Both internal and external validation data must meet the following inclusion and exclusion criteria.

(1) Diagnostic criteria: Sepsis diagnosis followed the Third In review International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3) (8).

(2) Inclusion criteria: Aged 18–85 years; Meets diagnostic criteria for sepsis.

(3) Exclusion criteria: Patients with pre-existing hematological disorders, malignancies, hepatic/renal insufficiency, or immune diseases; those who underwent continuous renal replacement therapy (CRRT) within 72 h of admission; non-mechanically ventilated patients; and patients with incomplete clinical data for accurate assessment.

A total of 55 mechanically ventilated sepsis patients aged 18–85 years, treated at The First Affiliated Hospital of Tianjin University of Traditional Chinese Medicine from January 2022 to January 2024, were included. The cohort comprised 37 males and 18 females, with 14 non-survivors and 41 survivors.

This study employed a single-center, retrospective case-control design. All patient data in this study were obtained from anonymous data from the medical database of Tianjin University of Traditional Chinese Medicine, and were retrospectively enrolled.

Both internal and external validation data were collected via anonymous data from the medical database of Tianjin University of Traditional Chinese Medicine: patient demographics: age, sex; History of In review underlying diseases: diabetes, hypertension, coronary heart disease; Laboratory markers: admission WBC count, Lactate levels at admission, 24h, and 72 h, Lactate clearance rates at 24 h and 72 h, PCT levels at admission, 24 h, and 72 h, PCT clearance rates at 24 h and 72 h, Clinical outcome: 28-day prognosis.

Statistical analysis was performed using SPSS 21.0. Measurement data conforming to a normal distribution were expressed as mean ± standard deviation (mean ± SD), while non-normally distributed measurement data were expressed as the interquartile range (IQR; Q25, Q75).

For intergroup comparisons: normally distributed data between two groups were analyzed using the independent samples t-test. Non-normally distributed data between two groups were analyzed using the Mann-Whitney U-test (rank-sum test). Multi-group comparisons were performed using the Kruskal-Wallis H-test.

Categorical variables were expressed as frequencies and percentages, and comparisons between categorical variables were conducted using the chi-square (χ²) test.

These statistical methods (independent samples t-test, MannWhitney U-test, and chi-square test) were applied to compare clinical parameters between groups, aiming to identify risk factors influencing In review the prognosis of sepsis patients.

SPSS21.0 was used to complete univariate and multivariate logistic regression analysis, identify independent risk factors affecting the prognosis of patients with sepsis in ICU, construct ROC curves, and complete the Hosmer and Lemeshow Test. The predictive value of individual independent risk factors vs. combined risk factors for ICU sepsis prognosis was compared. Based on the identified independent risk factors, the collinearity analysis of independent risk factors was completed using the car package, corrplot package, and Hmisc package of R software (version 4.3.1 (2023-06-16 ucrt), and then the nomogram model was developed using the rms package, rmda package, dcurves package, etc., and the clinical decision curve and Bootstrap internal validation (1,000 iterations) were generated. External validation was performed using clinical data from an additional 45 ICU sepsis patients to evaluate the accuracy, reliability, and clinical applicability of this model. Difference difference is defined as P < 0.05.

A total of 55 septic patients from the ICU of Tianjin University of Traditional Chinese Medicine were enrolled in this study. There were no statistically significant differences in baseline characteristics between the two groups (P > 0.05). For details, see Tables 1–3.

In the comparison of lactate and PCT-related indicators between the two groups, statistically significant differences were observed in 24-h lactate levels, 24-h lactate clearance rate, admission PCT levels, and 72-h PCT clearance rate (P < 0.05). Detailed results are shown in Table 4.

Statistically significant differences were observed between the survival and death groups in 24-h lactate levels, 24-h lactate clearance rate, admission PCT levels, and 72-h PCT clearance rate (P < 0.05). These variables were subjected to univariate logistic regression analysis. Results showed that 24-h lactate levels, 24-h lactate clearance rate, and 72-h PCT clearance rate were significantly associated with prognosis (P < 0.05), while admission PCT levels showed no significance (P = 0.104, P > 0.05; see Table 5).

Subsequently, multivariate logistic regression analysis was performed using these three risk factors as independent variables. Results demonstrated that 24-h lactate clearance rate and 72-h PCT clearance rate remained statistically significant (P < 0.05; see Table 6), identifying them as independent risk factors for 28-day prognosis in septic patients.

ROC curves were generated to evaluate the predictive value of individual and combined independent risk factors. The area value under the ROC curve is the core index to evaluate the discrimination ability of the prediction model, and the numerical size can reflect the model's ability to correctly distinguish between “true positive” and “true negative.” A higher AUC value represents a higher predictive value of the prediction model. The results showed: 24-h lactate clearance rate: AUC = 0.709, standard error (SE) = 0.091, P = 0.020, 95% confidence interval (CI): 0.531–0.887 (Figure 1); 72-h PCT clearance rate: AUC = 0.908, SE = 0.039, P < 0.001, 95% CI: 0.830–0.985 (Figure 2); Combined two independent risk factors: AUC = 0.920, SE = 0.036, P < 0.001, 95% CI: 0.850–0.990 (Figure 3).

The AUC value of the combination of two independent risk factors is greater than that of any one risk factor alone, so the combination of the two independent risk factors has a higher predictive value for the prognosis of sepsis patients.

The Hosmer and Lemeshow Test is the core test method of logistic regression model calibration, which is used to evaluate the degree of match between the predicted probability of the model and the actual observed outcome, if the test result is P > 0.05, and χ²/d < 2 indicates that the model is well calibrated and there is no local calibration defect.

The Hosmer and Lemeshow test results of the logistic regression model constructed with two independent risk factors alone and the two independent risk factors combined all showed P > 0.05 and χ²/d < 2, which supported the model fit well and the model calibration degree was acceptable (see Tables 7–9 for details).

The variance expansion factor (VIF) of the two predictors was 1.079493 less than 2, indicating that there was no significant multicollinearity problem. The highest correlation coefficient between variables was |r| = 0.08, which was much lower than the collinearity alert threshold of 0.7 (see Table 10 for details).

A nomogram model for predicting the prognosis of sepsis patients was developed using R language, based on the combination of two independent risk factors, as illustrated in Figure 4. In the nomogram, the 24-h lactate clearance rate and 72-h PCT clearance rate are displayed according to their respective weights. Each variable value corresponds to a point score, and the sum of all variable scores yields the total score. The total score is then mapped to a predicted probability, representing the survival probability of sepsis patients.

The calibration curve and clinical decision curve of the nomogram model were plotted, and Bootstrap internal validation was performed. External validation was conducted using clinical data from an additional 45 ICU sepsis patients to assess the model's accuracy, reliability, and clinical applicability. In Figure 5, the x-axis and y-axis of the calibration curve represent the nomogram-predicted and actual survival probabilities, respectively. The ideal curve is a 45° diagonal line, where closer proximity to the diagonal indicates better agreement between model predictions and actual observations, reflecting higher reliability. The results of this study show that the calibration curve of the model roughly coincides with the ideal curve in its overall trajectory, indicating that the model has certain predictive value.

In Figure 6, when the DCA curve lies above and to the right of the two extreme lines, it indicates superior predictive performance and clinical value. The DCA of this model suggests that this predictive model may provide some clinical net benefit for patients with sepsis.

For internal validation, the model achieved an AUC value of 0.901, sensitivity of 0.540, and specificity of 0.899. External validation yielded an AUC value of 0.681, sensitivity of 0.510, and specificity of 0.742. The large difference between the AUC values of internal and external validation indicates that there may be some overfitting or data heterogeneity in the nomogram prediction model, which may be caused by the small sample size of the prediction model and the systematic differences in the included population, which limits the generalizability of the nomogram prediction results.