Predictive Modeling: A Possible Answer to Spiraling Medical Costs
While using data to predict outcomes and manage potentially costly claims is desirable, it can be overwhelming and frustrating. Experts we spoke with offered their thoughts and suggestions to get the best outcomes from predictive modeling.
Defining predictive modeling. "The basic purpose of predictive modeling is to analyze data you have in hand that has a relationship between predictive factors and outcome factors to understand those relationships and extrapolate from them what might be expected to happen in the situation where you have only the input variables," said Dr. Laura B. Gardner, president and founder of California-based Axiomedics Research Inc. "Predictive modeling is not hocus-pocus, and it is not perfect either. It is not a complete substitute for human experience and human expertise. It is something not everyone can do."
What can be discouraging is determining what data to use and preparing it properly. Part of that involves having enough data.
"Unfortunately, in the health care system and workers' comp in general, there is no standard way of collecting or maintaining the data, and there is very little cooperation when it comes to sharing data," Gardner said. "Lots of predictive modeling efforts are going on, but many of them are frustrated by not having sufficiently rich data sources to work with or data that can be generalized to the point where the predictive information is useful enough."
Harnessing data for best outcomes is not impossible. But Gardner has some advice about entering into the predictive modeling fray.
"I would not discourage [workers' comp] payers from using predictive modeling on their data," she said. "What they need to be careful with is that they have done an adequate job of data quality assessment and they've compiled all available data with a sufficient number of outcome variables and predictor variables." She also advises using case mix adjustment.
Done correctly, predictive modeling can be invaluable in controlling medical costs in workers' comp. It can identify those claims that may be more troublesome.
"We used to have to wait for large quantities of treatment data to come through the system and trigger some sort of alarm," Gardner said. "Now we can, right when an injury first happens, with pretty good reliability say, 'This claim will benefit from very intensive management right away.' That can get started before medical treatment gets started. . . . The quality of care can be maximized from the beginning."
Descriptive vs. predictive analytics.
One way to get the most benefit out of predictive modeling is to take small steps into the realm of analytics, advises another expert. Leaping into it can be costly and inefficient.
"I'm a great proponent [of predictive analytics] but people need to start with descriptive analytics," said Karen Wolfe, president and CEO of MedMetrics, an Internet-based company. "Both are really powerful, but I don't like to see people skipping descriptive analytics. People don't understand their own data."
Wolfe defines descriptive analytics as the "what is" while predictive analytics are the "so what." Descriptive analytics identify critical business issues, trends and cost drivers, she said.
"The approach is essential to understanding relationships, the business process, and provides the platform for asking the right business questions," Wolfe posted on a blog. "Moreover, descriptive analytics are crucial to decision support and are the foundation for determining the right focus going forward."
Descriptive analytics also help payers understand what is in their data. "The analyses are very practical in everyday life," Wolfe said. "Don't skip over that."
Once a company has mastered descriptive analytics, predictive analytics can be used for forecasting, advanced reporting, and optimizing algorithms. "Advanced mathematical and actuarial analyses are used to predict the future based on the past," according to Wolfe. "If X is true, what is the probability Y will occur? Or when Y occurs, what are the factors that could have predicted it?"
Sufficient data. Wolfe cautions against reliance on insufficient data to get the best results. Data that exists in separate silos is a manageable problem technologically, Wolfe said. But one of her concerns is what she calls a resistance to gathering and integrating all the relevant data to perform adequate analytics.
"Many think bill review data is enough," Wolfe writes in a blog. "Bill review data alone will not reveal comprehensive actual claim costs or illuminate treatment effectiveness, provider performance, total cost of the claim, or outcomes because it represents only a portion of claim information." She advises at a minimum, combining claims level data with bill review data for "meaningful analytics."
Read more at the WorkersComp Forum homepage.
April 25, 2011
Copyright 2011© LRP Publications