OBJECTIVES: To assess the clinical and cost-effectiveness of magnesium sulphate compared with sotalol, and to assess the clinical effectiveness of magnesium sulphate compared with placebo in the prevention of atrial fibrillation (AF) in patients who have had a coronary artery bypass graft (CABG). DATA SOURCES: Major electronic databases were searched from December 2003 to May 2007. REVIEW METHODS: Selected studies were assessed, subjected to data extraction using a standard template and quality assessment using published criteria. A simple short-term economic model was developed, informed by a systematic review of economic evaluations and populated with data from a review of costing/resource-use studies and other published studies. The cost-effectiveness of magnesium sulphate as prophylaxis was estimated for a set of base-case assumptions and the robustness of these results was assessed using deterministic and probabilistic sensitivity analysis. RESULTS: Twenty-two papers met the inclusion criteria reporting 15 trials which all compared magnesium sulphate with placebo or control. They ranged in size from 15 to 176 patients randomised, and were conducted in Europe, the USA and Canada. The standard of reporting was generally poor, with details of key methodological attributes difficult to elucidate. No trials were identified that specifically aimed to compare magnesium sulphate with sotalol. Of 1070 patients in the pooled magnesium group, 230 (21%) developed postoperative AF, compared with 307 of 1031 (30%) patients in the placebo or (control) group. Meta-analysis using a fixed-effects model generated a pooled odds ratio (OR) that was significantly less than 1.0 [OR=0.65, 95% confidence interval (CI) 0.53 to 0.79, test for overall effect p<0.0001], but with statistically significant heterogeneity (I2=63.4%, p=0.0005). Two randomised controlled trials (RCTs) were notable as they had relatively lower ORs in favour of magnesium sulphate. When these were removed from the analyses the pooled OR remained statistically significant, but heterogeneity no longer remained significant. These two studies tended to impart a highly significant reduction in the odds of AF to whichever subgroup they were analysed in. When studies were ordered by total duration of prophylaxis, an apparent relationship between duration and odds of AF was evident, with decreasing odds of AF as duration of prophylaxis increased. This was confirmed by linear regression analysis (R2=0.743, p<0.001). When the data were grouped into three classes according to duration, a statistically significant intervention effect was only present for the longest duration (OR=0.12, 95% CI 0.06 to 0.23, p=0.00001). Statistically significant intervention effects were associated with the initiation of prophylaxis 12 hours or more before surgery (OR 0.26; 95% CI 0.16 to 0.44, test for overall effect p=0.00001, fixed-effects model) and less than 12 hours before surgery or during the surgery itself (OR=0.73, 95% CI 0.56 to 0.97, test for overall effect p = 0.03, fixed-effects model), but not when prophylaxis was initiated at the end of surgery or postsurgery (OR=0.85, 95% CI 0.59 to 1.22, p=0.37, fixed-effects model). When studies were ordered by total dose of intravenous magnesium sulphate (<25 g), the odds of AF were independent of the dose. A notable exception was that for a total dose of 9 g magnesium sulphate; here the odds of AF were significantly reduced relative to the control group, although this may be explained by the fact that these studies had excluded patients who were on antiarrhythmic drugs and so may have been at higher risk of AF. Sixty-three potentially relevant references about cost-effectiveness were identified, but no economic evaluations of intravenous magnesium alone as prophylaxis against AF following CABG, compared with sotalol as prophylaxis or no prophylaxis, were identified. Studies reporting resource use by patients with AF following CABG suggest that while AF significantly increased inpatient stays, by up to 2.3 days in the intensive care unit (ICU) and 3.4 days on the ward, differences in length of stay and costs between patients receiving prophylaxis and those not receiving prophylaxis were not statistically significant. In the base-case analysis, magnesium sulphate prophylaxis resulted in 0.081 fewer cases of AF at an incremental cost of 2.55 pounds sterling. The incremental cost-effectiveness ratio (ICER) was 32 pounds sterling per AF case avoided. The estimated difference in average length of stay between the prophylaxis and no-prophylaxis strategies was only 0.24 days, despite a large assumed difference of 3 days for patients experiencing AF in each group (1 extra day in the ICU and 2 extra days on the ward). In a deterministic sensitivity analysis the greatest variation in ICERs was observed for input parameters relating to the baseline risk of AF following CABG and the effectiveness of prophylaxis, cost of prophylaxis and the resource consequences of postoperative AF. The largest ICER (2092 pounds sterling) in the sensitivity analysis was associated with increasing the length of patients' preoperative stay. In the base case it was assumed that admission routines would be identical under both strategies. However, patients receiving prophylaxis by intravenous infusion may have longer preoperative stays. In a probabilistic analysis the majority of the simulations were associated with improved outcomes (in this case fewer cases of AF), but also higher costs. Prophylaxis was the dominant strategy (better outcome at lower cost) in about 41% of the simulations using the base-case assumptions. Under an alternative scenario where patients receiving prophylaxis are admitted for longer before their operation, to receive their initial infusion, the proportion of simulations where prophylaxis dominates fell to around 5%. The probability of being cost-effective was 99% at a willingness to pay (WTP) threshold of 2000 pounds sterling per AF case avoided and 100% at a WTP threshold of 5000 pounds sterling per AF case avoided under the base-case assumptions. Under the alternative scenario of longer preoperative stays the probability of being cost-effective at these two threshold values fell to 48% and 93%, respectively. It is unclear what the appropriate decision threshold should be, given that this model used intermediate rather than final outcomes. CONCLUSIONS: No RCTs were identified that specifically aimed to compare intravenous magnesium with sotalol as prophylaxis for AF in patients undergoing CABG. Intravenous magnesium, compared with placebo or control, is effective in preventing postoperative AF, as confirmed by a statistically significant intervention effect based on pooled analysis of 15 RCTs. It was also found that AF was less likely to occur when a longer duration of prophylaxis was used, and the earlier that prophylaxis is started; however, this finding was associated with two RCTs that had more favourable results than the other trials. No clear relationship between dose and AF was observed, although a lower constant dose rate was associated with the lowest odds of AF. Further research should investigate the relationship between dose, dose rate, duration of prophylaxis, timing of initiation of therapy and patient characteristics, such as degree of risk for AF. This will provide stronger evidence for the optimum delivery of intravenous magnesium in patients undergoing CABG. In the base-case analysis in the economic model, magnesium sulphate prophylaxis reduced the number of postoperative AF cases at a modest increase in cost. The results of the economic analysis are highly sensitive to variation in certain key parameters. Prophylaxis is less likely to be a cost-effective option if it requires changes in admission routines that result in longer preoperative stays than would be the case without prophylaxis.