Diagnostic digital phenotyping in schizophrenia-spectrum disorders: a systematic review

TL;DR

This review examines how digital phenotyping can help diagnose and predict outcomes in schizophrenia-spectrum disorders, finding promise but highlighting the need for better methods and standards.

Contribution

This is the first systematic review analyzing the diagnostic and predictive utility of digital phenotyping in schizophrenia-spectrum disorders.

Findings

Cognitive performance showed the largest effect size for differentiating SSD from controls.

Relapse prediction models reached AUC values of 0.8 but lacked standardization.

Most studies used smartphone or wearable data, with limited integration of active and passive methods.

Abstract

Digital phenotyping offers a promising but heterogeneous approach for assessing schizophrenia-spectrum disorders (SSD). This systematic review, the first of its kind, comprehensively analyzes the diagnostic and predictive utility of digital phenotyping in SSD. Following PRISMA guidelines, we synthesized data from 142 peer-reviewed studies (2004–2024; n = 6294 participants). Results show a predominance of smartphone and wearable-based approaches, with only ~20% of studies combining active and passive methods. Among six symptom domains, cognitive performance yielded the largest pooled effect size (Hedges’ g ≈ 1.20) for differentiating individuals with SSD from controls, followed by behavior and activity (g ≈ 0.62). However, both domains exhibited very high heterogeneity (I² > 70%). Correlations with the PANSS scale were scarce (<5% of studies), with coefficients reaching 0.6. Relapse prediction models showed promise, with some AUC values reaching 0.8, but lacked methodological standardization. This review highlights the potential of specific digital measures while underscoring the urgent need for improved reporting, multimodal data integration, and longitudinal studies with diverse populations to enhance diagnostic and predictive power in SSD.

Figures