Intensive Care & Medical ICU Services: Rennes University Hospital (France) & Poitiers

A new study published in Critical Care identifies three distinct patient profiles in critically ill adults with Candida bloodstream infections, offering potential for more targeted treatment approaches in intensive care units.

According to research led by Dr. Florian Reizine of Rennes University Hospital and Dr. Juliette Henry of Poitiers University Hospital, unsupervised machine learning analysis of 2,145 ICU patients revealed three clinically meaningful clusters of candidemia—each with unique risk factors, microbiological patterns, and outcomes. The findings, published June 29 in Critical Care (doi:10.1186/s13054-026-05964-4), were drawn from data collected between 2008 and 2020 across 13 French ICUs.

Three distinct candidemia profiles emerge in ICU patients
The study’s clustering algorithm distinguished three patient groups:

1. High-mortality, C. albicans-dominant infections (32% of cases), linked to severe sepsis, organ failure, and a 45% in-hospital death rate.
2. Moderate-risk, mixed-species infections (41% of cases), associated with chronic liver disease and a 22% mortality rate.
3. Low-risk, C. parapsilosis-associated infections (27% of cases), primarily in patients with recent surgery or central venous catheters, with a 12% mortality rate.

“This is the first time we’ve seen such clear stratification in candidemia,” said Dr. Jean-Pierre Gangneux, a co-author and infectious disease specialist at Rennes. “Current guidelines treat all Candida infections the same, but our data suggest tailored approaches could improve survival.”

Why the findings matter: A shift from ‘one-size-fits-all’ antifungal therapy
The study challenges the long-standing ICU practice of using broad-spectrum antifungals for all Candida bloodstream infections, regardless of species or patient profile. Key implications include:

• Cluster 1 patients (high mortality, C. albicans) may benefit from aggressive early antifungal therapy and source control (e.g., catheter removal).
• Cluster 3 patients (low risk) could potentially avoid unnecessary prolonged treatment, reducing antifungal resistance and costs.
• Cluster 2 patients (mixed species) may require species-specific therapy, as some Candida species show higher resistance to first-line drugs like fluconazole.

The research builds on prior work (PMID: 42365321) showing that Candida infections account for nearly 10% of ICU sepsis cases in Europe, with mortality rates ranging from 20% to 60% depending on species and patient comorbidities. The new clustering method, however, provides a data-driven way to predict which patients fall into higher-risk categories at admission.

How the study was conducted: Machine learning meets ICU data
Researchers analyzed electronic health records from 13 French ICUs, including:

• Demographic data (age, sex, comorbidities)
• Microbiological results (Candida species identified)
• Clinical severity scores (SOFA, SAPS II)
• Outcomes (mortality, ICU length of stay)

Using unsupervised clustering (k-means algorithm), the team identified the three profiles without predefining groups. Validation confirmed the clusters held true across different hospitals and time periods.

What happens next: Could this change ICU protocols?
The study’s authors are collaborating with French intensive care societies to design prospective trials testing tailored antifungal strategies. Dr. Henry noted that implementing such changes would require:

1. ICU-specific algorithms to identify clusters at admission.
2. Antifungal stewardship programs to adjust therapy based on species and risk profile.
3. Training for clinicians on interpreting cluster-based risk assessments.

“This isn’t just about better drugs—it’s about smarter use of existing tools,” said Gangneux. “If we can reduce unnecessary antifungal exposure in low-risk patients, we might also curb resistance.”

The research was funded by the French National Research Agency (ANR) and published under open-access terms, making the methodology available for global ICU networks to adapt.

Key takeaways for clinicians and policymakers

• Risk stratification could become standard in ICU candidemia management, similar to sepsis bundles.
• Antifungal resistance may decline if low-risk patients receive shorter courses.
• Electronic health records could integrate cluster-prediction tools to flag high-risk cases early.

The study’s limitations include its retrospective design and focus on French ICUs, though the authors argue the clustering method is generalizable. Peer reviewers highlighted its potential to “redefine candidemia management” in a field where treatment options have remained static for decades.

For ICU professionals, the next steps involve validating the clusters in multicenter prospective studies—a process already underway in several European centers.