Online Behavioral Digital Phenotyping for Depression within the RDoC Framework: A Scoping Review

Depression, affecting over 322 million individuals worldwide, is a leading cause of global disability [1]. Timely identification and assessment are prerequisites for effective treatment. However, conventional clinical screening primarily relies on retrospective self-reports (e.g., PHQ-9, HAMD). These methods are susceptible to recall bias [2], often fail to capture the heterogeneity and temporal dynamics of depressive disorders [3], and can be further compromised by stigma-related response bias [4]. Consequently, traditional approaches fall short of meeting the growing demand for objective and continuous depression assessment. The emergence of digital phenotyping offers a promising avenue to address these limitations. Torous et al. [5] define digital phenotyping as "the moment-by-moment quantification of the individual-level human phenotype in situ using data from personal digital devices." By passively collecting smartphone data—such as mobility (via GPS), sociality (via call and text logs), and sleep (via accelerometer and screen activity)—and linking it to self-reported and clinical measures [6], this paradigm enables continuous, ecologically valid monitoring. Behavior, in the psychological context, refers to any observable and recordable action of an organism [7]. This scoping review focuses specifically on "online behavioral markers," defined as passively collected, quantifiable sequences of user-device interactions at the smartphone interface and application layer (e.g., keystrokes, screen touches, app usage). It excludes offline contexts (e.g., GPS), behavioral derivatives (e.g., textual content), and physiological biomarkers, concentrating purely on fine-grained, operational behaviors in digital environments. These granular interaction features constitute a critical bridge connecting digital footprints to the psychopathology of depression. Despite rapid growth in the field, existing reviews often conflate online behaviors with their derivatives or physiological sensor data (e.g., heart rate, step count) [8-13], lacking a dedicated synthesis of pure online interaction modalities. High heterogeneity in feature definition and computational methods, coupled with an absence of standardization, hinders comparison, integration, and replication of findings. Furthermore, digital phenotyping data are well-suited for symptom-level screening, as they can be used to derive digital proxies for specific symptoms [14, 15]. This aligns with the shift towards dimensional and transdiagnostic frameworks in mental health research, exemplified by the U.S. National Institute of Mental Health's Research Domain Criteria (RDoC) framework, which emphasizes integrating units of analysis—including behavior—to understand mental disorders [16]. Systematically mapping online behavioral markers and interpreting them through theoretical frameworks like RDoC is therefore crucial for advancing depression assessment from label-based diagnosis towards a mechanistic, dimensional understanding. This scoping review aims to: Systematically identify? key online behavioral modalities (calls, SMS, screen use, application use, keystroke dynamics) and their specific features used for depression monitoring and assessment. Comprehensively summarize? the methodological landscape concerning data collection, preprocessing, feature engineering, and machine learning modeling in this field, advocating for the standardization of feature engineering. Construct a preliminary mapping? of behavioral features onto RDoC-based psychopathological dimensions to enhance the clinical interpretability of digital markers and propose directions for future research.