PITFALLS IN NONRANDOMIZED OUTCOMES STUDIES - THE CASE OF INCIDENTAL APPENDECTOMY WITH OPEN CHOLECYSTECTOMY

Objectives.-To assess the short-term outcomes of incidental appendecto my through analysis of hospital administrative data and determine the consistency and plausibility of the observed results. Design.-Populati on-based historical cohort study. Setting.-All general hospitals in On tario between 1981 and 1990. Patients.-Patients undergoing open primar y cholecystectomy with (7846 exposed) and without (191 599 unexposed) incidental appendectomy. Main Outcome Measures.-In-hospital fatality r ates, complication rates, and lengths of hospital stay. Results.-Crude comparisons showed a striking and paradoxical reduction in mortality after cholecystectomy when incidental appendectomy was performed (odds ratio [OR], 0.37; 95% confidence interval [CI], 0.23 to 0.57; P<.001) ; mean length of stay was also lower by -0.46 day (P<.001). After adju stment for confounding differences, such as comorbidity and nonelectiv e surgery, mortality and lengths of stay were similar for exposed and unexposed patients; but exposed patients showed a significant increase in nonfatal complications (OR, 1.53; 95% CI, 1.39 to 1.68; P<.001). A dverse effects from incidental appendectomy emerged consistently for a ll three outcomes only after restricting the analysis to subgroups of patients at low surgical risk. The increased mortality for exposed pat ients was largest among low-risk groups; for example, among those youn ger than 70 years undergoing elective surgery, the OR was 2.65 (95% CI , 1.25 to 5.64; P<.001). Conclusion.-These findings suggest that incid ental appendectomy is associated with a small but definite increase in adverse postoperative outcomes. However, plausible and consistent fin dings were only obtained after restricting the analysis to low-risk su bgroups in which unmeasured differences in patients' baseline characte ristics were less likely to confound adjusted outcome comparisons. Thi s exercise highlights the potential pitfalls in nonrandomized outcomes comparisons using data sources with limited clinical detail, such as hospital discharge abstracts,