A hybrid Bayesian network for medical device risk assessment and management

Abstract

Risk analysis methods for medical devices, including fault tree analysis, have limitations such as handling uncertainty and providing reasonable risk estimates with limited or no testing data. To address these limitations, this paper proposes a novel systematic method for medical device risk management using hybrid Bayesian networks (BNs). We apply the method to a Defibrillator device to demonstrate the process involved for risk management during production and post-production using 4 different scenarios: (1) where there are available testing data; (2) where there are limited or no testing data; (3) where it is a completely new device with no testing data; (4) where we are reassessing the risk of a previous model on the market based on reported hazards and injuries. In each scenario, the BN model, for the available data, provides the full probability of failure per demand distribution for each category of injury severity (fatal, critical, major, minor, negligible) and the probabilities associated with various risk acceptability criteria. The model results are validated using publicly available data for the LIFEPAK 1000 Defibrillator (PN: 320371500XX), which was recalled by Physio-Control in 2017. The results show that the device would fail the acceptability criteria for probability of fatal injury.