BACKGROUND: When the kidney fails the blood borne metabolites of protein breakdown and water cannot be excreted. The principle of haemodialysis is that such substances can be removed when blood is passed over a semipermeable membrane. Natural membrane materials can be used including cellulose or modified cellulose, more recently various synthetic membranes have been developed. Synthetic membranes are regarded as being more "biocompatible" in that they incite less of an immune response than cellulose-based membranes. OBJECTIVES: To assess the effects of different haemodialysis membrane material in patients with end-stage renal disease (ESRD). SEARCH STRATEGY: We searched Medline (1966 to December 2000), Embase (1981 to November 2000), PreMedline (29 November 2000), HealthStar (1975 to December 2000), Cinahl (1982 to October 2000), The Cochrane Controlled Trials Register (Issue 1, 1996), Biosis (1989 to June 1995), Sigle (1980 to June 1996), Crib (10th edition, 1995), UK National Research Register (September 1996), and reference lists of relevant articles. We contacted biomedical companies, investigators and we hand searched Kidney International (1980 to 1997). Date of the most recent searches: November 2000. SELECTION CRITERIA: All randomised or quasi-randomised clinical trials comparing different haemodialysis membrane material in patients with ESRD. DATA COLLECTION AND ANALYSIS: Two reviewers independently assessed the methodological quality of studies. Data was abstracted from included studies onto a standard form by one reviewer and checked by another. MAIN RESULTS: Twenty seven studies met our inclusion criteria and where possible data from these were summated by meta-analyses (Peto's odds ratio (OR) and weighted mean difference (WMD) with 95% confidence intervals (CI)). Twenty two outcome measures were sought in 10 broad areas. For two (number of episodes of significant infection per year and quality of life) no data were available. For the comparison of cellulose with synthetic membranes, data for 12/20 outcome measures were available in only a single trial. For modified cellulose and synthetic membranes, data for three outcome measures were available in one trial only and for 12 of the outcomes no data were found, crossover studies were analysed separately and studies which randomised by patient yet analysed by dialysis sessions adjusted for clustering. Pre-dialysis beta2 microglobulin concentrations were significantly lower at the end of the studies in patients treated with synthetic membranes (WMD - 14.5; 95% CI -17.4 to -11.6). One crossover study showed a lowering of beta2 microglobulin when low flux synthetic membranes were used. When analysed for a change in beta2 microglobulin across a trial a fall was only noted when high flux membranes were used. In one very small study the incidence of amyloid was less in patients who were dialysed for six years with high flux synthetic membranes (OR 0.05; 95% CI 0.01 to 0.18). In the single study which measured triglyceride values there was a significant difference in favour of the synthetic (high flux) membrane (WMD -0.66; 95% CI -1.18 to -0.14). Serum albumin was higher in patients treated with synthetic membranes (both low and high flux) although this just bordered statistical significance (WMD -0.09; 95% CI -0.18 to 0.00). Dialysis adequacy measured by Kt/V was marginally higher when cellulose membranes were used (WMD 0.10; 95% CI 0.04 to 0.16). There was no significant difference between these membranes for any of the other clinical outcomes measures but confidence intervals were generally wide. No differences were found between modified cellulose and synthetic membranes although many fewer trials were carried out for this comparison. REVIEWER'S CONCLUSIONS: For clinical practice This systematic literature review has generated no evidence of benefit when synthetic membranes were used compared with cellulose/modified cellulose membranes in terms of reduced mortality nor reduction in dialysis related adverse symptoms. Despite the relatively large number of RCTs undertaken in this area none of the included studies reported any measures of quality of life. End-of-study beta2 microglobulin values, and possibly the development of amyloid disease, were less in patients treated with synthetic membranes compared with cellulose membranes. Plasma triglyceride values were also lower with synthetic membranes in the single study that measured this outcome. Differences in these outcomes may have reflected the high flux of the synthetic membrane. Serum albumin was higher when synthetic membranes of both high and low flux were used. Kt/V and urea reduction ratio were higher when cellulose or modified cellulose membranes were used in the few studies that measured these outcomes. We are hesitant to recommend the universal use of synthetic membranes for haemodialysis in patients with ESRD because of; the small number of trials (particularly for modified cellulose membranes, most with low patient numbers), the heterogeneity of many of the trials compared, the variations in membrane flux, the differences in exclusion criteria, particularly relating to comorbidity and the relative lack of patient-centred outcomes studied. Such evidence as we have favours synthetic membranes but even if we assume extra benefit it may be at considerable cost, particularly if high flux synthetic membranes were to be used. For further research A further systematic review of RCTs comparing high and low flux haemodialysis membranes, subgrouped according to membrane composition (cellulose, modified cellulose, synthetic) and reporting clinical outcomes of major importance to patients needs to be undertaken. Further pragmatic RCTs are required to compare the different dialysis membranes available. We recommend that they: - Take into account other properties including flux as well as the material from which the membrane is made and test modified cellulose membranes as well as standard ones. - Record an agreed minimum dataset on primary outcomes of major importance to patients. - Explicitly record whether symptoms are patient- or staff-reported recognising that generally patient reporting will be more appropriate for evaluating effectiveness but staff reported data may be necessary for calculating the cost of treating complications. - Be multi-centre (and possibly multinational) to have sufficient patients to complete the study to allow for a considerable number of withdrawals and dropouts. - Have sufficient length of follow up to draw conclusions for important clinical outcome measures and continue to follow patients who have renal transplants. - Include older patients and those with comorbid illnesses and take into account age and comorbidity when assessing outcomes (possibly by stratification at trial entry). - Carry out, in parallel, an economic evaluation of the different policies being compared in the trial.