Comparison of methods to identify long term care nursing home residence with administrative data

Abstract

Background: To compare different methods for identifying a long term care (LTC) nursing home stay, distinct from stays in skilled nursing facilities (SNFs), to the method currently used by the Center for Medicare and Medicaid Services (CMS). We used national and Texas Medicare claims, Minimum Data Set (MDS), and Texas Medicaid data from 2011-2013.

Methods: We used Medicare Part A and B and MDS data either alone or in combination to identify LTC nursing home stays by three methods. One method used Medicare Part A and B data; one method used Medicare Part A and MDS data; and the current CMS method used MDS data alone. We validated each method against Texas 2011 Medicare-Medicaid linked data for those with dual eligibility.

Results: Using Medicaid data as a gold standard, all three methods had sensitivities > 92% to identify LTC nursing home stays of more than 100 days in duration. The positive predictive value (PPV) of the method that used both MDS and Medicare Part A data was 84.65% compared to 78.71% for the CMS method and 66.45% for the method using Part A and B Medicare. When the patient population was limited to those who also had a SNF stay, the PPV for identifying LTC nursing home was highest for the method using Medicare plus MDS data (88.1%).

Conclusions: Using both Medicare and MDS data to identify LTC stays will lead to more accurate attribution of CMS nursing home quality indicators.

Keywords: Long term care; Medicare; Minimum Data Set; Nursing home.

Fig. 1

Venn diagram illustrating overlap of the three methods with each other and with the gold standard of long term care (LTC) stays identified in Medicaid. Please note that the areas shown in the figure are not proportional to the numbers in each category. These analyses were done at the level of enrollee, determining whether the individual enrollee resided in a LTC nursing home for > 100 days in 2011. The analyses presented in Tables 2 and 3 are conducted at the episode level, comparing episodes in a LTC nursing home identified by the different methods. The results are similar with the two approaches

Fig. 2

Sensitivity and positive predictive value (PPV) of methods 2 and 3 for identifying a long term care (LTC) stay, as a function of the number of days identified by each method that are also identified with Medicaid data. The sensitivities and PPVs shown in Table 2 were generated with the rule that there was at least one day of overlap between the LTC episode identified by a method and an episode identified using Medicaid claims. The sensitivities of both methods decline as the number of days of required overlap with Medicaid increases. There are smaller declines in the PPVs