Predicting Future Geographic Hotspots of Potentially Preventable Hospitalisations Using All Subset Model Selection and Repeated K-Fold Cross-Validation.

Matthew Tuson,Berwin Turlach,Kevin Murray,Alistair Vickery,Mei Ruu Kok,David Whyatt

doi:10.3390/ijerph181910253

Abstract

Long-term future prediction of geographic areas with high rates of potentially preventable hospitalisations (PPHs) among residents, or “hotspots”, is critical to ensure the effective location of place-based health service interventions. This is because such interventions are typically expensive and take time to develop, implement, and take effect, and hotspots often regress to the mean. Using spatially aggregated, longitudinal administrative health data, we introduce a method to make such predictions. The proposed method combines all subset model selection with a novel formulation of repeated k-fold cross-validation in developing optimal models. We illustrate its application predicting three-year future hotspots for four PPHs in an Australian context: type II diabetes mellitus, heart failure, chronic obstructive pulmonary disease, and “high risk foot”. In these examples, optimal models are selected through maximising positive predictive value while maintaining sensitivity above a user-specified minimum threshold. We compare the model’s performance to that of two alternative methods commonly used in practice, i.e., prediction of future hotspots based on either: (i) current hotspots, or (ii) past persistent hotspots. In doing so, we demonstrate favourable performance of our method, including with respect to its ability to flexibly optimise various different metrics. Accordingly, we suggest that our method might effectively be used to assist health planners predict excess future demand of health services and prioritise placement of interventions. Furthermore, it could be used to predict future hotspots of non-health events, e.g., in criminology.

Highlights

Introduction published maps and institutional affilIn an environment with limited healthcare resources, it is essential to be able to accurately identify populations with excess burden of disease, in order to avoid inequality and effectively target interventions
Our method will utilise data aggregated to the level of small geographic areas. Such areas are typically administrative boundaries, which exist in many countries, e.g., censusblock groups (CBGs) in the US [42], middle layer super output areas (MSOAs) in the UK [43], and Statistical Areas Level 2 (SA2s) in Australia
Two phenomena are illustrated that are characteristic of all potentially preventable hospitalisations (PPHs) examined: first, positive predictive value (PPV) of the optimal model was lower at higher sensitivity thresholds, and second: clusters of models were observed

Summary

Introduction

Introduction published maps and institutional affilIn an environment with limited healthcare resources, it is essential to be able to accurately identify populations with excess burden of disease, in order to avoid inequality and effectively target interventions. Health service utilisation is often used as an indicator of health inequality; in particular, potentially preventable hospitalisations (PPHs), or ambulatory care-sensitive conditions (ACSCs), are widely used as an indicator of patients’ access to, and the provision and effectiveness of, primary care services [1,2,3,4,5,6] Such hospitalisations are characterised by being potentially avoidable, or preventable, through provision of nonhospital or ambulatory health services; high rates of PPHs may indicate poor functioning of the primary health care system or care inappropriately directed to hospitals [3,7,8].

Methods

Results

Discussion

Conclusion