Amazon Web Services (AWS), the technology and e-commerce giant’s cloud subsidiary, has launched Amazon Comprehend Medical, a cloud-based software that information technology (IT) developers, healthcare providers, pharmaceutical companies, and others can use to mine structured data from the unstructured clinician notes in electronic health records (EHRs) and other clinical databases, according to a company release.
The extracted data can be used for clinical decision support, revenue cycle management, clinical trials management, and population health management, Amazon said. The company has previously sold similar text-analysis software to companies outside healthcare for purposes such as travel booking and customer support, according to the Wall Street Journal .
The Journal story quoted Amazon executive Taha Kass-Hout, MD, MS, the former chief health informatics officer for the US Food and Drug Administration, as saying that Amazon’s software, during the testing phase, performed as well or better than similar applications that have been studied. However, it doesn’t appear, according to Kass-Hout, that any breakthrough has been achieved.
A number of other companies, including IBM Watson Health, Optum, and smaller developers and EHR vendors, use natural language processing to identify the key clinical data in text notes and convert that information into structured data that analytic applications can then interpret. The challenge is that doctors use many different terms and abbreviations for the same concept. As a result, even the most advanced forms of natural language processing have limited accuracy and are prone to making obvious errors, such as misinterpreting an order for a test to rule out diabetes as evidence that a patient has diabetes.
Comprehend Medical, says Amazon, “allows developers to identify the key common types of medical information automatically, with high accuracy, and without the need for large numbers of custom rules. Comprehend Medical can identify medical conditions, anatomic terms, medications, details of medical tests, treatments, and procedures.”
Amazon’s data mining application is not paired with clinical analytic software. Instead, the results of the data mining are made available to analytic solutions that can access the data in the Amazon cloud. “Through the Comprehend Medical API [application programming interface], these new capabilities can be integrated with existing services and health systems easily,” Amazon said.
Providers and developers have to send the unstructured data to Comprehend Medical’s cloud server before it can be mined. Amazon says that its service protects patient confidentiality and that the company has no access to the patient data, which requires a viewer to have an encrypted key. Moreover, Amazon notes that “the service is also covered under AWS’s HIPAA eligibility and BAA [business associate agreement].”
Skeptical Responses
It’s difficult to tell how much value the Amazon data mining software might have, Dean Sittig, PhD, professor of biomedical informatics at the University of Texas Health Science Center at Houston, told Medscape Medical News.
One reason, he said, is that Amazon has not released any data to verify the statement that the software is “highly accurate.” For the most part, he said, natural language processing applications are 75% to 80% accurate — far less than what would be required in clinical care, where even 95% accuracy isn’t considered good enough. But in areas such as screening patients for diseases and related medications, he said, the software might prove useful.
Greg Kuhnen, senior director and healthcare IT advisor for The Advisory Board Co, a healthcare consultancy firm, echoed that sentiment, saying that it would be difficult for any data mining application to be accurate and specific enough for clinical decision support.
“However, Amazon seems to have carefully selected the domains they want to target with the new software,” he told Medscape Medical News. “The domains they named — life sciences and clinical trials recruitment — are two of the easier areas to go after because they feed into a human review process that’s fairly tolerant of false positives.”
Building Care Pathways
Healthcare organizations clearly have an appetite for structured data, coupled with analytics, to support quality improvement and population health management. For example, Flagler Hospital, a community hospital in St. Augustine, Florida, uses an artificial-intelligence (AI) application to create optimal clinical pathways for patients with particular conditions and comorbidities, according to Healthcare IT News.
Flagler’s quality improvement team selected and applied the new pathways to order sets in the hospital’s EHR. In its pilot project on pneumonia, Flagler saved $1356 per patient in direct costs and lowered length of stay by 2 days. The hospital is now applying the same deep-learning software to other conditions and expects to save millions of dollars.
Similarly, Penn Medicine in Philadelphia has used deep learning to redesign its care pathways. Its first big success was in predicting which patients were likely to get sepsis in the hospital before they showed any signs of the often fatal illness. One challenge Penn had was that so much of its clinical data was unstructured.
Will data mining of the kind that Amazon is promoting significantly raise the accuracy of AI applications designed to improve the quality of care? Sittig was skeptical. While increasing the amount of structured data fed into AI algorithms might slightly improve the resultant care pathways, he said, it still wouldn’t address the challenge of human variability.
Nevertheless, if the majority of patients benefited, and physicians could alter the protocol to fit the individual patient, this approach would be valuable, he added.
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.