JMIR Publications

ABSTRACT

Background:

Artificial intelligence (AI)–based models for predicting mortality show potential among intensive care unit (ICU) patients, but evidence on their cost-effectiveness remains limited.

Objective:

This study aimed to evaluate the cost-utility of an AI-based mortality prediction model for ICU patients in Korea.

Methods:

A Markov model compared AI-assisted care with usual care from the payer’s perspective. Model inputs were derived from national claims data, published utility values, and cost references. Outcomes were expressed as incremental cost-effectiveness ratios (ICERs) in Korean won (KRW) per quality-adjusted life year (QALY). Deterministic and probabilistic sensitivity analyses were conducted. Descriptive analyses using the National Health Insurance Service (NHIS) cohort examined ICU transition pathways and mortality patterns.

Results:

The AI prediction model yielded an ICER of 8.37 million KRW per QALY, well below the societal willingness-to-pay threshold of 40 million KRW per QALY in Korea. Probabilistic sensitivity analysis showed an 83.94% probability of cost-effectiveness at the willingness-to-pay threshold. One-way sensitivity analysis identified the true positive rate, AI implementation cost, and post-ICU “Well” utility value as the most influential parameters. Analysis of the NHIS cohort confirmed that ICU transition pathways and patient age were major determinants of survival, with delayed ICU transfer associated with increased mortality.

Conclusions:

AI-based mortality prediction models appear to be both clinically beneficial and economically feasible in ICU settings. By integrating simulation-based economic evaluation with real-world data, this study provides policy-relevant evidence supporting the value of AI-assisted decision support in critical care. Clinical Trial: Not applicable.