How natural language AI speeds up patient incident reporting


ECRI and the PSO of the Safe Drug Practice Institute know that the thousands of patient safety incidents reported in 2021 will never be reviewed.

This Patient Safety Organization It is one of approximately 96 companies across the country, collecting erroneous data on patient injuries and near misses. According to Director Sheila Rossi, member hospitals sent more than 800,000 such reports to ECRI this year.

Federal agencies and PSO Insights can only be gained from a small number of incidents reported each year. The inability to filter all reports will have consequences, although not required by law. However, there is increasing integration between practitioners, PSOs, and the federal government to improve security technologies.

Even a small sample of safety reports can gather insights.Bureau of Healthcare Research and Quality analyze Approximately 300 safety incident reports involved COVID-19 patients seven months before the pandemic. A small sample shows that falls in COVID-19 patients are a problem.

“It takes a long time to obtain data and analyze it, so there are months of COVID-19 patients who will fall, and if we get this information early, the staff may come up with strategies to reduce the falls of COVID-19 patients,” Rossi said.

These delays are mainly due to the nature of the report. There are hard data—such as the age of the patient and the location of the incident—but there are also unstructured data in which the staff writes a summary of the incident and why it happened. Until recently, most insights from patient safety organizations came from analysts manually reading each report.

Raj Ratwani, vice president of scientific affairs at MedStar Health Research Institute, said: “They have to read through hundreds of reports, use tools such as Microsoft Excel to track them, and then rely on memory to establish connections.” “There is a great need for some kind of calculation support.”

Natural language processing can change the field of safety improvement, enabling PSOs and hospitals to quickly query millions of events, connect patient risk points faster, and take intervention measures in a shorter time. The method involves building an algorithm that is trained to understand keywords like a security analyst.

“We hope to shorten the period from problem discovery to notification to members [hospitals]”, Rossi said. “Ultimately, in the long run, it can improve patient safety.

AHRQ Wednesday Released A report to Congress on recommendations for improving patient safety. The agency said it is actively exploring natural language processing to help analyze unstructured narratives.

“Technical solutions…if feasible, can reduce the burden and accelerate data collection and analysis, and will become the preferred method for accelerating shared learning opportunities at the national level,” AHRQ wrote.

It should be noted that although natural language processing has great promise, the development of algorithms is not extensive enough to be widely used.

There are many NLP algorithms available to patient safety organizations, and Ratwani said his organization has developed some tools. But the algorithm is complex, and no one has created a user-friendly way for PSO to gain insights. He said this is similar to showing the GPS to the user, but only showing the work behind the scenes, without a map.

“Our Security Analyst [inside PSOs and health systems] Ratwani said: “They don’t necessarily have to be trained in data science, so we have to create the right layer for them to interact with. As a community of researchers and practitioners, this is what we have to really promote.”

It is also possible for the health system to use NLP on its own. Boston Children’s Hospital This technology has been applied to clinical practice for more than ten years. The hospital will not only check existing safety reports, but will also try to find unreported errors. For example, most emergency departments do not know how often workers fail procedures.

When clinicians perform spinal punctures but do not extract fluid, they do not need to record failed procedures. According to Dr. Amir Kimia, a pediatric emergency physician at Boston Children’s Hospital and an NLP researcher, this happened during “a large number of” spinal taps and still appears on medical records and consent forms.

Kimia and his team developed an NLP method in 2010 using data on failed spinal taps. At that time, when a child came to the emergency department with a seizure, the clinician would usually perform a spinal tap to rule out meningitis.Using NLP, Boston Children’s Hospital can Look for In almost all cases, children do not suffer from meningitis. Most unsuccessful spinal taps are unnecessary.According to the hospital’s research, the American Academy of Pediatrics changed its guide Recommended procedure.

Boston Children’s efforts to improve patient safety are mainly through Grant, And will not be included in operational finance. Kimia and others used these grants to build algorithms. Each institution has its own keywords, which can be related to medical procedures, dialects, and professions.

Although some companies produce NLP products, custom algorithms can be intimidating for hospitals without staff technical expertise. In addition, NLP products need to be integrated into electronic health records and patient safety incident reporting software, which has not happened yet.

“Until it is fully integrated into a platform that can be widely used by everyone, its adoption will be slow,” Ratwani said.



Source link