Machine learning (and its close cousin, artificial intelligence) are hot buzzwords on the health-care landscape right now. But below the surface, it often seems there is more theoretical discussion than practical application.
The core idea is simple. Machines are really, really good at poring over massive amounts of data and picking out patterns and anomalies within them. They can do it far better and faster than humans can.
Cognitive learning is an area that’s taking off, but it’s still in its infancy and is limited in not being able to infer the meaning of those patterns and anomalies. It’s still up to humans to determine whether those patterns are an indication of an issue that must be addressed immediately, a topic for discussion at the next executive or departmental meeting, or merely something to make you say “hmmm” before moving on.
Here are three examples of how machine learning is helping health-care organizations improve quality and reduce costs today.
Addressing opioid fraud, waste and abuse
We’ve all heard the stories about Florida’s infamous “pill mills” — pain clinics that were dispensing prescription opioids like candy (and reaping huge profits in the process). Yet that example of fraud, waste and abuse is only one of several drivers of the current opioid crisis, which is estimated to cost anywhere from $80 billion to $272 billion each year.
Another common cause is patients/members who see multiple physicians and fill multiple prescriptions at multiple pharmacies.
The quick answer, of course, would be to call out anyone who crosses a certain threshold for prescription opioids. Yet there can be legitimate reasons for that behavior, such as a cancer patient who is seeing multiple specialists. To go through each patient record manually to determine which prescriptions are allowable and which are examples of fraud, waste and abuse is well beyond the capabilities of any health payer, provider organization or government agency.
This is where machine learning can make a huge difference. By going through hundreds of millions of patient/member records, the machine can discern multiple patterns. Working with clinicians, data scientists can then use this information to develop predictive models that establish when a line has been crossed.
For example, by overlaying geographic information on the patterns, data scientists can see how far a pharmacy (or group of pharmacies) is from a patient/member’s home, place of business or physician. A pattern of going beyond what the model shows as a normal distance, or amount of time to travel, is a strong indicator of patient/member drug-seeking behavior.
The model can also predict how many different pharmacies a patient is likely to use. Again, cross that threshold, and the machine learning can trigger an alert, suggesting the organization investigate.
Reducing overbilling of patients
Machine learning is an extremely effective way of auditing hospital bills to determine whether patients are being overbilled, which can cause not only financial issues for the hospital and the patient but also harm the hospital’s reputation in the community if it happens consistently. In this era of consumer activism and patient-centered care, the impact on a hospital’s reputation cannot be underestimated.
Rather than auditing bills on a spot basis, or when a patient or insurer questions them, machine learning models can be used to review every bill generated. Applying business acumen, data scientists can create predictive models that score every bill prospectively, and continuously iterating and learning from outcomes to minimize false positives. These models can then be used to analyze the bills as they are generated so they can be corrected before they are sent.
A good example of what machine learning can catch are charges for anesthesia in a procedure that didn’t require any. Human reviewers who don’t have a clinical background may not find it unusual to charge for anesthesia in some instances. But machine learning can immediately spot the charge as an outlier and create an alert.
Understanding the total cost of care by market
The decision to expand into new markets, or expand services in existing markets, can be difficult for payers and providers. There are literally hundreds of metrics that must be understood in context with one another, which is more than any human brain can analyze.
Machine learning can help analyze the data to develop a risk score that acts like a credit score, based on overall parameters such as membership, total spend in the market, provider costs and member risks. With the scores in hand, business experts can use their experience to determine whether particular markets are worth targeting, and even where clinics should be located to draw the most patients/members.
While not always foolproof, this process of integrating man and machine can help payers and providers make the best possible decisions based on the information available to them.
Machine learning in health care is still in its early stages. It definitely holds tremendous promise for the future across all areas — clinical, financial, operational, supply chain and more.
This article was written by Lalithya Yerramilli from Information Management and was legally licensed through the NewsCred publisher network. Please direct all licensing questions to email@example.com.