Monitoring Provider Performance for Predictive Profiling

Part IV—Monitoring Provider Performance for Predictive Profiling

This is the fourth and final in MedMetrics’ series about rating medical providers, most specifically, rating physician performance. The series, available at www.medmetrics.org at the MedMetrics Blogs link includes:
Part I—Rating Medical Providers
Part II—How to Evaluate and Rank Medical Providers specifically for Workers Compensation
Part III—Transforming Workers’ Comp Provider Networks into Quality Networks
Part IV—Monitoring Provider Performance for Predictive Profiling

Transitioning from the concept of rating physician performance to the realm of predictive analytics might seem like a quantum leap, however using provider competence as a predictor of risk makes good sense. Poorly performing providers from a Workers’ Compensation vantage point can predict high cost and questionable outcomes. Rather than, or in addition to applying advanced, sophisticated mathematical formulas to predict risky claims, a little logic goes a long way. While sophisticated predictive modeling is invaluable and may be the ultimate answer to controlling Workers’ Compensation costs, some shorter term solutions are attainable, affordable, and valuable.

Predictive analytics is that area of data mining and business intelligence concerned with forecasting probabilities and trends. Advanced predictive modeling techniques are decidedly beneficial tools used to study the data to identify conditions or sets of conditions that will bring about a predictable outcome. Basically, predictive modeling is a process used to create a statistical model of future behavior.

The realm of predictive modeling includes multiple methods of testing assumptions and uncertainty while looking for patterns in the data. If X is true, then what is the probability Y will occur? Conversely, when Y occurs, what are the factors that could have predicted it? Find a correlation, look for causation, develop a theory, test the theory and apply it. Once implemented, the model must be continuously tested and adjusted.

A familiar example is auto insurance where actuaries take into account potential driving safety predictors in the data such as age, gender, and driving record when issuing auto insurance policies. The probability of an accident is calculated and the premium cost is rated by that analyzed intelligence in the data.

Multiple conditions or predictors are combined into a predictive model that when subjected to data analysis, can be used to forecast future probabilities with an acceptable level of reliability. Nevertheless, in Workers Compensation, we also have the opportunity to leverage existing knowledge to gain advantages in claim management and cost control. Prevailing knowledge (industry wisdom) is the untapped predictive resource in Workers’ Compensation!

For instance, nearly everyone would agree that poorly performing medical providers will almost certainly result in a complex and expensive claims with dubious outcomes. Research backs this up. In his article describing his research, “Impact of Cost Intensive Physicians on Workers’ Compensation”, Edward Bernacki, MD identified specific indicators of what he calls high intensity physicians. The research showed poor performance providers have higher medical costs, longer medical treatment durations, longer claims duration, and higher indemnity costs (increased lost time). Of course, that does not come as a surprise to anyone—and that’s the point.

Research literature describing generators of medical costs in Workers’ Compensation is not extensive, but we can learn from what is available. We can apply knowledge from the literature to claims handling procedures and medical management, thereby gaining advantages. If we know poorly performing providers produce unsatisfactory results, we should be aggressively measuring and carving out the deficient providers. Start by evaluating provider performance in context with the peculiarities of Workers’ using the parameters described in this series of articles about rating medical providers. (www.medmetrics.org, MedMetrics Blogs)

The Bernacki article suggests other predictive indicators, such as injury types that do not have precise treatment pathways. A lower extremity fracture has a specific course and duration of treatment with an expected outcome, whereas a low back strain does not. Treatment patters vary widely. Moreover, a low back strain diagnosis combined with surgery of any kind suggests complexity and cost. When further combined with a poorly rated provider, it guarantees trouble. Data combinations reflecting these conditions can identify (predict) complex, costly claims early in the course of the claim. Moreover, research combined with the general wisdom can be tapped for other predictive indicators.

People with experience in Workers Compensation have developed wisdom in these matters that should not be discounted. Such intelligence should be formally incorporated into the management process. Doctors known to be high-intensity, high-cost providers should be avoided if possible and certainly monitored aggressively. One management tactic is to use the data to compare a questionable provider’s performance with others of the same specialty or those who have treated the same kinds of injuries. Such objective comparisons based on data are far more palatable to physicians than other initiatives meant to influence treatment patterns. Moreover, physicians are more likely to adapt their treatment processes when they see comparisons of their performance to others like them.

Experienced claims adjustors and medical case managers know intuitively about predictors. Create predictive data models derived from research along with the knowledge of your colleagues. Statistically-based predictive modeling is a powerful tool just beginning to appear in Workers’ Compensation that should be taken seriously and planned for strategically. Nevertheless, before or while applying huge resources to it, we should leverage the untapped predictive knowledge already known to us.

View additional articles by Karen Wolfe under Blogs at www.medmetrics.org