Characteristics of patients
580 consecutive CRLM patients who underwent curative-intent hepatectomy at two medical centers were enrolled. 434 patients were from SYSUCC and 146 patients were from SYSUSAH. Clinicopathology and treatment characteristics were listed in Table 1. All enrolled patients were Chinese individuals, average age at diagnosis was 59. 6% of patients had CEA levels over 200ng/ml, while 16% of patients had LDH levels over ULN in in the whole population. The median follow-up time was 65.5 months in the SYSUCC (training) cohort and 42.1 months in the SYSUSAH (validation) cohort. OS was 59.5 months (95% CI, 58.4-70.6) in the pooled cohort, 58.9 months (95% CI, 46.2-71.6) in the training cohort and 63.3 months (95% CI, 61.3-67.8) in the validation cohort, respectively.
Overall, clinical features were well balanced between the two cohorts, except that patients in the training cohort had a higher proportion of synchronous CRLM (95% vs 69%), T4 stage of primary tumor (83% vs 64%), well to moderately differentiated pathology (88% vs 76%), and LDH levels above ULN (23% vs 14%) than patients in the validation cohort.
LDH levels and correlations with clinical characteristics
The relationship between serum LDH and clinicopathological parameters was detailed in Table 2. In summary, serum LDH levels showed no statistical difference when stratified by demography characteristics (age, gender), primary tumor characteristics (tumor location, pathology differentiation, T and N stage, KRAS and BRAF mutation), and metastatic sites characteristics (presence of extrahepatic disease, number of CRLM, perioperative chemotherapy).
However, we observed that in the traning cohort, patients with maximum diameter of CRLM > 5cm had a higher proportion of elevated LDH levels than patients with maximum diameter of CRLM ≤ 5cm (36.2% vs 10.3%, P < .001). Patients with preoperative CEA > 200 ng/ml also had a greater possibility of having elevated LDH levels than those with CEA ≤ 200 ng/ml (29.6% vs 12.3%, P = 0.018). In addition, patients with CRS of 4-5 had higher LDH levels than patients with CRS of 2-3 and 0-1 (LDH level > ULN, 42.3% vs 13.1% vs 3.9%; P < .001). While in the validation cohort, patients with left-sided primary tumor had a higher proportion of elevated LDH levels than those with right-sided primary tumor (36.7% vs 13.3%, P = 0.021).
Cox regression analysis of relapse-free survival and overall survival
Univariate and multivariate Cox regression analyses were performed in the training and validation cohort, elevated preoperative LDH level (defined as LDH > ULN) was found to be the strongest prognostic factor both for RFS and OS.
In the training cohort, total five variables were identified as adverse prognostic makers for OS in the univariate analysis: lymph node metastases of primary tumor (HR, 1.77; 95% CI, 1.31-2.39; P < .001), number of CRLM > 1 (HR, 1.85; 95% CI, 1.40-2.44; P < .001), maximum diameter of CRLM > 5cm (HR, 2.17; 95% CI, 1.53-3.08; P < .001), presence of extrahepatic disease (HR, 1.53; 95% CI, 1.03-2.27; P = 0.037), and preoperative LDH level > ULN (HR, 2.41; 95% CI, 1.72-3.39; P < .001). The above variables were subsequently introduced in the multivariate analysis, and all remained their independent prognostic values after adjusted for clinicopathologic parameters (Table 3).
In the validation cohort, three variables were confirmed to correlate with OS both in univariate and multivariate analysis: LDH level > ULN (univariate HR, 3.16; 95% CI, 1.75-5.70; P < .001), lymph node metastases of primary tumor (univariate HR, 1.96; 95% CI, 1.03-3.76; P = 0.042), and T4 stage of primary tumor (univariate HR, 2.26; 95% CI, 1.17-4.37; P = 0.015) (Supplementary Table 1). In pooled cohort, LDH level > ULN, lymph node metastases of primary tumor, number of CRLM, and maximum diameter of CRLM were independent predictors for OS in multivariate analysis (Supplementary Table 2).
In terms of RFS (supplementary Table 3), elevated LDH level (HR, 2.11; 95% CI, 1.54-2.89; P < .001), lymph node metastases of primary tumor, number of CRLM > 1, maximum diameter of CRLM > 5cm, presence of extrahepatic disease, and perioperative chemotherapy reversely correlated with RFS in the univariate analysis. While in the multivariate analysis, the first five variables remained independent predictive factors for RFS.
In the sensitivity analysis in cases with available data of KRAS mutation status, only number of CRLM > 1 was an independent predictor for OS in multivariable models, probably due to the limitation of sample size (Supplementary Table 4).
Survival outcomes according to LDH levels and subgroups analysis
RFS and OS was demonstrated by Kaplan-Meier curves according to LDH levels. In the training cohort, patients with LDH levels over the ULN had significantly shorter RFS (8.5 months vs 22.0 months; HR, 2.11; 95% CI, 1.54-2.89; P < .001) and OS (25.0 months vs 63.6 months; HR, 2.41, 95% CI, 1.72-3.39; P < .001) than patients with LDH levels under the ULN, as shown in Fig. 1 and supplementary Fig. 1. Survival rates at 5 years in the LDH-normal and LDH-high group were 52.9% versus 23.7% in the training cohort, and 62.5% versus 30.9% in the validation cohort, respectively. The validation cohort and pooled cohort demonstrated similar results, patients with elevated LDH levels showed impaired OS compared with patients with normal LDH levels (Fig. 1B-1C).
The forest plots provided a clear trend that patients with LDH levels under the ULN obtained better survival benefit from hepatectomy for OS (Fig. 2) and RFS (supplementary Fig. 2). Subgroup analyses revealed that LDH produced consistent prognostic value across patient subgroups stratified by sex, age, preoperative CEA level, perioperative chemotherapy, primary tumor characteristics (location, lymph node status), liver metastases characteristics (number, maximum diameter, surgical margin, disease-free interval from primary tumor, extrahepatic disease), even by Fong score.
Survival outcomes assessed by CRS model and modified CRS model
OS stratified by different risk scores (0-5) as defined by CRS and mCRS was demonstrated by Kaplan-Meier curves in each cohort (Supplementary Fig. 3). Then we defined three risk groups as following: the low-risk group (CRS or mCRS 0-1), the intermediate-risk group (CRS or mCRS 2-3), and the high-risk group (CRS or mCRS 4-5). The CRS and modified CRS prognostic system were evaluated by three risk groups. The detailed survival results are shown according to each cohort (Table 4) and pooled cohort (Supplementary Table 5).
In the training cohort, median OS of the high-risk group in CRS and mCRS model were 26.1 months and 21.8 months, respectively, and those of the intermediate-risk group were both 49.0 months, while those in the low-risk group were both not reached after median follow up of 65.5 months (Fig. 3A-3B). Median RFS of the high-risk group in the CRS and mCRS model were 10.2 months and 8.3 months, respectively, and those of the intermediate-risk group were 14.2 months and 15.2 months, while those in the low-risk group were both not reached (Supplementary Fig. 4).
When applying CRS and mCRS models to the validation cohort (Fig. 3C-3D) and pooled cohort (Supplementary Fig. 5), the survival distributions showed the same trend as in the training cohort.
Receiver operating characteristic (ROC) analysis for the comparison of CRS and mCRS in prediction ability
The modified Fong score, with a better prognostic discriminatory ability, outperformed the Fong score. Time-dependent ROC analysis displayed that the mCRS model exhibited a better predictive value than the CRS model in the training cohort, both for OS (P = 0.008) and RFS (P = 0.016). In the mCRS model, the C-index of 5-year OS probability forecast was 0.641±0.026, while the c-index of 24-months RFS probability forecast was 0.672±0.024 (Fig. 4). In validation cohort, mCRS was numerically superior to the CRS in predicting OS evaluated by area under curve (AUC) in the time-dependent ROC analysis, though the P-value did not reach statistical significance (Supplementary Fig. 6). Particularly, compared to CRS, mCRS identified a relatively higher proportion of patients in the high-risk group (the training cohort, 9.8% vs 7.1%; the validation cohort, 5.8% vs 2.2%). Moreover, the median OS of patients in the high-risk group was numerically more discriminative in mCRS than in CRS (the training cohort, 21.8 vs 26.8 months; the validation cohort, 27.6 vs 44.8 months) (Table 4). These results suggest that replacing CEA with LDH in the CRS scoring system demonstrated a better accuracy.
Association of LDH levels and immune/inflammation‑related indexes
In an exploratory analysis, it is interesting to note that LDH levels varied with a set of immune/inflammatory factors (Fig. 5). Specifically, patients with elevated LDH levels had higher preoperative neutrophil counts (P = 0.031), higher C-reaction protein (CRP) levels (P < .001), and lower lymphocyte counts (P = 0.022) than patients with normal LDH levels. Consequently, patients with elevated LDH levels also had a lower lymphocyte-to-monocyte ratio (LMR; P < .001) and lymphocyte-to-neutrophil ratio (LNR; P < .001). On the contrary, LDH level was not associated with preoperative total white blood cell counts or monocyte counts.