Among 26,710 hypertension patients, latent class trajectory modeling of electronic health record (EHR) visit patterns identified four reproducible engagement phenotypes across both 1-month and 3-month analytical intervals, with strong bootstrap reproducibility (mean ARI 0.966–0.969). Higher-contact phenotypes correlated with greater treatment documentation rates, but blood pressure control differences between groups were small and inconsistent after adjusting for age, sex, race, baseline BP, and comorbidity burden. Critically, 79.2% of patients had fewer than 24 months of follow-up, with dropout concentrated in lowest-contact groups — revealing that these phenotypes largely reflect administrative data availability rather than genuine patient disengagement.
This preprint, not yet peer-reviewed, raises an important methodological caution for the growing field of EHR-based phenotyping: trajectory patterns in real-world data are deeply entangled with how health systems generate and retain records. The finding that the highest-contact group did NOT achieve the lowest blood pressure — despite the largest first-to-last reduction — exposes observation-window bias that could mislead clinicians and policymakers into conflating frequent visits with better outcomes. For longitudinal cardiovascular research, this work underscores that administrative censoring is not random noise but a systematic confound. Practically, these phenotypes appear more useful for identifying documentation gaps than for predicting BP control. Until externally validated and peer-reviewed, applying these trajectory labels to guide clinical decision-making would be premature. The study is best read as a methodological warning about EHR-derived engagement metrics in chronic disease management.