We recently published a validation study on cancer cases identified in years 2004–2010 in persons treated for overactive bladder using primary care, hospital, and cancer registry data from the Clinical Practice Research Datalink.1 , 2 Primary care data contained cancer diagnosis records for 68% of the 720 confirmed cancer cases. The proportions of cases that could be identified in each of the three data sources varied by cancer type. We reported that primary care data were most complete for breast cancer and least complete for kidney cancer. Several researchers subsequently asked us to share these proportions for other cancer types to inform their decisions about which data sources within the Clinical Practice Research Datalink should be engaged for their oncology studies. We would like to make this information publicly available. This letter’s eAppendix (http://links.lww.com/EDE/B439) contains numbers and proportions of cases in each of the three data sources, and in combinations of data sources, for each of the 10 cancer types we studied: bladder, female breast, colorectal, corpus uteri, kidney and renal pelvis, lung and bronchus, non-Hodgkin lymphoma, pancreas, prostate, and skin melanoma; patient characteristics are also included. Only 14 of 144 cases (10%) of breast cancer and 31 of 151 cases (21%) of prostate cancer were not identifiable in primary care data. These proportions were considerably higher for other cancer types, particularly renal (18 of 28, 64%), pancreatic (11 of 21, 52%), and lung (48 of 91, 53%): over half of the cases would have been missed if hospital and cancer registry data had not been available. An analogous situation was described for nonfatal acute myocardial infarction: 74% of cases could be identified in primary care data.3 Our findings show that the use of hospital and cancer registry data increased the capture of cancer cases for common cancer types in this cohort of patients treated for overactive bladder. Linkage to these sources was available for approximately 65% of general practices contributing to the Clinical Practices Research Datalink at the time of our study, and the proportion of practices with linkages has increased since then; we assumed these cases represented valid cases. We have not assessed whether these missing cases (without use of the linked datasets) would affect relative risk estimates in the study we have conducted, but for any study in which the completeness of capture of cancer cases is critical, the findings reported here are important to highlight.
