BACKGROUND: In a postauthorization drug safety study, we assessed the validity of algorithms identifying acute outcomes in patients with type 2 diabetes mellitus (T2DM) initiating a glucose-lowering drug (GLD) other than insulin in the UK Clinical Practice Research Datalink (CPRD).
OBJECTIVE: Estimate positive predictive values (PPV) of diagnosis-coded algorithms for hospitalization for acute kidney injury (hAKI), acute liver injury (hALI), and severe complications of urinary tract infection (sUTI).
METHODS: Patients aged ≥ 18 years initiating a GLD (2012-2018) were identified from CPRD GOLD, a primary care medical records database, with linkage to Hospital Episodes Statistics where available. Case algorithms were: hAKI—hospitalized renal injury diagnosis (excluding chronic renal disease); hALI—hospitalized liver injury diagnosis (excluding chronic liver disease); and sUTI—hospitalized or emergency department visit diagnosis of (1) pyelonephritis or (2) urosepsis (or sepsis within 7 days of urinary tract infection). Hospitalized outcomes were identified as an inpatient diagnosis or a general practitioner (GP)–recorded diagnosis within 30 days of a hospitalization. Clinicians used patient profiles, GP questionnaires (when available), and prespecified case definitions to classify each algorithm-identified case (hAKI, n = 105; hALI, n = 27; sUTI, n = 96) as a confirmed case, noncase, or provisional case (insufficient information to assign case status). PPVs and 95% confidence intervals (CI) were estimated as the proportion of confirmed cases among all algorithm-identified cases: (1) excluding provisional cases from the denominator and (2) including provisional cases (treated as false positives) in the denominator.
RESULTS: Of the GP questionnaires requested, approximately 50% for each outcome were completed and returned. Among the algorithm-identified cases, 54 hAKI, 16 hALI, and 29 sUTI had sufficient information to assign case status; 33 hAKI, 8 hALI, and 14 sUTI were confirmed as cases. The PPVs (95% CI) excluding provisional cases from the denominator were: hAKI 63.0% (48.7%-75.7%), hALI 56.3% (29.9%-80.2%), and sUTI 48.3% (29.4%-67.5%). The PPVs decreased by 49% (hAKI), 41% (hALI), and 70% (sUTI) when including the provisional cases in the denominator.
CONCLUSIONS: Algorithms resulted in low to moderate validity for identifying hAKI, hALI, or sUTI in patients with T2DM in CPRD, with considerable variability in PPV estimates. Our hAKI and hALI results were consistent or higher (hALI) with other comparable algorithms. To our knowledge, there are no published algorithms for our sUTI case definition of pyelonephritis or urosepsis.
