Information on clinical activity in hospitals is collected through a coding process where data on individual episodes of care are summarized and classified using international classification systems such as ICD10 (with some country-specific variants of the classification system). The resulting data are used for a variety of purposes including for funding hospitals and in measures of quality of care. Therefore, data quality is essential.

The conduct of clinical coding audits is an essential component in quality assurance for clinical coding. Audits measure the accuracy of the clinical coding’s representation of the medical record. However clinical coding auditing is very time-consuming, resource-intensive and there are limited numbers of experienced auditors available to conduct audits. This presentation describes our method to balance statistical robustness of clinical coding audit sample sizes with the practicalities of auditing large numbers of medical records.

The approach is based on a disproportionate stratification by stratum method to allow us to examine each hospital individually, because we do not know if hospital size in terms of counts of episodes reflects the complexity of care at that hospital, or if the count of episodes impacts on the chance of errors in coding. In addition to disproportionate stratification, we deliberately bias the episode selection probability to select episodes from across the spread of complexity at the hospital.

In the analysis step we then calculate our findings for each stratum and extrapolate back to that hospital by adjusting for deliberate bias introduced in the sampling.

The result are sample sizes of episodes of care for clinical coding audit that are robust enough to extrapolate to the whole, while small enough to feasibly conduct the audit with limited resources.; facilitating the conduct of clinical coding audits of while maximising the efficient use of auditor resource. This has practical implications in real-life settings.