Changing landscape of nursing homes serving residents with dementia and mental illnesses

Huiwen Xu, Orna Intrator, Eva Culakova, John R. Bowblis

Research output: Contribution to journalArticlepeer-review

5 Scopus citations


Objective: Nursing homes (NHs) are serving an increasing proportion of residents with cognitive issues (e.g., dementia) and mental health conditions. This study aims to: (1) implement unsupervised machine learning to cluster NHs based on residents' dementia and mental health conditions; (2) examine NH staffing related to the clusters; and (3) investigate the association of staffing and NH quality (measured by the number of deficiencies and deficiency scores) in each cluster. Data sources: 2009–2017 Certification and Survey Provider Enhanced Reporting (CASPER) were merged with data on NHs in the United States. Study design: Unsupervised machine learning algorithm (K-means) clustered NHs based on percent residents with dementia, depression, and serious mental illness (SMI, e.g., schizophrenia, anxiety). Panel fixed-effects regressions on deficiency outcomes with staffing-cluster interactions were conducted to examine the effects of staffing on deficiency outcomes in each cluster. Data extraction methods: We identified 110,463 NH-year observations from 14,671 unique NHs using CASPER data. Principal findings: Three clusters were identified: low dementia and mental illnesses (Postacute Cluster); high dementia and depression, but low SMI (Long-stay Cluster); and high dementia and mental illnesses (Cognitive-mental Cluster). From 2009 to 2017, the number of Postacute Cluster NHs increased from 3074 to 5719, while the number of Long-stay Cluster NHs decreased from 6745 to 3058. NHs in Long-stay/Cognitive-mental Clusters reported slightly lower nursing staff hours in 2017. Regressions suggested the effect of increasing staffing on reducing deficiencies is statistically similar across NH clusters. For example, 1 hour increase in registered nurse hours per resident day was associated with −0.67 (standard error [SE] = 0.11), −0.88 (SE = 0.12), and −0.97 (SE = 0.15) deficiencies in Postacute Cluster, Long-stay Cluster, and Cognitive-mental Cluster, respectively. Conclusions: Unsupervised machine learning detected a changing landscape of NH serving residents with dementia and mental illnesses, which requires assuring staffing levels and trainings are suited to residents' needs.

Original languageEnglish (US)
Pages (from-to)505-514
Number of pages10
JournalHealth Services Research
Issue number3
StatePublished - Jun 2022


  • deficiency score
  • dementia
  • mental illnesses
  • nursing homes
  • staffing
  • unsupervised machine learning

ASJC Scopus subject areas

  • Health Policy


Dive into the research topics of 'Changing landscape of nursing homes serving residents with dementia and mental illnesses'. Together they form a unique fingerprint.

Cite this