If used properly, modeling can identify the high-cost claims that account for 80% of workers comp costs.                           

One-fifth of workers compensation claims account for about four-fifths of total claim costs. If a company could only identify these problem claims in advance, it could greatly reduce its workers comp expenses. However, most of these high-cost claims are not obvious at first; they emerge slowly over time. One method used by insurers to aid detection is predictive modeling. But to maximize any model's utility, there are several vital factors to consider.

Philosophy

It is critical to understand the carrier's claims philosophies in two areas even before diving into the specifics of an insurer's predictive model. First, does the insurer understand that managing workers comp costs requires treating the whole person rather than just the injury? Focusing only on the injury fails to take into account the factors that often determine the success, length and ultimate cost of treating the injured worker. 

For example, imagine an employee with a work-related knee injury.  Because pain is subjective, a host of psychological issues come into play. Does the injured worker believe he or she can manage the pain? Does the claimant believe the injury is a minor obstacle that will quickly heal? Is the worker motivated to return to work?  

Second, is the insurer committed to producing the best medical outcome for each injured worker? This is the best way to manage total claim costs over the long term. Any other philosophy will hamper the ability to cut workers comp costs regardless of whether or not modeling aids in early detection of the problem claims.

Data

Data is the fuel of predictive models; the larger the database, and the broader the range of information included, the more effective it will be. Ask about the size of the database used to develop the carrier's predictive model. Ask how many individual claims and medical billing transactions determined the variables the insurer uses to predict high-cost claims. How did the insurer validate the effectiveness of its model? And is the model based only on claims data or does it include other sources of information like psychosocial factors?

Analysis

How the insurer uses the data is arguably even more important than how thorough that data is. Can the insurer's predictive model look at that data holistically? Rather than the presence of one variable, it is far more often the interplay of a range of variables that helps identify potential high-cost claims. Find out if the insurer's model uses "multivariate analysis" to identify likely changes in the claim over time and their likely effect on costs.

Frequency

How often is the information on each claim updated? A model can only review the claim data that is available in the claim file. So it is vital that the process requires -- in fact, prompts -- claims professionals to gather specific information at certain points in each claim's life cycle.

This raises another key question: How often does the carrier's claim process run each claim through its model? Sure, it is important that the model reviews each claim at intake; the goal of the predictive model, after all, is to spot potential high-cost claims at the outset. But does the model look at each claim at other points in its life cycle? If so, how often and when? And why did the insurer select those points to run the claim through the model?

Operations

An insurer's predictive model must be integrated with its claims operations. It can be a useful tool to manage a claim -- but only if the model is embedded into the insurer's day-to-day claims management process. Furthermore, beyond simply alerting claims professionals of potential high-cost claims, the predictive model should provide them with guidance on how to manage the claim given its specific risk factors. Knowing something is a problem is not much help if you do not know how the problem can be solved.

Tracking

Find out how the insurer measures the effectiveness of its model. Does the insurer actually use this information to enhance and improve the model? Or did it simply start using one and presume it was working well? Without ongoing assessment, the only way to tell if the model is effective is by taking the carrier's word for it.

------------

George Neale is executive vice president and general claims manager at Liberty Mutual.