Do prospectively applied machine/deep-learning algorithms improve cardiovascular workflow, patient engagement, and clinical outcomes?
Data-driven clinical AI yields quantifiable efficiency improvements, enhances patient engagement, and reduces all-cause mortality when integrated with actionable decision support.
Artificial intelligence (AI) is progressively utilized in cardiology; nonetheless, the overarching advantages across various care domains remain ambiguous. We conducted a search of PubMed, Embase, CINAHL, and trial registries for randomized controlled trials up to January 16, 2026, assessing prospectively applied interventions based on machine/deep-learning algorithms while excluding rule-based systems. Endpoints were categorized according to NICE evidence tiers: workflow efficiency (Tier A), patient engagement/health promotion (Tier B), and clinical outcomes (Tier C). The risk of bias was evaluated using RoB 2.0. In 32 randomized controlled trials (27 of which were meta-analyzed), artificial intelligence improved all levels. Tier A: workflow time reduced (SMD - 0.71; 95% CI - 1.04 to -0.39), corresponding to a diagnostic time that is 30-120 s shorter and a decrease of 1.0-4.2 hospital days in trials reporting length of stay. Tier B: Behavioral nudging enhanced medication adherence (RR 1.59; 95% CI 1.01-2.50; NNT = 12). Tier C: decision-support implementations decreased all-cause mortality (RR 0.84; 95% CI 0.75-0.94; I² = 8%; NNT = 32). Limitations encompassed restricted blinding and insufficient sham-AI controls. Data-driven clinical AI yields quantifiable efficiency improvements, enhances engagement, and reduces adverse outcomes when integrated with actionable decision support, hence informing a structured framework for governance and implementation.
Lin et al. (Wed,) studied this question.