Statistical Methods in Monitoring Adverse Events for Medical Device Postmarketing Surveillance Studies

• Activity Number: 35
• Type: Topic Contributed
• Date/Time: Sunday, August 11, 2002 : 4:00 PM to 5:50 PM
• Sponsor: Biopharmaceutical Section*
• Author(s): Chang Lao*+
• Affiliation(s): Food and Drug Administration
• Keywords: Medical Device Reports ; Trend Analysis ; Model Fitting ; Signaling Detection

Abstract

Several statistical methods are used to analyze medical devices reports (MDRs) received by CDRH/FDA. Nonparametric regressions (polynomial, loess, kernel smooth, and cubic spline smoothing) are used for exploratory trend analysis. Nonparametric trend test, simple Poisson, zero-truncated Poisson, mixed binomial/Poisson, and negative binomial (Gamma-Poisson) probability models are used to signal /monitor sudden spikes of MDRs. Hip implant injury data and the intravenous tube total MDR data are used in our model fitting. Due to unavailable or unreliable denominator (device usage) data, only numerator (MDR) data are used in our analyses. Statistical time series models are also explored, but have not been successful in fitting relatively short, non-stationary time series data. Due to frequently observed over-dispersion of the MDR data, we conclude that the three models (zero-truncated Poisson, mixed binomial/Poisson, and negative binomial) generally fit the observed MDR data satisfactorily and can be used for MDR signaling and monitoring.