Randomization and cluster identification. Randomization occurs at the level of the provider cluster rather than the individual patient. Before the first patient is enrolled, investigators conduct educational sessions with primary care and geriatrics providers at each participating site to introduce study objectives and procedures. Eligible physicians and advanced practice providers are randomized after providing written informed consent. The randomization schedule is generated by the study statistician and revealed to providers only after assignment. A total of 36 provider clusters are randomized over 24 months. Each of the three participating health systems serves as a randomization stratum and contributes 12 clusters, 6 assigned to the AI-POCUS screening arm and 6 to usual care. Provider-level covariates are collected at baseline, including specialty, provider type, full-time equivalent status, and primary practice site.
Mitigation of contamination. Because randomization occurs at the provider level, contamination within shared clinic space is possible through informal knowledge exchange between providers or through patients being seen by more than one provider. Providers are encouraged to adhere to their assigned care pathway, and cross-cover events are recorded prospectively. The primary analysis follows the intention-to-treat principle, with prespecified sensitivity analyses addressing crossover and censoring.
Patient identification and enrollment procedures. Patients in both arms are identified, screened, and enrolled through identical procedures. At each cluster, research coordinators screen clinic schedules and review medical records to confirm eligibility, with questions resolved in consultation with a study principal investigator. Eligible patients receive an invitation letter and information sheet in advance of the visit, or at check-in if added to the schedule late, together with a phone or email opt-out option and notice that a coordinator will meet them in clinic. Coordinators approach patients who have not opted out, review study materials, address questions, and obtain written informed consent before baseline assessments. Study staff proficient in Spanish support enrollment of patients from diverse backgrounds. Coordinators abstract medical and surgical history and the medication list for all enrolled participants.
Symptom classification and accrual balance. Symptom status is determined at enrollment using a brief standardized questionnaire incorporating the Modified Medical Research Council dyspnea scale together with questions on orthopnea and lower extremity edema; participants with a dyspnea score of 1 or greater, or symptoms of congestion, are classified as symptomatic. Enrollment targets an approximate 70:30 ratio of asymptomatic to symptomatic participants in both arms. Because clusters are randomized and enrollment is rolling, participants are not matched at the point of enrollment. Coordinators instead monitor accrual in real time and apply identical symptom, age, and sex distribution targets across arms, adjusting recruitment to maintain balance as enrollment proceeds. To verify comparability, each usual-care participant is matched 1:1 to a screening-arm participant on age within 5 years, sex, and week of enrollment within 2 weeks; this matched set is used only to confirm balance, while the primary analysis includes all enrolled participants.
Study visits and follow-up. Study procedures are integrated into the routine clinic visit, and no additional in-person visits are required apart from a confirmatory echocardiogram when one is ordered by the primary care provider. Remaining study activities are completed during the index visit or remotely by computer, telephone, or mail, including reminders and optional interviews. Participants receive compensation for the screening visit and are offered travel vouchers or transportation reimbursement for follow-up echocardiography as needed. When a provider declines to order echocardiography after a positive screen, the study team ascertains and records the reason, which informs the implementation aim.
Statistical considerations. The sample size assumes a structural heart disease detection rate of approximately 6% in the usual care arm, which deliberately incorporates an allowance for a possible Hawthorne effect, and 12% in the screening arm. Calculations assume an intracluster correlation coefficient of 0.02, a coefficient of variation of 0.5 for cluster size, and that 90% of acquired scans will be interpretable. Under these assumptions, 18 clusters per arm with an average of 40 patients per cluster provide at least 80% power to detect the hypothesized difference at a two-sided alpha of 0.05. The primary comparison uses the generalized Mantel-Haenszel procedure adjusted for clustering, supported by a generalized linear mixed model with provider-level random effects and adjustment for patient-, provider-, and setting-level covariates. Missing data are addressed with multiple imputation and sensitivity analyses. A prespecified comparison of asymptomatic versus symptomatic participants is also planned.
Implementation evaluation. The trial uses a type 1 hybrid effectiveness-implementation design, so implementation is evaluated concurrently with effectiveness rather than sequentially. Adoption of AI-POCUS is assessed through structured observations of clinic workflow. Determinants and implementation outcomes, including acceptability, appropriateness, and feasibility, are examined through semi-structured interviews with providers and clinic staff in both arms and with patients in the screening arm. Findings are intended to inform transportability of the screening strategy to other settings and the development of implementation strategies supporting future dissemination, and will characterize barriers to adoption regardless of whether the screening strategy proves effective.
Oversight. A Stakeholder Advisory Board comprising patients, primary care and geriatrics physicians, imaging experts, health-system leaders, and community advocates meets quarterly to provide structured input on study implementation, recruitment, interpretation of findings, and dissemination. Board feedback is incorporated into study operations through meeting summaries and action plans.
