OBJECTIVE: During critical illness, physicians often provide estimates of the severity of underlying disease to aid patients and families when formulating care directives. We sought to determine whether factors such as the superimposed acute illness, the prognoses of other patients cared for by the same physician, or the phrasing of inquiry influence these assessments of underlying disease. DESIGN, SETTING, AND SUBJECTS: Internal medicine attending and resident physicians participated in a computerized, Web-available survey that described two case vignettes, one with cardiomyopathy and the other with lung cancer as underlying diseases. Using random assignment, each respondent was presented one case as having septic shock, and the other as an uneventful clinic visit. Respondents were explicitly asked to ignore the context and to assess the severity of the underlying disease alone to predict survival time and quality of life. The order in which subjects encountered the cases and phrasing of the survival question also were varied through randomization. MEASUREMENTS AND MAIN RESULTS: Mortality predictions for the cardiomyopathy case at 5 yrs were sensitive to both context (predicted survival, 39% +/- 23% when presented as septic vs. 52% +/- 24% when presented as a clinic patient; p < .001) and to whether a lung cancer case was presented before it (39% +/- 23% when presented after lung cancer vs. 52% +/- 24% when presented before; p < .001). These effects were independent and led to mean predicted 5-yr survival probabilities ranging from 31% to 59%. Predicted 1-yr survival from lung cancer was sensitive to phrasing (p < .05) but not to context. Quality of life predictions were also sensitive to context and case order. CONCLUSIONS: Physician appraisal of underlying disease severity is potentially vulnerable to a number of biases that may be relevant in the critical care setting. These biases appear to vary with the nature of the underlying disease.