A Machine Learning Algorithm to Predict Medical Device Recall by the Food and Drug Administration.

Area of Science:

• Medical device safety and regulation
• Artificial intelligence in healthcare
• Diagnostic accuracy studies

Background:

• Medical device recalls are critical in emergency medicine, impacting common devices like ventilators and infusion pumps.
• Early identification of unsafe medical devices is crucial to prevent patient harm.
• Existing FDA systems (MedWatch, MAUDE) can be augmented by other data sources and methods.

Purpose of the Study:

• To evaluate the sensitivity, specificity, and accuracy of a machine learning (ML) algorithm in predicting FDA medical device recalls.
• To assess the predictive capability of ML using publicly available data.
• To determine the lead time for predicting recalls using ML.

Main Methods:

• A random forest regressor ML algorithm was developed.
• The algorithm analyzed Google Trends and PubMed data for recalled (RMD) and non-recalled (NRMD) medical devices.
• The model was trained on 400 devices and tested on 100 devices, assessing performance at 3, 6, and 12 months prior to recall.

Main Results:

• The ML algorithm demonstrated high accuracy in predicting recall status.
• Sensitivity ranged from 75% to 90% at 3, 6, and 12 months prior to recall.
• Specificity was 100% across all tested time periods, with accuracy reaching 98% at 3 and 6 months.

Conclusions:

• Machine learning algorithms can accurately predict FDA medical device recall status with significant lead times.
• This approach offers a promising tool for proactive identification of potentially dangerous medical devices.
• Future research should explore longer lead times and additional data sources for enhanced predictive capabilities.