Analyzing cardiac medical device failures with a machine learning approach

Background: Cardiovascular disease is a prominent burden on modern day society. A wide variety of medical devices have been necessary in the treatment and therapy of disease that affect the cardiovascular system. Despite the ubiquitous nature of these devices, limited research exists regarding patient safety and device failure. The US Food and Drug Administration collects medical device data but the problem codes assigned to events as well as data integrity are problematic means of understanding the device failures and patient outcomes. Methods: Supervised machine learning and data filtering methods were used to create tags and identify cardiac medical device failures that were related to: migration, extrusion and expulsion between 1997 and 2017. The results were then used to analyze patient outcomes. Results: Approximately 20% to 21% of cardiac devices were identified from the base dataset, and of that .69% pertained to migrations, extrusions and expulsions. When evaluating results of cardiac and non-cardiac devices, injury was the most frequently occurring adverse outcome of the three failures. Death was an uncommon outcome in cardiac and non-cardiac failures, resulting in low percentages. Statistical significance between cardiac and non-cardiac injury was found. The problem codes associated with records as well as names for devices were unreliable within the realm of this research. Conclusions: Cardiac medical devices account for approximately 20% to 21% of overall medical device events reported to the FDA each year. The risk of injury for medical device failures are high, and data would be more useful for research if accessible and accurate.Ph.D.Includes bibliographical reference