Predictive analytics are the wave of the future. A form of data mining used in fields as varied as meteorology, supply chain analysis, and marketing, predictive analytics involves using variables, or predictors, to predict future behaviors and trends. Predictive modeling is an umbrella term used to refer to the use of predictive analytics, outlier analysis, behavioral analysis, and other statistical techniques to predict behavior. Predictive modeling is used by credit card companies to detect fraudulent charges, by IT administrators to detect spam, and now by CMS to uncover potential fraud and abuse.
CMS is under increasing pressure to move away from the "pay and chase" model, in which CMS pays first and determines potential fraud later, to a more proactive, preventative approach that scrutinizes providers before they are approved as a Medicare provider and claims before they are paid. CMS received $100 million through the Small Business Jobs Act of 2010 to further its experiment in predictive modeling, while the Affordable Care Act provides $350 million over 10 years to bolster anti-fraud efforts, including predictive modeling programs. CMS has contracted with Northrop Grumman, a global security firm known best for its defense contract work, to develop a predictive modeling system to identify high-risk claims, in conjunction with National Government Services and Federal Network Systems, a Verizon company. The Northrop Grumman contract is valued at $77 million over one year with three, one year option renewals.
While an advanced understanding of statistical modeling is necessary to fully understand and evaluate certain predictive models, the concept is simple. Predictors are selected that correlate to future behavior or results in a particular category. For example, likely predictors for an auto insurance company seeking to predict how many of its insureds will be involved in an automobile accident include gender, age, and history of driving infractions. Those predictors are then plugged into a statistical model to predict the likelihood of automobile accidents among the insured population. Using the resulting prediction, the auto insurer can determine how much it needs to raise its rates to cover the expected number of accidents. Credit card companies also use predictive modeling to develop a profile of how, where, and to what extent you typically use your card so that they can quickly identify charges that are out of the norm and thus potentially fraudulent, a fact that some travelers have unfortunately discovered when their credit cards are denied when trying to make purchases while on vacation in some far-flung locale.
On the health care front, CMS has indicated that some predictive models will be quite simple; for example, comparing where patients live to where they received treatment. If a provider in Rochester, New York treats a large number of patients from Reno, Nevada, such a pattern would draw CMS' attention. While one or two patients from Reno would not be statistically significant, a pattern of treating a certain type of patient may be an indication of fraudulent billing that would be flagged and investigated. CMS has also indicated that other predictive models will be much more complex, focusing on paid and denied claims to detect patterns indicative of potential fraud. The CMS predictive modeling program applies a risk score to "near real-time claims" and generates alerts regarding claims that carry a high risk score.
As with many new CMS initiatives, there are several unanswered questions. For example, while CMS has indicated that certain patterns determined through predictive modeling will be flagged and investigated, very little information is available regarding what those patterns or targets may be. It is also unclear how the investigations or audits will be carried out or whether claims payment will be put on hold until the investigation or audit is completed. We can safely say, however, that documentation "best practices" are critical. Sloppy, incomplete, or internally inconsistent documentation may impact the predictors that are used to construct the predictive models, and will certainly make it more difficult for providers to contest the results of the predictive modeling. Additionally, providers should consider monitoring their own claims for patterns and outliers that may raise a red flag for CMS and be prepared to explain and demonstrate why such patterns and outliers exist.
CMS' predictive modeling effort is just one more tool in the federal government's growing fraud and abuse arsenal. Providers cannot assume that because they provide medically necessary services and file Medicare claims in good faith that they will be immune from fraud and abuse scrutiny. CMS' efforts to screen all claims submitted by all providers on the front end, regardless of the providers' compliance history, demonstrates the need for providers to frequently reevaluate their documentation and billing practices and ensure they are prepared to respond to a finding of outliers or questionable patterns in their submitted claims.