3.1 Baseline clinicopathological characteristics

This retrospective study included 415 HCC patients who had received hepatectomy as an initial treatment in our hospital. Of the 415 patients enrolled, 372 (89.6) were male and 43 (10.4) were female; 346 (83.4) were HBsAg positive, 290 (69.9) had cirrhosis, and 65 (15.7) had preoperative diabetes mellitus. 313 (75.4) had stage I + II, and 102 (24.6) had stage III + IV. The median [IQR] follow-up was 30.34 [20.87–51.76] months. A total of 20.48% of the patient data were censored, meaning these patients were still alive at the time of the last follow-up and had not experienced the events defined by the study. Among them, 17.89% of the censored data occurred in stage I + II, while 28.43% occurred in stage III + IV. A total of 51 patients (13.7%) died, with 26 (8.31%) of these cases occurring in stage I + II and 31 (30.39%) in stage III + IV. The median [IQR] TyG, TG/HDL-c, and TyG-BMI indices were 7.05 [6.80, 7.44], 0.95 [0.69, 1.43], and 164.79 [150.34, 185.28], respectively. Table 1 presents the detailed clinicopathological characteristics of the high and low-TyG groups. BMI, hypertension, preoperative diabetes mellitus, HBsAg positivity, AFP, ALB, GGT, FPG, TG, HDL-C, tumor size, number of tumors, TG/HDL-c, and TyG-BMI differed significantly between the high and low TyG groups. There were no significant differences in other clinicopathologic features, especially TNM staging.

Table 1 Baseline characteristics of included patients3.2 Prognostic value of the TyG, TG/HDL-c, and TyG-BMI

We first assessed whether the TyG, TG/HDL-c, and TyG-BMI indexes could differentiate patients’ prognoses. The cut-off points of these three parameters were set as 6.58, 0.78, and 138.34. The results showed that low TyG (HR = 2.238, 95% CI 1.289–3.884, p = 0.002), low TG/HDL-c (HR = 1.868, 95% CI 1.111–3.140, p = 0.022), and low TyG-BMI (HR = 2.102, 95% CI 1.172–3.770, p = 0.004) were significantly linked to worse OS (Fig. 2A). The 1-, 3-, and 5 year AUC of the TyG were 0.734, 0.615, and 0.603, respectively. The 1-, 3-, and 5 year AUC of the TG/HDL-c were 0.616, 0.552, and 0.532, respectively. The 1-, 3-, and 5 year AUC of the TyG-BMI were 0.709, 0.596, and 0.625, respectively (Fig. 2B). All three parameters had modest prognostic efficacy, especially in 1 year OS of HCC patients.

Fig. 2

The Kaplan–Meier and receiver operator characteristic curves for different IR indexes. A The Kaplan-Meier curve for overall survival by different IR indexes; B The receiver operator characteristic curves for different IR indexes. IR insulin resistance; TG/HDL-c: triglycerides / high-density lipoprotein cholesterol ratio, TyG triglyceride-glucose index, TyG-BMI triglyceride-glucose index-body mass index.

3.3 TyG, but not TG/HDL-c and TyG-BMI, was an independent factor of OS.

Univariate analysis showed that preoperative AFP (HR = 0.51; 95% CI 0.29–0.90; p = 0.019), tumor size (HR = 2.70; 95% CI 1.51–4.82; p < 0.001), tumor number (HR = 0.49, 95% CI 0.28–0.86; p = 0.012), tumor capsule (HR = 3.00; 95% CI 1.66–5.42; p < 0.001), vascular invasion (HR = 2.26; 95% CI 1.32–3.88; p = 0.003), tumor grade (HR = 2.35; 95% CI 1.37–4.01; p = 0.002), major resection (HR = 2.62; 95% CI 1.56–4.40; p < 0.001), surgical approach (HR = 2.15; 95% CI 1.27–3.63; p = 0.004), low TyG index (HR = 2.25; 95% CI 1.33–3.78; p = 0.002), low TG/HDL-c (HR = 1.88; 95% CI 1.09–3.24; p = 0.024), low TyG-BMI (HR = 2.11; 95% CI 1.25–3.56; p = 0.005), and TNM stage (HR = 6.85; 95% CI 3.98–11.79; p < 0.001) significantly predicted OS. Furthermore, the multivariate analysis suggested that, in addition to the TNM stage (HR = 8.00; 95% CI 3.92–16.33; p < 0.001), low TyG index (HR = 2.43, 95% CI 1.39–4.24, p = 0.002) but not TG/HDL-c (HR = 0.90; 95% CI 0.41–1.96; p = 0.791) and TyG-BMI (HR = 1.51; 95% CI 0.81–2.82; p = 0.191), were independent prognostic factors for OS (Fig. 3). In exploratory analyses, Fig. 4 showed the impact of a lower TyG index on survival prognosis across different subgroups using Cox proportional hazards regression analysis. Specifically, the horizontal axis represented the HR, with larger values indicating higher risk; the vertical axis displayed the results for each subgroup. The P for interaction assessed the consistency of the proportional risk effect of the TyG index on mortality across different subgroups. The proportional effect of lower TyG index on mortality was consistent across 23 pre-specified subgroups but not in the ALT (P for interaction: 0.025) and major resection (P for interaction: 0.039) subgroups.

Fig. 3

Forrest plot of the univariate and multivariate Cox regression analysis in HCC. ALB albumin, ALT alanine aminotransferase, AFP alpha-fetoprotein, BMI body mass index, FPG fasting plasma glucose, HDL-c high-density lipoprotein cholesterol, TBIL total bilirubin, TC total cholesterol, TG triglycerides, TG/HDL-c triglycerides / high-density lipoprotein cholesterol ratio, TyG triglyceride-glucose index, TyG-BMI triglyceride-glucose index–body mass index, GGT glutamyl transpeptidase, HBsAg hepatitis B surface antigen, MVI microvascular invasion, TNM tumor node metastasis classification

Fig. 4

Forrest plot of the subgroup survival using univariate Cox regression analysis in HCC The horizontal axis represented the HR, with larger values indicating higher risk; the vertical axis displayed the results for each subgroup. The P for interaction assessed the consistency of the proportional risk effect of the TyG index on mortality across different subgroups using likelihood ratio tests. ALB albumin, ALT alanine aminotransferase, AFP alpha-fetoprotein, BMI body mass index, FPG fasting plasma glucose, HDL-c high-density lipoprotein cholesterol, TBIL total bilirubin, TC total cholesterol, TG triglycerides, TG/HDL-c triglycerides / high-density lipoprotein cholesterol ratio, TyG triglyceride-glucose index, TyG-BMI triglyceride-glucose index-body mass index, GGT glutamyl transpeptidase, HBsAg hepatitis B surface antigen, MVI microvascular invasion, TNM tumor node metastasis classification.

3.4 Cox regression analyses, nomogram building, and DCA curve for OS

A nomogram containing the TNM stage and TyG index was developed (Fig. 5A). The nomogram clearly contains four key nodes, which, upon careful observation, represent from left to right: TyGlowstageIII+IV, TyGhighstageIII+IV, TgGlowstageI+II, TgGhighstageI+II. Patients in different nomogram subgroup had significantly different OS, with TyGlowstageIII+IV showing the poorest prognosis, while TgGhighstageI+II had the best prognosis (P < 0.001) (Fig. 5B). Calibration plots suggest that column plots are better predictors of short-term prognosis, especially 1 year OS (Fig. 5C). The 1-, 3-, and 5-year AUC of the nomogram were 0.859 (95% CI 0.797–0.917), 0.884 (95% CI 0.827–0.940), and 0.789 (95% CI 0.693–0.886), respectively (Fig. 5D). The C-index was 0.612 (95% CI 0.575–0.709), 0.745 (95% CI 0.684–0.807) and 0.816 (95% CI 0.773–0.860) for the TyG model, TNM stage mode, and the nomogram model, respectively. The nomogram was further suggested to be better than the TNM and TyG index models alone by AUC and C-index curves. (Figs. 5E, F). DCA curve indicated that the nomogram provided clear clinical benefits in predicting HCC OS (Fig. 5G).

Fig. 5

Building the nomogram predicting overall survival for HCC patients A The nomogram plot was built based on three independent prognostic factors in HCC. B The Kaplan-Meier curve showed the survival difference between high and low-risk patients stratified by the nomogram. C The calibration plot for internal validation of the nomogram. D The time-dependent AUC curves of the nomogram. E The time-dependent AUC curves compare the nomogram, TNM, and TyG models. F The time-dependent concordance index curves compare the nomogram, TNM, and TyG models. G The DCA curves of the nomograms.

3.5 TNM stage I + II subgroup analysis

We then explored the clinical significance of the TyG, TG/HDL-c, and TyG-BMI in TNM stage I + II patients merely, which usually received hepatectomy as the standard treatment. Similar to that observed in the overall population BMI, hypertension, preoperative diabetes mellitus, HBsAg positivity, GGT, FPG, TG, HDL-C, number of tumors, TG/HDL-c, and TyG-BMI differed significantly between the high and low TyG groups (Supplementary Table 1). The Kaplan Meier analyses indicated that low TyG (HR = 2.532 95% CI 1.108–5.798, p = 0.014), low TG/HDL-c (HR = 2.73, 95% CI 1.261–5.910, p = 0.013), and low TyG-BMI (HR = 2.631, 95% CI 1.077–6.424, p = 0.010) were significantly linked to worse OS (Fig. 6A). The 1-, 3-, and 5-year AUC of the TyG were 0.799, 0.784, and 0.646, respectively. The 1-, 3-, and 5-year AUC of the TG/HDL-c were 0.739, 0.776, and 0.575, respectively. The 1-, 3-, and 5-year AUC of the TyG-BMI were 0.760, 0.717, and 0.664, respectively (Fig. 6B). In the stage I + II subgroup, the prognostic predictive performance of the three parameters was improved compared to the overall population. The univariate analysis revealed that a low TyG index (HR = 2.55; 95% CI 1.18–5.53; p = 0.018), low TG/HDL-c (HR = 2.76; 95% CI 1.20–6.38; p = 0.017), and low TyG-BMI (HR = 2.64; 95% CI 1.22–5.71; p = 0.014) were significant predictors of overall survival (OS). However, the multivariate analysis did not identify any independent prognostic factors (Supplementary Fig. 1).

Fig. 6

The Kaplan–Meier and receiver operator characteristic curves for different IR indexes in stage I + II patients. A The Kaplan-Meier curve for overall survival by different IR indexes; B The receiver operator characteristic curves for different IR indexes. IR insulin resistance, TG/HDL-c: triglycerides / high-density lipoprotein cholesterol ratio, TyG triglyceride-glucose index, TyG-BMI triglyceride-glucose index-body mass index.