Hype vs. reality in health care AI: Real-world approaches that are working today

Every major technological breakthrough has taken time to reach full potential. You might be reading this on your smart phone while flying across the country, but consider that the first computers took up entire rooms. Artificial intelligence (AI) is going through the same growing pains. Is AI the cure-all for all our problems? No. Does it offer great benefit in specific business applications? Yes, and with time these will expand – with particular promise already being realized in health care.

The first foray into AI was 1950’s “Turing Test” to see if a computer could “think” well enough to convince a human that the computer itself was human. From there, the potential for machines to learn and think like humans grew. But with the hopes and dreams of “augmented” intelligence also came unrealistic expectations and hype.

I would argue that today’s AI is more realistic, especially as it is already being applied in health care settings. Computers can augment professional expertise by automating repetitive, complex analyses and identifying patterns in large amounts of data. AI works best when it has been designed to use the right data to solve a specific problem. In the hands of skilled professionals, AI can support better clinical care, predict early signs of disease and reduce fraud and waste.


Where AI is most successful is in cases where it has access to broad data about individuals and situations and experts have given it a defined goal. For example, when you shop online, AI makes connections about your prior purchases, sites you have browsed and other individualized data to make suggestions about what you might like to buy. In the same way, AI can tie disparate pieces of health care information together and offer it up to people working in the field, who decide how to act on the information.

Natural-language processing (NLP) is one type of AI being used effectively in health care today. It helps computers understand and interpret human speech and writing. Electronic health records (EHRs) provide a rich repository for a person’s medical history, but about 80 percent of the information they contain is unstructured so doesn’t fit neatly into a database. NLP can:

  • Interpret EHR information to automate and verify billing coding, so medical coders can do their work more quickly and accurately.
  • Identify records that are lacking in clear and complete documentation, so clinicians can review them and fill in important gaps.
  • Flag indicators of undiagnosed conditions, which might be recorded in notes incidental to the primary reason for the visit, to facilitate connections between providers and visits.

In addition to improving outcomes, these applications decrease the time and effort health information management professionals, auditors and expert reviewers need to put into scouring files for the information relevant to their role.

Machine learning – another AI method – also is being implemented successfully in health care. It uses advanced statistical techniques to identify patterns in data and then make predictions. This method:

  • Goes further than NLP in being able to identify early indicators of diseases, taking multiple markers into account.
  • Can identify the right prescription at the right time to reach a desired outcome, by consulting pharmacy data, lab results and other health data.
  • Helps call center agents make actionable, data-driven decisions at an individual member level. Interventions are prioritized according to clinical value and an individual’s propensity to act on a given intervention.

The key to success with AI is knowing how and where to apply different models. For example, consider a project to identify people likely to have undiagnosed atrial fibrillation. The lessons from such a project could be used to build a disease prediction model that can include multiple prediction points throughout the progression of that condition. Such a model can potentially be used to predict if a person is more likely to develop diabetes and show his or her trajectory for progression or severity. Providers and patients could then focus on interventions that will keep them from progressing to the next stage.

These current applications may not be as robust and sexy as how we imagine AI in the future. They are, however, practical methods that are improving health care delivery and operations now. And they are just the beginning – serving as the foundation on which we will continue to build the next generation of AI applications.

Work on that next generation is already showing early success. For example, NLP is being tested to make the process of clinical documentation concurrent with a patient’s treatment. This combination of NLP and accurate voice capture promises to improve quality and streamline the clinical review, billing and coding process by providing a very sophisticated level of intelligence and a contextual understanding of the patient. Future applications of AI will provide even greater levels of context and automation, benefiting the entire health care ecosystem.

Photo: chombosan, Getty Images



Free Email Updates
We respect your privacy.


The Cheapest Hosting on the Net!


Bad Credit? No Problem…