Confusion Reigns on Predictive Analytics

It seems everyone in workers’ compensation wants analytics. At the same time, a lot of confusion persists about what analytics is and what it can contribute. Expectations are sometimes unclear and often unrealistic. Part of the confusion is that analytics can exist in many forms. Analytics is a term that encompasses a broad range of data mining and analysis activities. The most common form of analytics is straightforward data analysis and reporting. Other predominant forms are predictive modeling and predictive analytics. Most people are already doing at least some form of analytics and portraying their results for their unique audiences. Analytics represented by graphic presentations are popular and often informative, but they do not change behavior and outcomes by themselves. See Also: Analytics and Survival in the Data Age Predictive modeling uses advanced mathematical tools such as various configurations of regression analysis or even more esoteric mathematical instruments. Predictive modeling looks for statistically valid probabilities about what the future holds within a given framework. In workers’ compensation, predictive modeling is used to forecast which claims will be the most problematic and costly from the outset of the claim. It is also the most sophisticated and usually the most costly predictive methodology. Predictive analytics lies somewhere between data analysis and predictive modeling. It can be distinguished from predictive modeling in that it uses historic data to learn from experience what to expect in the future. It is based on the assumption that future behavior of an individual or situation will be similar to what has occurred in the past. One of the best-known applications of predictive analytics is credit scoring, used throughout the financial services industry. Analysis of a customer’s credit history, payment history, loan application and other conditions is used to rank-order individuals by their likelihood of making future credit payments on time. Those with the highest scores are ranked highest and are the best risks. That is why a high credit risk score is important to purchasers and borrowers. Similarly, workers’ compensation claim data can be collected, integrated and analyzed from bill review, claims system, utilization review, pharmacy (PBM) and claim outcome information to score and rank-order treating physicians' performance. Those with the highest rank are the most likely to move the injured worker to recovery more quickly and at the lowest cost. Both predictive modeling and predictive analytics deal in probabilities regarding future behavior. Predictive modeling uses statistical methods, and predictive analytics looks at what was, is and, therefore, probably will be. For predictive analytics, it is important to identify relevant variables that can be found in the data and take action when those conditions or events occur in claims. One way to find critical variables is to review industry research. For instance, research has shown that, when there is a gap between the date of injury and reporting or the first medical treatment, something is not right. That gap is an outlier in the data that predicts claim complexity. Another way to identify key variables is to search the data to find the most costly cases and then look for consistent variables among them. Each book of business may have unique characteristics that can be identified in that manner. Importantly, predictive analytics can be used concurrently throughout the course of the claim. The data is monitored electronically to continually search for outlier variables. When predictive outliers occur in the data, alerts can be sent to the appropriate person so that interventions are timely and more effective. For example, to evaluate medical provider future performance, select data elements that describe past behavior. Look at past return-to-work patterns and indemnity costs associated with providers. If a provider has not typically returned injured workers to work in the past, chances are pretty good that behavior will continue. For organizations looking to implement analytics, those who have already made the plunge suggest starting by taking stock of your organization’s current state. “The first thing you need to know is what is happening in your population,” says Rishi Sikka, M.D., senior vice president of clinical transformation for Advocate Health Care in Illinois. “Everyone wants to do all the sexy models and advanced analytics, but just understanding that current state, what is happening, is the first and the most important challenge.” The accuracy and usability of results will depend greatly on the quality of the data analyzed. To get the best and most satisfying results from predictive analytics, cleanse the data by removing duplicate entries, data omissions and inaccuracies. For powerful medical management informed by analytics, identify the variables that are most problematic for the organization and continually scan the data to find claims that contain them. Then send an alert. Structuring the outliers, monitoring the data to uncover claims containing them, alerting the right person and taking the right action is a powerful medical management strategy.