What Medicare Datasets Reveal About Care, Referrals, and Costs

CareSet’s DocGraph and federal hospital cost reports are providing new empirical insights into how policy changes, provider relationships and financial reporting play out across the U.S. health care system.

In a three-chapter dissertation at the MIT Institute for Data, Systems, and Society, researcher Carlo Duffy, Ph.D., used these large-scale administrative datasets to tackle three distinct questions: whether expanding nurse practitioners’ scope of practice affects patients and health care utilization, how primary care doctors choose cardiologists, and whether statistical rules can spot potential accounting anomalies in hospital filings.

Does Expanding Nurse Practice Harm Patients?

In his dissertation, Duffy examined an ongoing legislative debate: whether expanding nurse practitioners’ scope of practice harms patients.

Building on staggered state-level adoption of scope-of-practice laws as a natural experiment, Duffy used CareSet’s DocGraph data, which maps patient-sharing and referral relationships among Medicare providers, to examine how those expansions affected nurse practitioners and the broader referral patterns around them.

His analysis found largely null effects. While Duffy noted some evidence of increased referrals and Medicare Part B utilization in specific sub-analyses, he found little evidence that scope-of-practice expansions adversely affected total referrals, referrals to specific entities, overall utilization and spending.

As of 2026, 29 states and Washington D.C. grant nurse practitioners full practice authority. Yet the momentum is facing pushbacks. In 2026, legislation seeking to remove physician involvement for nurse practitioner practice was defeated in Florida, Indiana, Mississippi, Tennessee, Virginia and West Virginia, with opponents arguing that broader authority could harm quality of care and increase costs.

For lawmakers weighing whether to expand nurse practitioners’ authority, Duffy said his findings provide one more piece of evidence to set alongside the broader research consensus.

“My paper there would definitely support the idea that maybe we shouldn’t worry about this as much,” he said.

The Pattern Behind PCP-to-Cardiologist Referrals

Duffy’s research also examined homophily — the tendency of people to connect with others like themselves — in referrals from primary care physicians (PCPs) to cardiologists in Medicare.

Using DocGraph, he analyzed individual referral relationships across five characteristics: gender, medical school, medical school graduation year, and predicted race and ethnicity.

“I do actually find that there is strong evidence of associations,” he said, noting that PCPs referred more often to cardiologists who shared these same characteristics with them. 

But the finding raised another question: Could the pattern simply reflect the composition of local physician networks?

For example, if most cardiologists in a region graduated from the same state university, PCPs might naturally send most of their referrals to physicians from that school simply because those cardiologists are the most available.

To account for that possibility, Duffy compared the share of a physician’s referrals going to same-group cardiologists against the actual local availability of those cardiologists — a ratio that, in theory, should equal 1 if geography alone explained the pattern.

Instead, the ratio consistently diverged from 1 across nearly every characteristic he examined.

“There are behaviors going on here that simply are not predicted based on a clean theory going from same availability to same referrals,” he said, adding that future research could examine whether the homophily finding is associated with downstream outcomes such as health care spending or quality of care.

Can “Rule of Number” Spot Fraud?

Every year, thousands of hospitals submit detailed cost reports to Medicare. Duffy examined these filings from 2011 to 2023 to test whether Benford’s Law, also known as the law of anomalous numbers, could help regulators flag potential financial fraud in those filings.

Benford’s Law predicts that in most naturally occurring sets of numbers, the digit 1 should appear as the leading digit roughly 30% of the time, while 9 should show up less than 5% of the time. When humans manually alter numbers, they almost always break this pattern, making the law a gold-standard tool for forensic accountants.

Applying the rule to hospital cost reports, Duffy found that aggregate data did conform to Benford’s Law in most years. That, he said, is evidence that the overall system isn’t rife with fabricated numbers.

He also found that specific hospital characteristics, such as ownership, cost reporting, and uncompensated care, were systematically associated with conformity with Benford’s Law.

But failing the test is not, by itself, evidence of fraud.

“I can’t say that these hospitals are committing fraud,” he said. “I wish I could, but I can’t.”

Instead, he said, the test could serve as a screening mechanism, a way to identify hospitals whose reports warrant additional investigation.

“Just having an open mind when it comes to how to analyze the data,” Duffy said. 

He noted that CMS may already use Benford’s Law in some capacity, but said the research could point policymakers toward other ways of examining the data, including methods they may not have considered before.

Editor’s Note: Link to the full manuscript will be added upon publication