Prospective economic evaluation of a predictive artificial intelligence model for sepsis: Effects on hospital costs and return on investment
7 affiliations
- 1Multidisciplary Sepsis Unit, Son Llatzer Hospital, Palma, Spain
- 2Multidisciplinary Sepsis Group, Health Research Intitute of the Balearic Islands (IdISBa), Palma, Spain
- 3University of the Balearic Islands, Palma, Spain
- 4Fundación Código Sepsis, Valencia, Spain
- 5Faculty of Medicine, Finis Terrae University, Santiago, Chile
- 6Critical Care Center, Clínica Las Condes Hospital, Santiago, Chile
- 7BiometricsAI, Universidad Autónoma de Madrid, Madrid, Spain
Abstract
Early sepsis detection is essential for improving outcomes and reducing costs, but traditional rule-based systems have limited accuracy and real-world evidence for machine-learning alternatives remains scarce. In this context, BIAlert-Sepsis predicts sepsis risk within 24 hours using historical hospital data, and evaluating its implementation in a tertiary hospital setting provides an opportunity to quantify its clinical benefits and economic value. We conducted a retrospective quasi-experimental before–after study including all septic patients admitted from January 2011 to June 2024. The baseline period (Jan 2011 – Mar 2019) was compared with the AI implementation period (Apr 2019 – June 2024), excluding the COVID-19 interval. Outcomes were assessed using adjusted generalized linear models and interrupted time series regression. A hospital-perspective economic evaluation incorporated implementation and maintenance costs, and a 5-year model estimated net benefit and return on investment (ROI), supported by deterministic and probabilistic sensitivity analyses. A total of 8,039 patients were included (6,168 baseline period; 1,871 AI period). Demographic and clinical characteristics were comparable across periods. During the AI period, ICU admissions decreased from 34.4% to 30.4% (adjusted p = 0.001), accompanied by significant reductions of 0.35 ICU days and 0.59 ward days per patient. Mean admission costs declined from 26,517€ to 24,630€ (adjusted p = 0.005). After covariate adjustment, AI implementation was associated with a 26.1–31.1% reduction in mean admission costs across GLM models. Interrupted time series analysis identified a modest immediate cost level change after AI implementation and a larger sustained decline during the post-COVID period. The 5-year economic model projected a cumulative discounted net benefit of 3.55M€ and a 528% ROI. BIAlert-Sepsis was associated with favourable clinical outcomes and lower costs, with economic modelling suggesting early breakeven and positive financial returns. Author summary: Sepsis is a life-threatening condition in which early detection is critical to improving survival and reducing healthcare costs. However, traditional rule-based screening systems often fail to identify high-risk patients early enough. AI offers a promising way to support earlier and more accurate detection.
Related
- On the variability of generative artificial intelligence methods in conceptual modeling: an experimental evaluation on combining causal maps
- Robust Design and Evaluation of Predictive Algorithms under Unobserved Confounding
- Causal Inference with Generative Artificial Intelligence: Application to Texts as Treatments
- The Akaike Information Criterion and Artificial Intelligence
- Correctness and Predictive Uncertainty in Bayesian TSK Fuzzy Regression: A Reproducible Evaluation on Tabular and Building-Meter Data
BibTeX
@article{giglio2026prospective,
title = {{Prospective economic evaluation of a predictive artificial intelligence model for sepsis: Effects on hospital costs and return on investment}},
author = {Andres Giglio and Eric Macias-Fassio and Santiago Salas-Sosa and David Lopez and Cristina Pruenza and Aythami Morales and Marcio Borges-Sa},
journal = {PLOS Global Public Health},
year = {2026},
month = aug,
doi = {10.1371/journal.pgph.0006059},
url = {https://journals.plos.org/globalpublichealth/article?id=10.1371/journal.pgph.0006059},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Prospective economic evaluation of a predictive artificial intelligence model for sepsis: Effects on hospital costs and return on investment
AU - Andres Giglio
AU - Eric Macias-Fassio
AU - Santiago Salas-Sosa
AU - David Lopez
AU - Cristina Pruenza
AU - Aythami Morales
AU - Marcio Borges-Sa
PY - 2026
DA - 2026/08/31
JO - PLOS Global Public Health
DO - 10.1371/journal.pgph.0006059
UR - https://journals.plos.org/globalpublichealth/article?id=10.1371/journal.pgph.0006059
AB - Early sepsis detection is essential for improving outcomes and reducing costs, but traditional rule-based systems have limited accuracy and real-world evidence for machine-learning alternatives remains scarce. In this context, BIAlert-Sepsis predicts sepsis risk within 24 hours using historical hospital data, and evaluating its implementation in a tertiary hospital setting provides an opportunity to quantify its clinical benefits and economic value. We conducted a retrospective quasi-experimental before–after study including all septic patients admitted from January 2011 to June 2024. The baseline period (Jan 2011 – Mar 2019) was compared with the AI implementation period (Apr 2019 – June 2024), excluding the COVID-19 interval. Outcomes were assessed using adjusted generalized linear models and interrupted time series regression. A hospital-perspective economic evaluation incorporated implementation and maintenance costs, and a 5-year model estimated net benefit and return on investment (ROI), supported by deterministic and probabilistic sensitivity analyses. A total of 8,039 patients were included (6,168 baseline period; 1,871 AI period). Demographic and clinical characteristics were comparable across periods. During the AI period, ICU admissions decreased from 34.4% to 30.4% (adjusted p = 0.001), accompanied by significant reductions of 0.35 ICU days and 0.59 ward days per patient. Mean admission costs declined from 26,517€ to 24,630€ (adjusted p = 0.005). After covariate adjustment, AI implementation was associated with a 26.1–31.1% reduction in mean admission costs across GLM models. Interrupted time series analysis identified a modest immediate cost level change after AI implementation and a larger sustained decline during the post-COVID period. The 5-year economic model projected a cumulative discounted net benefit of 3.55M€ and a 528% ROI. BIAlert-Sepsis was associated with favourable clinical outcomes and lower costs, with economic modelling suggesting early breakeven and positive financial returns. Author summary: Sepsis is a life-threatening condition in which early detection is critical to improving survival and reducing healthcare costs. However, traditional rule-based screening systems often fail to identify high-risk patients early enough. AI offers a promising way to support earlier and more accurate detection.
SN - 2767-3375
ER -