Identification of risk factors of 30-day readmission and 180-day in-hospital mortality, and its corresponding relative importance in patients with Ischemic heart disease: a machine learning approach

ABSTRACT

Background: The primary objective of this study is to identify non-laboratory predictors for 30-day hospital readmission and 180-day in-hospital mortality rates among patients hospitalized with ischemic heart disease (IHD).

Research design and methods: This is a retrospective cohort study of hospitalized patients (≥ 40 years) with a primary diagnosis of IHD. Data were extracted from the Florida Agency for Health Care Administration dataset from 2006 to 2016. A machine learning approach was used to identify predictors of 30-day hospital readmission and 180-day in-hospital mortality.

Results: 346,390 patient records for incident IHD cases were identified. The top two predictors of 30-day readmission were the length of stay and the Elixhauser comorbidity index for readmission [ECI] (Area Under the Curve [AUC]=88%) using decision tree algorithms. For in-hospital mortality, the top two predictors were LOS and ECI (AUC=92%) using gradient boosting regressors. The cumulative 30-day readmission and the 180-day probability of mortality rates were 9.82% and 4.6% respectively.

Conclusions: Risk factors of 30-day readmission and 180-day mortality in hospitalized IHD patients identified by machine learning and their relative importance (value) will help pharmacists and other health care providers to prioritize their disease management strategies as they improve the care provided to IHD patients.

KEYWORDS:

• Cardiovascular disease
• ischemic heart disease
• machine learning
• quality measures
• risk factors

Article Highlights

• Ischemic Heart Disease (IHD) is associated with an increased risk of readmissions and mortality.

• Much is known about pharmacist use of laboratory variables in ascertaining the risk of mortality and readmission in patients with IHD

• Machine learning approach has been effectively deployed in identifying risk factors in several aspects of medical data analysis or image recognition

• Machine-learning algorithm was used to identify predictors of 30-day readmissions and in-hospital mortality and its relative ranking of importance

• The identified non-laboratory variables, including hospital LOS and the number of comorbidities, could be used by pharmacists and other health care providers to identify risk patients with IHD and develop a cost-effective intervention to optimize IHD outcomes.

Author contribution statement

“A.N.O obtained grant and conceived the presented idea. V.S, V.D performed machine learning analysis. H.A performed data cleaning and analysis. All authors discussed the results and contributed to the final manuscript.

Declaration of interest

The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Supplementary material

Supplemental data for this article can be accessed here.

Additional information

Funding

This project was funded by Florida A&M University Institutional Faculty Research Awards Program.