Comparative Algorithms for Identifying and Counting Hospitalisation Episodes of Care for Coronary Heart Disease Using...

Measures of disease burden using hospital administrative data are susceptible to over-inflation if the patient is transferred during their episode of care. We aimed to identify and compare measures of coronary heart disease (CHD) and myocardial infarction (MI) episodes using six algorithms that account for transfers.

We used person-linked hospitalisations for CHD and MI for 2000-2016 in Western Australia based on the interval between discharge and subsequent admission (date, datetime algorithms), pathway (admission source, discharge destination) and any combination to generate machine learning models (random forest [RF], gradient boosting machine [GBM]). The date and datetime algorithms used deidentified patient identifiers to identify records belonging to the same individual. We calculated counts, age-standardised rates (ASR) and age-adjusted trends for CHD and MI for each algorithm.

Counts of CHD increased from 11,733 in 2000 to 13,274 in 2016, while MI increased from 2605 to 4480 using the date algorithm. Correspondingly ASR for CHD decreased from 2086.2 to 1463.1 while MI increased from 468.2 to 498.1 per 100,000 person-years. ASR for CHD and MI for datetime algorithm were consistently 1-2% higher than the date algorithm. Differences in ASR of CHD and MI counts increased over time with the admission source, RF and GBM algorithms relative to the date algorithm. Age-adjusted trends in CHD and MI episode rates using RF and GBM differed significantly from all other algorithms. Only 86.7% and 87.6% of MI episodes identified by the date algorithm were identified by the admission source and discharge destination algorithms, respectively.

The date and datetime algorithms produced the most valid measures of CHD and MI episodes. Findings underscore the importance of identifying admission and discharge dates/times belonging to the same individual in enumerating these episodes.