Since the launch of Nursing Home Compare (NHC) in 2002, consumers have had access to information about the quality of care provided by most nursing homes (NHs) throughout the country. The intention is to help consumers distinguish among NHs and motivate informed decision making based on quality. For NHC to be useful, the quality measures that it publishes must be relevant, actionable, and scientifically sound.
The quality measures currently available on the NHC website address key quality domains of custodial and postacute care in general, such as functional improvement, pressure ulcer prevention, overuse of high-risk medications, and, more recently, community discharges, rehospitalizations, etc. However, none of the measures are designed specifically for residents at the end of life, whose care needs and goals are known to be different from the general population in NHs. At the end of life, the care priorities shift from ongoing curative measures to comfort, dignity, reliefs of pain, symptoms, and emotional stress. In particular, quality end-of-life (EOL) care emphasizes avoidance of unnecessary transfers, inappropriate prolongation of dying, adequate symptom management, and encourages death at places where patients live. With relatively fewer Americans spending the ends of their lives in hospitals (also here), information about the quality of EOL care in NHs is highly desired by both consumers and the NH industry. The lack of well-established EOL quality measures that can be readily constructed from existing data sources has impeded progress towards this goal.
A study newly published ahead of print in Medical Care investigates the feasibility of using administrative data to develop EOL quality measures that are substantively meaningful and statistically robust. The study used the Minimum Data Set (MDS) 3.0, a federally required resident assessment database that is routinely collected from all residents in Medicare/Medicaid-certified NHs. The study focused on decedents from the 626 NHs in New York State between October 2012 and September 2013.
Four outcome measures were selected according to relevance to EOL care and modifiability by providers. These outcome measures were: death in the hospital (instead of NH), number of hospitalizations, self-reported pain, and presence of depressive symptoms in the last 90 days of life. Quality measures at the nursing home level were defined as the difference between the observed versus expected rate after adjusting for risk factors that were shown in other analyses to influence individual patient outcomes, such as age, race, marital status, need for interpreter, length of stay, functional status, comorbidities, and treatments (slightly different groups of risk factors were relevant for the different quality measures).
According to the study, the four EOL outcome measures vary substantially across NHs, with the largest variation observed in the measure for the number of hospitalizations. Sufficient variation is an essential property of a good quality measure: it suggests ability to differentiate among providers while allowing room for improvement. The variation across NHs remains when focusing on NHs with at least 20 eligible residents, although the variation does decrease markedly for some measures, such as pain.
Most of the EOL outcome measures are positively but moderately correlated, as is the case for NH quality measures in general. According to the study, only 6% of NHs were outliers (defined as the top and bottom 10th percentile of the distribution) in two measures and only 1% were outliers in three measures. The authors (led by Dana Mukamel, PhD of the University of California-Irvine, with coauthors from UCI and the University of Rochester) suggest that these measures likely capture different constructs of EOL care in NHs, and that NHs may have to make decisions on which domains to invest on, given the constraints in resources.
EOL quality measures were further found to be positively related to the other NHC quality measures for the general NH population. For instance, NHs designated as four- or five-star facilities perform better in the management for both pain and depressive symptoms than one- or two-star facilities. When comparing four- or five-star facilities to those with one- or two-star ratings in nurse staffing, particularly registered nurses, four- or five-star facilities perform better on the prevention of hospitalizations and death in the hospital regardless of facility rating for overall quality. Performance on pain is higher in facilities with high performance on health inspection and other clinical quality measures. These results indicate that the EOL quality measures that were developed coordinate well with other domains of quality in NHs.
This study has profound policy implications. For the first time, this study provides evidence for the feasibility of using MDS 3.0 to construct meaningful EOL quality measures. Compared to its precursor (MDS 2.0), MDS 3.0 incorporates more information collected through resident interview directly, and hence provides more refined definitions for symptoms like pain and depression. However, some of the items highly relevant to the care provision at the end of life, such as advance directives, are absent. Another study, also published in Medical Care, has shown that advance directives change over time, particularly after hospitalizations and NH transfers, and argues that they should be monitored in a timely fashion along with other important variables for clinical and health status.
This study also provides insightful suggestions on how to augment MDS’s ability to construct meaningful EOL outcome measures. For outcome measures to be meaningful, risk adjustment is needed to prevent penalizing providers for factors that are beyond their control. MDS 3.0 contains a rich set of information on resident characteristics that may be employed as potential risk factors for this purpose. However, as discussed in the paper, some common risk factors, such as education, are not reported in the MDS 3.0.
Some of the risk factors identified to be highly predictive to the outcomes are available in only certain set of MDS assessments (e.g. admission assessment, annual assessment, assessment for significant changes of clinical status, etc.). Not all EOL residents will have these assessments. While majority of the NH residents have admission assessments, the information documented may not accurately reflect the NH quality of care for EOL residents. Additionally, MDS 3.0 does not incorporate resident and family preferences with regard to EOL care and advance directives, which, if properly monitored and fully respected by NH staff, should strongly predict EOL care practices and resident outcomes. Adding this information to the MDS data may considerably improve the performance of the risk adjustment model.
Despite these limitations, MDS remains to be a powerful tool and provides unparalleled opportunity for large scale quality assurance efforts for the vast EOL population residing in NHs. Further research is needed to evaluate the feasibility of adding additional information specific to EOL care to the MDS system.
