Prognostic significance of preoperative serum tumor markers in hepatoid adenocarcinoma of stomach (HAS)

we included patients with HAS who underwent the gastrectomy form October 2009 to July 2021 at the Center of Gastrointestinal Cancer of Peking University Cancer Hospital. All patients enrolled in the study were Chinese. Patients in the cases of the following would be excluded: existing intraperitoneal metastasis cancer (including Peritoneal metastasis and positive cytology rather than suspicious positive cytology) before the operation, not achieving R0 gastrectomy and unavailable pre-treat tumor makers data. All clinical data and materials collection were given informed consent from patients.

All HAS patients in the study were clinically confirmed by gastroscopic biopsy and histological examination. Computed tomography (CT) was also necessary for helping surgeon determine the tumor location and border, identify the lymph node stage and distant metastasis. Serving as an auxiliary diagnostic tool, tumor markers examination was often performed within 2 weeks before the gastrectomy. By the way, AFP was added to the routine tumor markers examination panel to avoid misdiagnosis and missed diagnosis of HAS Since 2016 and CA24–2 was not the routine tumor markers examination so far. Therefore, some AFP and CA24–2 data were absent.

Whether undergoing preoperative treatment (neoadjuvant chemotherapy) was based on the recommendations of The National Comprehensive Cancer Network (NCCN) guideline [10] and patients’ willing. Resectability and surgical resection protocol was formulated by a multidisciplinary team conference after communication with patients. The strategy of postoperative treatment was determined by the pathological evaluation [11]. S-1 and oxaliplatin based chemotherapy regiment was frequently used [11].

The diagnostic criteria of HAS in the study was the pathological “gold standard” that the area of hepatocellular differentiation was found in the primary gastric cancer tissue after gastrectomy, no matter the level of AFP value [6, 12, 13]. Patients would be diagnosed as HAS once the area of hepatocellular differentiation was detected. The two typical pathological pictures were exhibited on Fig. 1 a and Fig. 1 b. The hepatocellular differentiation area (right part) and adenocarcinoma area (left part) was showed in the same pathological section. AFP immunohistochemical stains was used in the histopathological diagnosis in this study (Fig. 1 c and Fig. 1 d). R0 resection was defined as pathologically negative margin≥1 mm [14]. The indicators of OS and RFS were regarded as the major end point. OS refers to the time from first treat (neoadjuvant chemotherapy or surgery) to death caused by any reason or the last follow-up, and RFS refers to the period from gastrectomy to the tumor recurrence or the last follow-up.

The examination of surveillance including abdominal and pelvic CT and tumor markers test would be performed to identify whether the tumor recur during follow-up. The patients would under the first follow-up in the 1 month after the gastrectomy. Then every 3 months follow-up would be performed in the next year. In the second and the third year after the gastrectomy, the follow-up would be conducted every 6 months. Telephone follow-up call was the major form.

The indicators of OS and RFS were regarded as the primary endpoint and the second endpoint respectively. If normally distributed, continuous variables would express as mean ± standard deviation(SD) and be compared by the Student’s t test, otherwise continuous variables would express as medians with interquartile range(IQR) and be compared by the Mann-Whitney U test. As for categorical variables, if they were ranked data, they would be compared using Mann-Whitney U test, otherwise they would be compared by Pearson’s chi-square test or Fisher’s exact test determined by the sample size. The optimal cut-off values of AFP and CA19–9 were determined by X-tile analytical tool [15]. Survival analyses including RFS and OS were estimated by Kaplan-Meier method (Kaplan & Meier, 1958) and the inter-groups difference were judged by log-rank test. If the risk factors were significant statistically in univariate analyses (P < 0.050), they would be recruited in multivariate Cox regression analyses to confirm the independent risk factors influencing RFS and OS with the widely accepted significant factors. The X-tile analyses used X-tile tool software® (Yale university 2003–05, USA; version 3.6.1). Statistical analyses used SPSS® (IBM Corp., USA, version 26) and relative curves were plotted by GraphPad (GraphPad software, USA). If P < 0.050 (or P < 0.100 in multivariate analyses), the differences would be deemed as significant statistically.

This study has received the approval of the Ethics Committee of Peking University Cancer Hospital and Institute. Informed consent was obtained for all participants.

Form October 2009 to July 2021, 159 patients pathologically diagnosed with HAS underwent gastrectomy were enrolled in our study at the Center of Gastrointestinal Cancer of Peking University Cancer Hospital. A total of 20 patients were excluded on the basis of exclusion criteria. Three patients had no enough clinical data, 10 patients with intraperitoneal metastasis cancer before the operation, 5 patients had no pre-treat tumor maker data and 2 patients did not achieve R0 gastrectomy. Ultimately, 139 patients were included in our study. Particularly, 2 out of them lost pre-treat CEA data and CA72–4 data, respectively. 33 out of them failed to test AFP and 56 out of them didn’t test CA24–2 before the surgery. Therefore, in order to take full advantage of the patients’ clinical data, we decided to analyses specific tumor markers using respective patients’ data. (Fig. 2).

We employed X-tile tool to find out the optimal cut-off value of tumor markers.

The optimal cut-off values determined by X-tile tool were the points where differences across groups exhibited the most significant for 3-year OS or RFS. In 106 patients whose AFP data were available, 516 ng/mL was the optimal cut-off value for 3-year OS (maximum χ2 log-rank values 4.841) and RFS (maximum χ2 log-rank values 6.510). In 139 patients whose CA19–9 data were available, 51.3 U/mL was the optimal cut-off value for 3-year OS (maximum χ2 log-rank values 6.796) and RFS (maximum χ2 log-rank values 15.457). In 138 patients whose CEA data were available, 9.95 ng/mL was the optimal cut-off value for 3-year OS (maximum χ2 log-rank values 2.855) and 1.36 ng/mL was the optimal cut-off value for 3-year RFS (maximum χ2 log-rank values 4.961). In 138 patients whose CA72–4 data were available, 4.89 U/mL was the optimal cut-off value for 3-year OS (maximum χ2 log-rank values 0.806) and 3.69 U/mL was the optimal cut-off value for 3-year RFS (maximum χ2 log-rank values 3.622). In 83 patients whose CA24–2 data were available, 9.1 U/mL was the optimal cut-off value for 3-year OS (maximum χ2 log-rank values 1.218) and 11 U/mL was the optimal cut-off value for 3-year RFS (maximum χ2 log-rank values 3.095). In addition, we defined high-level as the value above or equal to the cut-off point and low-level as the value below the point. (Fig. 3).

Of 106 patients for AFP analyses, 95 cases fell into low AFP group and 11 fell into high AFP group. Beyond expectations, the differences across two groups were not significant in the aspects of gender, age, BMI, history of tumor, T stage, N stage, pathological stage, cycle of perioperative chemotherapy, location of tumor, Borrmann types, Lauren classification, surge type, vascular tumor thrombus and neural invasion (Table 1).

In the group of patients for CA19–9, 125 patients fell into low CA19–9 group and 14 fell into high CA19–9 group. The pathological stage and Borrmann types showed significant differences between two groups. In low CA19–9 group, 22 patients were stage I, 52 patients were stage II and 51 patients were stage III. In high CA19–9 group, 0 patient was stage I, 4 patients were stage II and 10 patients were stage III. We use the Mann-Whitney U method to test P value and the result was 0.018. In low CA19–9 group, 36 patients were Borrmann I/II types and 68 patients were Borrmann III/IV types. And in high CA19–9 group, 0 patient was Borrmann I/II types and 11 patients were Borrmann III/IV types. And the P value was 0.018 analyzed by Fisher’s exact test. However, there were no prominent differences between two groups in gender, age, BMI, history of tumor, T stage, N stage, cycle of perioperative chemotherapy, location of tumor, Lauren classification, surge type, vascular tumor thrombus and neural invasion (Table 2).

In addition, the clinicopathologic characteristics didn’t show any differences in analyzing CEA, CA72–4 and CA24–2 patients. we didn’t show their clinicopathologic characteristics because they fail to predict survival in our later analyses.

Survival analyses indicated that AFP and CA19–9 could functioned as an index to predict the prognosis of HAS patients. We applied Kaplan-Meier method performing survival analyses for the perspective tumor markers. In the AFP group, the differences were significant between two groups for OS (mean OS: 30.8 m vs. 22.5 m, P = 0.028) and RFS (mean RFS: 29.7 m vs. 20.1 m, P = 0.011). Similarly, in the CA19–9 group, there was also significant differences between two groups for OS (mean OS: 30.9 m vs. 23.0 m, P = 0.009) as well as RFS (mean RFS: 30.7 m vs. 18.9 m, P<0.001). (Fig. 4).

However, the OS differences were not significant in CEA (mean OS: 30.9 m vs. 25.9 m, P = 0.091), CA72–4 (mean OS: 29.6 m vs. 32.1 m, P = 0.369) and CA24–2 groups (mean OS: 31.3 m vs. 28.8 m, P = 0.270). Similarly, the RFS differences were also similar in CEA (mean RFS: 24.2 m vs.30.2 m, P = 0.026), CA72–4 (mean RFS: 30.9 m vs. 26.4 m, P = 0.057) and CA24–2 groups (mean RFS: 31.1 m vs. 25.1 m, P = 0.079).

To identify the independent risk factors, we performed Univariate and multivariate analyses. Univariate analyses indicated that BMI, cycle number of perioperative chemotherapy, Borrmann types, serum AFP level and CA19–9 level were closely associated with OS. Furthermore, multivariate analyses confirmed that AFP level was an independent risk factor of OS (hazard ratio [HR] 7.264; 95% confidence interval [CI] 1.328–39.738; P = 0.022). Other independent risk factors included BMI (HR 1.377; 95% CI 1.048–1.810; P = 0.022), cycle number of perioperative chemotherapy (HR 0.273; 95% CI 0.064–1.157; P = 0.078), Borrmann types (HR 12.283; 95% CI 1.131–133.414; P = 0.039). As for RFS, univariate analyses manifest that pT stage, pathological stage, and serum AFP level and CA19–9 level were prominently associated with RFS. And multivariate showed that AFP level was also an independent risk factors of RFS (HR 2.688; 95% CI 0.922–7.836; P = 0.070). Moreover, pT stage is the other independent risk factor (HR 1.894; 95% CI 1.066–3.366; P = 0.029).

Considering the substantial overlap of patients in AFP and CA19–9 analyses, we put the indicators, AFP level and CA19–9 level, together to perform multivariate analyses in order to eliminate the eclipsing influences and ensure the objectivity of the consequence. Considering the pre-operative AFP testing was not universal, we sheltered AFP from multivariate analyses and found that CA19–9 could be a fair substitute to predict the OS (HR 7.816; 95% CI 2.084–29.308; P = 0.002) as well as RFS (HR 4.386; 95% CI 1.824–10.547; P = 0.001) when AFP data was unavailable. The univariates and multivariate analyses outcomes were showed on Tables 3 and 4.