JMIR Publications

ABSTRACT

The proliferation of telemedicine and mobile health technologies has generated unprecedented volumes of patient-generated health data. In practice, these datasets are frequently dominated by severe incompleteness as participant engagement declines over time. The dominant approach treats this missingness as a technical problem requiring imputation or exclusion. We argue this perspective is fundamentally misguided for patient-generated data: missing data in telemedicine is not noise to be corrected, but behaviorally informative signal to be modeled. We introduce the Adherence-Driven Sparsity Framework (ADSF), a conceptual and practical framework that reconceptualizes data incompleteness in telemedicine. The framework is grounded in four interconnected principles: (1) patient adherence behavior is the primary determinant of data availability; (2) missingness patterns carry clinically meaningful information; (3) adherence heterogeneity amplifies population differences in observed data; and (4) analytical methods must align with actual data characteristics rather than assumed ideals. The ADSF operates across three layers: a paradigm shift layer that establishes the conceptual foundation; a data characterization layer providing six dimensions for systematic assessment of adherence-driven data properties; and a method selection layer offering decision support for matching analytical approaches to data characteristics. This framework bridges behavioral understanding and methodological practice, providing researchers with tools to recognize, characterize, and appropriately analyze adherence-driven sparsity rather than treating it as a nuisance to be eliminated. By centering patient behavior as the generative mechanism underlying data characteristics, we enable more valid and clinically meaningful analysis of patient-generated health data.