When Missing Data Speak: An Adherence-Driven Sparsity Framework for Patient-Generated Telemedicine Data

Telemedicine and mobile health technologies increasingly rely on patient-generated health data collected outside traditional clinical settings. These data promise continuous monitoring, personalization, and early intervention, but in practice, they are often sparse, irregular, and incomplete because data availability depends on sustained patient participation. In this viewpoint, we argue that missingness in patient-generated telemedicine data should not be treated solely as a technical data quality problem. In many contexts, it may reflect patient adherence behavior and, therefore, carry behavioral or clinical meaning. We introduce the adherence-driven sparsity framework, an interpretive and reporting framework for reasoning about missingness in these data. The framework is grounded in 4 principles: patient adherence behavior is a central determinant of data availability, missingness patterns may carry clinically meaningful information, adherence heterogeneity creates differential representation in observed data, and analytical methods should be aligned with actual data characteristics rather than assumed data ideals. The adherence-driven sparsity framework does not propose a new algorithm or validated decision rule. Instead, it provides a structured workflow for preanalysis reasoning: identifying the data generation context; distinguishing adherence-driven from system-driven gaps; characterizing missingness across 6 reporting dimensions; assessing the plausibility of missing completely at random, missing at random, and missing not at random mechanisms; and aligning analytical choices with sparsity, irregularity, heterogeneity, and available evidence. The framework is intended for telemedicine researchers, clinical data scientists, and biostatisticians working with mobile health data. This viewpoint also cautions against the uncritical transfer of machine learning methods from dense, regularly sampled domains to patient-generated telemedicine data. In this setting, methodological sophistication alone is insufficient; researchers must first understand how the data were generated, why observations are missing, and whether missingness itself may be informative. Recognizing adherence-driven sparsity can support more transparent reporting, more defensible analytical decisions, and more clinically meaningful interpretation of patient-generated data.