The clinical characteristics of included patients. From January 2012 to December 2016, a total of 698 patients who fulfilled with our eligible criteria were included in this study. Among them, 482 patients who underwent curative LR before January 2015 were allocated into the training cohort, the later 216 patients were stratified into internal prospective validation cohort. The detailed information of these two cohorts were listed in Table 1. Except the neutrophil-to-lymphocyte ratio (NLR) (P = 0.007), platelet-to-lymphocyte ratio (PLR) (P = 0.011) and the presence of cirrhosis (P = 0.016), other baseline and clinicopathologic data were comparable between the training and validation cohorts. The median follow-up for all included patients was 36 months (range, 1–78 months). The incidence of ER was observed in 265 (55.0%) and 120 (55.6%) patients in the training and validation cohorts, respectively. For all included patients, the 1, 3 and 5-year OS rates in ER group were 66.47%, 23.85% and 13.18%, which was dramatically lower than that in late recurrence group with the 1, 3 and 5-year survival rate of 99.68%, 96.75% and 83.30%, respectively (P < 0.0001) (Fig. 1).
Table 1
The baseline and clinical characteristics of HCC patients in the training and validation cohorts
Clinical parameters
|
Training cohort (n = 482)
|
Validation cohort (n = 216)
|
P
|
Gender (male/famale)
|
404 (83.8%)/78 (16.2%)
|
185 (85.6%)/31 (14.4%)
|
0.538
|
Age, mean (SD)
|
49.58 (12.52)
|
51.68 (10.83)
|
0.698
|
HBsAg (positive/negative)
|
417 (86.5%)/65 (13.5%)
|
187 (87.0%)/29 (13.0%)
|
0.868
|
HBV-DNA (≥ 103/<103 copies/ml)
|
261 (54.2%)/221 (45.8%)
|
107 (49.5%)/109 (50.5%)
|
0.282
|
HBeAg (positive/negative)
|
97 (20.1%)/385 (79.9%)
|
45 (21.0%)/171 (79.0%)
|
0.323
|
AFP (< 20/20–400/>400 ng/mL)
|
157 (32.6%)/119 (24.7%)/206 (42.7%)
|
68 (31.6%)/54 (25.1%)/94 (43.3%)
|
0.966
|
NEU (> 3.56/≤3.56×109/L)
|
228 (47.4%)/254 (52.6%)
|
99 (46.0%)/117 (54.0%)
|
0.741
|
LYM (> 1.1/≤1.1×109/L)
|
362 (75.1%)/120 (24.9%)
|
171 (79.1%)/45 (20.9%)
|
0.255
|
PLT (> 100/≤100×109/L)
|
351 (72.8%)/131 (27.2%)
|
143 (66.2%)/73 (33.8%)
|
0.076
|
NLR (> 3/≤3)
|
153 (31.8%)/329 (68.2%)
|
47 (21.9%)/169 (78.1%)
|
0.007
|
PLR (> 111/≤111)
|
180 (37.3%)/302 (62.7%)
|
59 (27.4%)/157 (72.6%)
|
0.011
|
TBIL (> 28/≤28 umol/L)
|
15 (3.1%)/467 (96.9%)
|
8 (3.7%)/208 (96.3%)
|
0.689
|
ALT (> 50/≤50 IU/L)
|
161 (33.4%)/321 (66.6%)
|
76 (35.2%)/140 (64.8%)
|
0.646
|
AST (> 40/≤40 IU/L)
|
216 (44.8%)/266 (55.2%)
|
101 (46.8%)/115 (53.2%)
|
0.633
|
ALB (> 40/≤40g/L)
|
308 (63.9%)/174 (36.1%)
|
131 (60.6%)/85 (39.4%)
|
0.411
|
GGT (> 60/≤60IU/L)
|
240 (49.8%)/242 (50.2%)
|
114 (52.8%)/102 (47.2%)
|
0.466
|
PT (> 12.8/≤12.8 s)
|
119 (24.7%)/363 (75.3%)
|
39 (17.9%)/177 (82.1%)
|
0.344
|
INR (> 1.15/≤1.15)
|
101 (21.0%)/381 (79.0%)
|
50 (23.1%)/166 (76.9%)
|
0.515
|
Fib (> 2/≤2 g/L)
|
401 (83.2%)/81 (16.8%)
|
173 (79.9%)/43 (20.1%)
|
0.338
|
Tumor diameter (≤ 5/5–10/>10 cm)
|
219 (45.4%)/199 (41.2%)/65 (13.4%)
|
105 (48.4%)/89 (41.2%)/29 (13.4%)
|
0.531
|
Tumor number (1/2/3)
|
402 (83.4%)/58 (12.0%)/22 (4.6%)
|
170 (78.6%)/37 (17.2%)/9 (4.2%)
|
0.284
|
BCLC stage (A/B)
|
409 (84.9%)/73 (15.1%)
|
178 (82.3%)/38 (17.7%)
|
0.399
|
Cirrhosis (present/absent)
|
296 (61.4%)/186 (38.6%)
|
153 (70.8%)/63 (29.2%)
|
0.016
|
Differentiation (I + II/III + IV)
|
279 (57.9%)/203 (42.1%)
|
123 (56.9%)/93 (43.1%)
|
0.816
|
MVI (present/absent)
|
204 (42.3%)/278 (57.7%)
|
86 (39.8%)/130 (60.2%)
|
0.534
|
Statelite lesion (present/absent)
|
72 (14.9%)/410 (85.1%)
|
31 (14.4%)/185 (85.6%)
|
0.840
|
Resection (anatomic/non-anatomic)
|
250 (51.9%)/232 (48.1%)
|
110 (50.9%)/106 (49.1%)
|
0.818
|
Bold numbers indicate statistical significance |
SD, standard deviation; HBsAg, hepatitis B surface antigen; HBV-DNA, hepatitis B virus deoxyribonucleic acid; HBeAg, hepatitis B e antigen; AFP, alpha-fetoprotein; NEU, neutrophil; LYM, lymphocyte; PLT, platelet; NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; TBIL, total bilirubin; ALT, alanine transaminase; AST, aspartate aminotransferase; ALB, albumin; GGT, gamma-glutamyl transpeptidase; PT, prothrombin time; s, second; INR, international normalized ratio; Fib, fibrinogen; BCLC, Barcelona Clinic Liver Cancer staging system; MVI, microvascular invasion. |
The independent predictors for early recurrence. As shown in Table 2, univariate logistic analyses revealed that numerous variables including gender, age, hepatitis B surface antigen (HBsAg), HBV-DNA, hepatitis B e antigen (HBeAg), AFP, neutrophil (NEU), platelet (PLT), NLR, PLR, aspartate aminotransferase (AST), gamma-glutamyl transpeptidase (GGT), tumor diameter, tumor number, Barcelona Clinic Liver Cancer (BCLC) stage, differentiation, MVI, satellite lesions and resection type were significantly associated with the ER in training cohort. Subsequent multivariate analyses further revealed that four preoperative risk factors including age (odds ratio (OR) = 0.894, 95% confidence interval (CI): 0.987–0.990, P < 0.001), AFP level (OR = 1.328, 95% CI: 1.077–1.639, P = 0.008), tumor diameter (OR = 2.209, 95% CI: 1.645–2.967, P < 0.001) and tumor number (OR = 2.556, 95% CI: 1.474–4.434, P = 0.001), and five postoperative risk factors including age (OR = 0.981, 95% CI: 0.975–0.987, P < 0.001), tumor diameter (OR = 1.943, 95% CI: 1.438–2.624, P < 0.001), tumor number (OR = 1.826, 95% CI: 1.024–3.255, P = 0.041), differentiation (OR = 1.508, 95% CI: 1.059–2.358, P = 0.025) and MVI (OR = 2.904, 95% CI: 1.914–4.405, P < 0.001) were significantly associated with ER of HCC patients without macrovascular invasion after curative LR (Table 3).
Table 2
Univariate logistic analysis on clinical parameters in predicting early recurrence in the training cohort.
Clinical parameters
|
OR (95% CI)
|
P
|
Gender (male/famale)
|
1.27 (1.043–1.545)
|
0.017
|
Age (> 60/≤60 years)
|
0.596 (0.480–0.740)
|
< 0.001
|
HBsAg (positive/negative)
|
1.262 (1.075–1.482)
|
0.004
|
HBV-DNA (≥ 103/<103 copies/ml)
|
1.529 (1.164–2.010)
|
0.002
|
HBeAg (positive/negative)
|
1.771 (1.170–2.681)
|
0.007
|
AFP (< 20/20–400/>400 ng/ml)
|
1.337 (1.171–1.527)
|
< 0.001
|
NEU (> 3.56/≤3.56×109/L)
|
1.621 (1.241–2.117)
|
< 0.001
|
LYM (> 1.1/≤1.1×109/L)
|
1.221 (0.993–1.502)
|
0.059
|
PLT (> 100/≤100×109/L)
|
1.472 (1.189–1.821)
|
< 0.001
|
NLR (> 3/≤3)
|
2.000 (1.429–2.799)
|
< 0.001
|
PLR (> 111/≤111)
|
1.687 (1.247–2.282)
|
0.001
|
TB (> 28/≤28 umol/L)
|
2.000 (0.684–5.581)
|
0.206
|
ALT (> 50/≤50 IU/L)
|
1.368 (1.000-1.870)
|
0.05
|
AST (> 40/≤40 IU/L)
|
1.602 (1.218–2.108)
|
0.001
|
ALB (> 40/≤40g/L)
|
1.184 (0.947–1.482)
|
0.139
|
GGT (> 60/≤60IU/L)
|
1.697 (1.306–2.205)
|
< 0.001
|
PT (> 12.8/≤12.8 s)
|
1.164 (0.812–1.668)
|
0.41
|
INR (> 1.15/≤1.15)
|
1.244 (0.841–1.842)
|
0.275
|
Fib (> 2/≤2 g/L)
|
1.079 (0.873–1.335)
|
0.482
|
Tumor size (≤ 5/5–10/>10 cm)
|
1.791 (1.460–2.198)
|
< 0.001
|
Tumor number (1/2/3)
|
1.275 (1.108–1.466)
|
0.001
|
BCLC stage (A/B)
|
3.056 (1.795–5.203)
|
< 0.001
|
Cirrhosis (present/absent)
|
1.193 (0.949–1.499)
|
0.131
|
Differentiation (I + II/III + IV)
|
1.819 (1.365–2.426)
|
< 0.001
|
MVI (present/absent)
|
2.923 (2.133–4.005)
|
< 0.001
|
Statelite lesion (present/absent)
|
2.429 (1.461–4.037)
|
0.001
|
Resection (anatomic/non-anatomic)
|
1.475 (1.146–1.899)
|
0.003
|
Bold numbers indicate statistical significance |
OR, odds ratio; CI, confidence interval; HBsAg, hepatitis B surface antigen; HBV-DNA, hepatitis B virus deoxyribonucleic acid; HBeAg, hepatitis B e antigen; AFP, alpha-fetoprotein; NEU, neutrophil; LYM, lymphocyte; PLT, platelet; NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; TBIL, total bilirubin; ALT, alanine transaminase; AST, aspartate aminotransferase; ALB, albumin; GGT, gamma-glutamyl transpeptidase; PT, prothrombin time; s, second; INR, international normalized ratio; Fib, fibrinogen; BCLC, Barcelona Clinic Liver Cancer staging system; MVI, microvascular invasion. |
Table 3
Multivariate logistic analysis on clinical parameters in predicting early recurrence in the training cohort.
Clinical parameters
|
OR (95% CI)
|
P
|
Preoperative
|
|
|
Age (> 60/≤60 years)
|
0.984 (0.987–0.990)
|
< 0.001
|
AFP (< 20/20–400/>400 ng/ml)
|
1.328 (1.077–1.639)
|
0.008
|
Tumor diameter (≤ 5/5–10/>10 cm)
|
2.209 (1.645–2.967)
|
< 0.001
|
Tumor number (1/2/3)
|
2.556 (1.474–4.434)
|
0.001
|
Postoperative
|
|
|
Age (> 60/≤60 years)
|
0.981 (0.975–0.987)
|
< 0.001
|
Tumor diameter (≤ 5/5–10/>10 cm)
|
1.943 (1.438–2.624)
|
< 0.001
|
Tumor number (1/2/3)
|
1.826 (1.024–3.255)
|
0.041
|
Differentiation (I + II/III + IV)
|
1.580 (1.059–2.358)
|
0.025
|
MVI (present/absent)
|
2.904 (1.914–4.405)
|
< 0.001
|
OR, odds ratio; CI, confidence interval; AFP, alpha-fetoprotein; MVI, microvascular invasion. |
Construction of pre- and postoperative nomograms for predicting the early recurrence. Based on the results of multivariate analyses in training cohort, two nomograms for pre- or postoperatively predicting the ER were generated, respectively (Fig. 2). The C-indices of pre- and postoperative nomograms in training cohort were 0.712 (95% CI: 0.666–0.758, P < 0.001) and 0.850 (95% CI: 0.781–0.919, P < 0.001), respectively. The calibration plots showed ideal agreement on the incidence of ER between the prediction by the nomograms and actual observation in follow-up (Fig. 2).
For clinical use of present nomograms, as shown in Fig. 2, the projection of each variable on the point scale meant unique score of each variable. After summing the scores of all variables, total points for each patient could be calculated. Then the projection of total points on the probability scale represented the individual probability for ER.
Validation of the prediction models. Firstly, the performance of pre- and post operative nomograms was validated by a internal prospective validation cohort. The pre- and postoperative total points for each patient in validation cohort were formulated using above two nomograms, respectively. Then the pre- or postoperative total points was treated as a new risk factor to calculate the C-indices and produce calibration curves of ER. The results showed that the C-indices for pre- and postoperative prediction of ER in validation cohort were 0.754 (95% CI: 0.690–0.818, P < 0.001) and 0.857 (95% CI: 0.750–0.949, P < 0.001), respectively. The calibration curves also showed ideal consistency between prediction and observation in the probability of ER (Fig. 3).
In addition, the predictive performance of present nomograms was evaluated by ROC curves (Fig. 4). In training cohort, the area under the ROC curves (AUC) of pre- and postoperative nomograms were 0.721 (95% CI: 0.684–0.759, P < 0.001) and 0.848 (95% CI: 0.814–0.883, P < 0.001) respectively; in internal prospective validation cohort, the AUC of pre- and postoperative nomograms were 0.754 (95% CI: 0.690–0.817, P < 0.001) and 0.844 (95% CI: 0.790–0.897, P < 0.001), respectively, which were comparable with the C-indices of nomograms. These results indicated that present nomograms have good performance in predicting the ER for patients with HCC without macrovascular invasion after curative LR.
The predictive ability of nomograms. As shown in Table 4, the optimal cut-off values of the pre- and postoperative total nomogram scores were 88 (range: 4-284) and 110 (range: 5-356), respectively. For preoperative model, the sensitivity, specificity, positive predictive value, negative predictive value, positive likelihood ratio, and negative likelihood ratio in distinguishing ER were 0.611, 0.716, 0.704, 0.587, 2,151 and 0.543 in the training cohort, and 0.730, 0.677, 0.733, 0.674, 2.260 and 0.399 in the validation cohort. With respect for postoperative model, the sensitivity, specificity, positive predictive value, negative predictive value, positive likelihood ratio, and negative likelihood ratio in distinguishing ER were 0.706, 0.802, 0.793, 0.724, 3.564 and 0.367 in the training cohort, and 0.679, 0.800, 0.764, 0.699, 3.394 and 0.401 in the validation cohort.
Table 4
Predictive ability of the optimal cut off values on the risk of early recurrence
variables
|
Preoperative nomogram
|
Postoperative nomogram
|
Training cohort (482)
|
Validation cohort (216)
|
Training cohort (482)
|
Validation cohort (216)
|
AUC
|
0.721 (0.684–0.759)
|
0.754 (0.690–0.817)
|
0.848 (0.814–0.883)
|
0.844 (0.790–0.897)
|
Cut-off score
|
88
|
88
|
110
|
110
|
Sensitivity
|
0.611 (0.567–0.654)
|
0.642 (0.578–0.706)
|
0.706 (0.665–0.747)
|
0.679 (0.617–0.741)
|
Specificity
|
0.716 (0.676–0.756)
|
0.76 (0.703–0.817)
|
0.802 (0.766–0.838)
|
0.800 (0.746–0.853)
|
Positive predictive value
|
0.704 (0.663–0.745)
|
0.77 (0.714–0.826)
|
0.793 (0.756–0.829)
|
0.764 (0.707–0.821)
|
Negative predictive value
|
0.587 (0.543–0.631)
|
0.629 (0.565–0.693)
|
0.724 (0.684–0.764)
|
0.699 (0.638–0.760)
|
Positive likelihood ratio
|
2.151 (2.011–2.291)
|
2.675 (2.393–2.957)
|
3.564 (3.294–3.834)
|
3.394 (3.014–3.774)
|
Negative likelihood ratio
|
0.543 (0.499–0.587)
|
0.471 (0.404–0.538)
|
0.367 (0.324–0.411)
|
0.401 (0.336–0.466)
|
CI, confidence interval. |