Personalized Mood Prediction from Patterns of Behavior Collected with Smartphones

Study participants and treatment protocol

Participants (N = 437; 76.5% female, 26.5% white) are University of California Los Angeles (UCLA) students experiencing mild to severe symptoms of depression or anxiety enrolled as part of the Screening and Treatment for Anxiety and Depression^33–35 (STAND) study (Sup Figure 1). Participants are initially assessed using the Computerized Adaptive Testing Depression Inventory³¹ (CAT-DI), an online adaptive tool that offers validated assessments of depression severity (measured on a 0-100 scale). After the initial assessment, participants are routed to appropriate treatment resources depending on depression severity: those with mild (35 ≤ CAT-DI < 65) to moderate (65 ≤ CAT-DI < 75) depression at baseline received online support with or without peer coaching³⁶ while those with severe depression (CAT-DI ≥ 75) received in-person care from a clinician (Online Methods).

STAND enrolled participants in two waves, each with different inclusion criteria and CAT-DI assessment and treatment protocol (Sup Figure 2A). Wave 1 was limited to individuals with mild to moderate symptoms at baseline (N=182) and treatment lasted for up to 20 weeks. Wave 2 included individuals with mild to moderate (N=142) and severe (N=124) symptoms and treatment lasted for up to 40 weeks. Eleven individuals participated in both waves. Depression symptom severity was assessed up to every other week for the participants that received online support (both waves), i.e., those with mild to moderate symptoms, and every week for the participants that received in-person clinical care, i.e., those with severe symptoms.

Adherence to CAT-DI assessment protocol

In total, participants provided a total of 4,507 CAT-DI assessments (out of 11,218 expected by the study protocols). Participant adherence to CAT-DI assessments varied across the treatment groups (Likelihood ratio test [LRT] P-value < 2.2×10⁻¹⁶), enrollment waves (LRT P-value = 2.86×10⁻⁶), and during the follow-up period (LRT P-value = 1.29×10⁻⁶). Specifically, participants that received clinical care were more adherent than those which only received online support (Sup Figure 2A). Attrition for participants which received clinical care was linear over the follow-up period, with 1.7% of participants dropping out CAT-DI assessments within two weeks into the study. Attrition for participants that received online support was large two weeks into the study (33.5% of Wave 1 and 37.3% of Wave 2 participants) and linear for the remaining of the study.

For building personalized mood prediction models, we focus on 183 individuals (49 from Wave 1 and 134 from Wave 2) who had at least five mental health assessments during the study (Online Methods). For these individuals we obtained a total of 3,005 CAT-DI assessments with a median of 13 assessments, 171 follow-up days, and 10 days between assessments per individual (Figure 1A-C).

• Download figure
• Open in new tab

Figure 1: Overview of CAT-DI assessment frequency and source of variation in CAT-DI.

(A-C) Boxplot of the observed number of CAT-DI assessments (B), follow-up time in days (C), and median number of days between assessments (D) for each wave and treatment group. The numbers in the parentheses indicate the expected values according to study design (Sup Figure 2). (D) Proportion of CAT-DI severity variance explained (VE) by inter-individual differences and other study parameters with 95% confidence intervals. The proportion of variance attributable to each source was computed using a linear mixed model with the individual id, treatment group, season, year, and sex modeled as random variables and all other variables modeled as fixed (see Online Methods).

Computerized adaptive testing captures treatment-related changes in depression severity

We assessed what contributes to variation in the CAT-DI severity scores (Figure 1E). Subjects are assigned to different treatments (online with or without coaching and clinical care) depending on their CAT-DI severity scores, so not surprisingly we see a significant source of variation attributable to the treatment group (6.43% of variance explained, 95% CI: 5.20 - 9.40%, Online Methods). Once assigned to a treatment group, we expect to see changes over time as treatment is delivered. This is reflected in a significant source of variation attributable to the number of weeks spent in the study (10.87% of variance explained, 95% CI: 9.39 - 12.79%) and the improved scores for individuals with severe symptoms at baseline as they spend more time in the study (Sup Figure 3). We found no statistically significant effect of the COVID pandemic, sex, and other study parameters. The largest source of variation in depression severity scores is attributable to between-individual differences (42.26% of variance explained, 95% CI: 38.47 - 42.55%), suggesting that accurate prediction of CAT-DI severity requires learning models tailored to each individual.

Digital behavioral phenotypes capture changes in behavior

We set out to examine how digital behavioral phenotypes change over time for each person and with CAT-DI severity scores. For example, we want to know how hours of sleep on a specific day for a specific individual differs from the average hours of sleep in the previous week, or month. To answer these questions, we extracted digital behavioral phenotypes (referred to hereinafter as features) captured from participants’ smartphone sensors and investigated which features predicted the CAT-DI scores. STAND participants had the AWARE framework³² installed on their smartphones, which queried phone sensors to obtain information about a participant’s location, screen on/off behavior, and number of incoming and outgoing text messages and phone calls. We processed these measurements (Online methods) to obtain daily aggregate measures of activity (23 features), social interaction (18 features), sleep quality (13 features), and device usage (two features). In addition, we processed these features to capture relative changes in each measure for each individual, e.g., changes in average amount of sleep in the last week compared to what is typical over the last month. In total, we obtained 1,325 features (Online methods). Missing daily feature values (Sup Figure 4) were imputed (Online Methods), resulting in 29,254 days of logging events across all individuals.

Several of these features map onto the DSM-5 MDD criteria of anhedonia, sleep disturbance, and loss of energy (Supplemental Methods; Sup Figure 5). We found that these features in some cases do indeed correlate with changes in depression: Figure 2 illustrates an individual with severe depressive symptoms for whom we can identify a window of disrupted sleep that co-occurred with a clinically significant increase in symptom severity (from mild to severe CAT-DI scores). Subsequently, a return to baseline patterns of sleep coincided with symptom reduction. Quantifying this relationship poses a number of issues, which we turn to next.

• Download figure
• Open in new tab

Figure 2: Example of identifying window of potential sleep disruption using sensor data related to phone usage and screen on/off status.

The top panel shows estimated hours of sleep for an individual during the study while the bottom panel shows the depression severity scores during the same period. The dotted lines indicate the dates at which a change point is estimated to have occurred in the estimated hours of sleep as estimated using a change point model framework for sequential change detection (Online Methods).

Predicting CAT-DI scores from digital phenotypes

To predict future depression severity scores using digital behavioral phenotypes, we considered three analytical approaches. First, we applied an idiographic approach, whereby we build a separate prediction model for each of the participants. Specifically, for each individual, we train an elastic net linear regression model using the first 70% of their depression scores and predict the remaining 30% of scores. Second, we applied a nomothetic approach that used data from all participants to build a single model for depression severity prediction using the same analytical steps: we train an elastic net regression model using the first 70% of depression scores of each individual and predict the remaining 30% of scores (Online Methods). The result of this nomothetic approach was a single elastic net regression model that makes predictions in all participants.

The main difference between the nomothetic and idiographic approach is that the nomothetic model assumes that each feature has the same relationship with the CAT-DI scores across individuals, for example, that a phone interaction is always associated with an increase in depression score. However, it is possible, and we see this in our data, that an increase in phone interaction can be associated with an increase in symptom severity for one person, but a decrease in another (Sup Figure 6). The idiographic model allows for this possibility by using a different slope for each feature and individual. In addition, we know that large differences exist in average depression scores between individuals (Figure 1E). To understand the impact of accounting for these differences in a nomothetic approach, we also applied a third approach (referred to as nomothetic*) which includes individual indicator variables in the elastic net regression model in order to allow for potentially different intercepts for each individual.

To assess whether digital behavioral phenotypes predict mood, we have to deal with the problem that digital phenotypes are acquired daily, while CAT-DI are usually administered every week (and often much less frequently, on average every 10 days). We assume that the CAT-DI indexes a continuously variable trait, but what can we use as the target for our digital predictions when we have such sparsely distributed measures? We can treat this as a problem of imputation, in which case the difficulty reduces to knowing the likely distribution of missing values. However, we also assume that both CAT-DI and digital features only imperfectly reflect a fluctuating latent trait of depression. Thus, our imputation is used not only to fill in missing data points but also to be a closer reflection of the underlying trait that we are trying to predict, namely, depressive severity.

We interpolate the unmeasured estimates of depression by modeling the latent trait as a cubic spline with different degrees of freedom (Figure 3A). For many individuals, CAT-DI values fluctuate considerably during the study, while for others less so. To accommodate this variation, we alter the degrees of freedom of the cubic spline: the more degrees of freedom, the greater the allowed variation. For each individual, we used cubic splines with four degrees of freedom, denoted by CS(4df), degrees of freedom corresponding to the number of observed CAT-DI categories in the training set, denoted by CS(2-4df), and degrees of freedom identified by leave-one-out cross-validation in the training set, denoted by CS(cv). For comparison purposes, we also used a last-observation-carried-forward (LOCF) approach, a naive interpolation method which does not apply any smoothness to the observed trait. Because spline interpolation will cause data leakage across the training-testing split and upwardly bias prediction accuracy, we train our prediction models using cubic spline interpolation on only the training data (first 70% of time series of each individual) and assess prediction accuracy performance in the testing set (last 30%) using the time series generated by applying cubic splines to the entire time series (Figure 3B).

• Download figure
• Open in new tab

Figure 3: Interpolation of depression severity scores and latent trait inference.

(A) Illustration of different interpolation methods considered for imputing the depression severity scores and inferring the latent depression traits. The dashed horizontal lines indicate the depression severity score thresholds for the mild and severe depression severity categories. (B) Illustration of the prediction method. We first infer the latent trait on the full CAT-DI trajectory of an individual (continuous yellow line). We then split the trajectory into a training set (days 1 until t) and a test set (days t+1 until T), infer the latent trait on the training set (dashed yellow line), and predict the trajectory in the test set (yellow triangles). Finally, we compute prediction accuracy metrics by comparing the observed (yellow circles) and predicted (yellow triangles) depression scores in the test set. The vertical line indicates the first date of the test set trajectory, i.e., the last 30% of the trajectory. LOCF: last observation carried forward. CS(xdf): cubic spline with x degrees of freedom. CS(cv): best-fitting cubic spline according to leave-one-out cross-validation.

We evaluated the prediction performance of each model and for each latent trait across and within participants. We refer to the former as group level prediction and the later as individual level prediction. Looking at group level prediction performance, compared to within each participant separately, allows us to compute prediction accuracy metrics, e.g., R², and test for their statistical significance across all predicted observations. In addition, it allows us to study prediction accuracy as a function of how many days ahead we are predicting.

We first evaluated group level prediction accuracy. Figure 4A and Sup Figure 7 show group level prediction performance for each latent trait using the nomothetic, nomothetic*, and idiographic models. We observed that across all latent traits the nomothetic model shows very poor prediction accuracy (R²< 5% for all latent traits), compared to the nomothetic* (R²= 41-59%) or idiographic (R²=41-67%) models. This is in line with the large proportion of depression scores variance explained by between-individual differences (Figure 1E) which get best captured by the nomothetic* and idiographic models. We also compared the prediction performance for each of the different latent traits. We achieve a higher prediction accuracy for the cubic spline latent traits compared to the LOCF latent trait. For example, for the idiographic models, we obtained an R²= 67.95% for CS(2-4df) versus 41.34% for LOCF, implying that weekly patterns of depression severity, which are more likely to be captured by the LOCF latent trait, are harder to predict than depression severity patterns over a couple of weeks or months, which are more likely to be captured by the cubic spline latent traits with smallest degrees of freedom.

• Download figure
• Open in new tab

Figure 4: Idiographic models achieve higher group level prediction accuracy than nomothetic models.

(A-B) CAT-DI prediction accuracy in the test set as measured by R² (A) and mean absolute percent error (B) across all individuals from the idiographic and two nomothetic models for different latent depression traits. (B) Prediction accuracy versus the number of weeks ahead we are predicting from the last observation in the training set. Each dot indicates the prediction accuracy and bars indicate 95% confidence intervals. The color indicates the different interpolation methods considered for imputing CAT-DI scores. The size of the dots indicates the number of CAT-DI measurements we predict each week. The dotted line indicates 80% prediction accuracy. MAE: mean absolute error. LOCF: last observation carried forward. CS(xdf): cubic spline with x degrees of freedom. CS(cv): best-fitting cubic spline according to leave-one-out cross-validation.

To understand the effect of time on prediction accuracy, we assessed prediction performance as a function of the number of days ahead we are predicting from the last observation in the training set (Figure 4B). The idiographic models achieved high prediction accuracy for depression scores up to three weeks from the last observation in the training set, e.g., R²= 91.0% and 79.7% for the CS(2-4df) latent trait to predict observations one week and three weeks ahead, respectively. Prediction accuracy falls below 80% after four weeks.

We next evaluated individual level prediction accuracy. For this analysis, in order to be able to assess the statistical significance of our prediction accuracy within each individual, we only keep individuals with at least five mental health assessments in the test set (N=139). In accordance with the group level prediction performance, the idiographic model outperformed the other models at the individual level. Using an idiographic modelling approach, we significantly predicted the future mood for 65.5% of individuals (91 out of 139 with R> 0 and FDR <= 5% across individuals) for at least one of the latent traits (Figure 5A and Sup Figure 9), compared to 46.0% and 46.8% of individuals for the nomothetic and nomothetic* model, respectively. The median R² value across significantly predicted individuals for the idiographic models was 57.5% (Figure 5B and Sup Figure 10), compared to 31.7% and 35.9% for the nomothetic and nomothetic* model, respectively. In addition, for thirty-eight of these individuals, the idiographic model had prediction accuracy greater than 70%, demonstrating high predictive power in inferring mood from digital behavioral phenotypes for these individuals, compared to 13 and 9 for the nomothetic and nomothetic* model, respectively (Figure 5B and Sup Figure 10).

• Download figure
• Open in new tab

Figure 5: Idiographic models achieve higher individual level prediction accuracy than nomothetic models.

(A) Bar plots of the proportion of individuals with significantly predicted mood (FDR<5% and R>0) for each latent trait and prediction model. Computed across individuals with at least five assessments in the test set (N=139). Sup Figure 9 shows the overlap of significantly predicted individuals across different latent traits and prediction models. (B) Prediction accuracy (R²) with 95% CI for best predicted latent trait across all significantly predicted individuals (N= 118, FDR<5% and R>0). Sup Figure 10 contains results for each latent trait. LOCF: last observation carried forward. CS(xdf): cubic spline with x degrees of freedom. CS(cv): best-fitting cubic spline according to leave-one-out cross-validation.

To identify the features that most robustly predict depression in each person we extracted top-feature predictors for each individual’s best-fit idiographic model. We limit this analysis to the 91 individuals which showed significant prediction accuracy for at least one of the latent traits. Although no feature uniformly stood out, the variation within the last 30 days in the proportion of unique contacts for outgoing texts and messages (a proxy for social interaction), the time of first interaction with phone in the morning (a proxy for wake up time and sleep quality), and the proportion of time spent at home during the day (a proxy for activity level) were among the top predictors of future mood (Figure 6). The heatmap display of predictor importance highlights the heterogeneity of passive features for predicting the future across individuals. For example, poor mental health, as indicated by high CAT-DI depression severity scores, was associated with decreased variation in the hours the phone was off between midnight and 8 a.m. (a proxy for hours of sleep and sleep quality) in the past 30 days for one individual while for another individual it was associated with increased variation.

• Download figure
• Open in new tab

Figure 6: Most predictive behaviors for CS(4df) latent trait according to idiographic models.

Heatmap of idiographic elastic net regression coefficients for significantly predicted individuals (N=91 with FDR<5% and R>0). Columns indicate different individuals and rows indicate different features. To aid interpretation, we limit plot to features that are significant for at least two individuals and have an absolute coefficient value above one in at least one participant. The heatmap color indicates the elastic net coefficient for each feature and individual.

Evaluation of feature contribution to prediction performance

So far, our models are all of the form mood ∼ alpha + beta * phone features. This model implies that if the phone feature has no predictive accuracy, we may still be able to predict something about the future mood of an individual in the test set by using that individual’s stable (i.e., mean) mood in the training set, reflected by the value of ‘alpha’, the intercept. We assessed to what extent adding the phone features improves the prediction of the idiographic models above that achieved by a baseline model that includes just the intercept (i.e., mood ∼ alpha). In order to do that, we first identified the best fitting latent trait for each individual according to the idiographic model with features and compare that model to the equivalent baseline idiographic model without features in terms of prediction mean absolute error (MAE). Within each individual, the baseline idiographic model always predicts the same value, i.e., alpha, so we cannot compute R² metrics for the baseline model. As above, we limit this analysis to the individuals with at least five mental health assessments in the test set (N=139).

The feature-based model performed significantly better than the baseline model (Figure 7A, one sided Wilcoxon signed-rank test, p-value = 3.33×10⁻²) with a median decrease in MAE of 4.43% across all 139 individuals. The feature-based model reduced the MAE compared to a baseline model for 52% of individuals (70 out of 139) with 48 individuals having a decrease in the MAE of at least 20%. Figure 7B illustrates the prediction performance for one such individual in our study. These results suggest that the passive phone features enhance prediction, over and above past CAT-DI, for a subset of individuals.

• Download figure
• Open in new tab

Figure 7: Comparison of the feature-based prediction model to a baseline prediction model.

(A) Histogram of difference for each individual in mean absolute prediction error (MAE) between a baseline model that predicts based on past depression severity alone and a feature-based model which predicts based on past severity and past feature values. Negative values correspond to a decrease in prediction error when a prediction model with features is used, compared to the baseline model. (B) Prediction performance for an individual in our study for which features reduce the mean absolute prediction error of a baseline model by more than 10%. The lines indicate the observed LOCF (dark blue line) and cubic-spline interpolated (yellow line) CAT-DI scores for an individual in our study. The yellow dots are the predictions in the test set (30% of the time series; last six depression scores, two months ahead). The red line is the prediction of the baseline model which corresponds to the mean depression in the training set.

Factors associated with prediction performance

Using digital behavioral features to predict future mood was useful for about half of our cohort and the contribution of the features to the prediction performance varies across these individuals. What might contribute to this variation? Identifying the factors involved might allow us to develop additional models with higher prediction accuracy. To identify factors that are associated with prediction performance, we computed the correlation between accuracy metrics (prediction R² and MAE of feature-based model and difference in MAE between feature-based and baseline models) with different study parameters e.g., treatment group, sex, etc. (Figure 8). Larger differences in median depression scores between the training and test set for each individual were correlated with poorer prediction performance, as measured by MAE (Spearman’s ρ=0.65, p-value = 3.46 × 10⁻¹¹). This suggests that, for many of the individuals in the study, the training depression scores are not as representative of the test depression scores, e.g., if individuals have high CAT-DI in the training but low CAT-DI in the test. Indeed, this is the case for most individuals in our cohort who are in treatment that improves their symptoms during the course of the study (Sup Figure 3). The size of the training and test set as well as demographic variables were not strongly correlated to prediction performance. While we had poorer prediction performance for individuals whose mood changes between the training and test set, these are also the individuals for which using a feature-based model improves prediction accuracy compared to a baseline model that predicts based on past depression severity alone. Larger differences in median depression scores between the training and test set for each individual were correlated with better prediction performance of a feature-based model, compared to a baseline model (Spearman’s ρ=0.54, p-value = 1.53 × 10⁻⁷).

• Download figure
• Open in new tab

Figure 8: Factors associated with prediction performance of CAT-DI severity scores.

Correlation between prediction accuracy of an individual (metrics on the y-axis) and the number of CAT-DI assessment available in the training and test set, the difference in median CAT-DI severity between the training and test set, the number of the unique CAT-DI categories (normal to severe) observed (total and in training and test sets), age, sex, wave, and treatment group (a proxy for depression severity). MAE: mean absolute error of the feature-based model. MAE_diff: mean absolute error difference between the feature-based and baseline models of each individual, i.e., MAE_features – MAE_{no features}.

Study participants and protocol

Participants are University of California Los Angeles (UCLA) students experiencing mild to severe symptoms of depression or anxiety enrolled as part of the STAND program developed under the UCLA Depression Grand Challenge³⁵ treatment arm. All participants provided written informed consent for the study protocol approved by the UCLA institutional review board (IRB #16-001395 for those receiving online support and #17-001365 for those receiving clinical support). All groups are offered behavioral health tracking through the AWARE³² framework and had to install the app in order to participate in the study. STAND enrolled participants in two waves. The first wave enrolled participants from April 2017 to June 2018. The second wave of enrollment began at the start of the academic year in 2018 and continued for three years, during which time, from March 2020, a Safer-At-Home order was imposed in Los Angeles to control the spread of COVID-19.

Depression symptom severity was assessed using the Computerized Adaptive Testing Depression Inventory³¹ (CAT-DI), a validated online mental health tracker. Computerized adaptive testing is a technology for interactive administration of tests that tailors the test to the patient³⁹. Tests are ‘adaptive’ in the sense that the testing is driven by an algorithm that selects questions in real-time and in response to the ongoing responses of the patient. CAT-DI uses item response theory to select a small number of questions from a large bank, thus providing a powerful and efficient way to detect psychiatric illness without suffering response fatigue.

Depression symptom severity was assessed up to every other week for the participants that received online support (both waves), i.e., those with mild to moderate symptoms, and every week for the participants that received in-person clinical care, i.e., those with severe symptoms (Sup Figure 2A).

Participants that received in-person care had also four in-person assessment events, at weeks 8, 16, 28, and 40, prior to the COVID-19 pandemic. Thus, Wave 1 participants can have a maximum of 13 CAT-DI assessments while Wave 2 participants can have a maximum of 21 (online support) or 44 assessments, depending on severity and excluding initial assessments prior to treatment assignment.

CAT-DI was assessed at least one time for 437 individuals that installed the AWARE app. Here, we limit our prediction analyses to individuals that have at least five CAT-DI assessments (N=238; since we need at least four points to interpolate CAT-DI in the training set), have at least 60 days of sensor data in the same period for which CAT-DI data is also available (N=189), and show variation in their CAT-DI scores in the training set (N=183), which is necessary in order to build prediction models.

Overview of the treatment protocol

STAND use computerized adaptive testing to initially assess depressive mood severity at entry, from which students are routed to appropriate treatment resources. Individuals who are not currently experiencing symptoms of depression or anxiety are offered the opportunity to participate in the study with an active treatment component by contributing CAT-DI assessment. These individuals are excluded from our analyses as they do not show any variation in CAT-DI. Individuals who are mildly depressed or anxious or at risk for depression or anxiety are provided internet-based cognitive behavior therapy, which includes adjunctive support provided by trained peers or clinical psychology graduate students via video chat or in person. Individuals who are severely depressed, suicidal, or bipolar/manic and who need more intensive treatment, are offered an evaluation within the STAND clinic, including individualized treatment by a team of psychiatrists and psychologists.

Feature extraction from smartphone sensors

We describe feature extraction in detail in the supplement. Broadly, we extracted 23 features related to mobility, e.g., location entropy, 13 related to sleep and circadian rhythm, e.g., hours of uninterrupted sleep, 18 related to social interaction, e.g., duration of outgoing calls, and two related to mobile device usage, e.g., number of interactions with phone per day. Each of these features was calculated on a daily basis. Furthermore, each of these features was computed over three daily non-overlapping time windows of equal duration (night 00:00-08:00, day 08:00-16:00, evening 16:00-00:00), under the hypothesis that participant behavior may be more or less variable based on external constraints such as a regular class schedule during daytime hours.

In addition, considering a participant’s current mental state may be influenced by patterns of behavior from days prior, sliding window averages of each of the daily features were calculated over multiple sliding windows ranging from three days to one month prior to the current day, i.e., windows of length three, seven, 14, and 30 days. The variance of each feature was also calculated over these same windows, to estimate whether behavior had been stable or variable during that time, e.g., were there large fluctuations in sleep time over the past week?

Finally, under the hypothesis that recent changes in behavior may be more indicative of changes in mental state than absolute measures, a final set of transformations were applied to each feature. These transformations compared the sliding window means of two different durations against each other, to estimate the change in behavior during one window over that of a longer duration window (the longer window serving as a local baseline for the participant). This allowed estimates from the raw features of whether, e.g., the participant had slept less last night than typical over the past week or slept less on average in the last week than typical over the last month. All of these transformations were applied to the base features extracted from sensor data and included as separate features fed into subsequent regression approaches.

In total, 1,325 raw and transformed features were extracted and included in the final analysis.

Imputation of smartphone-based features

To address the missing features problem (Sup Figure 4), we considered two different imputation methods: matrix completion via iterative soft-thresholder SVD as implemented in the R package softImpute⁴⁰ and cubic spline interpolation with degrees of freedom equal to the number of days an individual participated in the study (to achieve the least amount of smoothing). Both approaches were applied separately to each individual. The former was applied across all features while the latter was applied separately to each feature. In the main text, we discuss results based on features imputed using cubic interpolation, which achieved the best prediction accuracy. We show results based on matrix completion interpolation in Sup Figure 8. When fitting prediction models without imputation of missing feature data, we remove days with more than 90% of features missing (within each individual for individual-level models and across individuals for the population-level model) and features that are missing for more than 90% of the days for which the individual(s) participated in the study. Before prediction, we normalize all features to have zero mean and unit standard deviation.

Variance partition of CAT-DI metrics

We calculate the proportion of CAT-DI severity variance explained by different study parameters using a linear mixed model as implemented in the R package variancePartition⁴¹ with the subject id, study id, season, sex, and year modeled as random variables while the day of the study, the age of the subject, and a binary variable indicating the dates before or after the safer at home order was issued in California modeled as fixed, i.e., where y is the vector of the CAT-DI values across all subjects and time points, X_j is the matrix of j^th fixed effect with coefficients β_j, Z_k is the matrix corresponding to the k^th random effect with coefficients a_k drawn from a normal distribution with variance . The noise term, ∈, is drawn from a normal distribution with variance . All parameters are estimated with maximum likelihood⁴². Variance terms for the fixed effects are computed using the post hoc calculation . The total variance is so that the fraction of variance explained by the j^th fixed effect is , by the k^th random effect is , and the residual variance is . Confidence intervals for variance explained were calculated using parametric bootstrap sampling as implemented in the R package variancePartition⁴¹.

Imputation and of CAT-DI severity scores

To get daily-level data, we interpolate the CAT-DI severity scores for each individual across the whole time series (ground truth) or only the time series corresponding to the training set (70% of the time series) by moving the last CAT-DI score forward, denoted by LOCF, or by smoothing the CAT-DI scores using cubic splines with different degrees of freedom (Figure 3A). Cubic smoothing spline fitting was done using the smooth.spline function from the stats package in R. We consider cubic splines with four degrees of freedom (denoted by CS(4df) and corresponding to the number of possible CAT-DI severity categories, i.e. normal, mild, moderate, and severe), cubic splines with degrees of freedom equal to the number of observed CAT-DI categories for each individual in the training set (ranging from two to four and denoted by CS(2-4df)), and degrees of freedom identified by ordinary leave-one-out cross-validation in the training set (denoted by CS(cv)).

Nomothetic and idiographic prediction of mood

We split the data for each individual into a training (70% of trajectory) and a test set (remaining 30% of trajectory). To predict the future mood of each individual in the test set from smartphone-based features in the test set, we train an elastic net linear regression model⁴³ in the train set. We set α, i.e., the mixing parameter between ridge regression and lasso, to 0.5 and use 10-fold cross-validation to find the value for parameter λ, i.e., the shrinkage parameter. For the idiographic models, we train separate elastic net models for each individual while for the nomothetic and modified nomothetic models we train one model across all individuals. To account for individual differences in the average CAT-DI severity scores in the training set, the modified nomothetic model fits individual-specific intercepts by including individual indicator variables in the regression model. This is similar in nature to a random intercept mixed model where each individual has their own intercept. Note that the test data are the same for all of these models, i.e., the remaining 30% of each individual’s trajectories. Predictions outside the CAT-DI severity range, i.e., [0,100], are set to NA and not considered for model evaluation. We compute prediction accuracy metrics by computing the Pearson’s product-moment correlation coefficient (R) between observed and predicted depression scores in the test set across and within individuals as well as the squared Pearson coefficient (R²). To assess the significance of the prediction accuracy we use a one-sided paired test for Pearson’s product-moment correlation coefficient, as implemented in the cor.test function of the stats⁴⁴ R package, and a likelihood ratio test for the significance of R². We use the Benjamini-Hochberg procedure⁴⁵ to control the false discovery rate across individuals at 5%.