Research Shows Machine Learning Can Help Identify Patient Safety Reporting Patterns to Prevent Harmful Events.

Research from the MedStar Health National Center for Human Factors in Healthcare demonstrates machine learning algorithms can identify patterns in patient safety reports,  allowing faster intervention and preventing harmful events.

Online banking has turned financial tracking on its head—when was the last time you balanced a checkbook or wrote a check? Our researchers are working toward the same paradigm shift in understanding patient safety reports. We’ve shown that machine learning algorithms can reduce and prevent harmful events.

Researchers from the MedStar Health National Center for Human Factors in Healthcare have developed natural language processing algorithms that securely search our databases of patient safety event reports to find trends based on the language in those reports. Published in Scientific Reports, our data showed that these methods save time by supporting analysis of medication error reports and are faster than manual processing.

There are more than 7 million patient safety events in the United States each year. Most safety events lead to reports documented in an electronic system to collect specific data, such as the type of safety issue, whether the patient was harmed, and contain free-text descriptions describing the event. 

Collectively across the nation’s health systems, these reports create a mountain of data in which opportunities to prevent safety events lie. Machine learning can help achieve that goal faster. 

Developing a machine learning approach to patient safety.

For the last seven years, partially supported by a grant from the Agency for Healthcare Research and Quality, our team has focused on developing algorithms to identify trends and prevent safety events. Here’s how it works:

• It searches thousands of reports to find standard connections. For instance, one report may mention a patient falling out of bed, a second details a fall on a slippery floor, and a third also mentions falling in the restroom. 
• The algorithm can identify a pattern related to patient falls and alert a patient safety analyst. 
• The human analyst follows up to determine whether there is a pattern and develops insights that could lead to an intervention, such as a new fall-prevention policy.

This process can also work in reverse. If an analyst is concerned about a particular issue, such as falls, they can ask the algorithm to scour safety reports to look for a pattern. The analyst can then examine selected reports to develop an intervention to reduce falls in the hospital.

Supporting efficiency and safety with machine learning.

Providing these tools complements the work of our human safety analysts, making their jobs easier and enabling them to find better the insights they need to develop interventions that can minimize current and future safety hazards.

Building more efficient processes can benefit providers, too. Improvements in reporting systems and methods can save time, reduce frustration, and build a positive feedback loop that encourages detailed reporting.

Our team has led this research and is excited to be at the forefront of significant advances. Building on MedStar’s longstanding research portfolio focused on patient safety and healthcare equity, we’re closing the gap between collecting data and implementing interventions, making healthcare safer for everyone.