We consecutively enrolled patients with HAS who underwent curative total or partial gastrectomy with D2 lymph nodes (LNs) dissection between November 2009 and October 2020. We selected patients who were pathologically diagnosed with HAS after radical gastrectomy and with clinical stage T3/T4 or N+ disease. The exclusion criteria were as follows: (1) perioperative death, (2) R0 resection was not performed, (3) preoperative or postoperative radiotherapy, (4) clinical stage IVb, and (5) pathological stage I patients without high risk factors, which included age below 40 years old, poor differentiation and lymphovascular invasion. Patients who received preoperative chemotherapy were defined as neoadjuvant chemotherapy-first (NAC-first) group, and patients who did not receive preoperative treatment were defined as surgery-first group. Clinicopathological features were retrospectively collected and all patients were followed up. We used abdominal and pelvic computed tomography (CT) to assess the clinical stage using the American Joint Committee on Cancer/Union for International Cancer Control 8^th classification system. Enlarged LNs over 8 mm at their largest axis or with internal necrosis were classified as cN+. This retrospective study was performed according to the tenets of the Declaration of Helsinki and was approved by the Institutional Review Board of Peking University Cancer Hospital.

A total of 125 HAS patients were eligible for the study. Twenty-five patients were excluded, of whom 13 patients were diagnosed with distant metastases, 2 patients had postoperative residual lesions, 4 patients accepted perioperative radiotherapy, one patient died perioperatively and 5 patients were diagnosed with pathological stage I without high risk factors or other types of tumor differentiation. Eventually, 29 patients were included in the NAC-first group and 71 patients were included in the surgery-first group, for a total of 100 patients ( Figure 1 ).

Pathological diagnosis was based on morphological features and immunohistochemistry, including hepatoid and/or adenocarcinoma components, by two independent pathologists ( Supplementary Figure 1 ). Clinical responses to NAC were assessed based on CT scans according to the Response Assessment Criteria for Solid Tumors version 1.1 (29). The assessment of target lesions was divided into the following four categories: complete remission (CR), partial remission (PR), stable disease (SD), and progressive disease (PD). According to the NCCN guidelines for gastric cancer (2018), the pathological response was graded according to the 3-point tumor regression grading (TRG) system (30). The tumors were divided into the following four grades: grade 0 (no visible cancer cells), grade 1 (single cells or small groups of cancer cells), grade 2 (residual cancer outgrown by fibrosis) and grade 3 (significant fibrosis outgrown by cancer or no fibrosis with extensive residual cancer). We also evaluated the toxicities related to NAC by the WHO standard criteria.

The patients underwent follow-up gastroscopy, abdominal and pelvic computed tomography, chest radiography and tumor biomarkers at our hospital or local hospital 3 months after the operation and every 3 or 6 months thereafter. Overall survival time (OS) was defined as the length of time from the date of first NAC treatment or radical gastrectomy to the date of the last follow-up or the date of death from any cause. Disease-free survival time (DFS) was defined as the length of time from the date of first NAC treatment or radical gastrectomy to the date of disease recurrence, metastasis or death from any cause or the date of last follow-up. The mean follow-up time was 30.5 months, ranging from 2.4 to 102.6 months.

To compare the clinicopathological features of the NAC-first and surgery-first groups, SPSS 23.0 was used for statistical analysis. Independent sample t-tests were used for continuous variables. The chi-square test or Mann-Whitney U test was used for categorical variables. Propensity score matching analysis (PSM) was used to reduce the impact of possible confounding factors. The 1:1 PSM method (match tolerance 0.2) was conducted to compare the NAC-first and surgery-first groups. To estimate the long-term OS and DFS outcomes, the Kaplan–Meier method and a log-rank test were used. To evaluate the independent predictors of OS and DFS, variables with P<0.10 in univariate survival analyses or with clinical significance were entered into the multiple regression analysis using the Cox proportional hazards model. P<0.05 was considered statistically significant. GraphPad Prism 5 was used to draw the Kaplan-Meier survival curve.

We found that there were differences in sex, tumor location, clinical T, N, TNM stage, lymphovascular invasion, nerve invasion, PDL-1 and SALL4 expression between the NAC-first group and the surgery-first group (P<0.1). The positive expression rate of Her-2 (staining by immunohistochemistry 3+ or with positive fluorescence in situ hybridization) in HAS patients was 22.89% (19/83). In the surgery-first group, 23% (14/61) of the patients were Her-2 positive. In the NAC-first group, 22.7% (5/22) of the patients were Her-2 positive. In addition, 91.7% (88/96) of the patients had AFP-positive cells as determined by immunohistochemistry. The number of patients with stage cIII/IVa disease in the NAC-first group was substantially higher than that in the surgery-first group (100% versus 78.9%, P=0.01). However, lymphovascular invasion were more prevalent in the surgery-first group than in the NAC-first group (67.6% vs 44.8%, P=0.03) ( Table 1 ). Nerve invasion levels were also similar (63.4% vs 41.4%, P=0.04) ( Table 1 ). To reduce confounding bias, 1:1 PSM was performed, and 56 patients were ultimately included. Most clinicopathological features were not significantly different between the two groups after 1:1 PSM ( Table 1 ).

In our study, 29 patients underwent NAC. Of these, 19 patients received S-1+oxaliplatin (SOX), 6 patients received oxaliplatin + capecitabine (XELOX), one received SOX+ paclitaxel + trastuzumab, one received XELOX+ trastuzumab, one received docetaxel+ cisplatin+ fluorouracil (DCF) and one received oxaliplatin+ calcium folinate+ fluorouracil (mFOLFOX). The median course of NAC was 3 cycles (1-5 cycles).

In the SOX regimen, one patient developed grade 1 gastrointestinal discomfort, and the main clinical manifestation was nausea and vomiting. One patient developed grade 1 gastrointestinal discomfort and neurotoxicity. Fourteen patients did not exhibit side effects during NAC. In the XELOX regimen, one patient experienced grade 3 gastrointestinal discomfort and grade 2 thrombocytopenia. One patient developed grade 1 neutropenia and leukopenia, grade 2 thrombocytopenia and slight numbness in the extremities. In the SOX+ paclitaxel + trastuzumab, DCF and mFOLFOX regimens, no toxicities were observed during NAC.

In the NAC-first group, a total of 6 patients achieved partial remission (PR), 20 patients achieved stable disease (SD), 1 patient had progressive disease (PD) and none achieved complete remission (CR). In the SOX regimen, pathological responses of TRG 1, 2 and 3 were observed in 2, 2 and 13 patients, respectively. In the XELOX regimen, TRG 1, 2 and 3 were observed in 0, 0, and 6 patients, respectively. Pathological responses of other chemotherapy regimens were shown in Table 2 .

The OS time of the surgery-first group was better than that of the NAC-first group ( Figure 2A , P=0.02). In particular, the 1- and 3-year survival rates of the NAC-first group were 92.7% and 68.2%, respectively. The 1- and 3-year survival rates of the surgery-first group were 97% and 83.4%, respectively. And univariate survival analysis was demonstrated in Supplementary Material . Multivariate Cox regression analysis indicated that clinical T4 (P=0.015), proximal gastrectomy (P=0.021), lymphovascular invasion (P=0.030) and CA199 (P=0.007) were independent risk factors for poor OS outcomes in HAS patients ( Table 3 ). However, no significant difference in OS times was found between the NAC-first group and the surgery-first group after PSM analysis, although the surgery-first group had a tendency toward better OS rates than the NAC-first group ( Figure 2B , P=0.105).

To explore the relationship between NAC and the recurrence of HAS, we also conducted a univariate survival analysis of the DFS rates. The most common site of metastases was the liver. In the NAC-first group, 5 patients had postoperative liver metastasis with a median time of 5 months (1-58 months), 1 patient had lung metastasis at 1 month after surgery, and 2 patients metastasized to other sites. In the surgery-first group, postoperative liver metastasis occurred in 6 patients, with a median time of 7.5 months (4-26 months). In addition, 2 patients had lung metastasis at an average time of 10.5 months after surgery, 1 patient had ovarian metastasis at 18 months after gastrectomy, and 2 patients had metastases to other sites. The DFS time of the surgery-first group was substantially longer than that of the NAC-first group ( Figure 2A , P=0.022). Specifically, the 1- and 3-year DFS rates of the NAC-first group were 80.4% and 71.5%, respectively. The 1- and 3-year DFS rates of the surgery-first group were 97% and 83.8%, respectively. Multivariate Cox regression analysis also revealed that clinical T4 (P=0.023) and lymphovascular invasion (P=0.046) was significant predictor of DFS outcomes ( Table 3 ). However, a difference in the DFS rates between the NAC-first group and the surgery-first group was not found after PSM analysis ( Figure 2B , P=0.108).

We also analyzed the relationship between the number of adjuvant chemotherapy cycles and the prognosis of HAS patients. Among the 61 patients who underwent surgery first, we found that the OS of the ≥6 cycles group were better than that of the <6 cycles group ( Figure 3 , P=0.023) and the DFS also had similar results ( Figure 3 , P=0.013).

As the clinical TNM stage was included in the survival analysis before and after PSM, there was confounding bias present. To reduce the confounding bias resulting from disease stage, we divided 100 patients into two groups, 29 patients in the NAC-first group and 71 patients in the surgery-first group. Their pathological TNM stages (pTNM, ypTNM) were used for univariate and multivariate survival analyses. In the NAC-first group, the results showed that the radiological response (P<0.01), the type of surgery (P = 0.032), and EGFR status (P=0.005) were related to the OS rate in the NAC-first group. Radiological response (P<0.01), number of LNs dissected (P = 0.039) and EGFR status (P=0.032) were related to the DFS rate ( Supplementary Table 1 ). Multivariate Cox regression analysis showed that EGFR status was an independent risk factor for poor OS (P=0.006) and DFS outcomes (P=0.036) ( Supplementary Table 2 ). In the surgery-first group, univariate survival analysis showed that age (p=0.03), lymphovascular invasion (P=0.045), CEA (P=0.044), and CA199 (P=0.003) were associated with the OS rate. Age (P=0.028), lymphovascular invasion (P=0.039), cycles of perioperative chemotherapy (P=0.029) and CA199 (P=0.001) were associated with the DFS rate ( Supplementary Table 3 ). Multivariate Cox regression analysis showed that age (P=0.049) and CA199 (P=0.001) were independent risk factors for the DFS outcome ( Supplementary Table 4 ).