Dynamic triage in hypertension: a quantitative framework for follow-up scheduling based on clinician heuristics and home blood pressure data

Background: Hypertension affects about a third of the global population and stays poorly controlled in many patients despite effective treatments being available. Clinical inertia, meaning the failure to intensify or reassess therapy when it is indicated, is one contributor. Home blood pressure (BP) monitoring and other mobile health (mHealth) tools generate frequent readings, but this adds to the interpretive work asked of clinicians. How clinicians turn home BP readings into follow-up scheduling decisions has had little empirical study. Objective: To elicit and characterise clinician scheduling preferences for follow-up appointments in newly diagnosed hypertensive patients, using simulated home BP data, as a first step toward mHealth-based scheduling support. Methods: An online questionnaire presented 15 simulated clinical scenarios depicting newly diagnosed hypertensive patients with 7 days of home BP measurements, and was distributed to physicians in two rounds. Physicians indicated their preferred timing for follow-up (0-5 weeks). We analysed 555 scheduling decisions from 37 physicians using a linear mixed-effects model with a random intercept by physician, which accounts for within-physician correlation. Results: = 0.83; median absolute difference 0.1 weeks). This is an internal goodness-of-fit measure and not out-of-sample validation. The most recent follow-up BP was the dominant driver of scheduling. Above the 140 mmHg systolic threshold, higher BP was associated with earlier follow-up (0.13 weeks earlier per 1 mmHg increase; 95% CI 0.09-0.17). Consultants and non-consultants differed: non-consultants scheduled later follow-up at lower BP but earlier follow-up at higher BP. Conclusions: In simulated scenarios, clinicians appear to use a small set of interpretable rules, chiefly the recent BP value relative to 140 mmHg, when stating preferences for follow-up timing in newly diagnosed hypertensive patients. The formalised rules show meaningful variation between clinicians and give a methodological basis for mHealth scheduling tools. Because the scenarios were simplified and the model has not been tested out of sample, the findings should be read as hypothesis-generating. Whether algorithm-assisted scheduling improves clinical outcomes or reduces workload needs prospective evaluation in real-world settings.