Safety, Censoring, and Intent‐to‐Treat Analysis: Dangers to Generalizability

To the Editor—Although the efforts of Makadzange et al [1] are commendable, several safety, methodologic, and reporting issues deserve clarification. The first one is safety. Fluconazole increases the area under the curve of nevirapine in a dose-dependent manner: it increases by 29% with 200 mg of fluconazole given 3 times weekly, by 50% with 200 mg of fluconazole given daily, and by 90%–100% with 400 mg of fluconazole given daily [2–4]. The effects of 800 mg of fluconazole per day on nevirapine pharmacokinetics are unknown. Because only the subjects who received immediate antiretroviral therapy (ART) were exposed to 800 mg of fluconazole per day and nevirapine for 10 weeks, the lack of safety monitoring data does not reassure the reader that increased hepatotoxicity was not contributory to excess mortality. The authors reported, “Liver function test results remained normal throughout the study…” [1, p 1536]. Although this may technically be true, baseline serum alanine aminotransferase levels were measured for only 44 subjects, and monitoring was reported as occurring every 6 months. When nearly one-half the deaths (17 of 35) occurred before first follow-up, 15 of 35 deaths occurred at home, and 33 of 35 deaths occurred prior to 6 months, an assumption of no hepatotoxicity is incorrect in the absence of actual data. Second, the way the censoring was performed is unclear. The reported mortality rate with immediate ART was 88% of 28 persons; however, there exists no integer that can be divided by 28 to yield 88%. On the basis of the text, the reported mortality in the immediate ART arm was 82.1% (23 of 28 participants), compared with 46.2% (12 of 26 participants) in the delayed ART arm. Of more concern is the fact that the outcomes displayed in [1, Figure 2] match neither the mortality rate (n = 12) and participants lost to followup (n = 3) for the delayed arm nor the mortality rate (n = 23) and participants lost to follow-up (n = 3) for the immediate arm as reported in the text and in [1, Figure 1]. In counting the “number at risk” in [1, Figure 2], the number of participants dead or lost to follow-up by 200 days was 15 for the delayed ART arm and 24 for the immediate ART arm, with 1 additional death in each arm after 200 days. Thus, nonsurvival rates are 61.5% (16 of 26 participants) for delayed ART, compared with 89.3% (25 of 28 participants) for immediate ART (P = .026).On the basis of reported mortality (ie, 12 and 23 deaths), the Kaplan-Meier figure thereby implies that 4 participants were censored in the delayed arm and 2 participants in the immediate arm. This censoring would yield mortality rates of 55% (12 of 22) and 88% (23 of 26) (similar to the rates reported by the authors). The implication is that participant numbers or outcomes in either [1, Figure 1 and Figure 2], and/or the text are incorrect, as well as that intent-to-treat analysis was not consistently used, although stated as such. One fundamental question for any clinical trial is appropriate outcome measures. We take the position that survival is a better outcome measure than known mortality. Cryptococcal meningitis has a near 100% mortality rate within 6 months in the absence of ART [5, 6]. In a time-to-event analysis, right-hand censoring is generally considered acceptable to evaluate participants who are lost to follow-up, yet we question the appropriateness of doing so in all scenarios. This is particularly germane to 3 (or possibly 4) participants in the delayed arm who were lost to followup before 2 weeks and who did not initiate ART. Considering them a success by using time-to-event analysis before censoring at ⩽2 weeks creates a subtle systematic bias. The overall concern is a type I error due to relatively small sample size, underpowered nature of the trial, inadequacy of randomization (eg, unequal CD4+ cell counts between the 2 groups [median, 27 vs 52 cells/µL), and post hoc ClinicalTrials.gov registration, which is further complicated by early stoppage. Unexplained shifts in 1 or 2 participants could become important. Nevertheless, the overall conclusion is likely correct that after cryptococcal meningitis, when treatment includes a regimen of nevirapine with 800 mg of fluconazole, immediate ART results in highermortality rates. The authors' conclusion that “early initiation of ART most likely resulted in increased rates of cryptococcal IRIS [immune reconstitution inflammatory syndrome]” [1, p 1536] is unsubstantiated speculation, because no attempt was made to assess patients for cryptococcal IRIS. The final question is whether this trial's result is generalizable, particularly if efavirenz were used instead of nevirapine, the nevirapine dosage were adjusted to 200 mg daily while 800 mg of fluconazole was given, amphotericin were used, or the degree of immunosuppression between the 2 groups were fully equal. Additional trials are needed. Financial support. National Institutes of Health National Institute of Allergy and Infectious Diseases (K23AI073192-01A2). Potential conflicts of interest. All authors: no conflicts.