A preoperative inflammatory score-based nomogram predicts overall survival after curative hepatectomy for hepatocellular carcinoma

Abstract

Objective: This study incorporated preoperative inflammatory scores to develop and validate a nomogram to predict overall survival in patients with hepatocellular carcinoma following curative resection.

Methods: The study included 402 postoperative hepatocellular carcinoma patients, divided into training (n = 281) and test (n = 121) cohorts. Variables were analyzed using Cox proportional hazards model. The nomogram's performance was assessed using receiver operating characteristic curves, calibration curves, and decision curve analysis.

Results: Multivariable Cox proportional hazards model analysis identified neutrophil-to-lymphocyte ratio-lymphocyte-to-monocyte ratio score (HR = 4.19, 95% CI 2.47-7.12), microvascular invasion (HR = 4.93, 95% CI 2.74-8.85), and total tumor volume (HR = 1.67, 95% CI 1.03-2.68) as independent prognostic factors (P < 0.05). The nomogram exhibited excellent discriminatory ability, with area under the curve values for 12-, 36-, and 60-month overall survival in the test cohort measuring 0.941, 0.810, and 0.881. Calibration curves verified a high degree of consistency, with a Brier score of 0.054, 0.120, and 0.102, between the predicted and observed survival probabilities in the test cohort. Decision curve analysis confirmed clinical utility across a wide threshold probability range (0.15-0.70).

Conclusion: The nomogram integrating neutrophil-to-lymphocyte ratio-lymphocyte-to-monocyte ratio score, microvascular invasion, and total tumor volume effectively identifies high-risk hepatocellular carcinoma patients with shorter overall survival. This tool provides clinicians with new evidence for risk-stratified interventions.

Keywords: Hepatectomy; Hepatocellular carcinoma; Inflammation indices; Nomogram; Overall survival; Prediction model.

Fig. 1

Flow chart of the study

Fig. 2

Maximal selection rank statistic was used to determine the optimal cut-off values. Patients (n = 402) were ranked into quartiles based on LMR and NLR values respectively, and the standardized statistic was calculated for each. a The largest standardized statistic corresponding to the dashed line in the figure based on LMR values is regarded as the optimal cut—off value. b Similarly, the NLR value corresponding to the dashed line in the figure represents the optimal cut—off value

Fig. 3

Kaplan–Meier survival curves and log-rank tests were used to validate the associations of LMR and NLR with OS in HCC patients following curative resection. a Training cohort LMR analysis. Patients with low LMR (≤ 1.7) exhibited significantly worse OS compared to the high LMR group (HR = 2.16, 95% CI 1.23–3.80, P < 0.001). b Test cohort LMR validation. The prognostic discrimination remained consistent (HR = 3.21, 95% CI 2.13–4.90, P < 0.001). c Training cohort NLR evaluation. Patients with high NLR (> 2.3) demonstrated significantly reduced OS compared to the low NLR group (HR = 1.94, 95% CI 1.72–3.68, P < 0.001). d Test cohort NLR confirmation. The prognostic significance persisted (HR = 4.56, 95% CI 3.24–6.13, P < 0.001). Risk tables below the Kaplan–Meier survival curves present the number of surviving patients at different time points

Fig. 4

LASSO and multivariate Cox regression analyses were used to identify independent risk factors associated with OS in HCC patients following curative resection. a Lasso coefficient trajectories for six variables (MVI, TTV, surgical approach, ALT, AST, NLR-LMR). Variables retaining non-zero coefficients at higher regularization intensities (λ ≥ 3) indicate stronger prognostic effects: MVI, TTV, NLR-LMR. b Cross-validation curve (y-axis: binomial deviance, lower x-axis: ln(λ), upper x-axis: non-zero coefficients) identifies λ_min = 3 (left dashed line) as the optimal parameter, retaining three predictors with optimal prognostic significance. c Multivariable Cox analysis: Tabular data (left) lists HRs (95% CIs), while the forest plot (right) visualizes associations. Horizontal bars (95% CIs) excluding the HR = 1 reference line (vertical dashed) denote statistical significance (P < 0.05). Final model integrates MVI (HR = 4.93, 2.74–8.85), NLR-LMR (HR = 4.19, 2.47–7.12), and TTV (HR = 1.67, 1.03–2.68), validated by 1,000 bootstraps (C-index = 0.76)

Fig. 5

Visualization and validation of nomogram. a Nomogram displays three predictors (MVI, TTV, NLR-LMR) with axis lengths proportional to variable weights. The total points are converted into the probabilities of 12-, 36-, and 60-month OS. b Training cohort ROC shows area under curves (AUCs) of 0.801 (12-month), 0.819 (36-month), and 0.808 (48-month). c Test cohort ROC replicates accuracy (AUCs: 0.941, 0.810, 0.881). d Training cohort calibration yields Brier scores (0 = perfect prediction) of 0.095 (12-month), 0.149 (36-month), and 0.158 (48-month). e Test cohort calibration maintains precision (Brier Scores: 0.054, 0.120, 0.102). f Training cohort DCA shows net benefit over universal treatment ("All") and no-treatment ("None") strategies at threshold probabilities of 0.05–0.72 (12-month), 0.10–0.97 (36-month), and 0.15–0.98 (48-month). g Test cohort DCA validates clinical utility (thresholds: 0.02–0.72, 0.05–0.97, 0.06–0.99)