Methodischer bias retrospektiver follow-up-untersuchungen und sein einfluss auf die erhebung von qualitätssicherungsdaten

Background: It seems reasonable to assume that standardized quality assurance and adherence to guidelines in the treatment of breast cancer will lead to an improvement in disease-free and overall survival. However there are still no prospective studies demonstrating this relationship. Because of the absence of a national clinical cancer registry in Germany, survival data are often collected using retrospective studies which are prone to different errors and biases. Material and Methods: Survival information was retrospectively collected on 710 breast cancer patients undergoing surgical treatment at the University of Frankfurt in the years 19971999. Complete data were obtained for 471 patients. The influence of selection bias and information bias on data quality was analyzed. Results: The study cohort was characterized by an elevated rate of patients with higher risk (lymph node positive and larger tumor size) compared to large population-based reference cohorts. Interestingly, the median follow-up time for the group of deceased patients was 1.7 times longer than that of censored cases. This is in contrast to other studies demonstrating an inverse relationship, whereby the proportion of drop-outs from follow-up is higher among disease-free patients in the study, and suggests an information bias. A simulation experiment was performed to estimate the influence of this bias and a possible difference of 13.9% in the 5-year overall survival was suggested (73.5±3.1% vs. 87.4±1.5%). Conclusion: Retrospective studies on survival are influenced by several confounding variables in addition to classical clinical parameters. These factors complicate comparisons with reference cohorts and could result in a misleading assumption of a worse quality of treatment.