":"daeaf5d7-fcd2-4a52-8af5-3a34d1acc9bb","added_by":"auto","created_at":"2025-12-08 16:11:50","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2401264,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/880c7539-b452-433c-be86-eed1c1b51e9d.pdf"},{"id":94887659,"identity":"abe4f998-9a75-4bbf-ab91-0273174ad5b2","added_by":"auto","created_at":"2025-10-31 19:04:21","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":2512952,"visible":true,"origin":"","legend":"","description":"","filename":"WangsupplementR1clean.docx","url":"https://assets-eu.researchsquare.com/files/rs-6585192/v1/79e4843fa745943bc628be79.docx"}],"financialInterests":"Competing interest reported. Over the past three years, Dr. Pizzagalli has received consulting fees from Boehringer Ingelheim, Compass Pathways, Engrail Therapeutics, Neumora Therapeutics (formerly BlackThorn Therapeutics), Neurocrine Biosciences, Neuroscience Software, Sage Therapeutics, Sama Therapeutics, and Takeda; he has received honoraria from the American Psychological Association, Psychonomic Society and Springer (for editorial work) and from Alkermes; he has received research funding from the BIRD Foundation, Brain and Behavior Research Foundation, Dana Foundation, DARPA, Millennium Pharmaceuticals, NIMH and Wellcome Leap MCPsych; he has received stock options from Compass Pathways, Engrail Therapeutics, Neumora Therapeutics, and Neuroscience Software. Dr. Pizzagalli has a financial interest in Neumora Therapeutics, which has licensed the copyright to the probabilistic reward task through Harvard University. Dr. Webb has received consulting fees from King \u0026 Spalding law firm. Dr. Pizzagalli’s and Dr. Webb’s interests were reviewed and are managed by McLean Hospital and Mass General Brigham in accordance with their conflict of interest policies. No funding from these entities was used to support the current work, and all views expressed are solely those of the authors. The other authors declare no competing financial interests.","formattedTitle":"Multimodal Prediction of Future Depressive Symptoms in Adolescents","fulltext":[{"header":"Introduction","content":"\u003cp\u003eRates of major depressive disorder (MDD) among adolescents and young adults have risen substantially over the last 15 years [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. More recently, evidence points to further increases in child and adolescent depression symptoms during the COVID pandemic [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. This notable increase underscores the critical need for early identification of adolescents at risk for depression, which could inform early intervention and preventative efforts.\u003c/p\u003e\u003cp\u003eDespite widespread awareness of a mental health epidemic afflicting youth, researchers have not yet succeeded in developing statistical models that can reliably forecast depressive symptoms in adolescents. Comparable systems already exist for certain physical health problems. For example, the Predicting Risk of Cardiovascular Disease Events (PREVENT) calculator is an algorithm incorporating several relevant predictors (e.g., age, sex, cholesterol, current smoking, and body mass index) as inputs to provide personalized estimates of a patient\u0026rsquo;s risk of developing cardiovascular disease outcomes such as coronary heart disease or heart failure [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. A similar predictive model for depressive symptoms would represent an important step forward for the field, as more reliable identification of early warning signs of future depression (i.e., \u0026ldquo;red flags\u0026rdquo;) may mitigate MDD\u0026rsquo;s negative consequences by allowing the timely delivery of preventive interventions. In addition, a fuller understanding of the extent to which certain factors (e.g., demographic and clinical characteristics, environmental/social variables or neural features) each contribute to depression risk might allow for the development of algorithms that offer personalized prediction of future symptoms, with the potential for tailoring preventative interventions to an individual\u0026rsquo;s particular risk profile (e.g., intervention recommendations may differ if neuroticism vs. rumination vs. anhedonia were especially elevated in a given individual) [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn an effort to better understand and predict the onset of depression during adolescence, previous studies have identified a range of risk factors that prospectively predict depression (for a recent review see [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]). Studies have typically only examined one or a few risk factors in isolation. For instance, previous studies have identified a range of predictors of depression, including demographic variables such as age and gender [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e], clinical characteristics like depressive symptoms, anhedonia, anxiety symptoms, and rumination [\u003cspan additionalcitationids=\"CR9 CR10\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], socioeconomic status [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], pubertal status [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], as well as neural markers such as decreased striatal activation to reward [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Considering each variable in isolation may account for only limited variance in future depression outcomes. Since depression is a heterogeneous disorder emerging from a complex interplay of environmental/social, psychological, and biological factors, it is difficult to predict the onset of depression based on any single variable [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Therefore, there is a need for a more comprehensive, data-driven approach that considers multiple risk markers simultaneously to improve predictive accuracy and to better reflect the reality of depression\u0026rsquo;s multifaceted etiology.\u003c/p\u003e\u003cp\u003eRecently, machine learning techniques have gained traction in predicting depression. These approaches are well-suited to handling a large number of predictor variables (including multicollinearity among variables) and modeling complex relations such as interactions and other non-linear associations that traditional statistical methods might overlook (e.g., in ordinary least squares (OLS) regression if relevant interactions and non-linear relations are not explicitly specified). Prior studies using machine learning have shown promising results. For instance, a recent large study achieved area under the receiver operating characteristic curve (AUROC) values of 0.72\u0026ndash;0.74 in identifying depression in 6,310 children and adolescents aged 4\u0026ndash;17 [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. However, while these studies incorporated several predictors, they typically collected data within one modality (e.g., all self-reports) anddid not combine multi-modal data from several relevant sources [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e], for an exception see [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. An important next step involves determining if combining measures of depression-relevant characteristics assessed from various modalities \u0026ndash; such as conventional self-report questionnaires (e.g., assessing clinical, demographic, and personality characteristics), ecological momentary assessments (EMAs; e.g., affect dynamics in daily life), behavioral tasks (e.g., reward learning and cognitive control abilities), and neuroimaging (e.g., reward-related brain activation) \u0026ndash; can enhance our ability to predict depression in adolescents.\u003c/p\u003e\u003cp\u003eIn addition, most recent studies employing machine learning to examine multiple predictors of depression tested associations with \u003cem\u003ecurrent\u003c/em\u003e depression. Longitudinal studies are required to study the predictive value of these factors for the onset of future depressive symptoms. One exception is a recent study that used machine learning to combine various modalities such as clinical data, cognitive assessments, environmental factors, and structural magnetic resonance imaging (MRI) to predict depression onset at a 2-year and 5-year follow-up [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. The study used penalized logistic regression and found that baseline depressive symptoms, female sex, neuroticism, stressful life events, and the surface area of the supramarginal gyrus were the strongest predictors. The predictive model achieved an AUROC of 0.68\u0026ndash;0.72. However, penalized logistic regression, while effective in handling multicollinearity and overfitting, does not account for complex interactions and nonlinear associations (unless they are explicitly modeled). Other algorithms, such as decision-tree based models (e.g., random forest) are well-designed to do so and may yield better predictive performance, highlighting the value of comparing a range of machine learning approaches.\u003c/p\u003e\u003cp\u003eThe current study aims to address the abovementioned limitations by combining multimodal data, including depression-relevant self-reports/EMA (e.g., assessing neuroticism, rumination, stress, and affect dynamics), behavioral assessments (reward learning), and neural measures (neural reward sensitivity), to prospectively predict future depressive symptoms in youth. The integration of different data sources may also help in identifying novel predictors and interactions that may be important for early detection and intervention. In addition, the current study will compare various machine learning approaches, as well as conventional regression, to identify which best predicts depressive symptoms in adolescents.\u003c/p\u003e"},{"header":"Methods and Materials","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eParticipants\u003c/h2\u003e\u003cp\u003eParticipants were 103 English-speaking adolescents aged 12\u0026ndash;18 (75 female, 28 male; \u003cem\u003em\u003c/em\u003e\u003csub\u003eage\u003c/sub\u003e = 16.0, \u003cem\u003eSD\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.9) recruited from the greater Boston area for two larger studies that recruited typically developing (non-anhedonic) adolescents (n\u0026thinsp;=\u0026thinsp;68 included in the current study), as well as adolescents with elevated levels of anhedonia (n\u0026thinsp;=\u0026thinsp;35 included in the current study). Exclusion criteria included history or current diagnosis of any of the following DSM-5 psychiatric illnesses: major depressive disorder, schizophrenia spectrum or other psychotic disorder, bipolar disorder, substance or alcohol use disorder within the past 12 months or lifetime severe substance or alcohol use disorder, and current diagnosis of anorexia nervosa or bulimia nervosa. Psychotropic medications were exclusionary, with the exception of stable-dose (at least one month) selective serotonin reuptake inhibitors (SSRIs) (n\u0026thinsp;=\u0026thinsp;2; see Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). See Supplement for additional exclusion criteria.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eProcedure\u003c/h3\u003e\n\u003cp\u003eAll procedures were approved by the Mass General Brigham IRB. Written informed consent was provided by participants who were 18 years of age, as well as from parents of participants who were under 18 years of age, along with the participant\u0026rsquo;s written informed assent. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. At the baseline session, either in-person or over Zoom, participants were administered a semi-structured clinical interview, the Kiddie Schedule for Affective Disorders and Schizophrenia (K-SADS; [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]), and completed self-report measures. Following the baseline session, participants completed an MRI scan session including a Reward Task [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] functional MRI (fMRI) probing neural response to the anticipation and receipt of, loss of, or no change in monetary reward vs. losses. After the MRI session, participants performed a Probabilistic Reward Task (PRT; [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]) assessing reward learning. See Supplement for further details on MRI acquisition, task design, and MRI data processing. At the MRI scan session, participants installed the MetricWire App on their smartphones to complete EMAs asking about current positive affect (PA) and negative affect (NA). As described in Murray et al. [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003ca class=\"FNLink\" href=\"#Fn1\" id=\"#FNLinkFn1\"\u003e\u003c/a\u003e, following the fMRI scan session EMA surveys were delivered 2\u0026ndash;3 times per day whereby participants were randomly signaled during two timeslots (4pm to 6:30pm and 6:30pm to 9pm) during a 5-day period (Thursday - Monday). A third survey was sent on weekends (11am-4pm). Three months after the baseline session, participants completed the baseline self-report measures again, including depression measures (see below).\u003c/p\u003e\u003cp\u003e\u003cb\u003eOutcome Measures\u003c/b\u003e: Two self-report depression outcome measures were assessed at the 3-month follow-up: Center for Epidemiological Studies Depression Scale (CES-D; [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]) total score, assessing past-week symptoms, and the Mood and Feelings Questionnaire \u0026ndash; Child Self-Report \u0026ndash; Long Version (MFQ; [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]) total score, assessing symptoms over the past two weeks. Both measures are widely used but differ on several dimensions, including content (e.g., the MFQ has greater coverage of DSM-5 symptoms), sensitivity (the CESD has been shown to be relatively sensitive in detecting depressive symptoms at the lower severity range), and timeframe (assessing symptoms over the past week vs. past 2 weeks) [\u003cspan additionalcitationids=\"CR27\" citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e\n\u003ch3\u003eBaseline Predictor Variables\u003c/h3\u003e\n\u003cp\u003e\u003cstrong\u003eClinical self-report scales\u003c/strong\u003e\u003cp\u003eDepression (CES-D and MFQ), anhedonia (Snaith-Hamilton Pleasure Scale [SHAPS; [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]] total score), rumination (Children\u0026rsquo;s Response Styles Questionnaire [CRSQ; [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]] rumination subscale score), perceived stress (Perceived Stress Scale [PSS; [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]] total score), physical anxiety (Multidimensional Anxiety Scale for Children [MASC; [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]] subscale score), social anxiety (MASC social subscore), separation anxiety (MASC separation subscore), harm avoidance (MASC harm subscore), negative life events (Adolescent Life Event Questionnaire-Revised [ALEQ-R; [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]] total score), extraversion (NEO Five-Factor Inventory-3 [NEO-FFI-3; [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]] extraversion factor), and neuroticism (NEO-FFI-3 neuroticism factor).\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eEcological Momentary Assessment\u003c/b\u003e: Participants rated emotions on a 5-point Likert scale, with mean PA and NA calculated from respective emotions. See Supplement for full details. Several affect dynamic measures previously linked to depression were also included: the variability in PA and NA were measured using the standard deviation (SD) and mean square successive difference (MSSD; [\u003cspan additionalcitationids=\"CR36 CR37\" citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]), and we also included temporal dependency (autocorrelation; [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]) of PA and NA as a measure of emotional inertia.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eClinical Interview\u003c/strong\u003e\u003cp\u003eParticipants were administered the K-SADS. As noted above, we included participants from a no anhedonia vs. elevated anhedonia group (binary variable). Elevated anhedonia was defined as having a K-SADS anhedonia item (from the MDD module) score greater than one.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eNeuroimaging\u003c/strong\u003e\u003cp\u003eParticipants completed an fMRI monetary reward task assessing neural responses during reward anticipation and outcome [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. See Supplement for task details.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eBehavioral\u003c/b\u003e: A reward learning task (Probabilistic Reward Task [PRT]) previously validated in adolescents [\u003cspan additionalcitationids=\"CR42\" citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e] was used. We included two variables as predictors: reward learning (change in response bias over the course of the task computed as Response Bias in block 2 minus Response Bias in block 1) and mean response bias (averaging the response bias score over the two blocks). For task details, see [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e].\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eDemographic\u003c/strong\u003e\u003cp\u003eSelf-reported age and sex.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003cstrong\u003ePubertal\u003c/strong\u003e\u003cp\u003ePubertal status (Tanner Staging Questionnaire [TSQ; [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]] total score).\u003c/p\u003e\u003c/p\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003eStatistical Analysis\u003c/h2\u003e\u003cdiv id=\"Sec7\" class=\"Section3\"\u003e\u003ch2\u003eConcurrent correlations with depressive symptoms at baseline\u003c/h2\u003e\u003cp\u003ePrior to testing the relation between baseline predictors and future (3 month) depressive (CESD and MFQ) symptoms, we first examined the concurrent correlations between depression scores and other predictor variables at baseline (see \u003cb\u003eFig.\u0026nbsp;1\u003c/b\u003e for correlation coefficients and \u003cem\u003ep\u003c/em\u003e-values).\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003ePredicting future (3 months) depressive symptoms\u003c/h2\u003e\u003cp\u003eSee Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003e for a schematic of the data processing and modeling pipeline.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eData Preprocessing\u003c/h3\u003e\n\u003cp\u003eWe standardized all variables to have means of 0 and standard deviations of 1 (Data type: \u003cem\u003eRaw\u003c/em\u003e). To reduce the potential instability in model training due to the high correlations between the baseline predictors, we re-ran analyses applying two separate approaches to data pre-processing: (1) To remove the confounding influence of baseline depression severity from our other predictors, for each variable other than the baseline depression scores (MFQ or CESD), we regressed the variable against baseline depression and took the residuals as the feature value. We trained the models using the residualized features along with baseline depression (Data type: \u003cem\u003eResiduals\u003c/em\u003e). (2) To reduce the number of predictors, we applied principal component analysis (PCA) to all baseline predictors (across all modalities), then evaluated the first 5, 10, or 20 principal component scores as features (Data type: \u003cem\u003ePC5, PC10, PC20\u003c/em\u003e). See \u0026ldquo;data type\u0026rdquo; column in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. See \u003cb\u003eSupplementary Fig.\u0026nbsp;3\u003c/b\u003e for the cumulative variance explained by the top principal components.\u003c/p\u003e\n\u003ch3\u003eModel Training and Validation\u003c/h3\u003e\n\u003cp\u003eWe fit five different predictive models with all baseline predictors: (1) conventional linear regression, (2) linear regression with elastic net penalty [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e], (3) random forest [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e], (4) XGBoost [\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e], and (5) Support vector machine (SVM, [\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e]). We further trained an ensemble learning model [\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e] which combines predictions from the base models (linear regression, elastic net, random forest, XGBoost and SVM) in an effort to maximize predictive performance. We also compared these models to a univariable model with baseline depression (CESD or MFQ) total score as the only predictor. We fit a linear regression to predict the depression outcome at 3 months with the corresponding baseline measure as the sole predictor variable. To perform model training with hyperparameter optimization while robustly evaluating the model performance, we performed \u003cem\u003enested\u003c/em\u003e cross-validation (CV; see Supplement for details). Preprocessing steps (standardization, residualization, and PCA) were performed within the training folds during nested cross-validation. This ensures that no information from the test folds was used to inform any transformation. See Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e (also see Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e3\u003c/span\u003e) for results.\u003c/p\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eFeature Importance\u003c/h2\u003e\u003cp\u003eTo assess the most influential features in the multivariable model for the prediction of depression, SHapley Additive exPlanations (SHAP) were implemented, and the top ten most influential features were reported for the random forest model, which achieved the highest performance using standardized predictors (data type: \u003cem\u003eRaw\u003c/em\u003e). See Supplement for additional details.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003eFactor Analysis\u003c/h2\u003e\u003cp\u003eWe conducted two exploratory factor analyses (EFA) to examine factors underlying the depression questionnaire (MFQ and CESD) items at baseline, in an effort to determine which depression features, if any, were most predictive of future depressive symptoms. See Supplement for details.\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003eParticipant characteristics\u003c/h2\u003e\u003cp\u003eParticipants were predominantly White (\u003cem\u003en\u0026thinsp;=\u003c/em\u003e\u0026thinsp;64, 62.1%), non-Hispanic (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;92, 89.3%) teens, most of whom were assigned female at birth (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;75, 72.8%). Baseline anhedonia (SHAPS: \u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;23.0, \u003cem\u003eSD\u003c/em\u003e\u0026thinsp;=\u0026thinsp;7.4) and depression symptoms (CES-D: \u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;12.3, \u003cem\u003eSD\u003c/em\u003e\u0026thinsp;=\u0026thinsp;11.1) were relatively mild in severity. We report additional demographic and clinical characteristics in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. At the 3-month timepoint, CES-D scores were slightly higher (\u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;13.3, \u003cem\u003eSD\u003c/em\u003e\u0026thinsp;=\u0026thinsp;13.1) than at the baseline assessment, but below the conventional cutoff score of 16 [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e], with substantial variability at both timepoints.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003eBaseline correlations\u003c/h2\u003e\u003cp\u003eWe report Spearman correlation coefficient between depression symptoms and other predictive variables at baseline, along with the p-values in \u003cb\u003eFig.\u0026nbsp;1\u003c/b\u003e. Both depression measures (CESD and MFQ) exhibit statistically significant correlations with a range of self-report measures: anhedonia (SHAPS), anxiety facets (MASC physical, social, and separation anxiety subscores), rumination (CRSQ) perceived stress (PSS), negative life events (ALEQR), and neuroticism (NEO), ranging from \u003cem\u003er\u0026thinsp;=\u0026thinsp;0.4 to 0.75\u003c/em\u003e. Strong negative correlations were found between both baseline depression measures and extraversion (NEO) (CESD: r = -0.58, p\u0026thinsp;=\u0026thinsp;2e-10; MFQ: r = -0.58, p\u0026thinsp;=\u0026thinsp;1e-10). Among the EMA measures, lower mean PA and higher mean \u003cem\u003eand\u003c/em\u003e variability in NA (both SD and MSSD) were significantly associated with higher depression. Among the neuroimaging variables, only blunted striatal (left caudate) response to the anticipation of rewards was significantly correlated with greater depressive symptoms on both measures. Our behavioral measure of reward learning (PRT) was not correlated with depression (nor anhedonia).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003ePrincipal Component Analysis\u003c/h2\u003e\u003cp\u003eThe PCA decomposition (see \u003cb\u003eSupplementary Fig.\u0026nbsp;3\u003c/b\u003e) showed that the first 5, 10, and 20 components explained 55.6%, 74.8%, and 91.3% of the total variance, respectively, and loadings of the first 5 PCs are summarized in \u003cb\u003eSupplementary Fig.\u0026nbsp;4\u003c/b\u003e.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\u003ch2\u003eModel Performance\u003c/h2\u003e\u003cp\u003eFor the model predicting future (3-month) depressive symptoms as measured by the CESD, the best performing model based on RMSE and R2 was the multivariable linear regression model with the first 5 PCs as features (RMSE\u0026thinsp;=\u0026thinsp;6.501, R2\u0026thinsp;=\u0026thinsp;0.688, Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e \u003cb\u003eand\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e3\u003c/span\u003e). It achieved better performance compared with the univariable model using baseline CESD as the only predictor (RMSE\u0026thinsp;=\u0026thinsp;6.750, R2\u0026thinsp;=\u0026thinsp;0.645). Across all models using \u003cem\u003eraw\u003c/em\u003e features as input, the random forest model and ensemble approach had the best prediction accuracy, achieving R2 of 0.614 and 0.616, and RMSE of 6.723 and 6.738, respectively.\u003c/p\u003e\u003cp\u003eFor the MFQ model, the best performing model based on RMSE and R2 was the simplest model: the univariable linear regression model with baseline MFQ total scores as the sole predictor (RMSE\u0026thinsp;=\u0026thinsp;8.054, R2\u0026thinsp;=\u0026thinsp;0.671). The best multivariable model was also the linear regression model with the first 5 PCs as features (RMSE\u0026thinsp;=\u0026thinsp;8.128, R2\u0026thinsp;=\u0026thinsp;0.647), but the performance was worse than the univariable model. Using raw features as input, the random forest model achieved the best prediction accuracy, with R2 of 0.562 and RMSE of 8.616.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\u003ch2\u003eFeature Importance\u003c/h2\u003e\u003cp\u003eTo evaluate feature importance of each predictor variable, we fit a random forest model using all raw features on the full dataset (i.e. outside of the nested cross-validation framework). We focused on the random forest model since that raw feature model performed the best across the two depression outcome measures. For the model predicting CESD scores at 3 months, depressive (MFQ) symptom severity at baseline appears as the most influential feature, with a mean SHAP value five times greater than the second most influential feature, the baseline CESD score (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e4\u003c/span\u003ea). Other features such as greater physical anxiety (MASC), anhedonia (SHAPS), and blunted striatal (right caudate) response to the anticipation of rewards also demonstrated predictive influence.\u003c/p\u003e\u003cp\u003eFor the model predicting MFQ scores at 3 months, greater baseline depressive (MFQ) symptoms again emerged as the most important contributor to the predicted outcomes (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e4\u003c/span\u003eb). The set of top influential features shows some overlap with those important for predicting MFQ scores, including greater physical anxiety (MASC), neuroticism (NEO), and blunted striatal (right caudate and putamen) response to the anticipation of rewards.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e\u003ch2\u003eDepression Symptom Factors\u003c/h2\u003e\u003cp\u003eGiven that baseline depression was the most robust predictor of future depression we sought to understand if there were particular factors or subsets of depression symptoms/items that were most predictive. For the CESD questionnaire, the first factor described depressed mood. The second factor was related to depressogenic social cognitions (e.g., \u0026ldquo;\u003cem\u003eI felt that people disliked me\u003c/em\u003e\u0026rdquo;). The third factor described anhedonia, while the last factor was related to sleep issues and fear (\u003cb\u003eSupplementary Fig.\u0026nbsp;6a\u003c/b\u003e). Within the MFQ questionnaire, the first factor reflected items related to melancholia, characterized by psychomotor changes (restlessness and psychomotor retardation), sleep difficulties, and difficulty concentrating. The second factor was related to low self-esteem and self-deprecation. The third factor points to items related to suicidal ideation, while the fourth factor included anhedonia symptoms and hopelessness (\u003cb\u003eSupplementary Fig.\u0026nbsp;6b\u003c/b\u003e).\u003c/p\u003e\u003cp\u003eWe explored whether certain factors were more strongly related to future depression. For the CESD model, only Factor 2 scores (depressogenic social cognitions) significantly correlated with the total CESD scores at 3 months (r\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.16, p\u0026thinsp;=\u0026thinsp;0.0014, \u003cb\u003eSupplementary Fig.\u0026nbsp;6a\u003c/b\u003e). In the MFQ model, only factor 1 (melancholia) baseline scores significantly correlated with total MFQ scores at three months (r\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.33, p\u0026thinsp;=\u0026thinsp;1.3e-6, \u003cb\u003eSupplementary Fig.\u0026nbsp;6b\u003c/b\u003e).\u003c/p\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eEarly identification of depression risk is crucial for the timely implementation of prevention strategies, particularly during adolescence, a developmental period marked by heightened vulnerability to depressive symptoms. A key challenge in predicting future depression is its complex, multidimensional nature. Depression is influenced by a wide array of factors spanning psychological, behavioral, biological, and environmental domains, making it important to integrate data from multiple modalities when developing predictive models of future symptoms.\u003c/p\u003e\u003cp\u003eIn this study, we explored machine learning approaches to predict future depressive symptoms in youth by integrating a wide range of theoretically-relevant predictors, including from conventional self-report scales (e.g., neuroticism and anhedonia), ecological momentary assessments (EMA; e.g., mean and variance in negative and positive affect), behavioral tasks (e.g., reward responsiveness), neuroimaging data (e.g., reward circuitry response), demographic characteristics (e.g., age and sex), and developmental measures (e.g., puberty status). Additionally, we employed machine learning (ML) techniques, combined with dimensionality reduction of baseline predictors, to enhance model accuracy and interpretability. In an effort to further improve predictive performance, this approach not only integrated numerous predictors, but also captured potential non-linear effects and interactions that studies using fewer predictors might overlook. Our use of a multivariable approach, combined with an analysis of feature importance, enabled a detailed assessment of which predictors exerted the most significant influence on depression outcomes.\u003c/p\u003e\u003cp\u003eFor the CES-D, the multivariable linear regression model that used the first five principal components as predictors achieved the best performance, with an RMSE of 6.501 and an R\u0026sup2; of 0.688. This model outperformed the univariable approach, which only included baseline CES-D scores as the sole predictor. Previous research on predicting depression outcomes in treatment-seeking populations has similarly shown that multivariable models outperform univariable ones [\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e]. Our findings contribute to this body of evidence by extending these results to an adolescent sample recruited from the community, using a wide range of predictors.\u003c/p\u003e\u003cp\u003eIn contrast, for the MFQ, the best-performing model was a simple univariable linear regression that used baseline MFQ scores as the sole predictor, achieving an RMSE of 8.054 and an R\u0026sup2; of 0.671. In line with this, feature importance analysis in the multivariable models indicated that baseline depressive symptoms, as measured by the MFQ, was the most robust predictor of future depressive symptoms. This finding aligns with previous research [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] and highlights the critical role that the severity of initial symptoms plays in predicting future depression outcomes (at least within the relatively short timeframe of 3 months). This emphasizes the importance of early symptom assessment as a key factor in forecasting long-term mental health trajectories.\u003c/p\u003e\u003cp\u003eInterestingly, feature importance showed that baseline MFQ was the best predictor of both CES-D and MFQ outcomes, even exceeding the predictive value of baseline CES-D for future CES-D scores. One explanation is that the MFQ is specifically designed for adolescents and covers a broader range of symptoms highly relevant to this age group. In addition, because the MFQ assesses symptoms over a two-week period rather than one week, it may provide a more stable index of severity. Together, these features may capture more trait-like aspects of adolescent depression vulnerability, whereas the CES-D may be somewhat more sensitive to short-term, state-related changes in symptoms.\u003c/p\u003e\u003cp\u003eWhile the univariable model using baseline depressive symptoms as the sole predictor performed best for MFQ, a random forest model incorporating all raw features also showed relatively strong performance in predicting both CES-D and MFQ depression scores. Feature importance revealed that the variables contributing most strongly to depression predictions were from conventional self-report measures (e.g., baseline depression severity, neuroticism and physical anxiety), EMA (mean positive affect) and neuroimaging (striatal response to the anticipation of rewards). These findings reinforce the multifaceted nature of depression, where a combination of psychological, biological, and behavioral factors contribute to its trajectory.\u003c/p\u003e\u003cp\u003eThe observed importance of physical anxiety and neuroticism emphasizes the need to incorporate psychological characteristics into predictive models, which are relatively easily assessed via low-cost self-report measures, supporting the idea that depression risk is influenced by a constellation of internal and external factors [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. These results align with previous research showing that depression is prospectively predicted by psychological variables, particularly neuroticism [\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e]. Moreover, findings also align with recent work suggesting that a blunted striatal response to reward is a hallmark of maladaptive reward processing in major depressive disorder [\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e], which may precede and predict future depression in youth [\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e]. However, when comparing across modalities, the bulk of the top predictors were self-report measures (CESD: 6/10 conventional self-report questionnaires and 1/10 EMA; MFQ: 6/10 conventional self-report questionnaires and 2/10 EMA). The relatively poor predictive performance of neural and behavioral measures may be due at least in part to fundamental differences in measurement reliability and error characteristics across modalities, as self-report measures have been shown to have higher reliability than single-session behavioral [\u003cspan additionalcitationids=\"CR55 CR56\" citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e] or neuroimaging tasks [\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e]. Lower reliability in these tasks is in part due to low between-person variance, which can attenuate predictor-outcome associations. This reflects a fundamental design feature of experimental tasks, which are optimized to detect robust group-level condition effects at the expense of minimizing individual differences (which ultimately hampers reliability). Also, laboratory-based neural and behavioral measures may not capture real-world functioning as effectively as self-reported symptoms and traits. These findings highlight both the robust predictive value of carefully selected self-report measures and the need for improved approaches to capturing meaningful individual differences in neural and behavioral functioning.\u003c/p\u003e\u003cp\u003eGiven the robust role of baseline depressive symptom severity in predicting future symptoms, with the univariable model performing the best for predicting MFQ scores, and baseline depression being the strongest predictor of CES-D scores in that multivariable model, we wanted to more closely examine which elements of depression were driving these effects. Therefore, we conducted a factor analysis of baseline MFQ and CES-D scores. Our analysis identified several distinct dimensions of depressive symptoms, including melancholia, low self-esteem, suicidal ideation, and anhedonia. These factors resemble those uncovered in previous research using the MFQ, namely vegetative symptoms, suicidality, cognitive symptoms, and agitated distress [\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e, \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e]. The strong correlation of the melancholia MFQ factor with future MFQ total scores, and the depressogenic social cognition CES-D factor with future CES-D total scores, suggest that it may be important to emphasize targeting these specific symptom dimensions in preventive interventions.\u003c/p\u003e\u003cp\u003eOur results showed that complex ML approaches did not outperform regression-based techniques, aligning with prior work [\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e]. However, tree-based machine learning models such as random forest performed relatively well and provided complementary insights via SHAP-based feature importance analyses. These models may still offer meaningful value, particularly in larger samples or when interactions and nonlinear relationships are more prominent. More research is needed to determine in which contexts ML approaches are likely to outperform simpler statistical models (e.g., see [\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e61\u003c/span\u003e, \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e62\u003c/span\u003e] for examples of ML models incorporating passive smartphone sensor data which outperformed simpler statistical approaches in predicting negative affective states). Future studies could benefit from assessing other depression-relevant baseline individual characteristics which were not assessed in this study that may improve predictive performance (e.g., emotion regulation abilities, sleep habits, home and school environment, early adversity, or executive function). Furthermore, there are of course factors in the intervening period \u003cem\u003ebetween\u003c/em\u003e baseline and follow-up assessments that were not captured in our study which are known to be robust predictors of depression (e.g., stressors and negative life events, especially in the interpersonal domain).\u003c/p\u003e\u003cp\u003eThe current study had several limitations. One notable limitation of our study is the relatively small sample size, in particular for the multivariable ML models. Although this risk was somewhat mitigated through dimensionality reduction of baseline predictors and the use of penalized ML models, a larger sample size is needed to more reliably estimate these models [\u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e63\u003c/span\u003e, \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e64\u003c/span\u003e]. Relatedly, while we used a nested cross-validation approach to mitigate overfitting, external validation in a larger, independent sample will be important to further evaluate the generalizability of these findings. Second, although our current sample is relatively diverse (38% non-White), it may not fully represent the broader adolescent population, which could limit generalizability. Another limitation is the short follow-up period, with only a single follow-up assessment after three months. This design restricts our ability to capture longer-term depression symptoms [\u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e65\u003c/span\u003e] and assess the stability of the predictive factors over time. Additionally, our study used two different depression outcome measures, CES-D and MFQ, analyzing them separately. This approach may complicate clinical interpretation and the integration of findings. Future research should explore methods for combining or integrating different outcome measures (including clinical interviews) to enhance the clinical applicability of predictive models and provide a clearer understanding of depression dynamics.\u003c/p\u003e\u003cp\u003eThese limitations notwithstanding, the present study demonstrated the utility of models that integrate diverse predictors from psychological, behavioral, developmental, and neural assessments to characterize risk for future depression and their relative predictive importance. Further research is needed to test the generalizability of these models across multiple time scales and populations.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eDemographic and Clinical Characteristics of the Sample\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e\u003cp\u003eSample Characteristics\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eN\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e%\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBiological Sex\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFemale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e75\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e72.8\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e28\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e27.2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eRace\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAmerican Indian or Alaska Native\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.0\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAsian\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e13\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e12.6\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBlack or African American\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e11\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e10.7\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNative Hawaiian or Other Pacific Islander\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.0\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWhite\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e62.1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eOther\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2.9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMore than one race\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e9.7\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eUnknown race\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.0\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eEthnicity\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHispanic or Latino\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e9.7\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNot Hispanic or Latino\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e92\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e89.3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eUnknown Ethnicity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.0\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eCurrent Diagnoses (DSM-V)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMajor Depressive Episode\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.0\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGeneralized Anxiety Disorder\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e4.9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSocial Anxiety Disorder\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2.9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePanic Disorder\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.0\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSpecific Phobia\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.0\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAttention-Deficit / Hyperactivity Disorder\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2.9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eObsessive Compulsive Disorder\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eMedication\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSSRI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cb\u003eM\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003eSD\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eAge (in years)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e16.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eFamily Income (dollars)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e165,844.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e92,242.4\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBaseline SHAPS Score\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e23.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e7.4\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBaseline CES-D Score\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e12.3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e11.1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cem\u003eNote\u003c/em\u003e. The SHAPS is scored on a 1\u0026ndash;4 scale where higher scores indicate greater anhedonia; the possible range is 14\u0026ndash;56. The CES-D is scored on a 0\u0026ndash;3 scale where higher scores indicate greater depression; the possible range is 0\u0026ndash;60.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003e\u003cb\u003eModel performances for prediction of depressive symptoms (CESD and MFQ) at 3 months follow-up.\u003c/b\u003e Results are summarized over 30 repetitions. RMSE: root mean squared error, R2: R squared, MAE: mean absolute error. Univariable: model with respective baseline depression as the sole predictor, Raw: all normalized baseline variables as predictors, Residuals: residualized baseline variables as predictors, PC5: first 5 principal component scores, PC10: first 10 principal component scores, PC20: first 20 principal component scores. The best performance (highest R2, lowest RMSE and MAE) across all data pre-processing procedures are bolded, and the best performance with \u0026ldquo;Raw\u0026rdquo; features are underlined.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"8\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eData type\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"3\" nameend=\"c5\" namest=\"c3\"\u003e\u003cp\u003eCESD\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"3\" nameend=\"c8\" namest=\"c6\"\u003e\u003cp\u003eMFQ\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eR2\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRMSE\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMAE\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eR2\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eRMSE\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c8\"\u003e\u003cp\u003eMAE\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eUnivariable\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.645\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e6.750\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e4.908\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e0.671\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e\u003cb\u003e8.054\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.769\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"5\" rowspan=\"6\"\u003e\u003cp\u003eMultivariable- Raw\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.502\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e8.589\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e6.631\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.469\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e10.812\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e8.210\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eElastic Net\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.570\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e7.098\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.217\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.521\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e9.019\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e6.002\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.614\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e6.723\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e4.808\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e0.562\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e8.616\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.764\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eXGBoost\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.545\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e7.476\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.209\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.543\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e9.039\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.966\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.510\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e8.134\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.937\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.459\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e9.985\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e7.101\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eEnsemble\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e0.616\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e6.738\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e4.819\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.549\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e8.755\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e5.697\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"5\" rowspan=\"6\"\u003e\u003cp\u003eMultivariable-Residuals\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.323\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e8.891\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e6.574\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.268\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e11.11\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e7.941\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eElastic Net\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.540\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e7.327\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.451\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.524\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e8.969\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e6.117\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.503\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e7.621\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.447\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.558\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e8.635\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.893\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eXGBoost\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.332\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e8.833\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e6.359\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.464\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e9.510\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e6.459\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.396\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e8.395\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e6.215\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.378\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e10.249\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e7.331\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eEnsemble\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.544\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e7.301\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.302\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.541\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e8.805\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.849\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"5\" rowspan=\"6\"\u003e\u003cp\u003eMultivariable-PC 5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003e0.688\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e6.501\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e4.873\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.647\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e8.128\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.931\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eElastic Net\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.591\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e6.912\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e4.919\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.511\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e9.093\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.937\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.606\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e6.785\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e4.932\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.535\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e8.866\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.961\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eXGBoost\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.463\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e7.891\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.651\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.343\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e10.523\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e6.753\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.602\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e6.823\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e4.806\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.509\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e9.106\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.817\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eEnsemble\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.600\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e6.836\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e4.846\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.528\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e8.933\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.931\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"5\" rowspan=\"6\"\u003e\u003cp\u003eMultivariable-PC 10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.588\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e7.204\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e4.823\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.521\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e9.356\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.812\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eElastic Net\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.584\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e6.976\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.014\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.534\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e8.869\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.792\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.604\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e6.807\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.022\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.541\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e8.801\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.883\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eXGBoost\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.453\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e7.982\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.672\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.426\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e9.792\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e6.404\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.616\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e6.705\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e4.708\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.557\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e8.649\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e\u003cb\u003e5.450\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eEnsemble\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.623\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e6.638\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e4.685\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.547\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e8.752\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.683\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"5\" rowspan=\"6\"\u003e\u003cp\u003eMultivariable-PC 20\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.546\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e7.468\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.078\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.477\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e9.248\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e6.289\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eElastic Net\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.562\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e7.156\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.312\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.529\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e8.922\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.811\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.580\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e7.004\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.254\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.506\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e9.132\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e6.234\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eXGBoost\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.425\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e8.190\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.911\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.364\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e10.341\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e6.779\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.589\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e6.929\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.007\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.550\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e8.722\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.662\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eEnsemble\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.575\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e7.046\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5.096\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.522\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e8.982\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e5.959\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003ch2\u003eEthics approval and consent to participate\u003c/h2\u003e\u003cp\u003eAll procedures were approved by the Mass General Brigham IRB. Written informed consent was provided by participants who were 18 years of age, as well as from parents of participants who were under 18 years of age, along with the participant\u0026rsquo;s written informed assent. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eConsent for publication:\u003c/strong\u003e\u003cp\u003eNot applicable\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003ch2\u003eCompeting Interests\u003c/h2\u003e\u003cp\u003eOver the past three years, Dr. Pizzagalli has received consulting fees from Boehringer Ingelheim, Compass Pathways, Engrail Therapeutics, Neumora Therapeutics (formerly BlackThorn Therapeutics), Neurocrine Biosciences, Neuroscience Software, Sage Therapeutics, Sama Therapeutics, and Takeda; he has received honoraria from the American Psychological Association, Psychonomic Society and Springer (for editorial work) and from Alkermes; he has received research funding from the BIRD Foundation, Brain and Behavior Research Foundation, Dana Foundation, DARPA, Millennium Pharmaceuticals, NIMH and Wellcome Leap MCPsych; he has received stock options from Compass Pathways, Engrail Therapeutics, Neumora Therapeutics, and Neuroscience Software. Dr. Pizzagalli has a financial interest in Neumora Therapeutics, which has licensed the copyright to the probabilistic reward task through Harvard University. Dr. Webb has received consulting fees from King \u0026amp; Spalding law firm. Dr. Pizzagalli\u0026rsquo;s and Dr. Webb\u0026rsquo;s interests were reviewed and are managed by McLean Hospital and Mass General Brigham in accordance with their conflict of interest policies. No funding from these entities was used to support the current work, and all views expressed are solely those of the authors. The other authors declare no competing financial interests.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e\u003cp\u003eThis research was supported by NIMH K23MH108752, the Tommy Fuss Fund, a NARSAD Young Investigator Grant from the Brain \u0026amp; Behavior Research Foundation and the Klingenstein Third Generation Foundation (CAW). CAW was partially supported by NIMH R01MH116969 and NCCIH R01AT011002. DAP was partially supported by the National Institute of Mental Health grants P50MH119467 and R014R37MH068376-17.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eCW acquired funding for the project. CW, LW, NZ, DP, and EF conceptualized the study. CW, NJ, KP, HF, and AT acquired and processed the data. CW, LW and NZ processed and analyzed the data. All authors contributed to interpretation of findings and drafting the manuscript. All authors approved the final version of the manuscript. Since completion of this study, DP has moved to the University of California, Irvine.\u003c/p\u003e\u003ch2\u003eAcknowledgements:\u003c/h2\u003e\u003cp\u003eNone\u003c/p\u003e\u003cp\u003e\u003cb\u003eClinical trial number\u003c/b\u003e: not applicable\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eMass General Brigham requires IRB approval and a signed Data Use Agreement for data sharing. Please contact the first author.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eTwenge JM, Cooper AB, Joiner TE, Duffy ME, Binau SG. Age, period, and cohort trends in mood disorder indicators and suicide-related outcomes in a nationally representative dataset, 2005\u0026ndash;2017. J Abnorm Psychol. 2019;128:185\u0026ndash;99.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMadigan S, Racine N, Vaillancourt T, Korczak DJ, Hewitt JMA, Pador P, et al. Changes in Depression and Anxiety Among Children and Adolescents From Before to During the COVID-19 Pandemic: A Systematic Review and Meta-analysis. JAMA Pediatr. 2023;177:567\u0026ndash;81.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLarkin H. New Cardiovascular Disease Risk Calculator Could Eliminate the Need for Statins for Millions. JAMA. 2024;332:186\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWebb CA, Dillon DG, Pechtel P, Goer FK, Murray L, Huys QJ, et al. Neural Correlates of Three Promising Endophenotypes of Depression: Evidence from the EMBARC Study. Neuropsychopharmacology. 2016;41:454\u0026ndash;63.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHankin BL, Griffith JM. What Do We Know About Depression Among Youth and How Can We Make Progress Toward Improved Understanding and Reducing Distress? A New Hope. Clin Child Fam Psychol Rev. 2023;26:919\u0026ndash;42.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eD\u0026uuml;rrbaum T, Sattler FA. Minority stress and mental health in lesbian, gay male, and bisexual youths: A meta-analysis. J LGBT Youth. 2020;17:298\u0026ndash;314.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePellicane MJ, Ciesla JA. Associations between minority stress, depression, and suicidal ideation and attempts in transgender and gender diverse (TGD) individuals: Systematic review and meta-analysis. Clin Psychol Rev. 2022;91:102113.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAbela JRZ, Hankin BL. Rumination as a vulnerability factor to depression during the transition from early to middle adolescence: A multiwave longitudinal study. J Abnorm Psychol. 2011;120:259\u0026ndash;71.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBennik EC, Nederhof E, Ormel J, Oldehinkel AJ. Anhedonia and depressed mood in adolescence: course, stability, and reciprocal relation in the TRAILS study. Eur Child Adolesc Psychiatry. 2014;23:579\u0026ndash;86.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKeenan K, Feng X, Hipwell A, Klostermann S. Depression begets depression: Comparing the predictive utility of depression and anxiety symptoms to later depression. J Child Psychol Psychiatry. 2009;50:1167\u0026ndash;75.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eStein MB, Fuetsch M, M\u0026uuml;ller N, H\u0026ouml;fler M, Lieb R, Wittchen HU. Social anxiety disorder and the risk of depression: a prospective community study of adolescents and young adults. Arch Gen Psychiatry. 2001;58:251\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eReiss F. Socioeconomic inequalities and mental health problems in children and adolescents: A systematic review. Soc Sci Med. 2013;90:24\u0026ndash;31.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHamlat EJ, McCormick KC, Young JF, Hankin BL. Early pubertal timing predicts onset and recurrence of depressive episodes in boys and girls. J Child Psychol Psychiatry. 2020;61:1266\u0026ndash;74.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNelson SK, Layous K, Cole SW, Lyubomirsky S. Do unto others or treat yourself? The effects of prosocial and self-focused behavior on psychological flourishing. Emot Wash DC. 2016;16:850\u0026ndash;61.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eToenders YJ, van Velzen LS, Heideman IZ, Harrison BJ, Davey CG, Schmaal L. Neuroimaging predictors of onset and course of depression in childhood and adolescence: A systematic review of longitudinal studies. Dev Cogn Neurosci. 2019;39:100700.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFried EI. The 52 symptoms of major depression: Lack of content overlap among seven common depression scales. J Affect Disord. 2017;208:191\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHaque UM, Kabir E, Khanam R. Detection of child depression using machine learning methods. PLoS ONE. 2021;16:e0261131.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHighland D, Zhou G. A review of detection techniques for depression and bipolar disorder. Smart Health. 2022;24:100282.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eToenders YJ, Kottaram A, Dinga R, Davey CG, Banaschewski T, Bokde ALW, et al. Predicting Depression Onset in Young People Based on Clinical, Cognitive, Environmental, and Neurobiological Data. Biol Psychiatry Cogn Neurosci Neuroimaging. 2022;7:376\u0026ndash;84.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKaufman J, Birmaher B, Brent D, Rao U, Flynn C, Moreci P, et al. Schedule for Affective Disorders and Schizophrenia for School-Age Children-Present and Lifetime Version (K-SADS-PL): initial reliability and validity data. J Am Acad Child Adolesc Psychiatry. 1997;36:980\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eForbes EE, Hariri AR, Martin SL, Silk JS, Moyles DL, Fisher PM, et al. Altered striatal activation predicting real-world positive affect in adolescent major depressive disorder. Am J Psychiatry. 2009;166:64\u0026ndash;73.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePizzagalli DA, Jahn AL, O\u0026rsquo;Shea JP. Toward an objective characterization of an anhedonic phenotype: a signal-detection approach. Biol Psychiatry. 2005;57:319\u0026ndash;27.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMurray L, Israel ES, Balkind EG, Pastro B, Lovell-Smith N, Lukas SE et al. Multi-modal assessment of reward functioning in adolescent anhedonia. Psychol Med. 2022:1\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRadloff LS, The CES-D, Scale. A Self-Report Depression Scale for Research in the General Population. Appl Psychol Meas. 1977;1:385\u0026ndash;401.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCostello EJ, Angold A. Scales to assess child and adolescent depression: Checklists, screens, and nets. J Am Acad Child Adolesc Psychiatry. 1988;27:726\u0026ndash;37.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOlino TM, Yu L, Klein DN, Rohde P, Seeley JR, Pilkonis PA, et al. Measuring depression using item response theory: An examination of three measures of depressive symptomatology. Int J Methods Psychiatr Res. 2012;21:76\u0026ndash;85.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOlino TM, Yu L, McMakin DL, Forbes EE, Seeley JR, Lewinsohn PM, et al. Comparisons across depression assessment instruments in adolescence and young adulthood: An Item Response Theory study using two linking methods. J Abnorm Child Psychol. 2013;41:1267\u0026ndash;77.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVilar A, S\u0026aacute;nchez-Mart\u0026iacute;nez N, Blasco MJ, \u0026Aacute;lvarez-Salazar S, Batlle Vila S, Forero G. Content agreement of depressive symptomatology in children and adolescents: a review of eighteen self-report questionnaires. Eur Child Adolesc Psychiatry. 2024;33:2019\u0026ndash;33.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSnaith RP, Hamilton M, Morley S, Humayan A, Hargreaves D, Trigwell P. A scale for the assessment of hedonic tone the Snaith-Hamilton Pleasure Scale. Br J Psychiatry J Ment Sci. 1995;167:99\u0026ndash;103.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAbela JRZ, Verroneau M, Taylor G, Bryan C. The development of a ruminative response style in early adolescents. Unpublished questionnaire. 2002. 2002.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCohen S, Kamarck T, Mermelstein R. A Global Measure of Perceived Stress. J Health Soc Behav. 1983;24:385\u0026ndash;96.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMarch JS, Parker JD, Sullivan K, Stallings P, Conners CK. The Multidimensional Anxiety Scale for Children (MASC): factor structure, reliability, and validity. J Am Acad Child Adolesc Psychiatry. 1997;36:554\u0026ndash;65.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHankin BL, Abramson LY. Measuring cognitive vulnerability to depression in adolescence: reliability, validity, and gender differences. J Clin Child Adolesc Psychol Off J Soc Clin Child Adolesc Psychol Am Psychol Assoc Div. 2002;53:31:491\u0026ndash;504.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCosta PT Jr., McCrae RR. Bridging the gap with the five-factor model. Personal Disord Theory Res Treat. 2010;1:127\u0026ndash;30.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSmit A, Helmich MA, Bringmann L, Oldehinkel A, Wichers M, Snippe E. Critical slowing down in momentary affect as early warning signal of impending transitions in depression. 2022.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003evan de Leemput IA, Wichers M, Cramer AOJ, Borsboom D, Tuerlinckx F, Kuppens P, et al. Critical slowing down as early warning for the onset and termination of depression. Proc Natl Acad Sci U S A. 2014;111:87\u0026ndash;92.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWichers M, Groot PC, Psychosystems, EG EWS Group. Critical Slowing Down as a Personalized Early Warning Signal for Depression. Psychother Psychosom. 2016;85:114\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWichers M, Smit AC, Snippe E. Early Warning Signals Based on Momentary Affect Dynamics can Expose Nearby Transitions in Depression: A Confirmatory Single-Subject Time-Series Study. J Pers-Oriented Res. 2020;6:1\u0026ndash;15.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKuppens P, Allen NB, Sheeber L. Emotional inertia and psychological maladjustment. Psychol Sci. 2010;21:984\u0026ndash;91.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWebb CA, Murray L, Tierney AO, Forbes EE, Pizzagalli DA. Reward-related predictors of symptom change in behavioral activation therapy for anhedonic adolescents: a multimodal approach. Neuropsychopharmacology. 2023;48:623\u0026ndash;32.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBelleau E, Kremens R, Ang Y-S, Pisoni A, Bondy E, Durham K, et al. Behavioral and Neural Markers of Reward Processing Deficits in Adolescents at High Familial Risk for Depressive Disorders. Biol Psychiatry. 2021;89:S27\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBoger KD, Auerbach RP, Pechtel P, Busch AB, Greenfield SF, Pizzagalli DA. Co-Occurring Depressive and Substance Use Disorders in Adolescents: An Examination of Reward Responsiveness During Treatment. J Psychother Integr. 2014;24:109\u0026ndash;21.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLuking KR, Pagliaccio D, Luby JL, Barch DM. Child Gain Approach and Loss Avoidance Behavior: Relationships With Depression Risk, Negative Mood, and Anhedonia. J Am Acad Child Adolesc Psychiatry. 2015;54:643\u0026ndash;51.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTanner J. Growth at adolescence. 2nd ed. Oxford: Blackwell; 1962.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZou H, Hastie T. Regularization and Variable Selection Via the Elastic Net. J R Stat Soc Ser B Stat Methodol. 2005;67:301\u0026ndash;20.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBreiman L. Random Forests. Mach Learn. 2001;45:5\u0026ndash;32.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen T, Guestrin C, XGBoost:. A Scalable Tree Boosting System. Proc. 22nd ACM SIGKDD Int. Conf. Knowl. Discov. Data Min., New York, NY, USA: Association for Computing Machinery; 2016. pp. 785\u0026ndash;794.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSteinwart I, Christmann A. Support Vector Machines. Springer Science \u0026amp; Business Media; 2008.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhou Z-H. Ensemble Learning. In: Zhou Z-H, editor. Mach. Learn. Singapore: Springer; 2021. pp. 181\u0026ndash;210.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWebb CA, Cohen ZD, Beard C, Forgeard M, Peckham A, Bj\u0026ouml;rgvinsson T. Personalized Prognostic Prediction of Treatment Outcome for Depressed Patients in a Naturalistic Psychiatric Hospital Setting: A Comparison of Machine Learning Approaches. J Consult Clin Psychol. 2020;88:25\u0026ndash;38.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFu Z, Brouwer M, Kennis M, Williams A, Cuijpers P, Bockting C. Psychological factors for the onset of depression: a meta-analysis of prospective studies. BMJ Open. 2021;11:e050129.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang X, Su Y, Yang F, Song Y, Yan J, Luo Y, et al. Neurofunctional mapping of reward anticipation and outcome for major depressive disorder: a voxel-based meta-analysis. Psychol Med. 2022;52:3309\u0026ndash;22.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMackin DM, Goldstein BL, Mumper E, Kujawa A, Kessel EM, Olino TM, et al. Longitudinal Associations Between Reward Responsiveness and Depression Across Adolescence. J Am Acad Child Adolesc Psychiatry. 2023;62:816\u0026ndash;28.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEnkavi AZ, Eisenberg IW, Bissett PG, Mazza GL, MacKinnon DP, Marsch LA, et al. Large-scale analysis of test\u0026ndash;retest reliabilities of self-regulation measures. Proc Natl Acad Sci. 2019;116:5472\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEnkavi AZ, Poldrack RA. Implications of the Lacking Relationship Between Cognitive Task and Self-report Measures for Psychiatry. Biol Psychiatry Cogn Neurosci Neuroimaging. 2021;6:670\u0026ndash;2.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHedge C, Powell G, Sumner P. The reliability paradox: Why robust cognitive tasks do not produce reliable individual differences. Behav Res Methods. 2018;50:1166\u0026ndash;86.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRouder JN, Haaf JM. A psychometrics of individual differences in experimental tasks. Psychon Bull Rev. 2019;26:452\u0026ndash;67.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eElliott ML, Knodt AR, Ireland D, Morris ML, Poulton R, Ramrakha S, et al. What Is the Test-Retest Reliability of Common Task-Functional MRI Measures? New Empirical Evidence and a Meta-Analysis. Psychol Sci. 2020;31:792\u0026ndash;806.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJeffreys M, Rozenman M, Gonzalez A, Warnick EM, Dauser C, Scahill L, et al. Factor Structure of the Parent-Report Mood and Feelings Questionnaire (MFQ) in an Outpatient Mental Health Sample. J Abnorm Child Psychol. 2016;44:1111\u0026ndash;20.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSupreeyaporn N, Watson R, Damrongpanit S, Orchard F, Reynolds S, Jenkins PE. Factor structure and measurement invariance of the Mood and Feelings Questionnaire: a cross-cultural study among Thai and British adolescents. Eur Child Adolesc Psychiatry. 2024;33:1563\u0026ndash;72.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRen B, Balkind EG, Pastro B, Israel ES, Pizzagalli DA, Rahimi-Eichi H, et al. Predicting states of elevated negative affect in adolescents from smartphone sensors: a novel personalized machine learning approach. Psychol Med. 2023;53:5146\u0026ndash;54.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWebb CA, Ren B, Rahimi-Eichi H, Gillis BW, Chung Y, Baker JT. Personalized prediction of negative affect in individuals with serious mental illness followed using long-term multimodal mobile phenotyping. Transl Psychiatry. 2025;15:174.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHuang AA, Huang SY. Exploring Depression and Nutritional Covariates Amongst US Adults using Shapely Additive Explanations. Health Sci Rep. 2023;6:e1635.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLuedtke A, Sadikova E, Kessler RC. Sample Size Requirements for Multivariate Models to Predict Between-Patient Differences in Best Treatments of Major Depressive Disorder. Clin Psychol Sci. 2019;7:445\u0026ndash;61.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGijzen MWM, Creemers DHM, Rasing SPA, Smit F, Engels RCME. Evaluation of a multimodal school-based depression and suicide prevention program among Dutch adolescents: design of a cluster-randomized controlled trial. BMC Psychiatry. 2018;18:124.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Footnotes","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003e Murray et al (2022) examined concurrent, baseline (not prospective) associations between reward-related variables in a subset of the current sample.\u003c/span\u003e\u003c/li\u003e\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-psychiatry","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bpsy","sideBox":"Learn more about [BMC Psychiatry](http://bmcpsychiatry.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bpsy/default.aspx","title":"BMC Psychiatry","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-6585192/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6585192/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e\u003cp\u003eDepression rates surge during adolescence. Early identification of youth at increased risk for depression is crucial for timely intervention and, ideally, prevention. This study aims to improve the prediction of future depressive symptoms in adolescents by using a multimodal approach that integrates relevant clinical, demographic, behavioral, and neural characteristics.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003e103 adolescents (ages 12\u0026ndash;18; 72.8% female) underwent a baseline assessment including self-report questionnaires, ecological momentary assessment, a clinical interview, and behavioral and neural measures of reward responsiveness. We used nested cross-validation to compare machine learning approaches as well as conventional linear regression in predicting depressive symptoms (Center for Epidemiological Studies Depression Scale [CES-D] and the Mood and Feelings Questionnaire [MFQ]) at a 3-month follow-up.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eFor the prediction of CES-D depression scores, the best performing model was a multivariable linear regression using as predictors five principal component scores from a principal component analysis of baseline variables (RMSE\u0026thinsp;=\u0026thinsp;6.501, R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.688). For the MFQ, the best performing model was a univariable linear regression with baseline MFQ scores as the sole predictor (RMSE\u0026thinsp;=\u0026thinsp;8.054, R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.671). A factor analysis revealed that items assessing melancholic features were most predictive of future depressive symptoms.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e\u003cp\u003eMore complex machine learning approaches did not outperform regression in predicting future depression. The integration of relevant multimodal predictors reveals which adolescent characteristics (e.g., melancholic features and physical anxiety) have a larger contribution to predicting short-term future depression. Future studies are needed with larger sample sizes and longer follow-up periods to provide a more comprehensive test of such models.\u003c/p\u003e","manuscriptTitle":"Multimodal Prediction of Future Depressive Symptoms in Adolescents","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-31 19:04:16","doi":"10.21203/rs.3.rs-6585192/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Accepted","date":"2025-11-25T10:02:55+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-18T12:02:22+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-11T19:48:51+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"117513995785006498381415091895766064832","date":"2025-11-08T11:23:42+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"59593919574998199385784800344316186874","date":"2025-11-03T08:11:45+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"933526329752274582778719169459485926","date":"2025-10-30T18:02:20+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-10-30T14:57:15+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-10-30T04:14:23+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Psychiatry","date":"2025-10-29T15:47:02+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-psychiatry","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bpsy","sideBox":"Learn more about [BMC Psychiatry](http://bmcpsychiatry.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bpsy/default.aspx","title":"BMC Psychiatry","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"f3be971c-9630-4e7f-a44f-6a2ad97ca83b","owner":[],"postedDate":"October 31st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-12-08T16:06:17+00:00","versionOfRecord":{"articleIdentity":"rs-6585192","link":"https://doi.org/10.1186/s12888-025-07665-8","journal":{"identity":"bmc-psychiatry","isVorOnly":false,"title":"BMC Psychiatry"},"publishedOn":"2025-12-03 15:57:57","publishedOnDateReadable":"December 3rd, 2025"},"versionCreatedAt":"2025-10-31 19:04:16","video":"","vorDoi":"10.1186/s12888-025-07665-8","vorDoiUrl":"https://doi.org/10.1186/s12888-025-07665-8","workflowStages":[]},"version":"v1","identity":"rs-6585192","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6585192","identity":"rs-6585192","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}