Updated on: June 2, 2016

Predictive Analytics: Curse or Cure?

By
Original story posted on: June 1, 2016

EDITOR’S NOTE: Dr. Shantanu Agrawal, Raymond Wedgeworth, and Kelly D. Bowman recently wrote a CMS blog post on the agency’s Fraud Prevention System. For context and perspective, RACmonitor asked New York attorney Edward Roche of Barraclough LLC for a reaction to the blog posting. 

Big data promises great benefits. But does it deliver? Computational power has increased so dramatically that it now is possible to analyze truly great volumes of data all at once. In every sector, big data is the flavor of the month. The market for big data analytics hardware, software, and services is expected to double from $27.4 billion in 2014 to more than $50 billion in 2018, according to The Channel Company. Big data works by pulling data from dozens of sources and performing analysis on the mix. It has applications in marketing, finance, healthcare, retail, and other sectors, including government, where it is used to monitor fraud and compliance with regulations. Like Medicare. Believe it or not, big data techniques allow the government to analyze millions of claims each day.

According to CAQH, the nonprofit alliance of healthcare companies, there are around 6 billion healthcare payment transactions processed per year, or about 16.5 million per day. Medicare is a large part of that. Considering that each transaction has a large number of data fields, that is a great deal of information available for data mining. 

For the past five years, CMS has claimed that its Fraud Prevention System (FPS) based on big data analytics combing through 4.5 million claims per day has paid off. In a recent CMS blog post, it is claimed that predictive analytics technology saved $1 billion over a two-year period starting in 2014. The system was started up in 2011, and in the two and a half-year period from mid-2011 until 2013, it is said to have saved half a billion dollars. It is claimed that for each dollar spent, the government is getting back $11.60, which is a wonderful payoff. And with such wonderful payoff numbers, CMS is investing even more in big data.

But what exactly is being measured here? In a 2012 OIG report[1] evaluating the system, the predictive analytics group claimed $7.3 million in savings, but the report found no support for the number, and evidence it was untrue. Another large cost saving of $6.7 million was counted as “changes in provider behavior.” But the OIG could not determine if it was accurate because it assumed that 100 percent of all claims identified were improper. Another $17.8 million in savings was claimed for denials by edits and payment suspensions. The OIG found the numbers were inconsistent with ZIPC and PSCs data. Many savings were claimed by counting hypothetical claims that might be submitted at some time in the future. The FPS team had included both actual and projected savings in its “return on investment” (ROI) calculation. A further complication was on the cost side. The full costs of FPS were not accounted for because all indirect costs were excluded. 

Nevertheless, there is never a government program that is too inefficient to kill, so in the hallowed tradition of Washington, D.C., the program continued and even grew. The CMS approach seems to be that if something is not working, then make it bigger.

There is yet another way to look at these “savings.” In the blog we are told that the Fraud Prevention System (FPS) has contributed to $1.5 billion in savings, but around two-thirds of that ($1 billion) was obtained by use of predictive analytics. If the system costs were $1 for each $11.60 benefit obtained , then the predictive analytics technology cost around $8.6m dollars. That seems like a very low figure. But there are other costs apart from the system costs. For example, for each round of enforcement, CMS must make a substantial investment in staff attorneys, analysts, evaluators, the administrative judicial system, and others to harvest the benefit. Plus indirect overhead costs. We can call all of this summed up together the cost of enforcement . 

There are other costs as well, on the provider’s side. For each provider that is found cheating, there are administrative and possibly legal and expert costs. But we also can be certain that the system will identify a number of “false positives”—providers called out as being cheats when they are not. In order to defend themselves, they too will be forced to expend large amounts of money for legal counsel, and other support such as medical, coding, and statistical experts. So there are at least two additional defensive costs to the healthcare system that are not being accounted for. We can call these and  and  for those who get properly caught and those who are false negatives, respectively. Keep in mind that much litigation in these matters easily can run to $250,000, or 4 providers per million dollars’ defense costs. So perhaps the question we should be asking is not whether the ratio of ): is is $1:$11.60, but instead whether the cost of )+++: is really so favorable. It most certainly is not. The only question is whether or not it is actually negative, i.e., whether the costs are actually greater than what are claimed as the benefits. And we still do not know how the costs were calculated. 

It is really just a question of perspective. Let’s use an analogy. In political economy, we are told that outsourcing of American jobs to overseas locations where the labor is cheaper is good because it lowers the cost to the enterprise. But this logic ignores the total systems cost of these actions. For example, if we added on the cost society must pay in benefits to unemployed workers and the longer-term costs of their social displacement, the savings would not look as good, and would probably be negative. So we can see that these supposed benefits of outsourcing are only possible because the corporation can shift costs to the public sector that outsourcing.

Here, with big data, we have cost shifting, but of a different type. In this payoff logic, CMS seems to be making great “profits” on its investment, but a claimed profit of this size is made possible only because the government is shifting the costs to the private sector. It would take much work to thoroughly analyze the economics of this issue, but my hypothesis is that when looked at from the point of view of total healthcare delivery system costs for the nation as a whole, the benefits may not be as great as claimed, and there may be no benefit at all.

What’s next? We can look forward to the day when all of the auditors are replaced with artificial intelligence. 

About the Author 

Edward M. Roche is the founder of Barraclough NY LLC, a litigation support firm that helps healthcare providers fight against statistical extrapolations. 

Contact the Author 

Roche@barracloughllc.com

Comment on this Article

editor@racmonitor.com 



[1] Daniel R. Levinson, Inspector General, The Department of Health and Human Services has Implemented Predictive Analytics Technologies but Can Improve Its Reporting on Related Savings and Return on Investment, Document A-17-12-53000, September 2012.

This email address is being protected from spambots. You need JavaScript enabled to view it.