Lipid Formulations of Amphotericin B and Solid Organ Transplant Recipients with Central Nervous System Cryptococcosis

To the Editor—Sun and colleagues [1] report that lipid formulations of amphotericin B are independently associated with less mortality, compared with the mortality associated with amphotericin B deoxycholate, in solid organ transplant recipients with central nervous system cryptococcosis. We have reservations about the validity of this conclusion. First, the article does not demonstrate how a number of baseline characteristics are distributed between the treatment groups. Could Sun and colleagues [1] provide these baseline data? Second, the data concern patients who were treated in various centers in various countries. Treatment center may be an important confounder, because it may influence both the doctor’s choice for a certain treatment and the clinical outcome. This variable should be reported and corrected for in the final analysis. Third, the decision of whether a variable is a confounder that should be adjusted for by means of a multivariate model is based on statistical criteria ( ). AlP ! .2 though it is common practice to correct for confounding in this way, this strategy may lead to bias from omission of important confounders or from inappropriate adjustment for nonconfounders [2]. For example, treatment with flucytosine, which is regarded as an important component of the treatment of central nervous system cryptococcosis, was not evenly distributed between the 2 groups: 67% of patients who received lipid formulations of amphotericin B were treated with flucytosine, compared with 40% of patients who received amphotericin B deoxycholate. Although treatment with flucytosine is not associated with the primary end point (ie, mortality) in a statistically significant way in this study, it is nonetheless a confounder that should be corrected for in the final analysis. The same argument holds for variables such as age, year of diagnosis, and treatment center. Many small confounders put together may cause a considerable degree of confounding. In fact, a substantial degree of residual confounding may also explain the lack of efficacy of flucytosine in this study. In conclusion, the reported difference in mortality between patients treated with lipid formulations of amphotericin B and those treated with amphotericin B deoxycholate may well be due to residual confounding.