Novel machine learning model for predicting multiple unplanned hospitalisations.

Area of Science:

• Health Services Research
• Predictive Analytics in Healthcare
• Public Health Policy

Background:

• Australian public hospitals face increasing demand outpacing funding.
• Identifying patients with high service usage is crucial for resource allocation.
• The Health Links Chronic Care (HLCC) program previously used an algorithm to predict multiple unplanned admissions.

Purpose of the Study:

• To develop a high-performance algorithm for identifying patients at risk of three or more unplanned hospital admissions within 12 months.
• To compare the performance of a new algorithm against existing models.

Main Methods:

• Evaluation of the HLCC and Hospital Unplanned Readmission Tool (HURT) models.
• Utilized data from 34,801 unplanned inpatient episodes (27,216 patients) from 2017-2018.
• Assessed the prevalence of 3+ unplanned admissions in the year post-discharge (8.3%).

Main Results:

• The HURT algorithm demonstrated a significantly higher Area Under the Receiver Operating Characteristic curve (AUROC) of 84% compared to HLCC's 71%.
• Statistical significance was confirmed using the Delong test (p<0.05).

Conclusions:

• The HURT algorithm is a strong predictor of future multi-admission risk, evidenced by its high AUROC, moderate sensitivity, and high specificity.
• Factors such as socioeconomic status and social support emerged as significant predictors of admission risk.
• The HURT algorithm can facilitate targeted support for patients at high risk of recurrent hospitalizations.