ProPublica and the Houston Chronicle have concluded that Medicare patients whose left ventricular assist devices were implanted by Dr. O.H. “Bud” Frazier from 2010-2015 fared worse than those whose surgeries were performed by other physicians. During that time, about half of the traditional Medicare patients who received a heart assist device from Frazier died within a year, nearly double the national mortality rate, ProPublica’s analysis showed. Here’s how we did it:

ProPublica obtained Medicare hospital inpatient data and mortality data for 2010-2015 (specifically, the Medicare 100% Standard Analytic Files and Denominator Files). The data was submitted by hospitals to Medicare for billing purposes, and hospitals are responsible for its accuracy.

Each record in the data represents a patient’s hospital visit, indicating when the patient was admitted, what diagnoses were present, which procedures were performed, and the identity of the operating surgeon, if applicable. The data conceals patients’ identities. The data does not represent all patients, only those who are enrolled in fee-for-service Medicare.

For the purpose of our analysis, we looked at records submitted under diagnosis groups 001 and 002, which cover heart transplants and implantation of mechanical heart pumps, and then further looked for specific procedure codes that correspond to ventricular assist device implantation (37.66 under the ICD-9 schema and 02HA0QZ under ICD-10). In total, our analysis looked at 7,552 such inpatient visits; we excluded 27 cases for which we lacked data on whether the patient survived.

A small percentage of those inpatient visits (about 1 percent) included both the implantation of ventricular assist devices and heart transplants. We included such cases when calculating outcomes. Results are very similar if those cases are excluded. Some patients (about 4 percent) had more than one inpatient visit to implant a ventricular assist device in the years covered by our data.

We looked at data for surgeons at Baylor St. Luke’s Medical Center in Houston, as well as all other hospitals across the country. The patients we noted for Frazier were those for which the hospital itself listed Frazier as the primary operating surgeon. Other physicians may have assisted with the procedures.

Frazier was one of the highest-volume surgeons in the country for ventricular assist device implantations. His patients were about the same age as typical Medicare patients receiving the procedure. (His median patient was 64 years old, compared to 65 nationally.) Medicare covers both seniors and disabled people, thus some patients are under age 65.

We tallied the number of patients who died within one year of discharge. Mortality data came from both inpatient records and from Medicare denominator files, which we obtained for 2010-2015. Survival rates for 2015 operations are incomplete. Without the 2016 denominator file, not all 2015 procedures have a full year of follow up.

Frazier and St. Luke’s disputed our analysis for a number of reasons. St. Luke’s said that Medicare only accounts for half of the total cases it performs. The hospital also said that some ventricular assist devices are not billed under diagnosis groups 001 and 002 and that our analysis did not include all types of ventricular assist devices. Finally, St. Luke’s said that a billing record may not reflect all of the physicians on a case.

While Medicare does not cover all surgeries, researchers commonly use Medicare data to compare outcomes because it represents tens of millions of patients and because data for privately insured patients is not publicly available. All doctors are compared on this basis, so no one is disadvantaged by it.

We examined whether St. Luke’s billed for ventricular assist device implantation codes under other diagnosis groups and found none. Also, we did not include hospital visits with only implants of short-term devices, which appear in the data with procedure codes such as 37.60, 37.65, 02HA0RJ, 02HA0RS, or 02HA0RZ, because such devices are generally used in extreme conditions and outcomes are generally worse. Experts said it would be inappropriate to include them in an analysis of outcomes for implanted ventricular assist devices.