Malfunction Reporting Patterns in Life-Supporting Versus Noninvasive Ventilators: A Stratified Text-Mining Analysis of the FDA Device MAUDE Database

Background: Mechanical ventilators are essential life-support devices, yet evidence regarding device-related vulnerabilities remains limited compared with the extensive literature on human-factor–related adverse events. This study applied text-mining techniques to a large post-market surveillance database to examine ventilator malfunction reporting patterns and the effect of stratifying reports by device category. Methods: Adverse event reports related to mechanical ventilators submitted to the U.S. Food and Drug Administration’s Manufacturer and User Facility Device Experience database between January 2010 and December 2025 were stratified into life-supporting and noninvasive devices and analyzed using KH Coder. A random sample of the life-supporting reports was drawn for the primary analysis. Frequency analysis, co-occurrence network analysis, hierarchical cluster analysis, and correspondence analysis were performed. Device brand and year were used as external variables in the correspondence analysis. Results: A total of 193,926 reports met the eligibility criteria, of which 177,641 were classified as malfunctions. The analyses were restricted to reports classified as malfunctions. Among life-supporting ventilators, the reported malfunctions spanned several broad themes, including power and battery failures, circuit-board and electronic faults, breathing-circuit and gas-delivery problems, and issues involving monitoring and control functions. Among noninvasive devices, reports were concentrated in terms associated with a single foam-related recall, as well as connector corrosion and patient-reported symptoms. Nearly 1 in 5 malfunction reports did not document a confirmed device failure during patient use. Conclusions: Text mining of ventilator-related adverse event reports revealed distinct reporting patterns across life-supporting and noninvasive devices. These findings are exploratory and do not establish the frequency or causes of device failure. Stratification of reports by device category helped prevent the reporting patterns of individual devices from being obscured by a single recall. This study demonstrates the potential value of combining post-market surveillance data with text-mining methods to support medical device safety surveillance.