Scholarlink

Background <p>Digital phenotyping — the moment-by-moment quantification of individual-level human behavior using data from personal digital devices — offers a novel approach to continuous, passive monitoring of psychiatric patients. Changes in behavioral digital phenotypes may serve as early warning signs of relapse or clinical deterioration, creating an opportunity for timely preventive intervention.</p> Methods <p>A systematic search of PubMed, PsycINFO, and IEEE Xplore was conducted for studies published up to January 2026. We included prospective and retrospective observational studies using passively collected smartphone or wearable data to predict relapse or clinical deterioration in individuals with diagnosed psychiatric disorders, with reported quantitative model performance metrics. Study quality was assessed using the modified Newcastle-Ottawa Scale [16] and the PROBAST [25, 26] tool. Data were synthesized narratively in accordance with the SWiM guideline.</p> Results <p>Fifty-two studies encompassing 4,814 participants met inclusion criteria. Disorders studied included schizophrenia spectrum disorders (35%), bipolar disorder (27%), and major depressive disorder (23%). Key predictive features included alterations in sleep patterns (83% of studies), physical activity (83%), GPS-derived mobility (75%), and social communication frequency (65%). Machine learning models reported AUC values ranging from 0.70 to 0.88 for predicting relapse one to four weeks in advance, although the majority of these estimates were derived from internal validation and are likely to overestimate real-world performance. Multi-modal data integration and individual-level modeling consistently outperformed single-modality and population-level approaches. High risk of bias was identified in 75% of studies, primarily attributable to inadequate analytic methodology and reliance on internal validation.</p> Conclusions <p>Passive digital phenotyping demonstrates significant promise for predicting psychiatric relapse across diagnostic categories, with moderate-to-good predictive discrimination (AUC 0.70–0.88) achievable up to four weeks prior to confirmed relapse. However, substantial methodological limitations — including reliance on internal validation, heterogeneous outcome definitions, and limited demographic diversity — must be addressed. Standardized outcome definitions, prospective external validation in diverse cohorts, and closed-loop intervention trials are required before widespread clinical implementation can be responsibly pursued.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Digital phenotyping for predicting relapse in psychiatric disorders: a systematic review of passive sensing approaches

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月9日

查看全文
  • Shih-Shuan Fang,
  • Sheng-Han Chen

摘要

Background

Digital phenotyping — the moment-by-moment quantification of individual-level human behavior using data from personal digital devices — offers a novel approach to continuous, passive monitoring of psychiatric patients. Changes in behavioral digital phenotypes may serve as early warning signs of relapse or clinical deterioration, creating an opportunity for timely preventive intervention.

Methods

A systematic search of PubMed, PsycINFO, and IEEE Xplore was conducted for studies published up to January 2026. We included prospective and retrospective observational studies using passively collected smartphone or wearable data to predict relapse or clinical deterioration in individuals with diagnosed psychiatric disorders, with reported quantitative model performance metrics. Study quality was assessed using the modified Newcastle-Ottawa Scale [16] and the PROBAST [25, 26] tool. Data were synthesized narratively in accordance with the SWiM guideline.

Results

Fifty-two studies encompassing 4,814 participants met inclusion criteria. Disorders studied included schizophrenia spectrum disorders (35%), bipolar disorder (27%), and major depressive disorder (23%). Key predictive features included alterations in sleep patterns (83% of studies), physical activity (83%), GPS-derived mobility (75%), and social communication frequency (65%). Machine learning models reported AUC values ranging from 0.70 to 0.88 for predicting relapse one to four weeks in advance, although the majority of these estimates were derived from internal validation and are likely to overestimate real-world performance. Multi-modal data integration and individual-level modeling consistently outperformed single-modality and population-level approaches. High risk of bias was identified in 75% of studies, primarily attributable to inadequate analytic methodology and reliance on internal validation.

Conclusions

Passive digital phenotyping demonstrates significant promise for predicting psychiatric relapse across diagnostic categories, with moderate-to-good predictive discrimination (AUC 0.70–0.88) achievable up to four weeks prior to confirmed relapse. However, substantial methodological limitations — including reliance on internal validation, heterogeneous outcome definitions, and limited demographic diversity — must be addressed. Standardized outcome definitions, prospective external validation in diverse cohorts, and closed-loop intervention trials are required before widespread clinical implementation can be responsibly pursued.