Decoding candidemia in critically ill patients: unsupervised clustering identifies three unique phenotypes
Abstract
Background: Candidemia displays significant clinical heterogeneity in critically ill patients. This study aimed to identify distinct clinical phenotypes and to assess their association with 90-day mortality. Methods: We conducted a multicenter retrospective cohort study of 492 intensive care unit (ICU) patients with candidemia from 16 French ICUs (2015-2023). We performed a factor analysis of mixed data (FAMD) incorporating both categorical and continuous baseline variables, followed by hierarchical clustering on principal components (HCPC). Survival analysis was performed with Kaplan-Meier curves and Cox proportional hazards models. Results: Overall, 90-day mortality for the 492 patients (median age: 64 years, 69.1% male) with candidemia was 62.6%. Three different phenotypes emerged from FAMD followed by HCPC: Phenotype 1 (n = 70, 14.2%) comprised patients with severe immunosuppression, mostly due to hematological malignancies (82.9%), and high severity scores (SAPS II:70); Phenotype 2 (n = 223, 45.3%) corresponded to elderly cirrhotic patients (19.3%) with early-onset digestive candidemia; Phenotype 3 (n = 199, 40.5%) comprised younger patients with lower severity scores and catheter-related candidemia. Mortality differed significantly between phenotypes: 72.9% (Phenotype 1), 70.4% (Phenotype 2), and 50.3% (Phenotype 3) (p < 0.001). Independent predictors of mortality included age (aHR: 1.01, 95% CI: 1.00-1.02; p = 0.003), cirrhosis (aHR: 1.90, 95% CI: 1.39-2.60; p < 0.001), SAPS II (aHR: 1.01, 95% CI: 1.01-1.02; p < 0.001), echinocandin use (aHR = 0.49, 95% CI: 0.39-0.63; p < 0.001) and proven catheter-related candidemia (protective; aHR: 0.48, 95% CI: 0.33-0.69; p < 0.001). Immunodepression was not associated with mortality. Conclusion: Unsupervised clustering identified three clinically different candidemia phenotypes with different outcomes. Cirrhosis, higher age and illness severity were associated with mortality, whereas a catheter-related source of infection was protective.
Related
- Unsupervised clustering identifies candidate metabolic profiles, including an acidotic profile detectable before calving, in periparturient dairy cows
- Concurrent Supervised-Unsupervised Representative Subspace Clustering (CSU-RSC)
- Unsupervised Clustering Enables Cohort-Specific Variant Recalibration in Crops
- Cumulative disease-course phenotypes in Crohn's disease: a retrospective single center FAMD clustering study
- Clinical Heterogeneity in Myotonic Dystrophy Type 1: Identifying Phenotype Categories Using Unsupervised Clustering.
BibTeX
@article{reizine2026decoding,
title = {{Decoding candidemia in critically ill patients: unsupervised clustering identifies three unique phenotypes}},
author = {Florian Reizine and Juliette Henry and Luc Desmedt and Christophe Camus and Valentin Coirier and François Arrivé and Estelle Burban and Gabriel Eustache and Tony Belleville and Antoine Marc and Jolan Malherbe and Pierre Bouju and Paul Jaubert and Olivier Lesieur and Maxime Leclerc and Pierre Fillâtre and Aurélien Frérou and Renaud Prével and Nolwenn Mainguy and Anne Cady and Florent Morio and Solène Le Gal and Estelle Perraud and Laurence Delhaes and Sébastien Imbert and Hélène Guegan and Julie Bonhomme and Marc Pihet and Clarisse Dupin and Agathe Delbove and Damien Du Cheyron and Yoann Launey and Cécile Aubron and Jean-Pierre Gangneux},
journal = {Critical Care},
year = {2026},
month = oct,
doi = {10.1186/s13054-026-05964-4},
url = {https://hal.science/hal-05675809v1},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Decoding candidemia in critically ill patients: unsupervised clustering identifies three unique phenotypes
AU - Florian Reizine
AU - Juliette Henry
AU - Luc Desmedt
AU - Christophe Camus
AU - Valentin Coirier
AU - François Arrivé
AU - Estelle Burban
AU - Gabriel Eustache
AU - Tony Belleville
AU - Antoine Marc
AU - Jolan Malherbe
AU - Pierre Bouju
AU - Paul Jaubert
AU - Olivier Lesieur
AU - Maxime Leclerc
AU - Pierre Fillâtre
AU - Aurélien Frérou
AU - Renaud Prével
AU - Nolwenn Mainguy
AU - Anne Cady
AU - Florent Morio
AU - Solène Le Gal
AU - Estelle Perraud
AU - Laurence Delhaes
AU - Sébastien Imbert
AU - Hélène Guegan
AU - Julie Bonhomme
AU - Marc Pihet
AU - Clarisse Dupin
AU - Agathe Delbove
AU - Damien Du Cheyron
AU - Yoann Launey
AU - Cécile Aubron
AU - Jean-Pierre Gangneux
PY - 2026
DA - 2026/10/14
JO - Critical Care
DO - 10.1186/s13054-026-05964-4
UR - https://hal.science/hal-05675809v1
AB - Background: Candidemia displays significant clinical heterogeneity in critically ill patients. This study aimed to identify distinct clinical phenotypes and to assess their association with 90-day mortality. Methods: We conducted a multicenter retrospective cohort study of 492 intensive care unit (ICU) patients with candidemia from 16 French ICUs (2015-2023). We performed a factor analysis of mixed data (FAMD) incorporating both categorical and continuous baseline variables, followed by hierarchical clustering on principal components (HCPC). Survival analysis was performed with Kaplan-Meier curves and Cox proportional hazards models. Results: Overall, 90-day mortality for the 492 patients (median age: 64 years, 69.1% male) with candidemia was 62.6%. Three different phenotypes emerged from FAMD followed by HCPC: Phenotype 1 (n = 70, 14.2%) comprised patients with severe immunosuppression, mostly due to hematological malignancies (82.9%), and high severity scores (SAPS II:70); Phenotype 2 (n = 223, 45.3%) corresponded to elderly cirrhotic patients (19.3%) with early-onset digestive candidemia; Phenotype 3 (n = 199, 40.5%) comprised younger patients with lower severity scores and catheter-related candidemia. Mortality differed significantly between phenotypes: 72.9% (Phenotype 1), 70.4% (Phenotype 2), and 50.3% (Phenotype 3) (p < 0.001). Independent predictors of mortality included age (aHR: 1.01, 95% CI: 1.00-1.02; p = 0.003), cirrhosis (aHR: 1.90, 95% CI: 1.39-2.60; p < 0.001), SAPS II (aHR: 1.01, 95% CI: 1.01-1.02; p < 0.001), echinocandin use (aHR = 0.49, 95% CI: 0.39-0.63; p < 0.001) and proven catheter-related candidemia (protective; aHR: 0.48, 95% CI: 0.33-0.69; p < 0.001). Immunodepression was not associated with mortality. Conclusion: Unsupervised clustering identified three clinically different candidemia phenotypes with different outcomes. Cirrhosis, higher age and illness severity were associated with mortality, whereas a catheter-related source of infection was protective.
SN - 1364-8535
SN - 1466-609X
ER -