Online Exclusive

Soft fraud's hard-dollar impact in workers' compensation

Using medical provider billing patterns to guide fraud examinations

On Feb. 10, 2015, Nova Healthcare Management pleaded guilty to felonious workers’ compensation fraud and agreed to repay Texas Mutual Insurance Company more than $6 million. (See Major Medical Chain Pleads Guilty, Repays Texas Mutual $6.5 Million, Texas Mutual Insurance Company, by Terry Frakes.) The investigation revealed that Nova Healthcare Management (also known as Nova Medical Centers or Nova) overbilled Texas Mutual for physical therapy by billing services as more expensive one-on-one therapy when it actually provided group therapy. This case is a significant reference point of medical provider fraud for three reasons.

First, the nature of the infraction falls under the amorphous category of “soft fraud.” (Defined by the Insurance Information Institute as “opportunistic fraud, occur[ing] when people pad legitimate claims.”) Nova didn’t conjure up patient visits or otherwise outright fabricate insurance claims submitted to Texas Mutual, but they did materially misrepresent the nature of care provided.

Second, it underscores the magnitude of fraud at the provider entity level; examples of individual employees defrauding the workers compensation system often make the news, but medical provider fraud tends to have a larger dollar impact even if it occurs less often.

Third, how insurance companies use (or don’t use) data can have a significant impact on their ability to take a proactive versus reactive stance to fraud.

Workers’ compensation soft fraud

The operative word in the soft fraud definition is “legitimate,” which in this context could mean there’s an actual injured employee who did, in fact, receive some sort of medical care. Soft fraud occurs when a medical provider milks additional dollars out of an entity paying for the injured worker’s care.

In the Nova case, the distortion of truth occurred when Nova upcoded therapeutic procedures and exercises, and submitted claims to Texas Mutual Insurance for one-on-one encounters that were actually done in a group setting. This billing pattern was rather subtle — consider that the current procedural terminology (CPT) code for “therapeutic exercises,” 97110, has a national average of $31.35 while the CPT code for “group therapeutic procedures,” 97150, has a national average of $18.74. (See License for Use of Current Procedural Terminology, Fourth Edition, Centers for Medicare & Medicaid Services.) The difference that the provider can bill, when extrapolated over thousands or tens of thousands of claims, can drive millions in incremental revenues.

A one-on-one code also can be billed multiple times because the dollar amount is per 15-minute increments, while the group therapy code is “untimed,” and as such would be billed only once per visit.

Medical providers (not claimants) are a major fraud driver

According to Texas Mutual, medical provider fraud “pack[s] a harder financial punch” than claimant fraud. Claimant-driven fraud remains a risk worthy of traditional surveillance work and rules-based checks. However, insurance companies also must protect themselves against provider-architected schemes. Estimates show that fraud accounts for approximately 10% of the property and casualty industry’s incurred losses and 3% to 10% of private and public medical care expenditures — the impact reaches far beyond workers’ compensation. Virtually any insurance product or line of business with medical claims exposure — whether it be general liability, disability, accident and health, auto bodily injury, employer-sponsored medical or government health programs (e.g. Medicare) — is exposed.

What the data says about Nova

I was curious to understand more about Nova’s billing practices, so I benchmarked against Concentra, the state’s leading provider for workers’ compensation rehabilitation. My hypothesis was that given Nova’s guilty plea, there might have been, in retrospect, some other noteworthy trends in their billing activity in the years preceding the announcement.

To complete the assessment, I used data from the Texas Department of Insurance that captures detailed medical billing transactions from all workers’ compensation providers in Texas. The data set revealed that the most common diagnosis across carriers is “sprain of ligaments of lumbar spine.” I used this diagnosis as a sort of litmus test to aggregate individual bill data across multiple years of physician and physical therapist visits to derive an average cost per claim. Controlling for diagnosis, Nova appears to cost 70%+ more than Concentra Occupational Health Centers — a difference that seems to be driven largely by visit frequency. (See figure 1 below.)

Diagnosis “sprain of ligaments of lumbar spine” (ICD-9 code 847.2, visits before ICD-10 conversion effective Oct. 1, 2015)
Concentra Occupational Health Centers Nova Healthcare Management
Claim count Average visit count per claim Average cost per claim Claim count Average visit count per claim Average cost per claim
Texas Mutual 822,665 6.5 $986 195,241 11.0 $1,680
Travelers Insurance 107,904 5.8 $872 82,609 10.5 $1,701

Figure 1

It would be fair to believe that taking a simple average across a basket of claims oversimplifies and ignores underlying factors like claimant/injury mix, geographical cost differences and quality of care. Nonetheless, such a staggering difference in the numbers, observed across different insurance carriers, is consistent with Nova’s guilty plea of overbilling for which we know is true. Ultimately, such benchmarking activities can be used at the provider level to prioritize fraud detection efforts based on statistical outliers.

Using data to manage soft fraud

Using data to manage soft fraud hinges on a systematic approach to curate raw claim transactions, analyze for patterns and anomalies and to ultimately communicate insights that are actionable. These three steps comprise the foundation to almost any analytics process. Here’s what they mean in the context of medical provider fraud.

Data curation

Data curation, simply put, means getting data into a shape that’s useful for analyzing. This ranges from basic data quality validations to more nuanced manipulations like deriving variables. If you think about a workers’ compensation medical claim, it’s comprised of potentially many transactions, each typically representing a medical provider encounter, whether it be with a physician, physical therapist or surgery.

While any one medical bill might alone seem suspicious, the entire premise of managing soft fraud is by quantifying provider activity across time, patients, geographies and other dimensions. Among other manipulations, first begin by aggregating the data to a level that makes sense, such as across a claim or a series of related transactions. Feature engineering, the data scientist’s term for deriving variables, is also crucial in extracting additional meaning by imputing values that aren’t explicitly found in raw data.


Having generated a “model ready” data set, the next step is to build a model or sequence of models. The goal of building such a model is to predict likelihood or propensity for fraud in a given bill, claim or provider. Assuming not all historical fraud has been identified, an approach that rests solely on predicting future fraud based on “labeled” historical cases will most certainly render an incomplete picture. Supplement this approach with a spectrum of analytical techniques appropriate for working with “unlabeled” data in parallel to create the complete picture. By identifying what’s statistically anomalous, the analysis is enriched by shedding light on previously undetected billing patterns that would be of interest to a fraud examiner.

Insights consumption

The ultimate goal of all this data and analytical work is to provide front-line fraud examiners with better resources to detect soft fraud patterns. At the individual examiner level, this includes a certain degree of end-user training, the goal of which is to provide transparency into how and why a case has been recommended for investigation by the so-called “algorithm.”

Other tools like visualizations and dynamic dashboarding can rapidly accelerate how end-users interact with insights. At the organizational level, the impact to fraud examiner workloads should be analyzed. When new analytics-driven insights are injected into a business process, often a degree of process re-engineering is required to recognize the full scope of possible operational and financial improvements.

Tying it all together

Soft fraud in workers’ compensation, while hard to explicitly define, is a big deal. Soft fraud comprises a material portion of the approximately $3 billion workers’ compensation fraud committed annually in the U.S. This figure expands when you broaden the scope to encompass other lines of business with medical claims exposure and additional regions across the globe.

Tim King is an insurance business consultant at Teradata. Contact him on LinkedIn or by email at