Using hospital claims data to identify ovarian cancer recurrence: A validation study.

Abstract

e15513 Background: The use of administrative hospital claims data in research studies offers a promising alternative to conducting expensive chart abstraction to assess utilization and health outcomes. However, the validity of using claims data to capture ovarian cancer recurrence is unknown. Methods: We used two strategies to identify recurrent ovarian cancer: 1) a previously validated algorithm of secondary malignancy diagnosis codes used to identify recurrent breast cancer, adapted to ovarian cancer and 2) a new algorithm we created based on the timing and utilization of secondary chemotherapy or secondary debulking surgery for ovarian cancer. To identify secondary debulking surgery, which does not have its own billing code, weobtained codes used by gynecologic oncology billers from across the nation to identify these procedures. We identified all ovarian cancer patients who had undergone primary debulking surgery followed by initiation of primary chemotherapy at a single academic medical institution in NYC, between 2003 and 2010. We then ran the two recurrence algorithms on this population and randomly selected a sample of 50 patients to assess the sensitivity and specificity of each algorithm versus gold-standard chart abstraction data. Results: When applied to our sample of ovarian cancer patients, the sensitivity of the previously validated algorithm was 91% (95% CI: 57% to 99%), the specificity was 28% (95% CI 16% to 45%), the PPV was 26% (95% CI: 14% to 43%) and the NPV was 92% (95% CI: 66% to 99%). In contrast, the sensitivity of our newly developed algorithm to detect ovarian cancer recurrence based on hospital claims data was 100% (95% CI: 68% to 100%), the specificity was 87% (95% CI: 72% to 95%), the PPV was 69% (95% CI: 41% to 88%) and the NPV was 100% (95% CI: 87% to 100%). Conclusions: An algorithm culled from billers and clinical experts identifying timing and utilization of procedures measured ovarian cancer recurrence with a greater degree of accuracy than a previously validated algorithm to detect cancer recurrence. When complex chart reviews are not feasible, our claims-based algorithm can be used by researchers to answer important questions about ovarian cancer treatment effectiveness.