Article #19
Newsletter list 


Previous Article 

Next Article 


Performance of Physicians at Judging Survival in Congestive Heart Failure

Wally R. Smith
Richmond, Virginia

Roy Poses, Donna McClish, and I have been investigating how physicians make judgments of the probability of important outcomes, especially for the common clinical problem of congestive heart failure (CHF), a clinical syndrome characterized by the inability of the heart to pump blood adequately for the body's needs. Acute CHF may be a medical emergency and may result in a decision to admit a patient to an intensive care unit (ICU). Current guidelines suggest that physicians should base ICU triage decisions in part on their judgment of the probability that the patient will survive in the short-term (and imply that patients with a very small likelihood of survival should not be admitted to an ICU because care for them there would be futile.) We have shown that physicians' judgments of survival for patients with acute CHF made at the time the triage decision has to be made are poorly calibrated (overly pessimistic) and have minimal discriminating ability. So our next questions were how do physicians use relevant clinical cues when making these judgments, how well does a model of their judgments based on such cues predict survival, and how well do such cues actually predict survival.

We enrolled a sequential cohort of patients visiting Emergency Departments (ED's) at one of three hospitals, an urban university hospital, a VA hospital, or a community hospital in one metropolitan area. We excluded patients with acute myocardial infarction (or "heart attack," who are physiologically and clinically different from other patients with acute CHF), and excluded patients who died or developed an acute complication requiring ICU care in the ED (because there would have been no question about the triage decision for them.)

We collected data about clinical cues from a chart review, about survival from multiple sources, and about physicians' judgments prospectively at the time of the ED visit using a standardized instrument. We selected as cues variables that previous research or our clinical judgments suggested might be related to survival for patients with acute CHF.

We modelled the relationship of these variables to 90-day survival using logistic regression (survival model), and the relationship of the same set of variables to the logit of the physicians' survival judgments (judgment model) using linear regression.

The R squared for the judgment model was .20. The area under the ROC curve for the survival model was .76. Of the eight variables that independently predicted judgments or survival (Table), one predicted only judgment, five predicted only survival, and two predicted both.

Varialbe p, Judgment p, Survival
Age .0001  .0007
Sodium .9982 .0003
Low systolic blood pressure .0602 .0411
Orthopnea or paroxysmal nocturnal dyspnea .7608 .0002
Acute Coronary Artery Disease symptoms .0120 .6402
Prior requirement of ACE inhibitor .1771 .0281
Charlson comorbidity score .2217 .0138
Functional status (ED judgment) .0001 .0001

Physicians' judgments of survival for patients with acute CHF may be inaccurate because they fail to use cues that predict survival while using others that do not predict survival. Developing better predictive models and teaching physicians how to use them may improve clinical prediction and thus clinical decision making.

Contact Wally R. Smith

Previous Article                Next Article

Newsletter list