Duke Pediatrics Improves Operations to Fund Research

By Walt Jordan

The Duke University Medical Center is one of the top academic medical and biomedical research institutions in the country. The Duke University Health System, of which it is a part, admits more than 60,000 patients a year to its hospitals and other facilities for in-patient care and manages nearly 1.5 million people on an outpatient basis.

But like many academic centers, Duke faces significant financial challenges. Training doctors is a costly process, and Duke also provides about $50 million in charitable care each year. At the same time, even for patients with health insurance, reimbursement rates are continually under pressure. The system is complex and any error in processing a claim can lead to its denial. Not only are medical institutions like Duke under enormous pressure to deliver first-rate medical care, the pressure is equally great to run as efficiently as possible.

These challenges reach into every aspect of the institution. Dr. David Tanaka is professor of pediatrics and associate director of neonatology at Duke University Medical Center. “It is an area that is very high in critical care, and sadly, very expensive,” he said. “We are trying to stretch every dollar to help as many kids as possible.”

To that end, about 10 years ago, Tanaka decided to bring his research skills to bear on the operation of his unit. At the time, he observed, the traditional way to look at the unit’s business operations was through spreadsheets and ratios. But the relevant data generated in the neonatal clinic was not better understood using ratios. Many of the assumptions used to generate meaningful ratios were just not applicable. For example, he said, “prospective payers may not pay on a continuous scale. They may pay only $25 even if you incurred $100 in costs. They are discontinuous so certain strategies are not useful.”

Instead, Tanaka looked for a tool to be able to visualize the data. “We wanted to examine the data visually rather than numerically,” he said. “By looking at the data visually, we can see the patterns, not just the average.” In this way, his team was able to identify the structure of the data and pinpoint those areas that needed scrutiny.

For that kind of data analysis, he used JMP, a dynamic desktop data visualization tool from SAS. “We use JMP to gain visual insights that we then follow up with more traditional numerical strategies,” he said.

One problem Tanaka tackled was third-party payments from organizations like Medicare, which accounts for about 50 percent of the payments for adult care in the U.S. The Medicare payment scheme is very complex. “There are all sorts of different components and a complex formula with 30 or 40 different steps,” he noted. “But it is definable and knowable.”

The real question is if the facility is getting paid as it anticipates getting paid - are the correct payment models in place? Using JMP, Tanaka plotted actual payments versus modeled payments. There are clear cases where the modeled payments do not match the actual payments. That is because some Medicare payments are made according to the disease and others are made per day. “If you don’t have a model that models on a daily basis, you can get less or more than you thought,” he said.

But there is more. Medicare pays different rates depending on how frequently a patient is admitted. “There is an off set,” Tanaka said. “And you can pick that out. And it might be important. You want to figure out why people are coming back.”

And even in the group where the payment model should have exactly matched what was actually paid, Tanaka found that Duke was consistently underpaid by about $2 per patient. Further investigation revealed that the underpayment was concentrated in certain areas specific to academic centers. The underpayment was the result of an error at the Federal level - with the Medicare intermediary. “When that was corrected, we got paid correctly,” he said. “There was system error. We want to get what we were owed so we could fulfill our academic mission.”

Interestingly, this kind of sophisticated analysis has nothing to do with Tanaka’s formal training in neural physiology. “In my research, I wanted to understand how airways contract in the context of a disease in children,” he said. But to do research, he said, “you need money and money is in tight supply.” The resources available for research into pediatrics are relatively meager. “We decided that there were opportunities to fund our research by looking at the margins; at the areas that people don’t look because it is too hard,” he said.

Indeed, funding research from clinical fees is becoming an increasingly attractive alternative. The challenge, Tanaka said, is that people aren’t aware of the tools that can help them. “The JMP/SAS tool allowed us to work on very small databases,” he said. The first database he analyzed consisted of only 300 patients but he found a multi-million-dollar error. “You don’t need 100,000 patients to find these errors,” he said.

Tanaka and his team at Duke are now exploring the use of other BI tools from SAS. “Some of the questions we are now asking ourselves are harder questions dealing with patient safety,” he said. “What are the factors that have an influence on an adverse event in drug administration or falls?” Events like that can have several factors involved and the challenge is to be able to extract those factors from databases consisting of tens of thousands of patients. “I need to go 50 million rows deep,” he said. “With SAS business tools, I can cut down the information to usable sizes and use JMP to do the finishing work.”

Tanaka has learned to do his analysis with virtually no IT support. “But that is how researchers do things,” he said. His main problem, he added, was gaining access to the databases he needed. “We had to educate people that there may be a different way to achieve our goals,” he said. “We need a fundamental understanding of how the business works.”