Abstract :
The measurement of quality outcomes is crucial in surgical care. Administrative data are increasingly used but their ability to provide clinically useful information is reliant on how closely the coding can define a particular cohort. In acute admissions for diverticular disease, it is important to differentiate between complicated and uncomplicated diverticulitis, and between diverticulitis and diverticular bleeding. We aim to develop a method to define clinically relevant cohorts of patients from an administrative database in acute diverticulitis. Codes for acute diverticulitis were found from the ICD-10- AM (Australia and New Zealand) coding system, and the accuracy was established with retrospective chart review and crossreferenced with a clinical database at a single institution. Coding of non-diverticular and missed diverticular cases was examined to determine non-diverticular codes that could differentiate these cases. These were combined into logic algorithms designed to differentiate between uncomplicated and complicated diverticulitis admissions derived from an administrative database. Specific K57 diverticular codes possessed sensitivity and positive predictive values of 0.92 and 0.69 for uncomplicated diverticulitis, respectively, with 0.61 and 0.92 for complicated diverticulitis, respectively, based on 153 cases. Most of the missing cases were usually complicated diverticulitis whilst some cases coded incorrectly as uncomplicated diverticulitis were often found as undifferentiated abdominal pain. Diagnostic codes combined into algorithms that accounted for predictable variations improved cohort definition. In conclusion, algorithms with combined codes improved definitions of clinically relevant cohorts for acute diverticulitis from an Australian or New Zealand administrative database. This method may be used to develop logic algorithms for other surgical conditions and enable widespread measurement of relevant surgical outcomes.
Keyword :
Validation; administrative data; diverticulitis; coding; algorithm