Underreporting of Quality Measures and Associated Facility Characteristics and Racial Disparities in US Nursing Home Ratings

This quality improvement study assesses facility-reported rates of major injury falls and pressure ulcers in nursing homes across the US and investigates whether nursing home characteristics are associated with underreporting of these measures.


A. Identification of major injury fall hospitalizations from MedPAR in 2016 and 2017
In the prior study on falls that we followed for linking assessments and hospitalizations, the study period was 2011 -2015. 1 In this paper, we included data through 2017. We used the same methods as the published fall study, adapted from an algorithm developed by Kim et al. 2 , to identify major injury fall hospitalizations from the Medicare Provider Analysis and Review (MedPAR) data, and cross walked fall-related ICD-9-CM codes to ICD-10-CM codes using CMS published mappings 3 . All diagnosis codes used can be found in the code link below.

B. Nursing home sample
Our initial sample consisted of 16,843 Medicare and/or Medicaid certified nursing homes from CASPER and LTCFocus during 2011 -2017, and we imposed some restrictions for inclusion in the final sample. First, we excluded nursing homes with fewer than 4 observations (n=226) in CASPER or LTCFocus during the study period for more complete data of nursing home characteristics. Next, we excluded nursing homes that were not in the sample continuously from the beginning to the end of the study period (n=1,952). We also dropped nursing homes that failed to match with any nursing home from Minimum Data Set data (n=7). Since this study focused on hospitalizations for falls and pressure ulcers, which were rare events in nursing homes, we excluded small nursing homes with fewer than 37 residents, the 10th percentile of the sample distribution (n=1,469). Our final sample included 13,179 nursing homes.

C. Sensitivity analysis: Nursing home hospitalization rates for falls and pressure ulcers
Most nursing homes had predominantly White residents and thus had small numbers of residents who were Black. This could result in extreme estimates of hospitalization rates for falls and pressure ulcers among residents who were Black. Therefore, in this sensitivity analysis, we included nursing homes only if they had more than 50 residents who were White or Black to calculate the hospitalization rates. eTable 1 shows the results. Compared with Table 3 in the paper, the estimates changed a little, but the patterns remained the same: within the same level of nursing home race mix, fall hospitalization rates were higher among residents who were White than among residents who were Black, whereas pressure ulcer hospitalization rates were lower among residents who were White. In addition, nursing homes with lower percentages of White residents had lower fall rates but higher pressure ulcer rates.

D. Sensitivity analysis: Nursing home characteristics by pressure ulcer reporting rate
In our final sample, most nursing homes had small numbers of hospitalizations: among the nursing homes that had pressure ulcer hospitalizations, 47.7% (n=4402) experienced only one or two such events. Reporting rates based on such small denominators could have high uncertainties, as they can only assume the values of 0, 0.5, or 1 (both 0 and 0.5 would be categorized as low reporting rates according to our cutoff). Therefore, in the paper, we restricted the sample for nursing homes to those with at least three pressure ulcer hospitalizations during 2011 -2017. In this sensitivity analysis, we present nursing home characteristics by the level of pressure ulcer reporting accuracy based on all nursing homes with at least one such hospitalization. As expected, the estimates in the medium reporting group remain the same, whereas many of those in low and high reporting groups changed. Some patterns in eTable 2 become less clear in this sensitivity analysis. For example, in eTable 2, moving from the lowest reporting to the highest reporting for pressure ulcers, the overall rating decreased; however, in eTable 2, the overall rating fluctuated with no clear trend. 3. ICD-9-CM to and from ICD-10-CM and ICD-10-PCS Crosswalk or General Equivalence Mappings. NBER. https://www.nber.org/research/data/icd-9-cm-and-icd-10-cm-and-icd-10pcs-crosswalk-or-general-equivalence-mappings. Accessed November 1, 2022. We used the same code from two studies previously published on falls and pressure ulcers (the authors of those studies publicly shared the code) to build the sample data for fall hospitalizations in 2016 and 2017 (the published study only included 2011 -2015). We identified major injury fall hospitalizations from MedPAR data, and linked the claims to MDS assessments for residents who were discharged from nursing homes, admitted to the hospital within one day, and returned to the same nursing home within one day after hospital discharge.
Scripts used in this step were listed under the "fall" folder.
2. Create nursing home variables using MDS assessments data and MBSF data By linking MDS data and MBSF data, we created several nursing home-level variables that were used in the paper: the number of Medicare fee-for-service residents in nursing homes stratified by short stay and long stay as well as by White and Non-White, the percent of White residents, the percent of residents dually eligible for Medicare and Medicaid, the percent of residents with ADRD. We also aggregated ratings and quality measures from Nursing Home Care Compare (NHCC) data for nursing homes. Below are the scripts in this step.