LAUSR

We have eagerly read a recent e-publication from your journal by Frea et al. [1]. This paper has set out to assess the efficacy of bolus and continuous infusion of furosemide—a commonly used loop diuretic—for a variety of renal and fluid outcomes, finding that continuous infusion led to more patients being free from congestion (48% vs 25% respectively) compared to bolus administration, and lower treatment failure (15% vs 38%). We have two comments on the paper. First, we note there are four groups in the trial with two stratifying conditions: continuous vs bolus dosage of furosemide and high vs low dose of furosemide. Within the bolus group, 25/40 of the patients were on high-dose furosemide, whilst 32/40 of the patients within the continuous infusion group were on high-dose furosemide. The potential for confounding—due to the differing treatment conditions within the two groups—was briefly discussed in the limitations section and was justified in the methods on the basis of “severe systemic congestion (wet score ≥ 14/18) and/or on daily outof-hospital oral furosemide equivalent dose ≥ 125 mg and/or on supplemental metolazone therapy on a long-term basis”. However, we believe the risk of confounding has not been addressed in the statistical analysis, and that high-/low-dose furosemide should have been introduced as a confounder. Second, the statistical methods contain the following statement on the power calculation: “A sample size of 78 patients was required to have a 70% chance of detecting, as significant at the 10% level, a decrease in the primary outcome measure from 48% in the bolus group to 25% in the infusion group.” This exactly matched the reported effects in the paper, except with the categories reversed: “The primary endpoint of freedom from congestion after 72 h occurred in 10 patients (25%) in the bolus arm and in 19 (48%) in the infusion arm (OR 2.71, 95% CI 1.05–7.00; p = 0.04, Fig. 2)”. We also note the chosen atypical thresholds: 70% power at a 10% significance level. It appears to us that the researchers have applied a post hoc power calculation and have not declared this. Post hoc power calculations have previously been discussed to be a problematic statistical procedure, as they are merely a transformation of the reported p value and do not offer any insight into the true power of the study [2]. If this is not the case, then we are curious as to why there is an exact match for the effect sizes between the trial results and their prior power calculation, save for the category reversal.