As healthcare organizations have moved from paper charts to electronic health records (EHR), they’ve gained access to oceans of patient data. Historically, organizations have only been able to mine structured data—information that lives in defined and tagged fields—to inform care and operations. But structured data only accounts for 20% of all healthcare data, and often doesn’t capture the nuances of patient-provider encounters. The other 80% is unstructured data—free-form text entries. While unstructured data contains critical details about patient health, conditions, and treatments, it’s much more challenging for computers to analyze, so it’s largely gone unused by payers and providers.
All that is changing with recent advances in artificial intelligence (AI) techniques such as machine learning. Today, technology can surface insights from unstructured text that were previously difficult to generate and use at scale. With this new wealth of machine learning-derived intelligence, physicians and administrators are empowered to make more timely, informed decisions about patient care and operational programs that impact millions of lives.
What is machine learning?
Machine learning is the science of programming computers to learn and make decisions on their own. Rather than relying on a set of pre-programmed rules, machine learning models are fed data and interactions they can use to develop their own sets of rules and classifiers. The more these models are used, the smarter they get.
Examples of machine learning are everywhere. Corporate chatbots such as Allstate’s Business Insurance Expert help agents better understand and sell their insurance products. It responds to a reported 25,000 inquiries a month. Personal assistants like Siri and Alexa also use machine learning to understand commands and questions.
Machine learning isn’t just being used in consumer applications. Healthcare payers and providers are employing machine learning to extract value from vast quantities of patient charts and other documents to inform operations and care delivery.
How machine learning helps improve healthcare operations
As healthcare organizations transition from fee-for-service to value-based care, they face more pressure than ever to improve outcomes and lower costs. Unstructured data analytics helps payers and providers solve problems more completely and efficiently.
For every hour a physician spends with patients, they spend nearly two hours on paperwork and EHR data entry according to an Annals of Internal Medicine study. Voice-to-text transcription, clinical note summary, auto-documentation, and other machine learning applications can help overworked physicians spend less time at the computer and more time with patients.
Healthcare organizations can also use machine learning to improve risk adjustment. Machine learning algorithms can extract information from clinical charts more quickly and accurately than manual review processes, and become smarter as they’re applied to more documents.
With machine learning, plans and providers can identify hidden risk factors and identify gaps in care, which improves risk score accuracy. A fuller picture of population risk can have a big impact on government reimbursements for Medicare Advantage and Commercial Exchange plans, as well as provider payment rates. By identifying gaps in care, machine learning gives healthcare leaders the information needed to better manage risk and improve the quality of patient care.
How machine learning helps improve clinical outcomes
Health behaviors and socioeconomic factors such as income, social support networks, and education are more important determinants of overall health than what happens inside the doctor’s office. Health organizations recognize that to improve overall health, they must address the whole person, including lifestyle and environment.
Unstructured data provides untapped information on social determinants of health (SDH). Physician, care manager, social worker, and behavioral health notes all provide valuable information about a patient’s health and environment. Machine learning models can be used to identify patients at higher risk for preventable, chronic conditions such as heart disease and diabetes. Health organizations can then reach these patients and provide resources on nutrition, smoking cessation programs, and community groups.
Machine learning also helps improve care inside the hospital. A 2018 study investigated how machine learning can help healthcare providers measure the quality of care for heart failure patients. Using an application called the Congestive Heart Failure Information Extraction Framework (CHIEF), researchers extracted EHR data from 1,083 patients at eight Department of Veteran Affairs (VA) medical centers to determine if physicians prescribed the appropriate medications upon discharge to select patients. Compared to the VA’s manual process, CHIEF identified mentions of heart failure medications and other metrics with a 99.7-99% success rate. The study showed that machine learning provided a cost-effective means to capture data and assess care.
Hospitals use composite risk scores to identify disease risk and health status in the general population. Yannis Paschalidis, professor of engineering and the director of the Center for Information and Systems Engineering at Boston University, used machine learning to improve upon the Framingham Risk Score—the gold standard for predicting heart disease risk—which predicts hospitalizations with 56% accuracy. Working with 50,000 EHRs, Paschalidis and team developed an algorithm that could predict heart disease and diabetes-related hospitalizations about a year in advance with an 82% accuracy rate—a marked improvement over the Framingham Risk Score.
Machine learning is an emerging technology in healthcare with exciting potential. Implementing machine learning gives healthcare organizations the potential to speed up processes, improve operational efficiency, and ultimately provide more effective patient care.