Preventing Fraud with Predictive Analytics

January 28th, 2013

HIPAA, ACA, ICD-10—the healthcare industry today is awash with regulatory events and government initiatives. Compliance has never been more essential across the healthcare spectrum, not least because fraud investigations—whether performed by RACs, investigations by the Department of Justice/OIG, or a payer—are becoming more and more frequent.  These investigations are often difficult to staff for and can be resource-intensive.

So, how do we remain compliant while also preparing for this onslaught of scrutiny? Using data and predictive analytics can help to identify and flag risk areas before they’re put under the microscope.  Even better, this information can prevent having to audit high-risk claims and assign scarce resources to audits post-discharge.

Not sure where to begin? Start by taking simple steps through analytics to identify risk, institute stop-gap measures, and let the data tell you where to focus.

Here are few suggestions to get you going:

ICD-9/HCPCS ­ – High-risk codes may be used to flag cases for coding review prior to billing. For example:

–          482.89 Pneumonia due to specified bacteria as principal diagnosis: Evaluate if the physician stated “bacterial pneumonia.” If so, then the appropriate code is 482.9.

–          486 Pneumonia as principal diagnosis with Sepsis 038.9 as secondary diagnosis: Sepsis is likely the principal diagnosis based on the coding rule and should be evaluated to determine.

–          According to the latest OIG reports, it is implied that J1950 or J9217 (both for Lupron) are based on the type of cancer being treated (prostate versus uterus). Validate if this is documented, and validate the units of service on these two codes.

Length of Staty (LOS) – May also be used to flag high-risk cases. For example:

–          One-two day LOS volume of Medicare cases compared to industry practice seen in MEDPAR 2011 of 30 percent.

–          Sepsis cases with LOS less than 48 hours that did not expire. Review to determine whether the documentation supports Sepsis as the principal diagnosis, as these cases typically require IV antibiotics and hospitalization for more than 48 hours.

–          CVA cases with LOS of one day that did not expire or get transferred. Determine if this is supported in the documentation or if the discharge disposition code is incorrect.

–          Knee replacements with one-day LOS that did not expire or get transferred. It is likely that the discharge disposition code is incorrect.

Benchmarking and Predicted Occurrence Rates – Both may be used to flag accounts. For example, review monthly total of cases with COPD compared to those with Pulmonary Edema and Respiratory Failure.

–          COPD volume (MS-DRGs  190, 191, 192)

–          Pulmonary Edema and Respiratory Failure volume (MS-DRG 189)— Compare to benchmark.  Twenty-eight percent of cases in the MEDPAR 2011 file are in Pulmonary Edema & Respiratory Failure, and the remaining 72 percent are in one of the three COPD MS-DRGs. If you are exceeding the 28 percent benchmark, you should audit these cases to determine if they are correct and if the documentation supports the codes and the principal diagnosis.

These examples are just the beginning. There are so many different scenarios to monitor, it can make your head spin or make you reach for the chocolate. For those of us in healthcare coding and auditing, though, this is where our focus needs to be. We need to take proactive steps to prevent non-compliance rather than having to be reactive, constantly pulling claims, writing appeals, or defending our actions. Data, such as having access to predictive information and benchmarks, is key to executing processes that will help avoid potential compliance issues before dropping a bill.

Garri Garrison is Director of Emerging Business with 3M Health Information Systems.