Should Do-Not-Resuscitate status be included as a mortality risk adjustor? The impact of DNR variations on performance reporting

Med Care. 2005 Jul;43(7):658-66. doi: 10.1097/01.mlr.0000167106.09265.4e.


Background: The practice of ordering "Do-Not-Resuscitate" (DNR) varies across hospitals. No research has explored how the DNR variation would affect cross-institutional performance reporting when DNR status is used as a risk adjustor.

Objective: We sought to assess the impact of DNR variation on performance reporting.

Research design: We used retrospective clinical data abstracted from chart review for our analysis.

Subjects: We studied a total of 184,057 adult patients admitted to 149 Pennsylvania acute-care hospitals in 2001 for ischemic stroke, hemorrhagic stroke, pneumonia, acute myocardial infarction, congestive heart failure, and sepsis.

Measures: DNR rate and DNR mortality rate per patient at the hospital level was assessed. DNR also was used as an additional covariate to predict mortality in logistic regression models. Change of rank and outlier-status at the hospital level based on adjusted mortality determined by multivariable logistic models with or without DNR was used to assess the impact of DNR on performance reporting.

Results: Large variations in DNR rates (1-37%) and DNR mortality rates (8-60%) existed across hospitals. There was a significant negative correlation between DNR rates and DNR mortality rates (r = -0.66, P < 0.0001). Adding DNR as a covariate resulted in a systematic shift in performance rank (r = 0.88, P < 0.0001) and change in statistical outlier-status (n = 33), which favored hospitals with higher DNR rates.

Conclusion: Using locally defined DNR as an additional covariate potentially introduces systematic bias in performance reporting. A more uniform definition and application of DNR is needed if it is to be included as a risk adjustor.

MeSH terms

  • Aged
  • Chi-Square Distribution
  • Female
  • Hospital Mortality*
  • Humans
  • Inpatients / statistics & numerical data
  • Logistic Models
  • Male
  • Pennsylvania
  • Resuscitation Orders*
  • Retrospective Studies
  • Risk Adjustment / methods*
  • Software