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

Abstract

Smartphone-based digital phenotyping enables potentially clinically relevant information to be collected as individuals go about their day. This could improve monitoring and interventions for people with Major Depressive Disorder (MDD). The aim of this systematic review was to investigate current digital phenotyping methods used in MDD. We searched PubMed, PsycINFO, Embase (20/07/2022), Scopus (21/07/2022) and Web of Science (22/07/2022) for articles including: (1) MDD population, (2) smartphone-based features, (3) validated ratings. Risk of bias was assessed using criteria from several sources. Studies were compared within prediction goals. Fourteen studies (3249 participants) were included, covering different goals (e.g., predicting symptom severity). Studies achieved moderate model performance. Common themes included challenges from missing and complex data, and a lack of external testing sets. There is a risk of bias from missing data. Studies made progress towards relating digital phenotypes to clinical variables, often focusing on time-averaged features. Methods investigating temporal dynamics more directly may be beneficial for patient monitoring.European Research Council consolidator grant: 101001118, Prospero registration: CRD42022346264, Open Science Framework:https://osf.io/s7ay4

Closing the loop between environment, brain and mental health: how far we might go in real-life assessments?

“…ESM can also be combined with sensors gathering objective data about internal states and real-life environments. Smartphones, with their integrated sensors and compatibility with, for example, wearables or smart home sensors (see section 2.2), are valuable tools [23,24 ▪ ]. Sensor-aided ESM for investigating the associations between environmental factors and mental well being can be categorized into three (not exclusive) groups: internal physiological sensing [25 ▪▪ ], external environmental sensing [26 ▪▪ ], and position tracking [27 ▪▪ ].…”

Uncovering social states in healthy and clinical populations using digital phenotyping and Hidden Markov Models

“…For example, the duration, circadian rhythm or statistical measures can be calculated (such as mean and standard deviation of a behaviour across time) or the occurrences of the behaviour counted. 9 This often leads to datasets with many features reflecting various different smartphone-measured behaviours. A major problem affecting digital phenotyping is that platforms are often prone to missing data due to the difficulties of real-world longitudinal data collection, leading to missing values across all or a subset of these features.…”

“…A major problem affecting digital phenotyping is that platforms are often prone to missing data due to the difficulties of real-world longitudinal data collection, leading to missing values across all or a subset of these features. 9 This gives rise to multiple analytic challenges: processing the collected feature sets, often representing a wide range of seemingly distinct observed behaviours with potentially similar underlying causes, requires many model decisions. Appropriate methods are therefore needed to analyse this multi-faceted data containing missingness, in order to produce meaningful, lowerdimensional data representations.…”

“…Many digital phenotyping studies have focused on time-averaged features and analyses, and a shift towards more direct investigations of temporal dynamics is expected to improve clinical utility. 9 Currently, digital phenotyping studies employ a broad range of approaches. For example, investigating associations between neuropsychiatric symptoms and summary measures (e.g.…”