You may have heard some buzz about machine learning (ML), but what exactly is it? And, more importantly, how does it relate to you? ML essentially refers to the phenomenon of computers and other devices that appear to be intelligent because they can learn from patterns in the environment to achieve a specified goal or predict an outcome. These patterns could come from a variety of sources, such as images, voice, free text or even structured data. The machine’s ability to anticipate or interpret the information given can make it seem almost human-like. For example, in the case of workers’ compensation claims, you can apply machine learning to feed the machine claim notes, and it will tell you which job class the claim is for.
What makes this possible is ML’s ability to look at combinations of features in data and learn from them to make a variety of associations. Imagine a machine looking at claimant age, diagnosis codes and Part, Nature and Cause (PNC) codes in tandem to instantly determine the likelihood of litigation. Machines now have this ability, thanks to the increasing variety and volume of available data points. They can generate specific algorithms to decode a seemingly infinite number of patterns in order to make massive amounts of information truly useful.
Why Does This Matter to Workers’ Compensation?
Teams are constantly trying to get in front of their claims so that they can be proactive rather than reactive. To do that, they need a variety of signals sent to them in real time in order to figure out the best strategy for a claim. ML helps provide these signals much faster than ever before, and there are four key reasons why.
1. Machine learning can handle many, many combinations in the data … fast.
Think of all the data you have to sort through to assess a claim. ML could be applied to increase speed and accuracy as well as to simplify the entire process and unlock invaluable predictive insights. Case in point: To get an accurate prediction of the projected cost of a claim, you might need to juggle as many as 45 different data features, including PNC codes, claimant information and diagnosis information. If you assume 10 values for each feature, which is a conservative estimate (International Classification of Diseases (ICD) codes have as many as 80,000+ potential values), the number of possible combinations can reach a mind-boggling 10^45. ML algorithms, however, can navigate these combinations and associate patterns with specific data characteristics in minutes. Not only that, they can also determine which set of features contributed most to the outcomes. Talk about the ability to improve claim teams’ efficiency!
The ability for ML to sift through all these combinations of features and utilize all their interactions is absolutely game-changing.
2. It can handle “holes” in the data.
Aside from quickly sorting through multiple data points and turning them into something meaningful and applicable, ML helps address another common problem: the presence of gaps in some of the fields in the claim or bill data, especially in the early stages of a claim. Techniques like data augmentation, where models are trained on various versions of a claim that expose different levels of gaps, can help ML models tolerate data “holes.” While it’s always important to get the best data one can for a claim, it’s equally important for models to be able to operate with incomplete information. ML makes it possible to move forward despite imperfect or incomplete claims.
3. It can handle changing data.
Workers’ compensation claims are also constantly evolving, sometimes dramatically. What started as a neck injury can evolve into a spinal injury. In an ideal world, a claims team would be notified as soon as a major shift happened in one of their claims instead of waiting until the traditional 30-day, 60-day and 90-day check-in points. ML helps here as well. Because ML models can handle gaps in the data, they can also instantly navigate the changing nature of claims, alerting claims teams immediately of changes.
Not only can ML tackle changes in claim data itself, it can also handle changes in the overall operations of a claims team. A robust retraining schedule helps the ML models stay current so that they are perpetually unearthing new patterns to better assist when claims operations or macro-effects, such as new regulations, occur.
4. It can work with more natural forms of data, such as free text, voice and images.
While this is all excellent news for claims teams interested in harnessing the power of ML to drive better care, I’ve saved perhaps the best benefit for last. One of the biggest advantages of ML is its ability to handle not only structured datasets but unstructured as well. What does this mean? ML can detect new patterns out of things like natural text and images. This has never been possible before.
A simple application of this is in the use of ICD codes. Claims that have a combination of ICD-9 and ICD-10, for example, can be really clunky to deal with because the mapping between them is a little complicated. But with ML and natural language processing (NLP) techniques, we can use the ICD descriptions instead of the pure codes to unearth relevant themes and topics. Those become features for the models, and no more mapping is required.
This is an area within ML that is evolving rapidly. We can see it all around us with devices like Amazon’s Echo and Apple’s Siri. In all likelihood, this will generate a hotbed of activity in workers’ compensation predictive analytics as well. Imagine a scenario where a claims examiner could get suggestions for doctors to recommend to the injured worker, all while they are typing in their claim notes.
Despite All Its Power, ML Is Just a Tool
No matter how awesome and transformative ML seems, it’s not a magic bullet. At its core, it’s a useful new set of tools meant to empower claims examiners to do what they do best — be a coach whose main goal is getting the injured workers through their recovery. ML can redefine the role of the examiner as a coach and a problem-solver while removing large parts of their “rote” work. This enables examiners to spend more time talking to injured workers.
Tack on the fact that ML can improve the efficiency of the entire claims process — from the ability to assess and process more claims faster and more accurately to getting employees the right care throughout their journey so that they can return to work and resume their lives — and ML has the capacity to provide data-driven insights that will help employees feel better faster, all while reducing costs overall.
Deploying a combination of tools like ML with empathetic examiners who have strong problem solving skills will elevate the entire workers’ compensation experience — for claims teams, employees and the companies they represent. It has the capacity to ultimately drive better care. While ML can’t do it all, it can modernize and fundamentally transform workers’ compensation.
Dr. Laura B. Gardner is a physician entrepreneur who founded and ran Axiomedics Research, Inc., a successful consulting company for 22 years before joining CLARA analytics as chief scientist. Her primary focus is providing claims teams and network managers with predictive analytics-based software tools that help to improve outcomes for injured workers. Visit www.claraanalytics.com and follow CLARA analytics on LinkedIn, Facebook and Twitter.