Estimating longitudinal depressive symptoms from smartphone data in a transdiagnostic cohort

Abstract

In modern psychopharmacology, the gold standard for measurement of depressive symptoms, as with most psychiatric outcomes, has been clinician-rated scales. However, reliance on such measures introduces substantial limitations: the need for trained clinician raters increases the cost of assessment (despite enthusiasm for measurement-based care and recognition of the importance

Smartphone accelerometer data as a proxy for clinical data in modeling of bipolar disorder symptom trajectory

“…Recent years have seen an explosion of researchers using smartphones to understand patterns of user behavior and their relationship to chronic health conditions [1][2][3] , including in particular typing dynamics for conditions such as bipolar disorder 4,5 . Typing dynamics refers to the speed and rhythm with which users type on their phone (e.g., when sending emails or text messages or posting to social media), which can be captured in various metrics that describe things such as the transition time between keypresses, repetitive presses of the same key, press duration, use of backspace and autocorrect, and so on.…”

From smartphone data to clinically relevant predictions: A systematic review of digital phenotyping methods in depression

“…Studies were grouped by analysis goal to allow for comparison of methods with similar objectives. To this end, we first compared studies that correlated passive smartphone features with depression symptom severity (Cao, et al, 2020; Sverdlov, et al, 2021; Zhang, et al, 2022), and then investigated the methods used for predicting symptom severity (Braund, et al, 2022; Cao, et al, 2020; Faurholt-Jepsen, et al, 2022; Pedrelli, et al, 2020; Pellegrini, et al, 2022; Sverdlov, et al, 2021; Zhang, et al, 2021; Zhang, et al, 2022). Somewhat unexpectedly, two studies aimed to predict specific smartphone features from ratings of depression (Laiou, et al, 2022; Tønning, Faurholt-Jepsen, Frost, Bardram, & Kessing, 2021).…”

“…The correlation between observed and predicted scores calculated in leave-one-out cross-validation was r = 0.43, showing a moderate correlation. Pellegrini, et al (2022) conducted a Principal Component Analysis on a set of passive smartphone features they created based on weekly summaries of GPS and accelerometer measures, and used the first principal component as a predictor in their linear mixed models. Pellegrini, et al (2022) investigated various models with and without this passive smartphone feature and a baseline depressive symptom score, demonstrating that including a smartphone feature did not improve the prediction of depressive symptom scores, but instead was comparable to predictions by models using only questionnaire data.…”

“…Pellegrini, et al (2022) conducted a Principal Component Analysis on a set of passive smartphone features they created based on weekly summaries of GPS and accelerometer measures, and used the first principal component as a predictor in their linear mixed models. Pellegrini, et al (2022) investigated various models with and without this passive smartphone feature and a baseline depressive symptom score, demonstrating that including a smartphone feature did not improve the prediction of depressive symptom scores, but instead was comparable to predictions by models using only questionnaire data. The model combining smartphone features and questionnaire data achieved a RMSE of 4.30, and the questionnaire-only model achieved an RMSE of 4.27 (with the possible MADRS score ranging from 0-60).…”

“…However, the shorter period of three days was found to be a more reasonable window for mood state prediction than longer periods during their parameter selection process, suggesting that later mood state prediction was difficult. This method differed from other studies that aimed to make predictions for mood at the end of smartphone data collection (e.g., Pellegrini, et al, 2022), and gives an example of how digital phenotyping research can shift towards predicting upcoming depression states.…”