The study drew on the CandidICU cohort, a retrospective collection of all adults hospitalized between January 1, 2015 and January 1, 2023 in participating intensive care units who had at least one positive blood culture for Candida within seven days before admission or during the stay. Of 492 patients initially enrolled, the investigators narrowed the analytic population to 237 individuals. The exclusions were deliberate and methodologically consequential: 122 patients who died within the first five days were removed because they could never have been eligible for either treatment strategy in the emulated trial, 76 patients treated for more than 21 days were excluded because such prolonged courses usually signal complicated infections such as endocarditis or deep-seated dissemination, 41 patients treated for fewer than five days were dropped because such brief exposure likely reflected discontinuation for unrelated reasons, and 16 patients lacked complete data. What remained was a focused population of patients with uncomplicated candidemia who survived long enough to plausibly receive either a short or a long course.

The population was severely ill. Median age was 62 years, roughly two-thirds were men, and the median SAPS II severity score was 53, a figure that places these patients among the sickest in intensive care. Candida albicans caused about two-thirds of infections, with Candida glabrata the most common non-albicans species, and echinocandins served as initial therapy in more than 84 percent of cases. Notably, patients excluded for early death were markedly sicker than those included, with higher SAPS II scores and greater rates of cirrhosis, renal replacement therapy, and acute respiratory distress syndrome—a comparison that supported the appropriateness of the five-day exclusion while clarifying that the findings apply only to patients who survive the initial phase of infection.

The central methodological challenge in any observational study of treatment duration is that duration is not known at the moment treatment begins. Patients who receive short courses may differ systematically from those treated longer, and clinicians often stop therapy early precisely because a patient is deteriorating, or conversely because the patient is improving rapidly. Worse, so-called immortal time bias arises because a patient must survive long enough to accumulate days of therapy. To confront these distortions, the researchers turned to target trial emulation, a framework in which the analysis is explicitly designed to mimic a randomized trial. The key instrument was a technique called cloning, censoring, and weighting: every patient was duplicated into two virtual clones, one assigned to a short-course strategy and one to a long-course strategy, and each clone was censored at the moment its observed treatment deviated from the assigned strategy, using a 10-day grace period for classification.

On top of the cloning structure, the team applied inverse probability of treatment weighting based on a propensity score model that incorporated 22 baseline covariates, including age, sex, comorbidities such as cirrhosis and diabetes, immunosuppression, neutropenia, mechanical ventilation, the Candida species involved, echinocandin use, and suspected source of infection. The model achieved a C-statistic of 0.704, and weighting brought covariate balance within acceptable limits, reducing the standardized mean difference for Candida albicans status from 0.417 to 0.108. Sensitivity analyses varied the grace period to 7 and 14 days, adjusted for treatment center, and applied inverse probability of censoring weights to address potential informative censoring, all yielding results consistent with the primary analysis.

The raw numbers told a dramatically different story from the adjusted ones. Among the 237 patients, 48 received short-course therapy and 189 received long courses, and crude 90-day mortality was 79.2 percent in the short-course group versus 39.7 percent in the long-course group, a difference so stark it would seem to condemn abbreviated therapy outright. But after cloning and weighting, that apparent chasm collapsed: the weighted hazard ratio for 90-day mortality was 1.14, with a 95 percent confidence interval of 0.82 to 1.58 and a p-value of 0.449—statistically indistinguishable from no effect. The authors argue that the crude difference was almost certainly driven by confounding by indication and selection effects rather than by any causal harm of shorter treatment, and they present the attenuation itself as a cautionary lesson about the magnitude of bias in conventional observational analyses of antimicrobial duration.

Secondary outcomes offered additional nuance. In weighted analyses, patients who received shorter therapy had significantly shorter hospital stays, a median of 31 days versus 46 days, and shorter intensive care unit stays, though the authors urge caution because early death shortens length of stay through competing risk. No meaningful differences emerged in ventilator duration, renal replacement therapy, vasopressor use, ventilator-associated pneumonia, or recurrent fungemia, and no cases of second fungemia occurred in the short-course group. Restricted mean survival time analyses hinted at a possible survival advantage for longer therapy at 30 days, with a p-value of 0.094, but the difference was not significant and faded at later time points.

The authors are emphatic that these findings should not change clinical practice. The confidence interval is wide, compatible with as much as a 58 percent relative increase in mortality from short courses, and the study was never designed as a formal non-inferiority trial. Residual confounding from unmeasured factors—time to blood culture clearance, microbiologic response, adequacy of source control, and the clinical reasons behind a clinician’s decision to stop therapy—cannot be excluded, and the modest residual imbalances in solid tumor status and SAPS II after weighting leave open the possibility of residual bias. With only 48 patients in the short-course group, the study had 69 percent power to detect a 25 percent absolute mortality difference, and the extreme censoring inherent to the cloning design limits the reliability of survival estimates in that arm beyond the grace period.

Nevertheless, the study fills a genuine evidence gap. Two randomized trials comparing 7 versus 14 days of therapy are registered but have not begun enrolling patients, and both exclude the sickest individuals—those in shock or requiring mechanical ventilation—precisely the population this French analysis addresses. From a stewardship perspective, the results are biologically plausible: prolonged antifungal exposure carries risks of hepatotoxicity, drug-drug interactions, and selection of resistant isolates, including echinocandin-resistant strains emerging after relatively brief exposure. The authors suggest that distinguishing uncomplicated from complicated candidemia may help clinicians and antimicrobial stewardship programs weigh abbreviated courses in selected patients who survive the early phase and show clear clinical improvement, while careful screening for ocular, cardiac, and deep-seated complications remains essential before any decision to shorten therapy. Until randomized trials specifically designed for intensive care populations deliver definitive answers, this target trial emulation offers the most rigorous glimpse yet into a question that has governed antifungal prescribing for decades without ever having been properly tested.

Subject of Research: Optimal duration of antifungal therapy for critically ill adults with candidemia in the intensive care unit

Article Title: Association of antifungal treatment duration with 90-day mortality in critically ill adults with candidemia: a multicenter target trial emulation in France

Article References: Reizine, F., Henry, J., Desmedt, L., Camus, C., Tadié, J. M., Coirier, V., Arrive, F., Burban, E., Eustache, G., Belleville, T., Marc, A., Malherbe, J., Bouju, P., Jaubert, P., Lesieur, O., Leclerc, M., Fillâtre, P., Frérou, A., Prével, R., … Gangneux, J.-P. (2026). Association of antifungal treatment duration with 90-day mortality in critically ill adults with candidemia: a multicenter target trial emulation in France. eClinicalMedicine, 99, Article 104178. https://doi.org/10.1016/j.eclinm.2026.104178

Image Credits: AI Generated

DOI: 10.1016/j.eclinm.2026.104178

Keywords: candidemia, antifungal therapy duration, intensive care unit, target trial emulation, cloning censoring weighting, 90-day mortality, Candida albicans, echinocandins, antimicrobial stewardship, immortal time bias, propensity score, eClinicalMedicine