Health insurance companies today are using artificial intelligence and machine learning in ways not possible just five years ago to better pinpoint at-risk individuals and to reduce costs.
"The applicability and opportunity on the insurers side is fantastic," said Mark Morsch, vice president of Technology for Optum360. "AI has gotten hot in the last few years.
The biggest breakthroughs are in more sophisticated machine learning. Being able to take that data and leverage it to drive algorithms and move towards being more predictive."
Optum, for instance, is now running a pilot program for insurers to take advantage of AI in processes done manually, according to Mark Morsch, vice president of Technology for Optum360.
Morsch -- co-inventor of the lifecode natural language processing engine with three patents on NLP technology for computer-assisted coding -- and his team are developing the project. He also serves as vice-chair of the HIMSS Health Story Project.
Specific areas to streamline include the medical record review process, prior authorization, pre-payment review and post-payment auditing.
Medical record review often relies on a nurse or physician to read through a patient's record and compare that to policies for what's authorized. A trained person needs to determine whether the patient qualifies for benefits.
"That is very manual," Morsch said, adding that it's just one use case "There's a range of processes insurers do today that are ripe to take advantage of AI to be smarter, more automated. There's a lot of interest from payers."
Payers manage risk
In addition to medical record review, payers are applying AI and machine learning algorithms to risk management.
"Managing and predicting risk is at the core of what payers do," said Frank Jackson, executive vice president of Payer Markets for Prognos.
Prognos is one example of a vendor using AI to model a more accurate level of risk to determine which members need the most care and will drive the highest cost, so insurers can expend their resources towards these beneficiaries.
Insurers must be able to assess risk correctly to set the right premium, Jackson said. If they miss slightly on pricing and go too low, it can be costly, he added. But if priced too high, they might lose that employer contract next year.
"One percentage point in premiums results in millions of dollars," Jackson said.
The traditional method in offering a premium price to an employer group is to use averages. For instance a male, 30-years-old, on average, costs ta certain amount, and then that figure is aggregated.
Payers usually start by using the most easily accessible data: claims. But claims have just one field, the primary diagnosis code. They don't record secondary diagnoses, which may reveal crucial information.
And it gets expensive. If a Medicare Advantage payer wants to pull a patient's chart for a clinical review, it can cost as much as $40 per chart. But greater risk in MA, insuring an unhealthier population, leads to greater reimbursement in the risk adjustment process. It's incumbent upon plans to identify their members' conditions.
Prognos uses a lab registry of 18 billion clinical records to stratify risk for a group of beneficiaries who have just enrolled. They can get identified data going two years back.
Applying artificial intelligence, they are able to let the insurers know which members need disease management.
"We're going to fill in the data gap," Jackson said. "AI is using the tools available like a deep narrow network and finding answers to difficult questions."
Five to 10 years ago, none of this was possible. AI requires significant computing power. A decade ago, running such models simply took too long.
AI to optimize health
Dr. Trishan Panch, chief medical officer at Wellframe, is using AI to optimize healthcare for chronic conditions.
AI and machine learning move from a custom and reactive approach to more standardized and proactive management of patient care.
"One of the biggest results is the high engagement rate we've been able to achieve," Panch said.
Patients get a personalized daily checklist on their mobile devices on all of the things they need to do. Data collected about medication compliance and other information is transmitted securely to a clinician or nurse plan manager through a dashboard.
Using a machine learning model, Wellframe can prioritize those patients who should be targeted, such as someone interested in weight loss, or smoking cessation. They know who has benefitted.
One surprising thing is that patients do not find it creepy to be in a remote relationship with a clinician, Panch said. Connections are formed by following a care program, asking benefit questions, discussing health concerns -- and some of it can be dark, emotional conversations.
"The fact that it's happening over a mobile device with some phone calls, the experience has held," Panch said. "The beauty of technology is, you can bring it to more people for a longer period of time."
In the payer realm, the financial incentives for value-based care are aligned. Population health is about which patients should be targeted for services. Payers have focused on historical cost.
Through AI, Panch said, "We have the opportunity to rethink how they reach out for clinical services."
Focus on Artificial Intelligence
In November, we take a deep dive into AI and machine learning.
Email the writer: firstname.lastname@example.org