This page describes the methodology used to calculate the skilled nursing facility (“SNF”) ratings reported in our website.

​

SNF Rating Categories

Rating categories between 1 to 5 were assigned based on their percent ranking (or percentile) on metrics analyzed. Specifically, SNFs in the top 20% for a particular metric were assigned a “5”, SNFs in the 60th to 80th percentiles were assigned a “4”, and similarly for rating categories “1” to “3”. As such SNFs in the lowest 20% for a given metric were assigned a “1”. When more than one metric was reported in a rating category, the percentiles for each metric were averaged before assigning the rating category. For example, in the health outcomes rating category, we averaged the fall, pressure ulcer and urinary tract infection percentiles. For the overall SNF rating, we consolidated all of the individual category ratings and assigned a composite rating based on the resulting data. The overall SNF score was not calculated if a majority of the individual metrics were missing or not calculated.

​

Patients per Nurse

This category used data from the Payroll-Based Journal, published by the Centers for Medicare & Medicaid Services (CMS). It contains data on staff hours and the number of patients at a skilled nursing facility. To calculate this metric, we compiled the total number of nurse hours at a given SNF, excluding days where the number of patients was recorded as 0. We excluded nurse roles that were primarily administrative (e.g. Director of Nursing; consistent with an analysis by Kaiser Health News) and added the total number of registered nurse and licensed vocational nurse hours together. We divided total nurse-hours by 24 to get the total nurse-days. We then took the total number of patient-days and calculated patients per nurse. This metric used data from 2018 to 2019.

 

Health Outcomes Metrics

We developed the three metrics below with the goal to provide more objective measures of SNF patient health outcomes, using third-party hospital data instead of self-reported data from SNFs. Please refer to our research report, Underreporting in Nursing Home Quality Measures, for a comparison of our metrics with those self-reported by SNFs.

​

Falls and Trauma

Our falls and trauma metric seeks to provide an objective, third-party metric which differs from the current falls metric reported by CMS which is based on SNF self-reported data. In its report, Medicare-Medicaid Eligible Beneficiaries and Potentially Avoidable Hospitalizations, CMS identifies falls and trauma as “conditions that could result in potentially avoidable hospitalizations for individuals in institutional settings.” This metric calculates the rate at which patients at a given SNF were admitted to a hospital with fall and trauma conditions. Using inpatient hospital claims data from the Medicare limited data set standard analytic files, we examined hospital diagnoses (primary and secondary) and external cause codes when patients were admitted to the hospital during their SNF stay or 1 day after discharge from the SNF. To identify falls and trauma conditions, we started with a list of ICD-9 diagnosis codes related to falls and trauma from the CMS report on potentially avoidable hospitalizations, then translated those codes to ICD-10. These include external cause codes indicative of falls and diagnoses such as broken bones. Pathological fractures that could have been caused by diseases such as osteoporosis were not included. Please refer to the diagnosis spreadsheet for a full list of codes used. To exclude injuries that could have occurred at the hospital, diagnoses were excluded if they were not marked ‘Present on Admission’ in the hospital claim. Consistent with other health metrics used by the Centers for Medicare & Medicaid Services (CMS), we only counted falls that occurred within a certain number of days after SNF admission, in this case 60 days. The denominator for this measure was the total number of admissions at the SNF from Medicare Part A during the time period. We chose not to report this measure for SNFs with less than 20 admissions during the time period. This metric used data from 2016 to March 31, 2019.

 

Pressure Ulcers

Similar to the falls and trauma metric, we used inpatient hospital claims data, and examined all hospital diagnoses when patients were admitted to the hospital during their SNF stay or 1 day after discharge from the SNF. See the appendix for a full list of codes used. To help exclude ulcers that could have occurred at the hospital, diagnoses were excluded if they were not marked ‘Present on Admission’ in the hospital claim. We only counted pressure ulcers that occurred within 60 days after SNF admission. The denominator for this measure was the total number of admissions at the SNF from Medicare Part A during the time period. We chose not to report this measure for SNFs with less than 20 admissions during the time period. This metric used data from 2016 to March 31, 2019.

 

Urinary Tract Infections

Similar to the falls and trauma metric, we used inpatient hospital claims data, and examined all hospital diagnoses and procedures when patients were admitted to the hospital during their SNF stay or 1 day after discharge from the SNF. Using lists of diagnosis codes provided by the Agency for Healthcare Research and Quality, we counted Urinary Tract Infection (UTI) diagnoses only when there was no indication of certain kidney diseases or an immuno-compromised state that could have caused the UTI. To help exclude ulcers that could have occurred at the hospital, diagnoses were excluded if they were not marked ‘Present on Admission’ in the hospital claim. We only counted pressure ulcers that occurred within 60 days after SNF admission. The denominator for this measure was the total number of admissions at the SNF from Medicare Part A during the time period minus admissions with kidney diseases or indications of immune-comprised state. We chose not to report this measure for SNFs when the denominator was less than 20. This metric used data from 2016 to March 31, 2019.

 

Incidents of Potential Abuse and Neglect

To identify incidents of potential abuse and neglect at SNFs, we used a list of diagnoses determined “to be high risk for potential abuse or neglect” by the U.S. Department of Health and Human Services, Office of the Inspector General 2019 Report titled Incidents of Potential Abuse and Neglect at Skilled Nursing Facilities were Not Always Reported and Investigated. In their sample based on emergency department claims, they report that 20% of these diagnoses could be traced to an incident of abuse. Please note that our calculated metric is not a claim that abuse occurred in the facility but is based on diagnoses (for example, bruises) that could potentially be the result of abuse. To calculate this metric, we used inpatient hospital claims data, and examined all hospital diagnoses when patients were admitted to the hospital during their SNF stay or 1 day after their discharge from the SNF. We also used outpatient claims from 2016. The denominator for this measure was the total number of admissions at the SNF from Medicare Part A. The numerator only includes diagnoses within 60 days after SNF admission, consistent with our health outcome metrics. This metric used data from 2016 to March 31, 2019.

 

Previous Risk of Excessive Treatment

The U.S. Department of Justice has taken action against SNFs for keeping their patients longer than medically necessary and for administering higher levels of rehab than medically necessary, arguing that SNFs provided therapy services to maximize profits rather than the interests of their patients. We calculated two metrics to capture the risk of excessive treatment: i) unexplained ultra-high rehab days and ii) unexplained length of stay.

Ultra-high rehab days are the number of days patients at a SNF received the highest level of rehab therapy, equivalent to over 720 minutes, and are reimbursed at the highest per diem rate by CMS. To calculate unexplained ultra-high rehab days, we ran a regression with ultra-high rehab days as the dependent variable. Explanatory variables included patient medical characteristics (e.g., principal and secondary diagnoses from the hospital stay prior to admission to SNF), patient characteristics (e.g., age, gender and race) and county-level demographic data (e.g., education and income). The unexplained ultra-high rehab days is the number of days that cannot be explained by the explanatory variables. Thus, a higher positive number of ultra-high rehab days at a SNF indicates that patients at that SNF on average received a higher number of days of ultra-high rehab than would be explained by their medical, personal, and demographic characteristics.

​

A similar methodology was used to calculate unexplained length of stay, which is the average additional number of days patients spent at a SNF that cannot be explained by their medical, personal and demographic characteristics.

For these metrics, admissions within the last 90 days of our data were excluded because the end of our data set would have artificially cut short the length of admissions during that time period. We also excluded SNFs with less than 20 admissions for this measure. These metrics used data from 2016 to March 31, 2019.

​

It should be noted that in the last quarter of 2019, CMS changed the reimbursement system for SNFs from RUG-IV (Resource Utilization Group IV) to the new PDPM (Patient Driven Payment Model) reimbursement system. This was to address the issue that the previous RUG-IV model “create[d] an incentive for SNF providers to furnish therapy to SNF patients regardless of the patient’s unique characteristics, goals, or needs.” According to CMS, the new PDPM reimbursement system “eliminates this incentive”. Nevertheless, we are including this metric to enable the public to see which SNFs had the prior risk of providing excessive treatment based on our calculated metrics.

​

Chain Ownership

Sometimes companies or individuals will own multiple SNFs. Using SNF ownership data published by CMS (updated in May 2020), we examined all owners who were ‘direct owners,’ ‘indirect owners’ or had ‘operational/managerial control.’ For chain determination, we only considered owners who owned more than 5 SNFs. To decide when the ownership of multiple SNFs was sufficiently similar to conclude they were in the same chain, we used a standard clustering algorithm, the Louvain algorithm. The Louvain algorithm is a type of ‘community detection’ algorithm used to partition networks or graphs into groups. For the algorithm we setup our data with SNFs connected to owners (i.e. an undirected bi-partite graph). To decide what to call the chain, if it had an organizational owner then we picked the organizational owner that owned the most SNFs, otherwise we chose the individual who owned the most SNFs.

​

List of Diagnoses Used

Click here to download the spreadsheet.

​