BACKGROUND: Previous studies validating cancer outcomes in the United Kingdom's Clinical Practice Research Datalink (CPRD) evaluated algorithms using primary Read (pR) and ICD-10 diagnosis codes. In an ongoing postauthorization drug safety study, we performed a pilot validation of algorithms that also included secondary Read morphology (sR [M]) codes to identify breast and bladder cancer outcomes among individuals with type 2 diabetes mellitus (T2DM) in CPRD.
OBJECTIVES: To estimate positive predictive values (PPV) for algorithms when combining codes and when using individual codes to identify female invasive breast cancer and in situ or invasive bladder cancer among individuals aged ≥40 years with T2DM and taking an antidiabetic drug in CPRD.
METHODS: The outcome-specific algorithms applied to CPRD (2012-2017) included at least one diagnosis code from (1) primary B...00 [Neoplasms] or (2) secondary BB* [M] chapters of the Read medical code listing in CPRD GOLD or (3) ICD-10 diagnosis code in patients linkable to the Hospital Episodes Statistics. For each outcome, a sample of provisional cases were reviewed for clinical case validation using information from clinical patient profiles, general practitioner questionnaires, and prespecified case definitions. The PPV was estimated as the proportion of confirmed cases among all provisional cases in the validation sample. PPVs were calculated for all codes combined with and without sR [M] codes and separately for each code type.
RESULTS: For female breast cancer, 110 provisional cases were reviewed; 98 were confirmed. The PPVs (95% confidence interval [CI]) were: all codes combined, 89.1% (81.7%-94.2%); ICD-10 code C50.9, 69.2% (38.6%-90.9%). The PPVs (95% CI) for the pR codes ranged from 93.6% (85.7%-97.9%) for B34..00 to 100.0% (54.1%-100.0%) for B34z.00. No cases were identified through sR [M] codes. For bladder cancer, 74 provisional cases were reviewed; 56 were confirmed. The PPVs (95% CI) were: all codes combined, 75.7% (64.3%-84.9%); ICD-10 code C67.9, 90.5% (69.6%-98.8%); pR code B49..00, 85.7% (67.3%-96.0%); sR [M] codes, 18.2% (2.3%-51.8%); ICD-10 and pR codes combined and excluding sR [M] codes, 85.7% (74.6%-93.3%).
CONCLUSIONS: Overall, the validity of the algorithms was high (breast, 89.1%) or moderately high (bladder, 75.7%). However, sR [M] codes contributed no additional cases for the breast cancer algorithm and had a low PPV and were detrimental to the overall validity of the bladder cancer algorithm. We recommend to not include sR [M] codes to identify breast or bladder cancers in CPRD.
