Baseline characteristics
BoxPlots grouped by ANLR quartiles (Figure 2) show that NLR levels gradually decreased and albumin levels gradually increased in quartiles. The baseline characteristics stratified by ANLR quartiles are shown in Table 1.p <0.001), gender (p= 0.031), and weight (p<0.001). All vital signs, including heart rate, breathing rate, temperature, SBP, and Spo₂, were significantly different across quartiles (All p<0.001). Among clinical scores, sofa, APS III, oasis, and CCI were significantly different (All p<0.001). There were significant variations for comorbidities, cerebrovascular disease, chronic pulmonary disease, hypertension, septic shock, and AKI (All p<0.001). All laboratory indicators except potassium showed significant differences between groups (p<0.05), including RBC, WBC, platelets, lymphocytes, neutrophils, albumin, sodium, ALT, AST, anion gap, INR, PT, BUN, and creatinine (All p<0.001). Therapeutic measures such as propofol, midazolam, GC, antibiotics, norepinephrine, dopamine, vasopressin, EN, MV, and CRRT also varied significantly across groups (All p<0.001). Furthermore, all outcome indicators - hospital stay, ICU stay, hospital mortality rate, ICU mortality rate, 30-day mortality rate, and 90-day mortality rate were significantly different for the ANLR quartile (All p<0.001).
nlr's boxplots (a) and albumin (b) Stratified by ANLR quartile
Key Results
Kaplan-Meier survival analysis reveals significant differences in survival rates for patients with ANLR quartile grouped, indicating the most favorable survival outcome. Statistical comparisons using the log-rank test showed very significant differences between the quartiles in both 30-day and 90-day mortality rates (p<0.001;Figure 3).
Kaplan-Meier curve for 30 days (a) and 90 days (b) mortality rates in patients with sepsis stratified by ANLR quartiles
Variables included in the multivariable COX regression analysis (Table S1 Supplementary Material) were selected based on univariable COX regression results and clinical expertise. The results are shown in Table 2. In the fully adjusted model (Model 3), ANLR as a continuous variable was independently associated with reduced risk for both 30-day mortality (HR = 0.68, 95% CI = 0.59–0.79; p<0.001) and 90-day mortality rate (HR = 0.85, 95% CI = 0.76–0.94; p= 0.002). When ANLR was analyzed as a categorical variable (quartile), quartile patients had a significantly lower risk of mortality compared to those in quartile 1 for 30-day mortality (HR = 0.44, 95% CI = 0.37–0.53; p<0.001) and 90-day mortality (HR = 0.45, 95% CI = 0.38–0.53; p<0.001). Important trends were observed across the ANLR quartiles (pTrends <0.001) at both times.
After adjusting for all covariates, RCS analysis revealed a very important overall association between ANLR and both 30- and 90-day mortality (overall overall p<0.001), evidence of significant nonlinear patterns (nonlinear p<0.001) (see Figure 4).
RCS analysis of ANLR and mortality rates in sepsis patients (a) 30-day mortality rate (b) 90-day mortality rate
Subgroup analysis
After adjusting for related covariates, interaction analyses were performed within a set of prescribed subgroups, including gender, race, cerebrovascular disease, chronic pulmonary disease, diabetes, and the use of mechanical ventilation. The findings showed that the relationship between ANLR and mortality was stable in these categories as no statistically significant interactions were detected (no all interactions were detected) p– Value exceeded 0.05. (See Figure 5).
Subgroup forest plots of adjusted covariates for mortality in patients with sepsis (a) 30 days of all-causal mortality rate, (b) 90-day all-causal mortality rate
ROC analysis of ANLR and other predictive markers
ROC analysis demonstrated that ANLR achieved an AUC of 0.66 to predict 30-day mortality rates and the optimal threshold was identified at 0.27. Notably, this predictive ability exceeded the predictive ability of other individual biomarkers, including NLR, SOFA score, neutrophil count, albumin, and lymphocyte count. In particular, combining SOFA and ANLR as a ratio (SOFA/ANLR) further improved discriminatory power, with the AUC of SOFA/ANLR reaching 0.68 and the optimum cutoff value of 12.56. Similar findings were observed at 90-day mortality rate, with only ANLR of 0.65 (cutoff = 0.30), and SOFA/ANLR of a higher AUC of 0.67 (cutoff = 10.75), indicating that the SOFA/ANLR ratio significantly promoted post hoc performance compared to SOFA alone (Fig. 6).
a ROC curve for 30-day mortality rate; b ROC curve for 90-day mortality rate;
Volta
According to a pre-established protocol, the dataset was divided into training and validation cohorts. Comparative analyses revealed no statistically significant differences in baseline characteristics between these two groups (all p-values> 0.05; see Table S2 in Supplementary Materials). A total of 24 variables were identified as important predictors of 30-day mortality using the Boruta algorithm for feature selection. Of these, ANLR was ranked as the second most important feature. Other important variables include weight, platelets, RBC, AKI, SBP, heart rate, septic shock, fluid balance, potassium, ALT, WBC, cerebrovascular disease, AST, respiratory rate, SPO₂, PT, creatinine, sodium, age, anion gap, pan, and temperature. All these variables were identified as “confirmed” by the Boruta algorithm and highlighted in green in the feature importance plot (Fig. 7). These variables were included in subsequent machine learning modeling.
a Boruta algorithm to identify key variables for predicting 30-day mortality (green indicates important variables), b ROC curves of the validation set, c Decision curves for multiple models in the validation set. d Calibration curves for multiple models in the validation set. e SHAP summary plot with dependencies showing various importance and feature effects of LightGBM models
Establishing and verifying predictive models
The predictive model was constructed using 24 key variables highlighted as important features by the Boruta algorithm. All of these were marked green. The training set consisted of 4715 patients, and the validation set included 1573 patients. In the training set, 968 patients (21%) died, and 323 deaths (21%) were observed in the validation set. These event counts ensured compliance with generally accepted criteria for a minimum of 10 outcome events per predictor variable.
Various models were built using selected features such as SVM, Ridge, LightGBM, KNN, ENET, XGBOOST, DT, MLP, and more. LightGBM was identified as the best model, demonstrating the lowest Brier score (0.1250), the highest net profit in decision curve analysis, and the highest AUC (0.821) in the validation set (Fig. 7). The LightGBM model was further interpreted using SHAP values to identify ANLR as the second most important variable (Fig. 7).
Sensitivity analysis
To assess the robustness of the findings, multivariable Cox regression analysis was performed using the original (Imputed) dataset. The results for both 30-day and 90-day mortality were consistent with those obtained from the main analyses (Table S3, Supplementary Material).
