-2,0,1
Search

Your Healthcare Focus

Program Integrity: Analytics & Predictive Models

Fraud and abuse continues to take its toll on programs in Federal and State government around the country — with an estimated $1 out of every $10 spent on healthcare attributed to questionable activity.   With the onslaught of new covered patients coming on board in 2014, and constrained budgets and resources, it’s more important than ever to leverage sophisticated analytics and predictive models to help identify and prioritize opportunities that offer the highest ROI. 

Healthcare “Smarts”

Our approach to FWA analytics is to infuse our solution with healthcare intelligence.  As one example, Truven Health pioneered a ‘medical model’ for the healthcare fraud detection process.  A key component of this model is our Medical Episode Grouper™ (MEG).  MEG is an advanced healthcare analytic methodology that groups all the services for a clinically-relevant episode of illness, including inpatient, outpatient, and drug.  The MEG episodes are constructed using rigorously defined clinical classifications based on current medical literature.  Episodes analysis can expose patterns of clinical and billing abuse that are otherwise difficult to detect such as wasteful or unnecessary services.  Using MEG, the analyst can understand the entire range and cost of services provided to a patient during a single episode of illness, which then is aggregated to profile a provider’s entire practice.    The method supports best practice case-mix adjusted peer group comparisons to identify aberrant behaviors.  Finally, services that do not group logically into clinical groups, like seeing diabetic test strips for a patient who is not diabetic, can be easily detected using this method.    This episodes approach is just one example of how we apply years of healthcare knowledge and experience to our FWA analytics to produce superior results.

Program Integrity Rules, Algorithms and Predictive Analytics

Our approach to Program Integrity analytics is an outgrowth of our understanding that there is no one ‘silver bullet’ for fighting healthcare fraud.  Fraud schemes evolve and change over time—become more sophisticated—requiring not only detection rules and algorithms that have been proven, but also requiring detection methods that uncover completely new schemes and variations on old schemes.  All available techniques should be brought to bear on the problem. 

Our claim edit analytics include industry standard clinical edits to identify incremental savings during the payment authentication scoring process.  In addition, we offer our PI Fraud Algorithm Library - a vast store of knowledge acquired over the past decade from hundreds of successful rules-based algorithms, supervised predictive models, and findings from hundreds of fraud investigations.   We work with each customer to select algorithms that align with their areas of focus, and implementing those algorithms in the desired toolset to ensure that the they work correctly with their data and business rules. 

We also apply proven parametric statistical methodologies and healthcare expertise for fraud and abuse detection, including sophisticated peer group comparisons to detect aberrant patterns.  We help users quickly eliminate “false positive” leads that are explained by unique specialty or patient-mix factors from those that warrant further investigation for potential fraud, waste, and abuse.

Finally, Truven has experience in development and deployment of a variety of successful predictive analytics.  From supervised learning techniques that include generalized linear regression models, classification and regression tree (CART) models, multivariate adaptive regression spline (MARS) models, and neural network models, among others.  From an unsupervised model perspective, we have developed temporal (event sequence) models as well as probability-based models that produce threshold scores.  One such patent pending predictive model utilizes robust, non-parametric statistical techniques that are not constrained by data distribution assumptions as are parametric statistical techniques or independence to evaluate claims, providers and beneficiary data to determine if behavior is “abnormal”.   This model is not restricted to known cases, but adapts and identifies new and emerging abnormalities as new information is received.  This model is used in our Prepay Review and Prevention solution.