JMIR Publications

ABSTRACT

Background:

Background:

Mental health crises occur in five out of every 20 adults throughout the year, but medical professionals do not detect 60% of these cases until the condition reaches its most severe point. Current psychiatric diagnosis relies on categorical frameworks and episodic clinical assessments, missing critical transitions in mental state and creating barriers to early intervention.

Objective:

: This research presents a privacy-preserving framework for continuous mental health crisis detection using digital shopping behavior. We introduce the Decision Paralysis–Impulsivity Spectrum (DPIS) as a transdiagnostic behavioral model grounded in dual-process neuroscience and demonstrate its application for detecting episodes of depression and anxiety through e-commerce behavioral signatures.

Methods:

We generated synthetic cohorts (N=12,847 participants) with embedded causal mechanisms producing 847,392 shopping sessions over 12 months. We extracted 67 behavioral features across five domains: temporal patterns, cart dynamics, purchase characteristics, search behavior, and interaction patterns. Machine learning models (XGBoost, Temporal Convolutional Networks, Random Forest) were trained with federated learning and differential privacy (ε=1.0, δ=10⁻⁵). Causal relationships were established through Granger causality analysis, instrumental variables regression, and propensity score matching.

Results:

The ensemble model achieved 85.3% sensitivity and 86.1% accuracy in detecting mental health crises 8-14 days before onset, with an AUC of 0.913 and 18.7% false positive rate. Granger causality analysis demonstrated that DPIS temporally precedes PHQ-9 changes (F=47.3, p<0.001). Instrumental variables regression controlling for confounders yielded a causal coefficient of β=4.2 (SE=0.7, p<0.001). The system maintained strong performance across demographic subgroups with no significant bias (Kruskal-Wallis H=2.31, p=0.68). Adaptive Personalized Differential Privacy improved utility by 23% compared to fixed-ε approaches while maintaining privacy guarantees.

Conclusions:

This work establishes Commercial Behavioral Psychiatry as a new interdisciplinary domain and demonstrates the feasibility of continuous, privacy-preserving mental health monitoring through ubiquitous e-commerce platforms. The DPIS framework provides a neurocognitively grounded, transdiagnostic alternative to categorical psychiatric diagnosis, enabling early intervention before crisis onset. With strong causal validation and robust privacy protections, this approach offers a scalable, stigma-free pathway to addressing the global mental health diagnostic gap.