Artificial intelligence
Sergei Tarasov – stock.adobe.com
Over the years, there has been a lot of talk about artificial intelligence (AI) and the healthcare industry. Much of it has been focused on two extremes. On the one hand, there is the fairly mature use of neural networks for radiological analysis. On the other hand, the focus is on fraud management. These have become “must haves” in my opinion. It is the filling of the golden mean that interests me. Medical insurance is, as patients, providers and payers can all agree, is often complicated and complex. There’s a business problem with making processes more efficient, and robotic process automation (RPA) is just one step in the right direction. More robust AI can help all three stakeholder groups meet their medical insurance management needs.
The general medical insurance industry deals with radiology and imaging. However, it is usually in the specialties. In the dental industry, radiology is a regular tool, using x-rays to understand the condition of teeth and gums, and then to document the work that has been done. The basics of AI and radiology have been covered, in this column and many other places, so this article will not cover concepts, it is important to realize how important this analysis is in dental care. The evaluation of periodontitis, for example, can be very subjective,” says David Rock, CEO of NovoDynamics. “In one of our NovoHealth Dental studies, sixteen percent of participating clinicians disagreed with their own initial diagnosis. Using AI to support clinical decisions with consistent, statistical X-ray analysis can help remove ambiguity and improve outcomes for clients. Further, Mr. Rock said human staff capacity limits the review of the flood of complaints to only a small percentage, and that sixty percent of those cases reviewed had no issues.
This is the point to briefly discuss fraud. Yes, insurance companies are concerned about fraud but, like in most other industries, fraud is rare. Much more common are simple user errors, mainly in recording details. For enterprise usability, what interests me is the ability to speed up claims processing while improving accuracy. This is even more valuable than fraud detection and helps all three stakeholder groups.
While AI is a job hazard in many use cases, it doesn’t seem to be there. As has been pointed out, very few applications are processed because there are not enough people. Psychology tells us that boredom drives mistakes. Consider the number sixty percent given above. If existing staff can focus on exceptions, they stay more interested. At the same time, automating the approval of the majority of claims can solve the cash flow problems that most dentists, who work in small practices, have as their main concern. In addition, patients have problems resolved much faster and are also more satisfied with faster treatment.
In order to achieve these benefits, robust systems are required. As a reminder, no individual AI tool is a solution, they are components of a larger system. “At NovoHealth Dental, we use different AI technologies to analyze different types of data. There is no one-size-fits-all solution,” says Sean McMillan, Principal Scientist at NovoDynamics. “To make accurate clinical assessments on radiographs submitted to payers without scale information, we use neural networks as well as advanced probabilistic methods such as random forests and statistical regression. When looking for the nuanced behaviors that distinguish outlier vendors, we go beyond known sources of bias such as zip codes, using fuzzy logic and causal inference techniques to automatically group more sophisticated sets of similar vendors for comparison with peer groups .
For example, as anyone involved in insurance knows, each payer has their own set of processing guidelines and procedure codes. Periodontitis (gum disease, but it’s fun typing the medical term) can be coded for the number of teeth in a specific segment of the mouth, and each insurer may use a different set of codes. It may take a combination of techniques to identify the segments in the X-ray image and then notice that the code is for a different number of teeth or a different segment of the mouth.
This does not imply fraud, just that the people in a dental practice are very busy and no one is perfect. What machine learning can do is automatically process many more correct claims than people can handle, and then flag the exceptions for human review. Human-centered AI (HCAI) is a growing problem, and it’s a prime example. The NovoDynamics system does not say anything upfront that it must be a fraud. It uses intelligence to classify risks based on type and frequency, providing humans with the information needed to act. The claims processor can then either approve, knowing that the suggested correction is correct, send the claim back to the vendor for clarification, or escalate the issue, all as needed from an expert’s point of view.
This decision is the next step in the evolution of claims handling. Radiology and fraud become known quantities in artificial intelligence. As difficult as they are to manage, these are the “easiest” problems from a business perspective. The integration of human and artificial intelligence combines the areas where both are best suited to solve a real business problem. In this case, it increases the accuracy and speed of dental insurance processing, resulting in better medical oversight, better bottom lines for providers and payers, and improved care and customer service for the patient. Artificial intelligence can be a win-win solution for everyone involved in medical insurance.
