Back in February, senators expressed dismay at a multi-million dollar anti-fraud computer system installed by the Centers for Medicare and Medicaid Services. CMS hoped to prevent fraudulent payments, reversing its standard mode of paying, discovering and chasing after money that wrongly went out the door.

In April, the Health Care Fraud Prevention and Enforcement Action Team, or HEAT, from Health and Human Services, made announcements in Chicago. The Attorney General and the HHS secretary highlighted their high-tech war against CMS fraud, and announced a slew of procedural and legal changes. But most of it focused on stronger fraud penalties, prosecutions, and suspensions or debarments of Medicaid contractors. Nothing was said of the $77 million system.

Earlier this month, the multi-agency HEAT made 107 arrests in a CMS fraud case in which a slew of phony care providers bilked the government of nearly $500 million. That’s great, but it isn’t on the surface a case of prevention via predictive analysts. Perhaps it’s a case of pattern matching. Most likely, it’s a simple case of tip-offs by snitches – a perfectly legitimate way to crack these types of cases.

The problem with the moniker “big data” so much in vogue is that it’s imprecise, and implies that some software tool can be thrown at – what? – and come up with – what?

This article originally appeared on

Even in so-called big data, human beings must have a definitive idea of what it is they want as an outcome. Application of algorithms to big data may not be the same process as reporting writing against structured databases, but that doesn’t mean it lacks discipline.

A good example of success is the Navy and the Defense Finance and Accounting Services (DFAS). To prevent improper payments, the Navy runs software to compare requisitions from its contracting shops, invoices from its vendors and the processes that take place at DFAS, which pays on behalf of the Navy. Catching erroneous payments means knowing what data elements must match. It can be done by hand in theory, but in large organizations the volume exceeds the ability to hand process in time.

(A wonderful anecdote in David Halberstam’s 1988 The Reckoning describes how the accounting department worked at Ford Motor Company after World War II, when the company was virtually bankrupt. The accountants somehow figured out the average amount of all of their invoices, and how many invoices there were in a stack so many inches high. They multiplied the average times the height of the stack to determine the company’s obligations.)

What the Navy example shows is the need to be specific about what the organization wants from the data. CMS has a bigger challenge, to be sure. It is a payor organization, but it doesn’t initiative the service that, once rendered, qualified for reimbursement. A patient sees a provider, and the provider bills Medicare. Given the open-ended nature of the initiations of service, CMS has visibility only as far back in the chain as the provider, whereas the Navy initiated the orders it intends to pay for.

CMS operators seem to be getting better at identifying the patterns of billing that indicate fraud – large numbers of repetitious transactions, long-distance transactions for things ordinarily provided locally, that sort of thing. Its challenge is how to use that big data analysis in such a way as to prevent them, and not merely discover them.

Tom Temin is editor in chief of and co-host of The Federal Drive on WFED AM 1500.