Abstract / Summary
Malnutrition coupled with chronic inflammation substantially accelerates disease progression in critically ill patients with colorectal cancer. However, reliable biomarkers for the evaluation of acute kidney injury (AKI) in this population remain lacking. Eligible critically ill patients with colorectal cancer were firstly enrolled from the eICU database ( N = 706) and stratified into three groups according to the tertiles of red cell distribution width-to-albumin ratio (RAR). Four machine-learning algorithms were applied to establish a RAR-based prediction model for AKI. Finally, clinical data of critically ill colorectal cancer individuals from Wuhan Union Hospital were used for external validation ( N = 259). The primary clinical endpoint of this study was defined as AKI among critically ill colorectal cancer individuals, while the secondary outcome was set as in-hospital mortality. Restricted cubic spline (RCS) analysis indicated a linear trend between RAR and prognosis of critically ill colorectal cancer individuals in eICU database ( P = 0.078). Low RAR was linked to lower AKI incidence ( P = 0.004) and better survival ( P < 0.001). Multivariate regression validated elevated RAR as an independent AKI risk factor (OR = 2.79, 95%CI:1.67–4.66, P < 0.001). RAR showed better predictive ability for AKI and in-hospital mortality than SOFA, OASIS and APSIII scores. External validation in Wuhan Union Hospital cohort ( N = 259) confirmed such association ( P < 0.001). The machine learning-established RAR-based score yielded satisfactory AKI predictive performance, with AUCs of 0.769 and 0.743 in eICU database and Union cohort, respectively. This retrospective study indicated that RAR level is linearly correlated with the AKI of critically ill colorectal cancer individuals, and high RAR acts as an independent predictor of AKI. The RAR-derived model can markedly improve prognostic predictive efficiency, which is promising for clinical risk stratification and therapeutic decision-making.