Related Experiment Video
Updated: Jan 5, 2026
06:55
Inverse Probability of Treatment Weighting Propensity Score using the Military Health System Data Repository and National Death Index
Published on: January 8, 2020
15.0K
An Interpretable Algorithm on Post-injury Health Service Utilization Patterns to Predict Injury Outcomes
Hadi Akbarzadeh Khorshidi1,2, Behrooz Hassani-Mahmooei3, Gholamreza Haffari4
1School of Computing and Information Systems, The University of Melbourne, Melbourne, Australia. hadi.khorshidi@unimelb.edu.au.
Journal of Occupational Rehabilitation
|October 18, 2019
Summary
This study groups transport accident patients by health service utilization (HSU) to predict recovery costs. Eight distinct patient groups were identified, significantly improving cost prediction accuracy for better healthcare management.
Area of Science:
- Health Services Research
- Data Science
- Injury Prevention
Background:
- Post-injury health service utilization (HSU) impacts injury outcomes, yet its relationship with patient stratification and cost prediction remains under-explored.
- Limited research exists on grouping injured patients based on HSU patterns to predict recovery costs and outcomes.
Purpose of the Study:
- To develop a method for grouping injured patients from transport accidents based on early HSU data.
- To identify patient clusters that are meaningfully associated with total recovery costs and patient outcomes.
- To create predictive models for injury outcomes using discovered HSU patterns and classify future patients into these clusters.
Main Methods:
- A hybrid approach combining unsupervised and supervised machine learning methods was applied to a dataset of 20,692 injured patients.
- Clustering was performed using the first week's post-injury health service utilization data to identify patient groups associated with total cost to recovery.
- Decision tree classifiers were utilized to classify future patients into the discovered clusters based on their initial HSU information.
Main Results:
- Eight distinct patient groups were identified with well-defined clusters, indicated by an Average Silhouette Width of 0.71.
- The identified patient groups significantly improved the predictability of injury costs, more than doubling the accuracy compared to traditional predictors like gender, age, and injury type.
- These patient groups demonstrated a substantial association with patient recovery trajectories.
Conclusions:
- The study provides a novel framework for discovering actionable insights into HSU patterns for transport accident survivors.
- The developed patient grouping and classification system offers valuable tools for health service providers and policymakers to manage injury outcomes and reduce accident severity.
- The transparency of decision tree classifiers facilitates the integration of findings into operational healthcare processes.
