Saati, Amer. An analysis of the effectiveness of a commercial hospital artificial intelligence system in reducing readmissions in high-risk medicare patients. Retrieved from https://doi.org/doi:10.7282/t3-94kf-ct65
DescriptionReadmission within 30 days of hospital discharge and avoidable emergency room (ED) visits have shown to result in an overall increase in healthcare cost and increased risk to patients. Evolving payment models under the ACA focus on reducing unnecessary costs and decreasing short-term readmissions for Medicare patients with common comorbidities. Recent artificial intelligence (AI) models may have the potential to help care coordination team members accurately identify high-risk patients electronically and embed that application within their clinical workflow. This research aimed to evaluate the use of the artificial intelligence-based application in a standalone webform to reduce unplanned 30 days hospital readmissions and unnecessary ED utilization.A commercial vendor AI application was implemented in an enterprise health system in a large metropolitan area between October 2019 and February 2020 that assessed all Medicare admitted patients to 10 major hospitals and generated interventions intended to reduce unplanned 30 days readmissions at those hospitals. In addition, readmission rate change was evaluated for the same hospitals compared with a five months pre-implementation period of a similar population.
Over five months post-implementation, 30 days ED re-visits decreased from 7.7% to 6.5% (p <0.001), and readmission rate decreased from 15.7% to 13% (p <0.001) with 7% absolute risk reduction and 21% relative risk reduction among the high-risk subgroup of patients based on AI application risk stratification algorithm with the number needed to treat to avoid one readmission of 15 for that subgroup. This study may provide valuable insight into creating proper models of population health risk stratification electronic tools by combining risk identification and tailored interventions to that population and establishing best practices for future implementations of AI in population health.