Predictive modeling has long been used in personal lines, especially auto insurance. It’s only in the last 8 or 9 years that we’ve seen it squeezing through the workers’ compensation front door in the areas of underwriting and claims administration. In this period, the major risk management consultants, TPAs and insurers have been developing sophisticated models to, in consultant-speak, “use advanced statistical techniques (e.g., multivariate analyses, generalized linear models) to simultaneously evaluate numerous potential explanatory risk factors for maximum amounts of knowledge from available data sources” (from a TowersPerrin 2006 paper) (PDF).
To translate, in the claims process, the purpose of predictive modeling is to identify injured workers who are most at-risk of delayed recovery or malingering. The best time to do this, of course, is at the time of the injury. As my friend and colleague Mike Shor, of Best Doctors, puts it, “Think of it as being no different from the triage process that occurs in combat medicine or an emergency room…. the military talks about the golden hour….it’s what happens in that first 60 minutes that drive outcome. In WC we believe there is a golden 24-48 hours where the claim decisions that get made determine the ultimate outcome. It is here where claims that have the potential to run off the rails actually do.”
To a certain degree, predictive modeling systems can suggest which injured workers are most at risk for staying out of work longer than is medically necessary. Predictive models use advanced statistical techniques to perform multivariate analyses that suggest the degree of risk associated with any one underwriting risk or any one injured worker claim. Some predictive models use hundreds, even thousands, of univariates, but, in the claims arena, as you can probably imagine, there are a limited number, perhaps 10 to 15, that are of most value, and many of these are of the common sense variety. For example, co-morbidities such as obesity, diabetes and diseases that affect oxygen intake, all of which hinder healing. Others are demographic, such as age, education, marital status and distance from the worksite. For example, if you have a 55-year old divorced Type-2 diabetic male who lives alone more than 20 miles from the worksite and who suffers a crushing injury to the foot you more than likely have an employee at high risk for extended absence. Of course, any claims adjuster worth his or her salt intuitively knows this, but a predictive modeling system can examine all of the appropriate variables and spit out a ranking with recommendations in a nanosecond or two. Predictive modeling doesn’t come cheap, and it doesn’t replace the experience and judgment of a seasoned claims specialist, but, if used wisely, it offers a significantly sharp, relatively new arrow in the claims quiver.
“Used wisely” is the key phrase, because if that happens the claims adjuster can quickly link the at-risk injured worker with a clinician skilled in dealing with the bio-psychosocial risk factors associated with delayed recovery. In other words, the full-court claims press can be applied very early in the claims cycle.
Add to this mix an educated employer injury coordinator who projects a caring and compassionate approach to injured workers and who offers a well-thought-out modified duty program, and the likelihood of successful return to work is increased substantially. The goal is to remove excuses for staying out of work longer than is medically necessary. This type of approach assures that injured workers, the vast majority of whom are motivated to return to productive lives as fast as possible, do so on the fast track. Even more important, those who are not so motivated, those with other agendas, are identified almost immediately.
We recommend that you ask your insurer or TPA claims executives to explain their firm’s approach to and usage of predictive modeling. Employers should know to what degree and in what way their claims adjusters are using this tool.