The large effort by the Intercontinental Antimicrobial Study Group to compare single-agent and combination therapy for febrile neutropenia [1] appears long on data and short on analysis. Consequently, this otherwise impressive report leaves unanswered several important issues that should be extracted from the raw data.

First, how is the reported odds ratio (1.07) for a satisfactory response (ceftazidime compared with tobramycin plus piperacillin) affected by important patient variables such as age, underlying disease, and granulocyte count? Although the authors list and compare such items in a table, the odds ratio appears to be unadjusted for any of these potentially important outcome variables. (This impression is easily verified: If P₁ = 0.627 [the probability of success given ceftazidime] and P₂ = 0.611 (the probability given combination therapy), the odds ratio = (P_1/1-P₁)/[P_2/1-P₂] = 1.07.) Simply comparing potential confounders using a table of demographics (or even using direct hypothesis testing) does not absolve such variables, either alone or in combination, from significant effects on the odds ratio [2]. Modeling the data by logistic regression (binary or polychotomous), however, would give the clinician a more useful assessment of the odds ratio by adjusting the antimicrobial effects for important patient-related factors.

Second, aside from "satisfactory response," did the study have sufficient power to draw conclusions regarding any of the other nonsignificant results?

Third, odds ratios with confidence intervals can convey much information. Unfortunately, the authors' Figure 4 is difficult to interpret because the abscissa is scaled differently with respect to therapy (1 to for ceftazidime, 0 to 1 for combination therapy). Plotting the data against the log odds ratio, however, would make this Figure visually interpretable.