BACKGROUND The success rate of cardiopulmonary resuscitation (CPR) varies with the patient population studied. Prearrest variables have been used to identify groups of patients with a particularly low rate of survival following CPR. The purpose of this study was to use the technique of meta-analysis to identify prearrest variables associated with a decreased rate of survival to the time of discharge following CPR of hospitalized patients. METHODS The MEDLINE database was searched using the following key words: resuscitation, survival, and 1980-1991. The bibliographies of studies identified in the computerized search as well as those of appropriate studies in the author's personal files were reviewed. The following inclusion criteria were used: study published since 1980 (data collected after 1975), in-hospital resuscitation, retrospective or prospective design, consecutive adult patients from both the general wards and intensive care units, an end-point of survival to hospital discharge, and data reported for at least one variable measured before cardiac arrest. Of the 22 studies initially identified, 14 met the above criteria. Data were taken directly from the published reports. In two cases where no specific survival data were reported but the studies met all other criteria for inclusion, the authors were contacted and agreed to supply the necessary information. Two-way variables were combined across studies using the Mantel-Haenszel statistic. RESULTS The following prearrest variables were associated with a decreased rate of survival to discharge following CPR (P less than .005): age over 70 years, serum creatinine greater than 130 mumol/L, serum creatinine greater than 220 mumol/L, homebound lifestyle, presence of cancer or metastatic cancer, and a primary diagnosis of sepsis or pneumonia. A primary diagnosis of myocardial infarction was associated with an increased rate of survival to discharge following CPR (P = .005). Based on these results, modifications to the Pre-Arrest Morbidity Index initially devised by George and colleagues have been proposed. CONCLUSIONS The identification of prearrest variables that are associated with decreased survival following CPR will assist clinicians when they counsel their patients regarding do-not-resuscitate (DNR) orders. In addition, the further refinement of a predictive tool such as the modified Pre-Arrest Morbidity Index can help clinicians to identify patients for whom CPR is futile. Such an instrument must be validated on an independent data set before it can be considered for clinical application.