AI-Based Triage Decision Support: Multisite Economic Evaluation in the United States.

BACKGROUND: Emergency department (ED) visits have risen in the United States, with demand for emergency care exceeding supply. Resultant ED crowding harms patients, causes staff burnout, and places financial strain on hospitals, payers, and patients alike. Digital tools, including those leveraging artificial intelligence (AI), offer promise for driving efficiency and mitigating the harms of crowding. However, economic frameworks for evaluating these tools remain underdeveloped, limiting adoption. OBJECTIVE: This study aimed to develop and apply a generalizable economic model to assess the financial impact of AI-driven efficiency (improved patient throughput) in emergency care. The model contrasts hospital management versus public policy cost-modeling approaches. METHODS: We applied an economic model to operational and financial data collected from 3 EDs (170,723 visits across 180-day preintervention and postintervention periods) following the implementation of an AI triage clinical decision support system. Revenue was directly measured, and costs and operating margin were estimated using 2 cost-modeling frameworks: hospital management (accounting for fixed, variable, and modifiable costs) and public policy (cost-per-visit approach). Sensitivity analysis was used to generalize the model to changes in patient throughput spanning -10% to +10% for 2 ED scenarios: capacity-constrained EDs with volume increases and volume-stable EDs. Break-even analyses determined the maximum sustainable AI tool cost per visit under each scenario that would preserve operating margin for each scenario. RESULTS: Postintervention revenue increased US $15.4 million (US $138.3 million to US $153.7 million) after AI triage clinical decision support implementation and was accompanied by a 9.6% (7797/81,466) increase in ED visit volume. The financial impact on operating margin from this revenue increase differed substantially by cost-modeling framework. Under the hospital management perspective, costs increased US $2.9 million (US $130.3 million to US $133.2 million), attributing most incremental revenue to an operating margin gain (US $12.6 million). Under the public policy perspective, costs rose proportionally with revenue (US $14.5 million; US $130.3 million to US $144.8 million), generating only US $0.9 million in incremental operating margin. This yielded remarkably different break-even thresholds: US $66.02 per visit (hospital management) versus US $4.69 per visit (public policy) for a 5% (16/311) gain in efficiency (patient throughput) in capacity-constrained ED settings. CONCLUSIONS: The financial impact of tools that promote increased ED efficiency varied greatly based on cost-modeling framework. Traditional policy-level approaches substantially underestimate value from a hospital management perspective. The economic model provides a pragmatic and generalized framework for evaluating AI-driven efficiency in the ED, addressing a critical gap that can limit adoption. The economic model is publicly available and adaptable to local contexts.