We retrospectively analyzed patients who received curative-intent radical cholecystectomy at West China Hospital, Sichuan University, between September 2010 and September 2020 from a retrospective electronic database. Having ruled out patients who had lost follow-up as well as those without adequate clinical data, only patients with pathologically confirmed GBASC (Figure 1A) and pure GBAC (Figure 1B) were finally identified. The diagnostic and inclusion criteria for GBASC in our cohort is a percentage of squamous components higher than 30%, which is consistent with the diagnostic standard for ASC in pancreatic cancer in the 2010 WHO criteria (5). Tumor staging was based on the eighth American Joint Committee on Cancer (AJCC) criteria.

All patients received a standard radical cholecystectomy, including a gallbladder incision, a partial hepatectomy, and a regional or more extended lymph node dissection. The extent of liver resection and lymphadenectomy were mainly driven by surgeons depending on the preoperative imaging, gross intraoperative findings, and intraoperative frozen biopsy. The extent of liver resection can be roughly divided into two categories: minor resection (wedge resection, SIVB + V resection) and major resection (extended hemi-hepatectomy, right or tri-segmentectomy). Lymph node dissection was performed in all patients, including regional (lymph nodes along with the cystic duct, hilum of the liver, and hepatoduodenal ligament) or extended (lymph nodes along with the pancreatic head, duodenum, and celiac artery) dissection. Node status (N0, positive lymph node = 0; N1, positive lymph nodes ≤3; and N2, positive lymph nodes >3) were evaluated based on the latest 8th AJCC criteria. When there were no enlarged lymph nodes detected pre- or intra-operatively, the surrounding fibro-fatty tissue would be cleared and submitted for pathological evaluation. Bile duct excisions were performed when obvious bile duct or hepatoduodenal ligament invasion was detected or there was intraoperatively pathologically confirmed positive cystic duct margin. When direct infiltrations of adjacent organs and structures were detected, concurrent multi-visceral resections were also performed to achieve a clear margin. R0 resection indicates a completely clear margin with no microresidual tumor. R1 resection indicates a relatively clear margin with no gross tumor or that the tumor is closely adjacent to the margin. R2 resection indicates a macroscopic residual tumor, which a specialized pathologist confirms.

Patients were regularly followed up via blood tumor biomarkers and abdominal radiological examination to detect disease progress every 1 to 2 months in the first year after surgery and every 3 to 6 months thereafter.

All relevant preoperative, intraoperative, and postoperative clinico-pathological data were retrospectively collected and recorded within a database. Preoperative details included sex (male vs female), age (≥60 vs <60), preoperative CA199 level (≥37 U/ml vs <37 U/ml), preoperative jaundice, and preoperative biliary drainage. Intraoperative details included the extent of liver resection (minor vs major), the number of harvested lymph nodes and positive lymph nodes, bile duct resection, portal vein or hepatic artery reconstruction, combined multi-visceral resection, and the R0 resection rate. The size of the resected tumor was also analyzed when provided. Postoperative details include pathologically confirmed tumor biological features (liver invasion, neural invasion, lymph-vascular invasion, node metastasis, and tumor differentiation status), tumor stage, morbidities, mortalities, postoperative chemotherapy, overall recurrence rate, and recurrence rate within 6 months.

IBM SPSS version 22.0 (IBM SPSS, Chicago, IL, USA), R Studio 3.6.3 software, and Graph-Pad Prism 7 were used for statistical analysis. Continuous data are recorded as medians (or ranges). Categorical data are recorded as numbers (percentages). Categorical variables were evaluated via Chi-squared and Fisher’s exact tests. Continuous variables were analyzed using a non-parametric test. Overall survival (OS) was defined as the period from the date of receiving radical surgery to the date of death or last follow-up. Disease-free survival (DFS) was defined as the living period from the date of receiving surgery to the date of recurrence or progression. Kaplan–Meier curves were used for evaluating survival differences. The Cox-proportional hazards model was used to create a multivariate model for independent prognostic factors for survival, which were presented was the Hazard ratio (HR) with its 95% confidence interval (CI). P-values lower than 0.05 indicated the existence of statistical significance. To evaluate the solo impact of GBASC on the entire cohort, propensity score matching (PSM) was performed via controlling age, sex, and various independent prognostic factors that significantly influenced prognosis (R Studio 3.6.3 software, Ratio 1:1, standard deviation 0.1).

For further validation of the differences between GBASC/SC and GBAC, we searched the following databases, including PubMed, Embase, and the Cochrane Library. RevMan5.3 software was applied in the data analysis. Eligible studies were restricted to comparative studies between GBASC/SC and GBAC. The following terms were used: gallbladder, adenosquamous, squamous, and carcinoma. Inclusion criteria: published English literature; comparative studies provided relevant survival information. Exclusion criteria: abstracts, meetings, conferences, letters, and case reports; studies that shared the same database (the study with the largest cohort would be incorporated); studies that failed to provide survival information. The HR with its 95% CI was applied in the survival analysis, and Tierney’s method would be used for a rough estimate of HR when not directly provided (11). The odds ratio (OR) was applied to dichotomous variables, and the weighted mean difference (WMD) was used for continuous variables. The method by Luo et al. was used to have a rough estimate of means and standard deviations if they were not directly provided (12). A random-effects model would be used if the heterogeneity was >50%; otherwise, a fixed-effects model would be applied.

As Figure 2 illustrates, according to the inclusion and exclusion criteria, 304 patients with post-surgical, pathologically confirmed GBC were identified, including 34 patients with GBASC and 270 patients with conventional GBAC. Ninety-nine patients were male, and 205 were female. The median age of the entire cohort was 60 (30–82). All patients received curative-intent surgery, and cases with distant metastasis were ruled out (Table 1).

The clinico-pathological features of the entire cohort, grouped by pathological types, are summarized in Table 1. Regarding preoperative details, comparable sex (male vs female), age (≥60 and <60), preoperative jaundice (P = 0.278), and preoperative biliary drainage (P = 0.568) were observed between the two groups. However, patients with GBASC had a significantly higher incidence of preoperative CA199 ≥37 U/ml (P <0.0001). Regarding intraoperative details, patients with GBASC shared comparable incidences of receiving bile duct resection (P = 0.554), major hepatectomy (P = 0.350), and major vascular reconstruction (portal vein or hepatic artery) (P >0.05). Combined multi-visceral resections were more frequently performed in patients with GBASC, although the result did not reach a statistical difference (20.6% vs 11.5%, P = 0.111). A comparable R0 resection rate was also acquired (88.2% vs 91.9%, P = 0.328). Regarding postoperative pathological outcomes, patients with GBASC had comparable incidences of neural invasion (P = 0.124), lymph-vascular invasion (P = 0.455), node metastasis (P = 0.822), and well-to-moderate differentiation status (P = 0.065). A comparable number of harvested lymph nodes as well as the number of positive lymph nodes were also acquired. However, the incidence of liver invasion was significantly higher in patients with GBASC (70.6% vs 37.8%, P <0.0001). Besides, the percentage of patients with T3–4 disease (91.2% vs 59.3%, <0.0001) or III–IV disease (91.2% vs 62.2%, P = 0.003) was significantly higher in patients with GBASC. Patients with GBAC received postoperative chemotherapy more frequently than patients with GBASC (27.4% vs 14.7%, P = 0.075). Comparable incidences of overall morbidities (P = 0.173) and mortalities (P = 0.788) were acquired between two groups. Two patients died peri-operatively due to an abdominal pseudoaneurysm rupture. The overall recurrence rate (79.4% vs 57%, P = 0.010) and recurrence rate within 6 months (32.4% vs 9.6%, P = 0.001) were significantly higher in patients with GBASC (Table 1).

The median survival time of the entire cohort (n = 302; two cases of death peri-operatively were excluded) was 31 months, ranging from 2 months to 124 months. The 1-, 3-, and 5-year survival rate of the entire cohort was 85.8%, 40.4%, and 13.9%. Patients with GBASC had a significantly worse OS (median survival time: 15.5 months vs 36 months, P = 0.0002) (Figure 3A) as well as DFS (median recurrence time after surgery: 15 months vs 30 months, P = 0.0002) (Figure 3B) than patients with pure GBAC. Moreover, we also evaluated the survival impact of postoperative adjuvant chemo-radiotherapy in patients with GBAC and GBASC, respectively. Patients with GBAC could benefit from adjuvant chemo-radiotherapy. Patients who received adjuvant chemo-radiotherapy shared a better OS (median survival time: 47 months vs 30 months, P = 0.0132) (Figure 4A) or DFS (median survival time: 41 months vs 26 months, P = 0.0436) (Figure 4B) than those who did not. As for patients with GBASC, only five patients received postoperative adjuvant chemo-radiotherapy and its survival impact seemed to be vague. For patients with GBASC, comparable OS (median survival time: 50 months vs 12 months, P = 0.1022) (Figure 4C) and DFS (median survival time: 36 months vs 8 months, P = 0.0680) (Figure 4D) were obtained between those who received adjuvant chemo-radiotherapy and those who did not.

A Cox-proportional hazards regression model was used to identify the independent prognostic factors in patients with GBC. The results indicated that preoperative CA199 (≥37 U/ml vs <37 U/ml) (P <0.0001), bile duct resection (P <0.0001), major hepatectomy (P = 0.011), combined multi-visceral resection (P <0.0001), R0 resection (P <0.0001), node metastasis (P <0.0001), neural invasion (P <0.0001), lymph-vascular invasion (P <0.0001), liver invasion (P <0.0001), tumor differentiation status (low to moderate vs high) (P <0.0001), T stage (T1–2 vs T3–4, P <0.0001), pathological type (GBASC vs GBAC, P <0.0001), and postoperative adjuvant chemo-radiotherapy (P = 0.002) were prognostic factors for patients with resected GBC. The results of multivariable analysis revealed that R0 resections (P = 0.001), node metastasis (P <0.0001), T stage (T1–2 vs T3–4) (P <0.0001), and postoperative adjuvant chemo-radiotherapy (P <0.0001) were all independent prognostic factors (Table 2).

Considering the impact on survival brought by the inherent bias between two groups, a PSM was performed by controlling the following factors: age, sex, and acquired independent prognostic factors for OS.

After matching, as illustrated in Table 1, 34 patients with GBASC and 87 patients with GBAC were identified. Comparable surgical margins, tumor stage, node status, and the proportion of patients receiving postoperative chemotherapy were collected. However, neural invasion (31% vs 11.8%, P = 0.022) was more frequently detected in patients with GBAC. Moreover, a significantly higher percentage of patients with low to moderate differentiation status was also observed in patients with GBAC (93.1% vs 67.6%, P = 0.001). Comparable overall recurrence rates (P = 0.386) and the recurrence rate within 6 months (P = 0.222) were also determined. Moreover, comparable OS (median survival time: 15.5 months vs 17 months, P = 0.9093) (Figure 3C) and DF (median survival time: 15 months vs 13 months, P = 0.6991) (Figure 3D) were found between patients with GBASC and GBAC.

According to the inclusion criteria mentioned above, six comparative studies were finally incorporated (7–9, 13–15). As summarized in Table S1, with our own cohort incorporated, 1,434 patients with GBASC/SC and 29,767 patients with pure GBAC were included. The primary endpoint of our additional meta-analysis was the survival difference between patients with GBASC/SC and GBAC. As summarized in Table S2, pooled results revealed that patients with GBASC/SC had significantly worse OS (HR = 2.27, 95% CI 1.80 to 2.85, P <0.00001) (Figure S2A) and DFS (HR = 2.76, 95% CI 1.81 to 4.20, P <0.00001) (Figure S2B) versus patients with pure GBAC. Moreover, we also analyzed the inconsistencies in the clinico-pathological features between GBASC/SC and GBAC. As illustrated in Figure S1, patients with GBASC/SC had a significantly larger tumor size (WMD = 1.41, 95% CI 0.44–2.37, P = 0.004) (Figure S1B), a significantly higher incidence of node metastasis (24.9% vs 19.5%, P = 0.02) (Figure S1C), a significantly higher incidence of liver involvement (78.3% vs 61.3%, P = 0.0001) (Figure S1F), and a significantly lower R0 rate (59.1% vs 67.2%, P <0.00001) (Figure S1A). Patients with GBASC/SC were more frequently in an advanced stage that the proportion of patients with T3–4 or III–IV disease were significantly higher in the GBASC/SC group (65.6% vs 52.5%, P <0.0001) (Figure S1G). The incidences of neural invasion (P = 0.76) (Figure S1D) and lymph-vascular invasion (P = 0.55) (Figure S1E) were comparable between the two groups.