Send to

Choose Destination
J Gerontol A Biol Sci Med Sci. 2018 Nov 12. doi: 10.1093/gerona/gly260. [Epub ahead of print]

Predicting Mortality in Nursing Home Residents with Dementia and Pneumonia Treated with Antibiotics: Validation of a Prediction Model in a More Recent Population.

Author information

Department of Epidemiology and Biostatistics, Amsterdam Public Health research institute, Amsterdam UMC, VU University Medical Center, Amsterdam, The Netherlands.
Centre for Health Protection, National Institute for Public Health and the Environment (RIVM), Bilthoven, the Netherlands.
Department of Family and Community Medicine, School of Medicine, University of Missouri, Columbia, MO USA.
Leiden University Medical Center, Department of Public Health and Primary Care, Leiden, The Netherlands.
Radboud University Medical Center, Department of Primary and Community Care, Nijmegen, The Netherlands.



We aimed to find the best predictive model for 14-day mortality in antibiotic-treated nursing home residents with dementia and pneumonia by first applying an existing model to the recent PneuMonitor study. Second, we evaluated whether model performance improved by revising variables or adding variables related to recent changes in the care for older people.


The original prognostic model included gender, respiratory rate, respiratory difficulty, pulse rate, decreased alertness, fluid intake, eating dependency and pressure sores. This model was applied to 380 recent pneumonia episodes in nursing home residents with dementia, updated by considering revising and/or adding variables, internally validated using bootstrapping, and transformed into a simplified risk score that can be used in clinical practice. Model performance was evaluated by Hosmer-Lemeshow statistics (HL) and calibration graphs to assess calibration; and area under the receiver operating characteristic curve (AUC) to assess discrimination.


The newer cohort had lower 14-day mortality and was less often dehydrated or malnourished. Median AUC of the original model over the imputed datasets was 0.76 (IQR: 0.76-0.77), compared to 0.80 in the old cohort. Extending the model with dehydration, bowel incontinence, increase in eating dependency and cardiovascular history, while removing pressure sores, improved AUC: 0.80 (IQR: 0.80-0.81) after internal validation. Calibration remained adequate (HL: p=0.67).


In the newer cohort with less severe illness, model performance of the existing model was adequate, but a new extended model distinguished better between residents at low and high mortality risk.


Supplemental Content

Full text links

Icon for Silverchair Information Systems
Loading ...
Support Center