Machine Learning Identifies Digital Phenotyping Measures Most Relevant to Negative Symptoms in Psychotic Disorders: Implications for Clinical Trials

Abstract

Background Digital phenotyping has been proposed as a novel assessment tool for clinical trials targeting negative symptoms in psychotic disorders (PDs). However, it is unclear which digital phenotyping measurements are most appropriate for this purpose. Aims Machine learning was used to address this gap in the literature and determine whether: (1) diagnostic status could be classified from digital phenotyping measures relevant to negative symptoms and (2) the 5 negative symptom domains (anhedonia, avolition, asociality, alogia, and blunted affect) were differentially classified by active and passive digital phenotyping variables. Methods Participants included 52 outpatients with a PD and 55 healthy controls (CN) who completed 6 days of active (ecological momentary assessment surveys) and passive (geolocation, accelerometry) digital phenotyping data along with clinical ratings of negative symptoms. Results Machine learning algorithms classifying the presence of a PD diagnosis yielded 80% accuracy for cross-validation in H2O AutoML and 79% test accuracy in the Recursive Feature Elimination with Cross Validation feature selection model. Models classifying the presence vs absence of clinically significant elevations on each of the 5 negative symptom domains ranged in test accuracy from 73% to 91%. A few active and passive features were highly predictive of all 5 negative symptom domains; however, there were also unique predictors for each domain. Conclusions These findings suggest that negative symptoms can be modeled from digital phenotyping data recorded in situ. Implications for selecting the most appropriate digital phenotyping variables for use as outcome measures in clinical trials targeting negative symptoms are discussed.

Digital Phenotypes for Early Detection of Internet Gaming Disorder in Adolescent Students: Explorative Data-Driven Study

“…Some studies have even captured emotional states through keyboard and touch-stroking activities [ 20 ], with associations between keyboarding and various types of emotions identified [ 21 , 22 ]. Moreover, passive data, such as geolocation and accelerometer information, have proven successful in classifying negative symptoms in psychotic disorders [ 20 , 23 ]. Leveraging these sensor data, a handful of studies suggest the potential of digital phenotyping for screening IGD.…”

The positivity offset theory of anhedonia in schizophrenia: evidence for a deficit in daily life using digital phenotyping

“…Second, the accelerometry and geolocation variables included in the study belong to the ‘third generation’ of negative symptom assessments that are still being validated. Additional work is needed to extend findings from preliminary validation studies (Narkhede et al, 2021 ; Raugh et al, 2020 ; Strauss et al, 2022 ) and identify which are the strongest, most reliable measures of negativity symptoms. Third, the current sample included adult outpatients with chronic, stable SZ.…”

“…geolocation, accelerometry) (Onnela & Rauch, 2016 ). Preliminary psychometric studies support the reliability and validity of active and passive digital phenotyping measures of negative symptoms in SZ, as well as their feasibility and tolerability (Depp et al, 2019 ; Fulford et al, 2021 ; Granholm et al, 2019 ; Harvey et al, 2021 ; Miller, Raugh, Strauss, & Harvey, 2022 ; Narkhede et al, 2021 ; Raugh et al, 2020 ; Raugh et al, 2021 ; Strauss et al, 2022 ). When used in tandem, active and passive digital phenotyping methods offer promise for exploring questions regarding the nature of emotion-motivation interactions in SZ since the same computational approaches validated for the ESM can be used in conjunction with objectively measured and self-reported behaviors.…”

The future of psychopharmacology: a critical appraisal of ongoing phase 2/3 trials, and of some current trends aiming to de‐risk trial programmes of novel agents

“…More than one outcome in a trial is desirable, as one outcome only can hardly provide a comprehensive clinical picture, yet adjusting for multiple comparisons in the statistical analyses is needed in case that more than one primary outcome is be­ing assessed or in case that inferential statistical testing is desired even of key secondary outcomes. For secondary and exploratory, hypothesis‐generating outcomes and those requiring a lot of multidimensional data, such as for functioning, modern tools including digital phenotyping and ecological momentary assessment can be of great value and should be progressively introduced in assessment of trials 218‐228 . Digital phenotyping and ecological momentary assessments can be repeated mul­tiple times, and can be even continuous in case of passive monitoring.…”