Decrease in the average duration of hemodialysis treatment time is a continuing phenomenon. We investigated the relationship of 3-year mortality to duration of dialysis in a 1984-1985 national random sample of 600 hemodialysis patients from 36 dialysis units. Mortality was negatively associated with duration of dialysis treatments, as shown by the Cox model, adjusted for other patient and dialysis unit covariates. With adjustment for other covariates, patients receiving an average dialysis treatment duration of less than 3.5 hours had relative mortality risks of 1.17 to 2.18 compared with those with treatments longer than 3.5 hours (mortality risk of 1.0). Reverse causation (the possibility that more seriously ill patients received dialysis for a shorter time) appears unlikely. We conclude that duration of the dialysis procedure is an important element in determining patient mortality as one of the factors determining the adequacy of dialysis. (JAMA. 1991;265:871-875) SINCE hemodialysis first became an accepted therapy for end-stage renal disease (ESRD), treatment times have decreased substantially.1 The National Cooperative Dialysis Study2"6 repre¬ sented a definitive approach to the pre¬ scription of hemodialysis, including the duration of the dialysis procedure it¬ self. 3 Length ofdialysis is a key determi¬ nant of the procedure's effectiveness.4 This article reports on 3-year patient mortality, patient outcomes, and re¬ lated issues, by duration of dialysis pro¬ cedures, using a large national sample and controlling for a large number of clinical, demographic, and other covariates. MATERIALS AND METHODS The analyses used longitudinal data from approximately 600 patients under¬ going hemodialysis in 36 dialysis units nationwide, including home but not peritoneal dialysis. The dialysis units were all chosen from metropolitan areas with five or more Medicare-approved dialysis providers in 1984, excluding the two largest areas. Two of these 46 met¬ ropolitan areas were chosen purposive- ly for another project, and all 15 dialysis units in these two areas were included in the study. Then a matching number of patients were chosen through a ran¬ domized process in the other 46 metro¬ politan areas (including the two largest areas) so that every patient in the other areas had an equal probability of being selected, resulting in participation by patients from 21 other dialysis units. The analysis controlled for purposive vs random inclusion. The response rate by dialysis units was in excess of 70% in both the purposive and randomly se¬ lected sites. Fifty-two percent of the final sample came from the two purpo¬ sive sites and 48% from the other ran¬ domly selected ones. The 48 metropoli¬ tan areas contained 64% of all Medicare ESRD patients in 1984. Over three quarters of the randomly sampled hemodialysis patients were ac¬ tually included in the analyses. A full description of the sampling technique is available.7 These patients accurately represented the national ESRD popula¬ tion. Patient age, sex, and diagnosis data closely matched census data from the Medicare data files8; moreover, mor¬ tality risks for this population by age and diagnosis were comparable with the norms in other national samples.9 During late 1984 and early 1985, data were obtained on characteristics of the sampled patients, their dialysis units, and their dialysis treatments. We inter¬ viewed patients and medical directors, and abstracted data from the immedi¬ ately preceding 6 months of medical records (random selection of up to 18