Healthcare - Flipbook

An Introduction to Processing Unstructured Medical Data Using Amazon Comprehend Medical

Issue link:

Contents of this Issue


Page 0 of 1

Unstructured clinical documents such as hospital admission notes and patient medical histories contain critical health information necessary for efficient, high-quality care delivery. However, this data can prove difficult to extract in an efficient and meaningful way. Healthcare organizations generate about 1.2 billion unstructured clinical documents each year. In 2018, Amazon Web Services (AWS) released Amazon Comprehend Medical, a natural language processing (NLP) service makes it easy to use machine learning to extract relevant medical information from unstructured text. In a recent webcast, AWS explained how healthcare providers and health IT companies can use Amazon Comprehend Medical to retrieve unstructured patient data and leverage this information to satisfy a variety of use cases. Currently, healthcare organizations often process unstructured data through coding. "If you look at healthcare in general, a lot of retrospectives are done through coding whether it's an ICD-10 code, RX, SNOMED," said Arun Ravi, Senior Product Manager for Amazon Comprehend Medical. "There's a loss of meaning as you go from the unstructured text to actual codes." Amazon Comprehend Medical is a HIPAA-eligible service that allows users to easily index patient data at the source to create more powerful applications based on accurate information. The service does not store any consumer data that runs through its service. Providers and developers can use the tool to enhance existing health IT products or assist in building new products more quickly. The tool's capacity to facilitate innovation is key as the healthcare industry transitions to a value-based care system. The service extracts and contextualizes unstructured medical data through a three-step process, Ravi explained. First, Amazon Comprehend Medical analyzes unstructured clinical documents and extracts data in five categories: medications, medical conditions, test treatment procedures, anatomical terms, and protected health information (PHI.) When Amazon Comprehend Medical pulls data about medications, it may extract data about dosages, route modes, strength, and frequency. "Now, if I give you all these entities at once, it's valuable, but 15 milligrams by itself really doesn't tell you much," explained Ravi. "So we realized we had to take it one step further. The next step that we process through the API is called relationship extraction." Relationship extraction attaches information within subtypes, such as dosage or frequency information, to its parent category. "The dosage, router mode, strength, and frequency is actually tied to the medication," explained Ravi. "If there is a dosage and route mode associated with the medication, you'll see it associated to the right parent, and you'll see a relationship score that allows you to see if it's the right relationship for you to take in your downstream applications." The third step of the process provides contextual information to fill out a complete picture of the extracted data. Contextual data are referred to as traits. One example of a trait is negation. "Negation could be that a patient denies taking a medication," said Ravi. "That's really important when you're trying to understand what data elements you need to take for your downstream applications," he noted. Another trait differentiates data that can be classified as a diagnostic sign or symptom. A sign is patient-reported, while a symptom is physician-reported. "Being able to take different features from existing NLP platforms and put it into a simple API call is unique, and we see this as very powerful and as an enabler for our customers," said Ravi. An Introduction to Processing Unstructured Medical Data Using Amazon Comprehend Medical

Articles in this issue

Links on this page

view archives of Healthcare - Flipbook - An Introduction to Processing Unstructured Medical Data Using Amazon Comprehend Medical