Boston-based Liberty Mutual Insurance Co. today is among the leaders of those companies using advanced predictive modeling tools to not only manage more effectively costs stemming from those core claims but, even more critically, provide the best quality of care possible.
For the past four years, the carrier has been developing and using three data-driven claims tools aimed at the so-called "slow emerging" comp claims that do not always scream out for attention. Yet when they have it lavished on them in the right way, it can make all the difference, according to company executives.
These tools assist specialized so-called High Exposure Medical Teams in looking at not only those claims, but also at the traditional high-cost catastrophic claim such as a roofer falling off a building and becoming paralyzed for life.
But the claims slow to emerge present the main challenge because the impeding factors are often hidden and also tend to have many failed medical procedures, prescriptions and chronic pain associated with them.
Joe Hooyboer, manager of medical claims data, said that Liberty's approach focuses on managing medical costs by managing medical outcomes, "as opposed to looking at which individual workers are likely to file a claim, and when they file a claim, are more expensive than maybe the guy sitting in the next cubicle doing the same job."
"What we are looking at are injured workers who are already being treated, and by looking at demographic and bill payment and clinical information on those cases, systematically eye which of those claims could very well turn into a high, complex claim," he said.
Liberty's Medical Loss Data Mart came into being in 2005 and is the driving force behind two of the three main prongs of the data modeling program, with the exception being the first tool--the predictive model--that is driven by Liberty data fed to an outside vendor twice a year:
This tool is unleashed on all claims after 90 days and looks at more than 50 attributes, such as the number of specialists and the demographics of the claimant that are red flags for high costs, and thus pointing to candidates for focused intervention by claims and managed care experts. The model has more than 2.5 million claims for history to back up the data, which is run twice a year. While maybe 100,000 cases might go under computer review, about 1 percent will be singled out for special attention, Hooyboer estimated.
Cost Driver Model: Each quarter, all noncatastrophe claims are run through this program that flags claims with three specific items showing costs above a certain figure for intervention. It is used to quickly identify claims where current treatment patterns and cost suggest problems that could lead to longer and costlier recovery periods unless different actions are taken.
Chronic Pain Model: Uses nine claims attributes, such as type and frequency of drugs, to identify candidates for a chronic pain program. It also serves as another tool to make sure the right resources are working on the claims and could point to the need for involvement of a medical director.
But the first tool, the predictive model tool, is what claims and case managers have been doing instinctively for decades to determine what claims need special treatment.
"It comes anecdotally, and from the knowledge that claims managers might have. They are going to look at the attributes of a case to see if that will blow up into a major high cost or complex claim," Hooyboer said. "This is the kind of case that will require intervention sooner or later."
Such an alert could involve a case with a localized injury. But over time, the claimant starts to feel chronic pain involving body parts other than the one originally injured.
The first step might be to see if the right specialist is looking at the case.
"So if you still have an orthopedic surgeon looking at a shoulder, and all of a sudden they are starting to do work on a lower back, the question might be is it time to bring a neurologist in," Hooyboer said.
In addition, Liberty has set up a system of 10 regional medical directors to perform a peer review of a case with the attending physician.
"One of the main points of the predictive model tool is to identify those cases where an additional level of expertise may be warranted," Hooyboer said.
As this example and others show, while cutting costs is one goal, quality of care remains paramount.
"It is fundamentally a quality-of-care issue," Hooyboer said. "This is especially true when you might be looking at surgeries that could be unnecessary."
He said that predictive modeling could in the short run sometimes result in more expensive treatments than at first envisaged. But in the long run, the correct treatment is always the most cost-effective.
Another example of a case not resolving as quickly as possible could be increasing lengths of time that durable medical equipment is being leased. But no matter what the red flag may be, a predictive model only sounds an alarm and does not prescribe any action.
"All a predictive model is going to do is say that these cases have a higher likelihood of becoming a highly complex case," Hooyboer said. "And believe me, if I have a choice of looking through 100,000 claims or a thousand, I will look though those thousand."
Measuring the return on investment of any predictive modeling program remains challenging for, among other reasons, any carrier's reluctance to spell out in detail the facts.
"We could go into a lot of detail, but we don't want to give away the store," Hooyboer said.
In addition, Hooyboer noted that predictive modeling is just one part of a broader claims management operation, and therefore it's difficult to pinpoint any possible ROI figure.
"The whole goal is to identify claims that apply appropriate management tools," Hooyboer said. "After all, I can apply a tool to identify a claim, but if you did not apply the appropriate management, then you did not save any money at all."
Joseph Paduda, a Madison, Conn.-based healthcare consultant, said that, of all the major workers' compensation carriers, Liberty appears to lead the pack in terms of using predictive modeling based in part on the arsenal of data that is behind the program
"I would say a lot of them are in kindergarten, while Liberty is at least a sophomore in high school," he said.
One possible metric for success for predictive modeling could be in the accuracy of reserve setting. But Paduda said that carriers would be reluctant to disclose such information.
As for the competition, Hooyboer said he knows of no other carrier that has the capabilities of the Medical Loss Data Mart.
"How we have applied that to predictive modeling is another question. I am not sure who all has done what with predictive modeling," he said.
STEVE TUCKEY has written on insurance issues for a decade for several national media outlets.
August 1, 2008
Copyright 2008© LRP Publications