Identifying episodes of care in hospital admissions data for measures of disease burden: A tutorial and protocol for individual-level data analysis

Background

We are not aware of any comprehensive, publicly available, standardised protocol or syntax for the processing of hospital admissions data for individual-level analysis. Failure to appropriately process and analyse data in a standardised manner could lead to misestimation of event rates, inconsistency between studies, and incorrect findings informing clinical practice and health policy.

Aim

To develop an open source, standardised protocol for processing of admitted episodes data that can be regularly updated.

Methods

We identified common data structures that require processing to define single episodes of care (i.e., events) and developed Stata code to address these. We then presented a full worked example using UK admission data. The code is stored on a public online platform that allows living updates. Results: Common data structures requiring processing include duplicated records, shorter records within a longer period of care, and mis-coded transfers. Using the UK admission data sample, data processing resulted in 33,170 records with myocardial infarction as the primary diagnosis being refined to 18,289 episodes of care (i.e., events). The ratio of records to episodes of care varied for different primary diagnoses: for example, for lung cancer, there were 29,274 records and 26,389 events; for pneumonia, 21,029 records and 12,334 events; and for head injury, 21,957 records and 17,736 events.

Conclusion

Appropriate data processing is vital to derive accurate results from hospital admissions data. We have presented open source, live syntax for this purpose.