Predicting Future Geographic Hotspots of Potentially Preventable Hospitalisations Using All Subset Model Selection...

Area of Science:

• Health Services Research
• Spatial Epidemiology
• Predictive Analytics

Background:

• Identifying geographic areas with high rates of potentially preventable hospitalisations (PPHs) is vital for targeted health interventions.
• Place-based interventions are resource-intensive and require accurate long-term predictions, as hotspots can naturally fluctuate.

Purpose of the Study:

• To introduce and validate a novel statistical method for predicting future hotspots of PPHs.
• To compare the proposed method's performance against common alternative prediction strategies.

Main Methods:

• Utilized spatially aggregated, longitudinal administrative health data.
• Developed a method combining all subset model selection with repeated k-fold cross-validation.
• Optimized models by maximizing positive predictive value while maintaining a minimum sensitivity threshold.

Main Results:

• The proposed method demonstrated superior performance in predicting three-year future hotspots for type II diabetes mellitus, heart failure, chronic obstructive pulmonary disease, and high-risk foot in Australia.
• The method showed favorable performance compared to predictions based on current or past persistent hotspots.
• The approach offers flexibility in optimizing various performance metrics.

Conclusions:

• The developed method effectively assists health planners in predicting excess future demand for health services.
• This approach can prioritize the placement of health interventions and potentially be applied to predict future hotspots in other fields, such as criminology.