Damage Extrapolation and Sample Size Determination in the Healthcare Sector

Investigating authorities and health insurance companies note time and again incorrect or implausible invoices for nursing services and contract physicians. We use statistical extrapolation to assist them in reaching a more efficient settlement. Since May 30, 2016, a law to combat corruption (StGB § 299a, § 299b, SGB V § 197a), has made bribery and corruption illegal in the health services industry.


Example: Home Care Billing Fraud

The public insurance and law enforcement agencies experience great difficulties in investigating conspicuous cases. This is especially true in suspected fraud cases concerning home care. It is very costly to  check all of the individually billed services for correctness while, at the same time, difficult to undoubtedly prove the faultiness of a single bill. This is due to the special  situations in outpatient care (possibly demented patients, many "small" services).

Example of evaluated guaranteed damage
© Fraunhofer ITWM
Example of evaluated guaranteed damage in an observation period of 20 quarters with respect to two considered performance figures
Example of a receipt profile
© Fraunhofer ITWM
Example of a billing profile on a population and the number of receipts in a reviewed sample, in light blue the faulty receipts from the sample

To avoid a review of all services, we determine on behalf of the investigating authorities a so-called »guaranteed damage« as a lower limit for the total damage due to the billing fraud. For this, we take a statistical sample from the whole set of billed services. Only this much smaller number of claims is then evaluated by the investigating authorities.On this basis of a few claims, taking into account the resulting statistical uncertainty, we determine a lower bound to extend to the whole population. Only with a very low probability of error will the true total damage lie below this predetermined bound. Our procedure is legally established in court for cases of medical billing fraud. Usually, a confidence interval of 99.5 percent is used. Then on average only one in 200 cases is below the calculated guaranteed damage.


New Method Enables Faster Billing Review

We always take into account the specific structure of the  invoices. Nurse and physician cases differ in the choice of a meaningful sample unit. A new method, developed in close cooperation with the investigating authorities, makes it possible to check invoices dealing with nursing care much faster. The time period(s) for which services are billed is important for the legal proceedings. Our method also covers this necessary periodization of the claim.

The approach can be used both in home care cases (SGB V and SGB XI) and in medical billing fraud cases (in-patient and out-patient). In specific cases, we specifically adapt the statistical methodology. Our calculation of a guaranteed damage gives investigative agencies the ability to track cases involving a number of small individual claims that cannot be fully investigated simply because of the lack of time and staff. Because of our additional optimization of the number of cases prior to the sample survey, we are able to provide substantial relief to the investigative resources of the authorities.