VARIABILITY IN PHYSICIAN PROGNOSIS AND RECOMMENDATIONS FOR LIFE SUSTAINING TREATMENT AFTER INTRACEREBRAL HEMORRHAGE

Purpose: The role of physicians in the observed variability in end-of-life treatment decisions remains under debate.  We investigated physician prognostic estimates and treatment recommendations in intracerebral hemorrhage (ICH), a particularly severe type of stroke where early limitations in life-sustaining treatments are common. 

Methods: A written survey was mailed to 3727 practicing US neurologists and neurosurgeons consisting of two scenarios of moderate to severe ICH. Selected factors were randomly varied including patient characteristics (age, clinical severity) and presence (versus absence) of a validated prognostic score indicating probability of 90-day functional recovery. All patients were described as being functionally independent at baseline and having no explicit advance directives but a general preference to avoid long-term dependence on machines. Physicians were asked to indicate their predictions of 30-day mortality (free text write-in from 0-100%) and initial treatment recommendations (6-point ordinal scale from 1: comfort only to 6: full treatment, dichotomized as 1-3 vs. 4-6 for analysis). Multilevel marginal regression models were used to investigate predictors of physician-predicted mortality and treatment recommendations.

Results: A total of 816 physicians responded (response rate 22%), with complete data available for 742. Mean age was 52, 32% were neurosurgeons, and 17% were female. Physician predictions of 30-day mortality varied widely (Figure 1). Physician factors associated with mortality prediction included surgical specialty (p<0.001; surgeons more optimistic than non-surgeons), geographic region (p=0.02; West was most optimistic), and number of ICH cases seen in the prior year (p<0.01; 16+ cases more pessimistic than 1-15 cases). Other physician factors including age, race, sex, and personality characteristics (empathy, religious importance, optimism) were not associated with mortality predictions. Treatment recommendations also varied widely (Figure 2), though none of the investigated physician demographic or personality characteristics were associated with treatment recommendations. Providing the results from a validated prognostic score did alter physicians' overall treatment recommendations (p<0.001), though this effect was mostly seen in the younger, moderately severe case (odds ratio for predicting limited treatment 0.22, 95% CI 0.10, 0.50, p<0.001).

Conclusions: Physicians vary substantially in their prognostic estimates and treatment recommendations for ICH. Providing physicians with a formal prognostic score does change their treatment recommendations, though the impact of the prognostic score depends on patient characteristics.