Using graph theory to flexibly construct patient journeys in linked healthcare data

Area of Science:

• Health Informatics
• Epidemiology
• Data Science

Background:

• Linked admitted patient data is crucial for epidemiological studies.
• Understanding the complete patient journey across multiple records is essential for accurate health system use analysis.

Purpose of the Study:

• To develop a flexible and generalizable method for aggregating administrative admitted patient records.
• To create distinct periods of hospital care that accurately track patients from admission to discharge.

Main Methods:

• A graph theoretic algorithm was developed to group patient records into hospital care periods.
• The algorithm accommodates complex hospitalisation patterns, including transfers and overlapping records.
• An R package, 'journeyer', was created to implement the algorithm.

Main Results:

• The algorithm was applied to the New South Wales Admitted Patient Data Collection.
• It identified 21,405,451 periods of hospital care from 22,794,746 hospital records.
• The algorithm's parameters were assessed and found suitable for the dataset, with guidance provided for broader application.

Conclusions:

• The developed method enhances data preparation for epidemiological research.
• The approach is generalizable to inpatient data from other jurisdictions.
• Future extensions can incorporate ambulance and emergency department data for a more comprehensive patient journey analysis.