PATIENT-LEVEL SIMULATION-BASED SENSITIVITY ANALYSIS TO EVALUATE THE IMPACT OF UNDER-RECORDING OF UNDERLYING COMORBIDITIES IN A RETROSPECTIVE CLAIMS DATABASE ANALYSIS

• Published in: Value in health Vol. 20; no. 5; p. A318
• Main Authors: Zorina, O, Ferreira, A, Korn, JR, Bezlyak, V, Sagkriotis, A, Skelly, A, Zangvil, E, Snow, H, Zuurman, L, Holekamp, N
• Format: Journal Article
• Language: English
• Published: Lawrenceville Elsevier Science Ltd 01.05.2017
• Subjects: Age differences; Analysis; Bevacizumab; Bias; Comorbidity; Confidence intervals; Confounding factors; Demography; Drug therapy; Macular degeneration; Medical history; Monoclonal antibodies; Ophthalmology; Population studies; Sensitivity analysis; Simulation; Standardized patients; Statistics
• ISSN: 1098-3015; 1524-4733
• DOI: 10.1016/j.jval.2017.05.005

OBJECTIVES: To evaluate the impact of under-recording of comorbidities in a retrospective claims database analysis of patients with claims for neovascular age-related macular degeneration (nAMD) treated with ranibizumab or unlicensed bevacizumab. METHODS: A retrospective analysis was conducted in the QuintilesIMS Integrated Data Warehouse (IDW) database to compare reported incidence of stroke between ranibizumab (n=44949) and unlicensed bevacizumab (n=131718). Event rates were compared using rate ratios (RR) within a multivariate Poisson regression, adjusted for baseline demographics and comorbidities. Frequencies of reported comorbidities were much lower than those reported in similar studies of nAMD populations (Curtis 2010; Gower 2011), potentially due to the open nature of the IDW database, which may not integrate all patient-level medical encounters. Over 40% of patients did not have any comorbidity recorded. This finding was unexpected owing to the typical age profile of nAMD patients. A patient-level sensitivity analysis was conducted to adjust for under-recording of medical history in the IDW database. Comorbidity records from the patients with at least one recorded comorbidity were randomly sampled with replacement and used to substitute records for the patients without recorded comorbidities, stratified by index drug, age group, gender and stroke event RESULTS: The adjusted RR for the reported rate of prevalent stroke for ranibizumab was comparable with unlicensed bevacizumab (0.993 (95% CI: 0.961, 1.025)). Following the simulation-based sensitivity analysis, the adjusted RR for the reported rate of prevalent stroke was 0.964 (95% CI: 0.932, 0.997). A similar effect was observed for incident stroke. CONCLUSIONS: Under-recording of potential confounding factors may lead to biased estimates even if non-differential between groups. With large sample size resulting in narrow confidence intervals, even small magnitude of resulting bias may affect the statistical interpretation of the study results. Therefore, it is important to utilize appropriate statistical methods to detect and correct for comorbidity under-recording.