A substantial proportion of cancer patients presenting to an emergency center (EC) or clinic with acute dyspnea survives fewer than 2 weeks. If these patients could be identified at the time of admission, physicians and patients would have additional information on which to base decisions to continue therapy to extend life or to refocus treatment efforts on palliation and/or hospice care alone. The purpose of this study was to identify risk factors for imminent death (survival </= 2 weeks) and short-term survival (1, 3, or 6 months) in cancer patients presenting to an EC with acute dyspnea and to combine these factors into a model to help clinicians identify patients with short life expectancies. A random sample of 122 patients presenting to an EC with acute dyspnea was selected for a retrospective analysis. Data that were available to physicians during the initial EC visit included patient histories, triage and discharge vital signs, chest radiographs, and laboratory results. These variables were used in univariate and logistic regression models to develop predictive models for imminent death and short-term survival. Variables and interactions meeting a univariate criterion of P < 0.10 were included in stepwise regression by using forward and backward stepping. Models were compared with the use of Hosmer-Lemeshow statistics and receiver operating characteristics curves. Underlying cancers were 30% breast, 37% lung, and 34% other cancers. Triage respiration greater than 28/min., triage pulse greater than or equal to 110 bpm, uncontrolled progressive disease, and history of metastasis were found to be statistically significant predictors (alpha </= 0.05) of imminent death. Patients with uncontrolled progressive disease had a relative risk of imminent death of 21.93. Relative risks for triage respiration, pulse, and metastases were 12.72, 4.92, and 3.85, respectively. Cancer diagnosis was not predictive of imminent death but was predictive when longer time periods were modeled. It may be possible to identify patients whose death is imminent from a group of cancer patients with acute dyspnea. Some factors that predict imminent death (triage pulse and respiration) differ from those (cancer diagnosis) that predict short-term survival. Extent of disease/response to treatment is common to all models. These factors need further examination and validation. If these findings are confirmed, this quantified information can help physicians in making difficult end-of-life decisions.