Malignancy validation in a United States registry of rheumatoid arthritis patients

Background: Physician reporting is commonly used to ascertain adverse events or outcomes measured in epidemiologic studies. However, little is known on the accuracy of physician reported malignancies compared to pertinent medical record review in large cohort studies. Methods: The Consortium of Rheumatology Researchers of North America (CORRONA) registry gathers physician-completed questionnaires for rheumatoid arthritis (RA) patients, including request for information on incident malignancies, approximately every three months. For incident malignancies reported from October 1st, 2001, through December 31st, 2007, we retrospectively requested completion of a Targeted Adverse Event (TAE) form for additional information as well as primary source documents to adjudicate the malignancy reports. CORRONA has employed a prospective request for source documentation for these events since 2008. We classified each malignancy as definite, probable, possible, or not a malignancy. Results: From 20,837 RA patients enrolled in CORRONA, 461 incident malignancies were initially reported on physician questionnaires. After review of returned source documents with adjudication, 234 were deemed definite, 69 probable, 101 possible, and 57 not an incident malignancy. The positive predictive value (PPV) of initial physician report of a malignancy versus "definite or probable" malignancy based on adjudication was 0.66 (95% CI 0.61 - 0.70). The PPV was 0.68 (95% CI 0.63 - 0.72) when the subsequent TAE form also confirmed the presence of malignancy. When possible malignancies were included, the PPV of physician-reported malignancies without a subsequent TAE form increased to 0.86 (0.83 - 0.89), and with a subsequent TAE form, 0.89 (0.85-0.91). Conclusion: Twelve percent of initial physician reports of incident malignancy could not be confirmed with review of source documents. The most common reason for lack of confirmation was inability to obtain documents or insufficient data in source materials. These results suggest that timely collection of relevant medical records and an adjudication process are required to improve the accuracy of cancer reporting in epidemiologic studies.