With Medicare-aged Americans accounting for over 15% of the population and significant health care costs, stakeholders have leveraged various approaches, including the Medicare Current Beneficiary Survey, for estimating costs.
Researchers are urging caution when using survey or claims data alone for estimating medical costs. In a recent study, prevalent discrepancies were identified between claims and survey data from a widely used cost estimation resource for Medicare patients.
With Medicare-aged Americans accounting for over 15% of the population and significant health care costs, stakeholders have leveraged various approaches, including the Medicare Current Beneficiary Survey (MCBS), for estimating costs.
“The MCBS provides a unique opportunity to investigate health care costs from both claims (bill records files) or adjusted survey data (survey and administrative records files). The adjusted survey data is a synthesized, matched version of the original survey data and claims,” wrote the researchers of the current study in Health Science Reports. “The inclusion of claims information addresses the underreporting of health events from the survey data due to recall bias, which is a severe issue often seen in health surveys. However, the matching process of survey and claims data is not straightforward and does not guarantee complete accuracy.”
Drawing on data from 5000 Medicare patients whose cancer was diagnosed between 2006 and 2012, the researchers found significant discrepancies between survey and claims data cost estimates despite MCBS adjusted survey data being matched with original claims. These discrepancies occurred regardless of whether the patient also had diabetes, which was present in 26%.
In most cases, claims data yielded higher cost estimates than survey data. For example, in 2007, there was an $8000 difference in annual medical cost per capita; this was estimated at $29,000 from claims data and $21,000 for survey data. Although differences were less prominent in other years, discrepancies were still prevalent.
Throughout the study period, the annual medical cost per capita was higher for Medicare patients with both cancer and diabetes than for patients with cancer alone, with costs peaking at a high of $29,000 for patients with both conditions and $21,000 for patients with cancer alone.
Patients included in the study were enrolled in Medicare Parts A, B, and D; Medicare Advantage was not included.
“Our findings seek to raise awareness about the potential implications of methodological assumptions,” explained the researchers. “A small body of literature examined methodological assumptions when estimating medical expenditures using MCBS. For instance, MCBS has been used to compare medical expenditure estimates with publicly available data sources (Medical Expenditure Panel Survey, MEPS) to determine survey underreporting in MEPS.”
The group highlighted the use of MCBS to determine the accuracy of data for Medicare-covered veterans, which revealed inconsistencies with data from the Veterans Health Administration. Differences have also been identified in condition-specific medical expenditure estimates, such as those for Alzheimer disease due to use of different combinations of survey and claims data.
Within their study, the researchers flagged that, although unlikely, there’s a possibility that patient diagnoses may have been miscategorized as they were not directly followed across different years of inclusion. However, they noted that miscategorization would have had minimal effects on their large cohort due to use of both claims and survey data.
Chou C, McDaniel C, Lai T, McDonald C, Rockwell D, Loh FH. Reproducibility of the medical cost estimation from the Medicare Current Beneficiary Survey: comparing claims and survey. Health Sci Rep. Published online April 13, 2023. doi:10.1002/hsr2.1193