Neurosurgical software device safety in the US: analysis of 2017–2025 FDA Recall and Adverse Event Data

OBJECTIVE Neurosurgeons rely on multiple software applications for surgical planning, intraoperative navigation, and neurophysiological monitoring. Combined with the proliferation of AI-based clinical adjuncts, software as a medical device (SaMD) represents an evolving frontier with profound implications for patient safety. The aim of this study was to systematically analyze all software-related neurological medical device recalls initiated in the US from 2017 to 2025 and to assess the implications of recalls and adverse events for software devices currently in use, as well as the risks posed by emerging software platforms. METHODS Recall and adverse event information related to neurological SaMDs (2017–2025) was extracted from the FDA recall, MAUDE, AccessGUDID, and openFDA databases. Product codes were identified using the FDA Product Classification database. Associated adverse event data were extracted using a custom Python script with the openFDA application programming interface. RESULTS There were 20 unique software-associated neurological device recalls identified across 7 medical device companies and 4 distinct device categories, namely surgical navigation (55%), neurophysiological monitoring devices (25%), surgical planning (15%), and stereotactic devices (5%). More than half of the neurological devices that were recalled had FDA class I severity. Recalls categorized as class I displayed a significant upward trend in annual frequency from 2017 to 2024 (p = 0.027). The median recall times were shorter for class I versus class II recalls (4 vs 10 years, IQR 2–7 vs 6–11 years; Mann-Whitney U = 28, p = 0.182). There were 16,333 adverse event notifications, with 93.4% related to malfunction and 5.9% related to injury. Two surgical navigation devices alone were responsible for 88.8% of all documented adverse events. There was a significantly higher rate of severe adverse harm reported among class II than class I recalls (14.6% vs 3.6%, p < 0.001). More than 10,000 adverse events were reported after formal recall notification for two medical devices. System instability (55.0%) and software defects (43.5%) comprised 98.5% of adverse event notifications. CONCLUSIONS Device recalls and adverse events associated with neurological SaMDs trended toward greater severity of harm, were concentrated in a few high-throughput platforms, and generated a significant volume of adverse events that persisted even after the recall procedure took effect. The observed discordance between recall class and reported patient harm, as well as the prevalence of runtime instability as the most common software failure mode, represent potential areas of improvement for current regulatory frameworks. Adaptation by regulatory bodies and vigilance by providers is critical, particularly for identifying SaMD-related adverse events.