Basic Research Of The Entire Cohort
As is summarized in Table 1, a total of 337 patients with GBC who received curative surgery were identified. Among them, ninety-seven patients were detected with preoperative fibrinogen abnormalities (preoperative fibrinogen > 4g/L). Patients with an elevated level of preoperative fibrinogen were generally in a more advanced stage, sharing a significantly higher percentage of T3-T4 (84.6% vs 57.3%, P < 0.0001) or TNM III-IV disease (85.7% vs 59.8%, P < 0.0001). Significantly higher incidences of preoperative CA199 level (P < 0.0001) and preoperative obstructive jaundice (P = 0.018) were detected in patients with hyper-fibrinogen. Moreover, liver invasion (P < 0.0001) as well as neural invasion (25.3% vs 17.9%, P = 0.089) were also more frequently detected in those with hyper-fibrinogen. Compared with those with hypo-fibrinogen, patients with hyper-fibrinogen more frequently received major hepatectomies (P = 0.04) and combined multi-visceral resections (P = 0.001) while the R0 rate was conversely lower (86.8% vs 93.5%, P = 0.044). Patients with hyper-fibrinogen had a much longer postoperative hospital stay (P = 0.017) and shared a significantly higher rate of mortalities within 90 days after surgery (P = 0.006). The overall recurrence rate (P < 0.0001) as well as the recurrence rate within six months after surgery (P < 0.0001) was significantly higher in patients with hyper-fibrinogen. Patients with hyper-fibrinogen shared a significantly worse OS (median survival time: 19 vs 39 months, P < 0.0001) (Fig. 1A) as well as a DFS (median survival time: 16 vs 33 months, P < 0.0001) (Fig. 1B). Considering the inherent bias between two groups, a PSM analysis was performed via controlling the following factors: age, sex, postoperative chemotherapy, T stage, Node stage, and surgical margins. After PSM, as is summarized in Table 1, eighty-seven patients with hyper-fibrinogen and eighty-seven patients with hypo-fibrinogen were identified (ratio: 1:1). After PSM, all these factors reached a balance without significant bias while the recurrence rate within six months after surgery was still higher in patients with hyper-fibrinogen (25.3% vs 12.6%, P = 0.018). However, comparable OS (P = 0.4995) (Fig. 1C) and DFS (P = 0.7893) (Fig. 1D) were acquired.
Table 1
Baseline characteristics of all patients with GBC according to the level of preoperative fibrinogen before and after PSM
Variable, n(%) | Preoperative plasma fibrinogen(Before PSM) | P value | Preoperative plasma fibrinogen(After PSM) | P value |
> 4g/L (n = 91) | ≤ 4 g/L (n = 246) | > 4g/L (n = 87) | ≤ 4 g/L (n = 87) |
Age > 60 | 52(57.1%) | 122(49.6%) | 0.134 | 49(56.3%) | 48(55.2%) | 0.500 |
Sex, female | 54(59.3%) | 175(71.1%) | 0.028 | 53(60.9%) | 53(60.9%) | 0.562 |
Preoperative CA199 > 37U/ml | 47(51.6%) | 71(28.9%) | < 0.0001 | 44(50.6%) | 39(44.8%) | 0.272 |
Preoperative Jaundice | 18(19.8% | 25(10.2%) | 0.018 | 15(17.2%) | 13(14.9%) | 0.418 |
Preoperative biliary drainage (PTCD or ENBD) | 1(1.1%) | 6(2.4%) | 0.394 | 1(1.1%) | 4(4.6%) | 0.184 |
Bile duct resection | 23(25.3%) | 45(18.3%) | 0.104 | 19(21.8%) | 22(25.3%) | 0.361 |
Major hepatectomy | 16(17.6%) | 24(9.8%) | 0.04 | 15(17.2%) | 12(13.8%) | 0.338 |
Major vascular resection and reconstruction | 7(7.7%) | 9(3.7%) | 0.109 | 6(6.9%) | 6(6.9%) | 0.617 |
Combined adjacent organ resections | 21(23.1%) | 23(9.3%) | 0.001 | 20(23.0%) | 12(13.8%) | 0.085 |
Negative margins | 79(86.8%) | 230(93.5%) | 0.044 | 79(90.8%) | 76(87.4%) | 0.314 |
Liver invasion | 59(64.8%) | 87(35.4%) | < 0.0001 | 55(63.2%) | 44(50.6%) | 0.063 |
Neural invasion | 23(25.3%) | 44(17.9%) | 0.089 | 20(23.0%) | 28(32.2%) | 0.117 |
Lymph-vascular invasion | 13(14.3%) | 34(13.8%) | 0.519 | 12(13.8%) | 22(25.3%) | 0.042 |
Lymph node metastasis | 39(42.9%) | 92(37.4%) | 0.215 | 38(43.7%) | 38(43.7%) | 0.561 |
T3-4, (8th AJCC) | 77(84.6%) | 141(57.3%) | < 0.0001 | 73(83.9%) | 74(85.1%) | 0.500 |
TNM III-IV, (8th AJCC) | 78(85.7%) | 147(59.8%) | < 0.0001 | 74(85.1%) | 75(86.2%) | 0.742 |
Low to moderate differentiation status | 79(86.8%) | 195(79.3%) | 0.075 | 75(86.2%) | 74(85.1%) | 0.500 |
Postoperative chemotherapy | 16(17.6%) | 69(28.0%) | 0.04 | 16(18.4%) | 21(24.1%) | 0.229 |
Postoperative hospital stay ≥ 7 days | 70(76.9%) | 158(64.2%) | 0.017 | 66(75.9%) | 53(60.9%) | 0.025 |
Morbidities | 19(20.9%) | 36(14.6%) | 0.114 | 18(20.7%) | 19(21.8%) | 0.500 |
Mortalities within 90 days | 5(5.5%) | 1(0.4%) | 0.006 | 4(4.6%) | 0(0%) | 0.060 |
Overall Recurrence | 70(81.4%) | 135(55.1%) | < 0.0001 | 66(79.5%) | 68(78.2%) | 0.489 |
Recurrence within 6 months | 25(29.1%) | 24(9.8%) | < 0.0001 | 22(26.5%) | 11(12.6%) | 0.018 |
GBC: gallbladder carcinoma; PTCD: percutaneous transhepaticcholangial drainage; ENBD: endoscopic nasobiliary drainagel; AJCC: American Joint Committee on Cancer; PSM: propensity score matching |
The Diagnostic Performance For Survival Of Preoperative Fibrinogen
Our previous research has indicated that preoperative fibrinogen was closely correlated with GBC prognosis. Hence, the diagnostic performance of preoperative fibrinogen in long-term survival in patients with GBC was furtherly explored. The area under curve (AUC) of 1-, 3-, and 5-year survival rate was 0.68, 0.68, and 0.66 respectively (Fig. 2A), which was similar to the findings reported by Yang ZY et al(13). The diagnostic performance of CA199 was also analyzed at the same time and the AUC was 0.74, 0.69, and 0.67 respectively (Fig. 2B). Obviously, the diagnostic performance of CA199 was superior to preoperative fibrinogen. However, a combination of CA199 and fibrinogen also showed promising diagnostic performance (AUC: 0.73) (Fig. 2C).
Lasso Regression-based Predictive Model
The Lasso model was developed with all potential prognostic factors, including postoperative chemotherapy, age, sex, preoperative CA199, preoperative fibrinogen, surgical margin, node metastasis, neural invasion, lymph-vascular invasion, liver invasion, tumor differentiation status, and T stage. As presented in Fig. 3A and 3B, after Lasso regression, a total of eight variables (postoperative chemotherapy, age, preoperative fibrinogen, surgical margin, node metastasis, lymph-vascular invasion, liver invasion, and T stage) were finally selected for model development and their coefficients were summarized in Table S3. Subsequently, the predictive accuracy of Lasso model was tested via a scatter diagram (Fig. 3C) and a ROC curve (Fig. 3D). A ROC curve revealed an extremely high predictive accuracy of Lasso model (AUC: 0.9). We also performed a time-dependent ROC curve to evaluate the risk score-based predicative accuracy in terms of 1-, 3-, and 5-year survival rates after Lasso regression and the results were much more satisfactory (AUC at 1-year: 0.81, AUC at 3-year: 0.88, AUC at 5-year: 0.9) (Fig. 3E). Moreover, based on the risk score, patients were roughly divided into the high and low risk groups. Patients belonged to the high risk group had a significantly worse OS versus those in the low risk group (P < 0.001) (Fig. 3F).
Cox Regression-based Predictive Model And Nomogram
To identify the optimal parameters in the Nomogram, we performed a backward stepwise selection procedure (AIC) to find the applicable predictors. A total of six parameters were identified via stepwise selection, including fibrinogen, age, surgical margin, node metastasis, postoperative chemotherapy, and T stage (lowest AIC value: 1914.41). Next, the common predictors in the Lasso regression and AIC were adopted for model establishment. However, in the subsequent proportional hazard test, the P value of postoperative chemotherapy was lower than 0.05 and therefore was ruled out. Finally, a total of five predictors were incorporated for the final cox regression-based predictive model. Their coefficients were summarized in Table S3. The individualized Nomogram was presented in Fig. 4A. To evaluate the predictive accuracy of the final model, its primary C-index was calculated as 0.77. However, a bootstrap validation was also performed (300 resamples) and its predictive accuracy reached 0.82 with a Kappa value with 0.62. When the predictive accuracy of bootstrap validation was converted to a specific value of AUC, the result was much more satisfactory (AUC: 0.88). Additionally, calibration curves were performed in terms of 1-, 3-, and 5-year survival rates (Fig. 4B-4D). A forest-plot was also performed to reveal the percentage of contribution of five predictors and the result revealed that T stage had the most powerful impact (Fig. 4E).
Comparison Of The Predictive Accuracy Among Three Models
In order to compare the predictive accuracy of two model established based on our own cohort with the latest 8th AJCC TNM staging criteria, a time-dependent ROC curve was furtherly performed according to the 8th AJCC TNM staging criteria. As presented in Figure S2, the AUC reached 0.85, which showed the high predictive accuracy of 8th AJCC TNM staging criteria. The predictive accuracy of three models were compared according to the AUC. Obviously, Lasso model shared the most favorable predictive accuracy (AUC: 0.9) while the latest 8th AJCC TNM staging system shared the worst predictive accuracy (AUC: 0.85) (Table 2).
Table 2
The predictive accuracy of Lasso, Cox-based nomogram, and the lastest 8th AJCC TNM criteria
| Three models |
Prognostic factors | 8th AJCC-based TNM model | Lasso-based predictive model | Cox regression-based Nomogram |
AJCC TNM stage (I/II/III/IV) | Selected | —— | —— |
T stage (T1-2/T3-4) | —— | Selected | Selected |
Node status (N+/N-) | —— | Selected | Selected |
Liver invasion (+/-) | —— | Selected | —— |
Fibrinogen | —— | Selected | Selected |
Postoperative Chemotherapy (+/-) | —— | Selected | —— |
Age | —— | Selected | Selected |
Surgical margin (+/-) | —— | Selected | Selected |
Neural invasion (+/-) | —— | —— | —— |
Lymphvascular invasion (+/-) | —— | Selected | —— |
Predictive accuracy (AUC) | 0.85 | 0.90 | 0.88 |
AJCC: American Joint committee on Cancer; AUC: area under curve |
Meta-analysis
After a comprehensive literature researching, a total of four studies were finally included (Table S1). With our own cohort incorporated, pooled results revealed that GBC patients with hyper-fibrinogen shared a significantly higher percentage of T3-T4 or III-IV disease (P < 0.00001) and therefore shared a significantly worse OS than those with hypo-fibrinogen (P < 0.00001) (Table S2 and Figure S1).