Multimodal brain-gut-sleep phenotypes predict delirium, long-term cognitive decline, and survival after colorectal cancer surgery.

Unsupervised clustering of preoperative multimodal assessments identified four biobehaviorally coherent phenotypes in colorectal cancer surgery patients.

• The prospective cohort included n=300 patients undergoing colorectal cancer surgery.
• Preoperative assessments included circadian actigraphy, gut microbial diversity and short-chain fatty acids, inflammatory cytokines (IL-6, CRP), nocturnal heart rate variability, sleep-wake characteristics, and psychological symptoms.
• Unsupervised clustering was used to derive the four phenotypes.
• Phenotypes ranged from a resilient reference (Phenotype A) to a convergent multidomain dysregulated group (Phenotype D).

Postoperative delirium incidence increased across phenotypes from 4.7% in Phenotype A to 21.6% in Phenotype D.

• Delirium incidence in Phenotype A was 4.7%.
• Delirium incidence in Phenotype D was 21.6%.
• Delirium incidence followed a gradient across the four phenotypes.
• Analyses used mixed-effects models, Cox regression, and gradient boosting.

Cognitive decline over 36 months followed the same phenotype gradient, with Phenotypes C and D showing progressive deterioration.

• Cognitive trajectories were assessed over 36 months postoperatively.
• Phenotypes C and D showed progressive cognitive deterioration across the follow-up period.
• Phenotype A (resilient reference) showed the most favorable cognitive trajectory.
• Mixed-effects models were used to analyze cognitive trajectories.

Fatigue and sleep recovery displayed parallel phenotype-dependent stratification across the 36-month follow-up.

• Fatigue and sleep recovery trajectories were assessed alongside cognitive outcomes.
• Both fatigue and sleep recovery outcomes followed the same gradient pattern as delirium and cognitive decline.
• Phenotype-dependent stratification was observed across all major outcome domains.

Three-year disease-free survival and overall survival were significantly worse in higher-risk phenotypes compared to the resilient reference phenotype.

• Three-year disease-free survival (DFS) ranged from 86.7% in Phenotype A to 69.3% in Phenotype D.
• Three-year overall survival (OS) ranged from 91.2% in Phenotype A to 78.8% in Phenotype D.
• Relative to Phenotype A, phenotypes with convergent multidomain dysregulation were independently associated with worse survival (HR 2.11 for DFS; HR 1.96 for OS).
• Cox regression was used to assess survival associations.

Machine-learning models identified circadian amplitude, microbial diversity, SCFA concentrations, IL-6, and nocturnal heart rate variability as the dominant predictive contributors.

• Gradient boosting was among the machine-learning methods applied.
• The top-ranked features were circadian amplitude, microbial diversity, short-chain fatty acid (SCFA) concentrations, IL-6, and nocturnal heart rate variability.
• These features spanned circadian, microbial, inflammatory, and autonomic domains.
• Feature importance ranking was used to identify dominant contributors to outcome prediction.

Preoperative assessments incorporated multiple biological and behavioral domains reflecting circadian, gut-brain, autonomic, inflammatory, and psychological pathways.

• Assessments included circadian actigraphy, gut microbial diversity and short-chain fatty acids (SCFAs), inflammatory cytokines IL-6 and CRP, nocturnal heart rate variability, sleep-wake characteristics, and psychological symptoms.
• The study framed perioperative resilience as reflecting coordinated regulation across these multiple pathways.
• Assessments were conducted preoperatively in the prospective cohort of 300 patients.