Response to "Fragility of Observed Associations in Selection of Empirical Antifungal Therapy for Critically Ill Non-Neutropenic Patients” [Response to Letter]

We thank Manzoor et al for their thoughtful evaluation of the statistical fragility of selected findings from our multicenter observational study.1,2 We agree that statistically significant unadjusted comparisons should not be interpreted as evidence that one empirical antifungal agent is causally superior to another, particularly in a retrospective observational study with non-random treatment allocation, unequal treatment-group sizes, and potential confounding by indication.The Letter correctly notes that our original multivariable analysis focused on in-hospital mortality, whereas the associations between antifungal selection and clinical improvement, hepatotoxicity, and nephrotoxicity were reported as unadjusted comparisons.We therefore conducted a post-publication reanalysis of the original patient-level dataset using R.Because several treatment groups and outcome cells were relatively small, we used Firth bias-reduced logistic regression.Covariates were selected parsimoniously based on clinical relevance, baseline organ function, illness severity, and the available number of outcome events.The original article defined clinical improvement as a composite clinical outcome and defined nephrotoxicity using creatinine change or urine-output criteria.Treatment allocation was clearly non-random.Antifungal groups differed significantly by participating hospital, mechanical ventilation status, and Candida score, with all three comparisons yielding p<0.001.Mechanical ventilation was present in 78.9% of fluconazole-treated patients, compared with 93.6% receiving caspofungin and 98.6% receiving anidulafungin.These imbalances support the likelihood of site-specific prescribing practices and confounding by indication.In the mortality model (324 patients; 255 events), after adjustment for hospital, age, Charlson comorbidity index, mechanical ventilation, Candida colonization, and sepsis, antifungal selection was not independently associated with inhospital mortality overall (penalized likelihood-ratio chi-square=0.43,3 degrees of freedom, p=0.935).Compared with fluconazole, the adjusted odds ratios were 1.40 for caspofungin (95% CI 0.49-3.87),1.37 for anidulafungin (95% CI 0.45-3.96),and 1.33 for liposomal amphotericin B (95% CI 0.26-8.19).During this reanalysis, we also identified that several estimates in the published mortality table had been labelled as adjusted odds ratios even though their confidence intervals and p-values were internally incompatible, suggesting that logit-scale coefficients and confidence limits may have been mixed with exponentiated odds ratios.When the coefficients and confidence limits were consistently exponentiated, the corrected adjusted odds ratios were 0.40 for chronic pulmonary disease (95% CI 0.15-1.10;p=0.075), 1.99 for Candida colonization (95% CI 1.03-3.82;p=0.040),Infection and Drug Resistance 2026:19 639697 1