Survival prediction in terminal cancer patients: a systematic review of the medical literature

The clinical significance of studies on survival predictors in terminal cancer patients is hindered by both methodological limitations and the difficulty of finding common predictors for all final events in cancer related deaths. To evaluate the published medical literature concerned with the survival of patients with terminal cancer and identify potential prognostic factors, major electronic databases including MEDLINE (1966-), CANCERLIT (1983-) and EMBASE (1988-) were searched up to September 1999. Studies were included in our review if published in English, were cohort studies, addressed the identification of clinical prognostic factors for survival and looked at samples with median survival of < or = 3 months. Data extracted from selected papers included: sample size, median survival, type of study, sampling frame, cohort type, type of statistical analysis (univariate or multivariate), choice of models and underlying assumptions, predictors examined and their reported level of statistical significance. A total of 24 studies were found and reviewed. On the basis of these studies, performance status and the presence of cognitive failure, weight loss, dysphagia, anorexia and dyspnoea appear to be independent survival predictors in this population. Clinical estimation of survival by the treating physician appeared independently associated with survival but the magnitude of the association generally appeared small. Clinical predictions should be considered as one of many criteria, rather than as a unique criterion by which to choose therapeutic interventions or health care programmes for terminally ill cancer patients. The use of convenient samples as opposed to more representative inception cohorts, the inclusion of different variables in the statistical analyses and inappropriate statistical methods appear to be major limitations of the reviewed literature. Methodological improvements in the design and conduction of future studies may reduce the prognostic uncertainty in this population.