Did you know that a single mom burdened with three kids and memories of the man that done her wrong is more than likely to string out the same workers' compensation claim that a Brady Bunch dad who goes off to work with a smile on his face and a peck on his cheek from his equally contented missus might not even file?
Don't bet the farm on that one particular supposition. But every day workers' compensation insurers are entertaining such theories in an effort to cut costs and provide better and more efficient care for injured workers.
While predictive modeling has been part of industry for decades and group health for nearly 10 years, only now in the workers' compensation realm has it come into its own as soaring medical costs have employers and their carriers looking for ways of doing things better.
All stakeholders in the comp realm are taking a deep breath and examining the science of predictive modeling and how much resources should be spent and for what aims.
As founder and principal of Los Altos, Calif.-based Axiomedics Research Inc., Laura Gardner has developed a workers' comp niche and today is in the vanguard of advocates for a more effective use of predictive modeling.
Nonetheless, challenges remain.
"Comp has a very different disease model," she says. "It does not lend itself to episodes of care or a chronic versus acute distinction that group health does."
But the more companies accept the premise behind predictive modeling, the more the science improves.
"I have seen increasing awareness of what can be learned from predictive modeling and how it can be used in managing claims and getting workers back on the job," she says. "It is an investment that returns more dollars than it consumes."
Some experts remain skeptical. Joe Paduda, a Madison, Conn.-based healthcare consultant, says predictive modeling for the most part remains a "black art."
"Predicting what is going to happen is like predicting whether a baseball player is going to hit the next particular at-bat," he says. "You just don't know."
The jury is still out on predictive modeling, says Paduda. "There is an awful lot of knowledge that is contained in the brains of longtime claims people that is probably just as effective as some computer programs."
While a batting average and recent indication that such a batter has success against left-handed pitchers when the humidity is high is helpful, "the person still has to get up there and swing, and you just don't know."
Moreover, false positives and negatives could possibly gum things up more than help.
"I would say predictive modeling is really in its early stages," says Paduda. "To a certain extent, it has been overhyped and oversold. People are going to be disappointed by the results because the folks who have been pushing it have not been managing expectations properly."
PASSIONS OF THE CONVERTED
Among the so-called "folks who have been pushing it" is Karen Wolfe, who speaks with the same passion and conviction of the need to "integrate the data" as civil rights activists in the 1950s might have spoken about the public school system.
"Workers' comp has lagged behind other industries for decades in using analytics to analyze their data," she says.
"But when it comes to medical data, comp has lagged behind even further. Group health has been analyzing their data to look at quality and best procedures and outcomes and even taking that data and creating predictive models," she says.
Any workers' compensation claim contains a stream of data from the first notice of injury to the final resolution.
"Predictive modeling is way down the line from getting the data and integrating it properly," she says. "If you really want to do medical case management and cost containment, I don't see how you can do it without all the data."
Wolfe is founder and principal of Bend, Ore.-based MedMetrics, a firm that analyzes workers' compensation data with the aim of identifying cost drivers and reducing overall administrative costs. At the heart of the Med-Metrics universe is a risk alert system that will let carriers and their providers know that a claim may need special attention.
"We set up scenarios of data elements and combinations of data elements that we think are potentially high-risk combinations," she says. "Or it could be just a single data item."
For example, a combination of a lower back pain diagnosis with surgery could cause such an alert. Likewise, a second MRI procedure could be another such alarm bell.
"These are the kinds of things you glean from the data to force action on the part of claims adjusters and medical case managers," she says.
As for medical case management in general, "it is a great concept, especially if you give them some tools to work with."
Predictive modeling serves a purpose beyond cost containment. Gardner says other goals could include identifying providers with a more challenging patient load and provide support and educational assistance.
Another use involves evaluation of the medical provider networks that have recently become an important part of the comp landscape.
"A year or two after implementation of these MPNs, carriers want to know if they have made a difference in not only costs, but quality of care," she says.
As a physician, Gardner also says that she believes one of the most critical concerns to carriers is quality of care and the degree of impairment to workers. "The most expensive patients are the ones who never made it back to work, so predictive modeling can look at what factors contributed to that and how they can be addressed."
That relatively limited field of variables has been a source of controversy, Gardner admits.
Today the data that Gardner examines for her carrier clients comes primarily from billing documents, whereas sources such as doctors charts and hospital admitting forms could provide a more complete picture.
Even the human resources department of the employer could provide useful nuggets of information in the form of performance evaluation. "There has been some conjecture that workers who have been given a poor performance evaluation are more likely to file a claim for a given injury than those who have not," Gardner says.
In addition, there has been an effort to identify the individual professional such as a doctor or physical therapist in any one given billing entity to provide a fuller picture of the care given than can come just treating the entity as a whole.
THE PRICE OF VARIABLES
But each new data element, of course, adds to the total cost of the process.
"In the past, predictive modeling was something that was not done, so carriers typically had a limited array of variables," Gardner says. "But now that predictive modeling has shown such great value, carriers are starting to add more variables."
In the end, Gardner will tell any client basically what question she can answer with what data they plan on providing, "which will hopefully bring them one step closer to what they are looking for."
Thus, the term predictive modeling covers activities varying widely in scope and complexity.
Paduda says that some predictive modeling consists merely of gleaning the data from the first notice of loss and assigning some sort of score to assess the potential difficulty for the claims. "There are different types of predictive modeling that are done at different points in the claim, and some are more resource intensive than others," he says.
So just what happens to that single mom, the person who got a poor performance evaluation and or any other such individual whose particular characteristic raised a red flag in the predictive modeling scheme of things?
"There are a number of things you can do about it, and it just depends on how assertive the claims-payer wants to be," Paduda says.
For example, the employer could conduct an investigation into such a claim, as well as perform additional studies, to see if there is truly a soft tissue injury, in the case of a back injury.
"As an employer, I can call this person every day and say, 'Gee, we really need you back, we miss you here, what else can we do to help?'" Paduda says. "By having an employer on one of those cases that might seemingly go off the tracks, it could have a significant impact."
Factoring in pharmaceutical usage also provides a good hint of cases headed for long and costly outcomes, particularly if they are using a lot of pain medicines and antidepressants.
"There is a very strong co-relation, at least in the studies I have seen, between the use of those kinds of drugs and the person not coming back to work," he says.
In that case, you might not want to settle the claim quickly, but rather conduct a peer review to determine how injured the person is and whether all the treatment, including drug use, makes sense. "In almost all cases, something can be done slightly differently, especially from the pharmaceutical perspective," he says.
But Paduda asserts that predictive modeling does not impede any care for those cases earmarked as having potential for "falling off the track," as he says.
"I think there is a relatively low risk of that," he says. "It is not like they are going to paint a target on someone's head and shoot."
For example, in that case of that "mythical single mom working in food service in some godforsaken town in upstate New York in wintertime, if worse comes to worse they do everything to make that person feel good and get the right medical care for them."
"I would argue they ought to be doing that anyway," he says.
Over the past couple of years, several vendors have begun to offer predictive modeling services that complement those provided by Wolfe and Gardner.
Last year, Boston-based Urix announced the launch of its Integrated Disability Management System, which according to its vice president, Kathleen Larson Day, will help identify the best cases for either settlement or early referral for case management. "Predictive modeling results help focus a claims adjuster's case load on the most 'actionable' or resolvable claims," she says.
In addition, the data can be used for reserve setting and detecting fraud and abuse.
Urix isn't alone, of course. Every major third-party administrator is involved to some degree in trying to harness data analytics and modeling.
Atlanta-based Crawford & Co., a provider of claims management services, offers its Crawford Claims Advantage, which helps claims professionals conduct initial three-point contact and ongoing interviews to "quickly gather important information on every file, every time."
Further north, Philadelphia-based ESIS Inc., a claims management company owned by ACE, has spent millions of dollars upgrading its claims systems to help clients lower the costs of claims using advanced data analytics techniques.
In an interview with Risk & Insurance®published in April, Dave Patterson, CEO of ESIS, said the firm has been able to lower clients' claims costs by about 12 percent by merging managed-care and medical bill review data with claims data.
Meanwhile, all this medical data being used for the greater good of improving care and quality and containing costs has raised some privacy concerns.
Wolfe says that, even though workers' compensation is not covered under the 1996 U.S. Health Insurance Portability and Accountability Act, her firm takes pains to de-indentify all data that come from her clients. Not only is it the right thing to do, she says, but she predicts that comp will eventually fall under HIPAA.
Some products out in the market today, such as the Deloitte Claims Predictive Model offered by Deloitte & Touche, use external public databases to score claims at the first report of injury by using up to 75 variables. They include age, years of employment, relationship with an attorney, marital status, prescription drug patterns and obesity or other negative health factors on the basis of their potential to incur severe costs or even possible fraud.
"Protecting privacy is a responsibility for every company, and as long as the appropriate steps are taken a breach of privacy should not be a concern," says a Deloitte spokesman.
Paduda says there is a lot of publicly available information and that people have signed off on certain clearinghouses sharing information, although he says he had no specific knowledge of the Deloitte product.
"I am not an attorney, but to my knowledge it is not because folks have already signed off saying, 'Yes, you can have access to this data,' " he says.
The Insurance Services Office in Jersey City, N.J., maintains a database of all property/casualty claims, including those of workers' compensation, and makes it available to its members, which include most property casualty carriers. In addition several, Paduda says, several states like Florida and Nebraska have databases of workers' comp claims that are available to carriers.
Carriers can check such databases whenever a claim is filed to see if there was a previous such incident that could have a bearing on the handling of the claim, he adds.
STEVE TUCKEY has written on insurance issues for a decade for several national media outlets.
August 1, 2008
Copyright 2008© LRP Publications