Importance of Point-of-Care Ultrasound for Early Detection of Valvular and Cardiac Diseases (IMPROVE)

Structural heart disease (SHD) - principally degenerative valvular heart disease (VHD) and left ventricular systolic dysfunction (LVSD) - affects an estimated 21.7% of adults aged 65 and older and carries substantial morbidity and mortality, yet detection relies on passive case finding through auscultation and subsequent echocardiography referral. AI-guided point-of-care ultrasound (AI-POCUS) offers a means of systematic screening at the point of routine ambulatory care by operators without formal sonography training. IMPROVE is a hybrid type 1 effectiveness-implementation, stratified, cluster-randomized trial. The unit of randomization is the provider cluster: each participating primary care or geriatrics physician or advanced practice provider, with their patient panel, constitutes one cluster. Thirty-two clusters are randomized 1:1 to AI-POCUS screening or usual care, stratified by hospital system (UT Southwestern Medical Center and Parkland Health \& Hospital System), with approximately 1,088 patients aged 65-85 enrolled over 24 months. In intervention clusters, trained clinic staff perform protocolized AI-guided POCUS using the Kosmos Torso-One device, together with a 2 mL NT-proBNP blood draw. Images are reviewed by a blinded echocardiographic core laboratory; participants with at least moderate VHD or LVEF 50% or less are referred for confirmatory transthoracic echocardiography, with all subsequent management at the treating provider's discretion. Usual care clusters receive standard care and are monitored for echocardiography referral within 90 days of the index visit. All echocardiograms performed within 90 days in either arm are interpreted in the core lab in a blinded fashion. The primary outcome (Aim 1) is a new diagnosis of at least moderate VHD or LVSD confirmed by complete echocardiography within 90 days of screening. Secondary outcomes include downstream care processes and major adverse cardiovascular events over 2 years. Aim 2 evaluates implementation determinants and outcomes - adoption, acceptability, appropriateness, and feasibility - using CFIR and Proctor's taxonomy, through structured workflow observations and semi-structured interviews. The primary analysis is intention-to-treat using a generalized linear mixed model with provider-level random effects.