Track: Methods in Pharmacoepidemiology -> Measurement Methods - Validation methods to confirm data (including Patient Reported Outcomes development and validation); methods to deal with missing data
(PO-3811) Validation Of Acute Outcomes Among Patients With Type 2 Diabetes Mellitus In Medicare: A Pilot Study
Monday, September 14, 2020
Heather E. Danysh, J. Bradley Layton, Daniel C. Beachler, Alejandro Arana, Manel Pladevall-Vila, Rita Schmid, Brian Calingaert, Ryan Ziemiecki, Phillip R. Hunt, Alicia Gilsenan and Catherine Johannes
Background: In an ongoing postauthorization drug safety study, a pilot assessment was conducted in the United States (US) Medicare claims database to evaluate the validity of algorithms to identify acute outcomes among individuals with type 2 diabetes mellitus (T2DM) who initiated an antidiabetic (AD) drug.
Objectives: To estimate the positive predictive values (PPV) of claims-based algorithms for hospitalization for acute kidney injury (AKI), acute liver injury (ALI), and severe complications of urinary tract infection (UTI).
Methods: Patients were fee-for-service US Medicare enrollees aged ≥ 65 years initiating an AD drug (2014-2015). Algorithms were: AKI-hospital diagnosis of acute renal failure or dialysis, ALI-hospital diagnosis of ALI or liver transplant, UTI-inpatient or emergency department diagnosis of (1) pyelonephritis or (2) sepsis within 7 days of a UTI diagnosis. Patients with chronic renal or liver disease were excluded from AKI and ALI analyses, respectively. Medical records were requested for algorithm-identified cases (AKI n = 150, UTI n = 150, ALI n = 59). Up to 75 received patient records were included in each outcome-specific validation sample for clinical review. Using abstracted medical record information and prespecified case definitions, adjudicators classified each algorithm-identified case as a confirmed case, confirmed noncase, or postreview provisional case (insufficient information to assign case status). PPVs and 95% confidence intervals (CI) were estimated as the proportion of confirmed cases: (1) among all cases in the validation sample (postreview provisional cases treated as false positives); (2) removing postreview provisional cases from the denominator.
Results: We obtained 62.7% (AKI n = 94), 76.3% (ALI n = 45), and 53.3% (UTI n = 80) of requested patient records. Algorithm-identified cases included in the validation samples: AKI n = 75, UTI n = 75, ALI n = 38. The PPV (95% CI) for the algorithms (1) including all cases and (2) excluding postreview provisional cases were: AKI (1) 46.7% (35.1%-58.6%), (2) 58.3% (44.9%-70.9%); ALI (1) 52.6% (35.8%-69.0%), (2) 55.6% (38.1%-72.1%); UTI (1) 68.0% (56.2%-78.3%), (2) 78.5% (66.5%-87.7%).
Conclusions: Claims-based algorithms resulted in moderate validity for identifying hospitalizations for AKI, ALI, or severe complications of UTI among older patients with T2DM in Medicare, with considerable variability in PPV estimates. Our AKI and ALI results are consistent with other comparable algorithms in the literature, and, to our knowledge, there are no published algorithms for our UTI case definition of pyelonephritis or urosepsis.