Development of a staging system for hepatoid adenocarcinoma of the stomach based on multicenter data: a retrospective cohort study

Abstract

Background:

Hepatoid adenocarcinoma of the stomach (HAS) is a rare subtype of gastric cancer (GC) with a poor prognosis. Furthermore, the current pathological staging system for HAS does not distinguish it from that for common gastric cancer (CGC).

Methods:

The clinicopathological data of 251 patients with primary HAS who underwent radical surgery at 14 centers in China from April 2004 to December 2019 and 5082 patients with primary CGC who underwent radical surgery at two centers during the same period were retrospectively analyzed. A modified staging system was established based on the differences in survival.

Results:

After 1:4 propensity score matching (PSM), 228 patients with HAS and 828 patients with CGC were analyzed. Kaplan–Meier (K–M) analysis showed patients with HAS had a poorer prognosis compared with CGC. Multivariate analysis identified pN stage, CEA level, and perineural invasion (PNI) as independent prognostic factors in patients with HAS. A modified pT (mpT) staging was derived using recursive partitioning analysis (RPA) incorporating PNI and pT staging. The modified pathological staging system (mpTNM) integrated the mpT and the eighth American Joint Committee on Cancer (AJCC) pN definitions. Multivariate analysis showed that the mpTNM stage outperformed other pathological variables as independent predictors of OS and RFS in patients with HAS. The mpTNM staging system exhibited significantly higher predictive accuracy for 3-year OS in patients with HAS (0.707, 95% CI: 0.650–0.763) compared to that of the eighth AJCC staging system (0.667, 95% CI: 0.610–0.723, P<0.05). Analysis using the Akaike information criterion favored the mpTNM staging system over the eighth AJCC staging system (824.69 vs. 835.94) regarding the goodness of fit. The mpTNM stages showed improved homogeneity in survival prediction (likelihood ratio: 41.51 vs. 27.10). Comparatively, the mpTNM staging system outperformed the eighth AJCC staging system in survival prediction, supported by improvements in the net reclassification index (NRI: 47.7%) and integrated discrimination improvement (IDI: 0.083, P<0.05). The time-dependent ROC curve showed that the mpTNM staging system consistently outperformed the eighth AJCC staging system with increasing observation time.

Conclusion:

The mpTNM staging system exhibited superior postoperative prognostic accuracy for patients with HAS compared to the eighth AJCC staging system.

Introduction

Highlights

• The eighth AJCC pT staging system failed to effectively differentiate the prognosis of patients with hepatoid adenocarcinoma of the stomach (HAS).

• This study represents the first investigation of the applicability of the eighth AJCC staging system in postoperative patients with HAS. The modified pTNM (mpTNM) staging system suitable for patients with HAS based on nationwide multicenter data demonstrated superior predictive performance compared with the eighth edition of the AJCC pTNM staging system.

• There is no specific content about HAS in the current guidelines, and the results of this study could be used as a supplement to the guidelines.

Hepatoid adenocarcinoma of the stomach (HAS) is a rare subtype of gastric cancer (GC) characterized by elevated levels of alpha-fetoprotein (AFP) and shares pathological similarities with hepatocellular carcinoma (HCC)^1,2. Studies have demonstrated that HAS exhibits more aggressive biological behaviors than non-HAS, including lower differentiation, increased invasion of nerves and blood vessels, more prominent liver metastasis, and a poorer prognosis^3–5. Due to the low incidence of HAS in GC, ~0.3–1%, comprehensive studies with large-scale data on HAS are scarce^6,7.

The staging criteria for HAS are based on the common gastric cancer (CGC) guidelines. However, some researchers believe that the staging of malignant tumors cannot be determined by the TNM classification, particularly because the depth of tumor invasion alone may not adequately reflect the extent of local tumor infiltration^8–11. The current staging system incorporates pathological factors such as tumor size, lymphatic vascular embolism, perineural infiltration (PNI), and degree of differentiation to enhance prognostic prediction for specific cancers, including gastric neuroendocrine tumors (G-NETs) and penile cancers^12–15. However, the eighth edition of the American Joint Committee on Cancer (AJCC) pT staging for GC may not effectively predict the prognosis of patients with HAS based on preliminary study results from the China Hepatoid Adenocarcinoma of the Stomach Study Group (China-HASSG) database, comprising primary HAS data from 14 centers^16–18. Therefore, this study aimed to establish a modified pTNM (mpTNM) staging system using a national multicenter data sample to enhance the accuracy of prognostic evaluation of patients with HSA after radical surgery.

Methods

Study design and patients

The clinicopathological data of patients with primary HAS admitted to 14 centers of the China-HASSG between April 2004 and December 2019 were retrospectively analyzed. Patients with CGC who underwent radical gastrectomy at two centers during the same period were included. The inclusion criteria for this study were the pathological confirmation of HAS or CGC, including simple HAS (SHAS), mixed HAS (MHAS), and radical gastrectomy. The exclusion criteria were the presence of other malignant tumors, gastric stump cancer, neoadjuvant therapy, and postoperative pathology-confirmed distant metastasis (Supplementary eFig. 1, Supplemental Digital Content 2, http://links.lww.com/JS9/D483). A total of 251 patients with HAS and 5082 patients with CGC were included in this study.

Procedures

All patients underwent radical surgery, including complete resection (R0) of the primary lesion and standardized lymph node dissection (D2, D1+, or D1). Tumor staging was performed according to the eighth edition of the AJCC criteria. For patients with advanced GC, adjuvant chemotherapy is recommended, consisting of a fluorinated chemotherapy regimen with either two drugs or a single drug, for 6–8 cycles.

Pathological diagnosis

All pathological specimens were independently reviewed by two experienced pathologists to ensure consistency and accuracy of diagnostic results. In cases with disputed or inconsistent pathological diagnoses, a panel of pathologists conducted pathological evaluations to determine the final diagnosis. In addition, all pathological reports and diagnostic results were audited and verified to ensure the completeness and accuracy of the pathological diagnoses. Pathology serves as the ‘gold standard’ for diagnosing HAS, primarily relying on morphology¹⁹. In GC, the presence of hepatic differentiation areas in pathological morphology, confirms a diagnosis of HAS, regardless of their quantity²⁰. Typically HAS tissues exhibit both hepatic and common adenocarcinoma areas, often with a transitional zone and occasionally with entirely hepatic differentiation (Supplementary eFig. 2A, Supplemental Digital Content 2, http://links.lww.com/JS9/D483). Immunohistochemical staining assists in diagnosis: areas of hepatic differentiation typically exhibit positive or strongly positive staining for AFP, whereas common adenocarcinoma areas show negative staining. However, positive immunohistochemical staining for AFP is not mandatory for the diagnosing of HAS¹⁶. When tumor cells infiltrate the perineurium of nerve fibers or enter the nerve bundle, they are classified as PNI (Supplementary eFig. 2B, Supplemental Digital Content 2, http://links.lww.com/JS9/D483)²¹.

Follow-up

Patients underwent follow-up examinations every 3 months for the initial two 2 years postradical surgery, semi-annually from the third to fifth year, and annually thereafter. Follow-up was conducted regularly through outpatient visits, phone calls, text messages, or home visits. Recurrence was diagnosed based on the radiological findings or biopsies of suspicious lesions. Recurrence patterns were classified as local (anastomotic or gastric remnant), lymph node, peritoneal, hematogenous, or unclear. Tumors involving the ovaries (Krukenberg’s tumors) are considered peritoneal metastasis²². Overall survival (OS) was calculated from the date of surgery to the date of death or censorship at the last follow-up date²³. Recurrence-free survival (RFS) was calculated from the date of surgery to recurrence (locoregional or systemic), death from any cause, or last follow-up²⁴.

Statistical analysis

Statistical analysis was performed using SPSS 23.0 software and R software (Version 4.2.2). Categorical variables were represented using frequency and composition ratios. Group differences were compared using the χ ² test or Fisher’s exact test. Logistic regression models were used to calculate propensity scores and adjust for covariates, including BMI, ASA, CEA, tumor location, tumor size, PNI, lymphovascular invasion (LVI), pT, pN, and postoperative complication. Survival analysis employed the Kaplan–Meier (K–M) method, with group differences assessed using the log-rank test. Cox proportional hazards models were used for prognostic factors analysis, including statistically significant variables from the univariate analysis in the multivariate analysis using stepwise regression. Recursive partitioning analysis (RPA) using the ‘rpart’ package in R, incorporated independent prognostic pathological factors into the improved pT staging system. Harrell’s concordance index (C-index) was used to assess the discriminative ability of different staging systems. The likelihood ratio χ ² score in the Cox regression analysis was used to measure homogeneity, where higher scores indicated better homogeneity. The Akaike information criterion (AIC) in the Cox regression was used to compare the efficacy of the two staging systems, and a smaller AIC value indicated a more optimistic prognostic stratification. The net reclassification index (NRI) and integrated discrimination improvement (IDI) quantified the improvements in the new staging system. Time-dependent receiver operating characteristic (ROC) analysis was performed to evaluate the discriminative ability of the prognostic model for time-dependent disease outcomes. Statistical significance was set at P<0.05. This study was conducted in line with the strengthening the reporting of cohort, cross-sectional, and case–control studies in surgery (STROCSS) criteria (Supplemental Digital Content 1, http://links.lww.com/JS9/D482)²⁵.

Results

Clinicopathological characteristics

A total of 251 patients with HAS and 5082 patients with CGC were included in this study. Compared to the CGC group, the HAS group had a higher proportion of patients with BMI ≥25 (25.5 vs. 17.5%), ASA score of 2/3 (78.6 vs. 67.0%), AFP ≥20 (60.4 vs. 2.9%), and CEA ≥5 (30.7 vs. 23.8%). The HAS group also had larger tumors and a higher prevalence of PNI (40.2 vs. 30.0%) and LVI (57.4 vs. 34.1%). Additionally, the proportion of patients with stage III disease was higher in the HAS group than in the CGC group (66.1 vs. 56.0%). However, the proportion of patients receiving adjuvant chemotherapy was smaller in the HAS group (48.6 vs. 56.6%, P<0.05, Supplementary eTable 1, Supplemental Digital Content 2, http://links.lww.com/JS9/D483). There were no statistically significant differences between the two groups in terms of age, sex, or pN stage (all P>0.05; Supplementary eTable 1, Supplemental Digital Content 2, http://links.lww.com/JS9/D483). After performing 1:4 propensity score matching (PSM), 228 patients with HAS and 828 with CGC remained. No statistically significant differences were observed between the two groups in terms of age, sex, BMI, ASA, CEA level, tumor location, tumor size, LVI, PNI, pT, pN, pTNM, adjuvant chemotherapy, or postoperative complications (all P>0.05, Supplementary eTable 2, Supplemental Digital Content 2, http://links.lww.com/JS9/D483).

Prognostic analysis of patients with HAS and CGC

The median follow-up time was 73 months for patients with CGC and 40 months for patients with HAS. Before PSM, the 3-year OS and RFS of patients with HAS were significantly lower than those of patients with CGC (3-year OS, 62.1 vs. 73.9%; 3-year RFS, 55.9 vs. 72.3%; both P<0.05; Supplementary eFig. 3, Supplemental Digital Content 2, http://links.lww.com/JS9/D483). The analysis of survival difference between patients with HAS and patients with CGC after PSM was similar to those of the raw cohort (3-year OS, 62.5 vs. 72.7%; 3-year RFS, 57.0 vs. 71.3%; both P<0.05; Supplementary eFig. 4, Supplemental Digital Content 2, http://links.lww.com/JS9/D483). Cox univariate and multivariate analyses showed that histological type (HAS/CGC) was independent prognostic factor for OS and RFS in the raw cohort (Supplementary eTables 3–4, Supplemental Digital Content 2, http://links.lww.com/JS9/D483). The results were confirmed after PSM (Supplementary eTables 5–6, Supplemental Digital Content 2, http://links.lww.com/JS9/D483). Cox univariate and multivariate analyses showed that pT stage, pN stage, LVI, age, CEA level, tumor size, and postoperative complications were independent prognostic factors for OS in patients with CGC (Supplementary eTable 7, Supplemental Digital Content 2, http://links.lww.com/JS9/D483). In the analysis of RFS prognostic factors, pT stage, pN stage, LVI, age, BMI, CEA level, tumor size, and postoperative complications were independent prognostic factors for patients with CGC (Supplementary eTable 8, Supplemental Digital Content 2, http://links.lww.com/JS9/D483). For patients with HAS, only the pN stage, CEA level, and PNI were independent predictors of both OS and RFS, whereas the pT stage was not an independent prognostic factor (Tables 1–2).

Table 1.

Univariate and multivariate Cox analyses of risk factors associated with OS in patients with HAS.

Variables	Univariate analysis	P	Multivariate analysis	P
	HR (95% CI)		HR (95% CI)
Age (years)		0.171
<60	Ref
≥60	1.372 (0.872–2.157)
Sex		0.976
Male	Ref
Female	1.007 (0.615–1.650)
BMI (kg/m2)		0.615
<25	Ref
≥25	0.881 (0.538–1.443)
ASA		0.339
1	Ref
2	0.726 (0.446–1.182)	0.198
3	1.013 (0.496–2.070)	0.972
AFP (ng/ml)		0.787
<20	Ref
≥20	1.070 (0.656–1.745)
CEA (ng/ml)		0.018		0.004
<5.0	Ref		Ref
≥5.0	1.708 (1.098–2.655)		1.922 (1.098–2.655)
Tumor location		0.434
Upper
Middle	0.572 (0.263–1.244)	0.159
Lower	0.788 (0.480–1.295)	0.348
Mixed	1.035 (0.551–1.945)	0.914
Tumor Size (mm)		0.684
<50	Ref
≥50	1.094 (0.711–1.682)
Lymphovascular invasion		<0.001
Negative	Ref
Positive	2.456 (1.520–3.968)
Perineural invasion		<0.001		0.002
Negative	Ref		Ref
Positive	2.818 (1.818–4.368)		2.088 (1.325–3.289)
Histologic type		0.605
MHAS	Ref
SHAS	0.893 (0.581–1.372)
pT		0.041
1	Ref
2	1.274 (0.359–4.517)	0.708
3	2.295 (0.790–6.666)	0.127
4a	3.336 (1.201–9.265)	0.021
4b	2.520 (0.630–10.088)	0.191
pN		<0.001		0.003
0	Ref
1	1.928 (0.799–4.653)	0.144	1.464 (0.598–3.588)	0.404
2	2.643 (1.169–5.976)	0.020	1.762 (0.756–4.108)	0.190
3a	3.538 (1.594–7.855)	0.002	2.136 (0.923–4.939)	0.076
3b	7.281 (3.148–16.842)	<0.001	4.543 (1.895–10.890)	0.001
pTNM Stage		0.001
IA	Ref
IB	0.377 (0.034–4.154)	0.425
IIA	2.25 (0.412–12.301)	0.350
IIB	2.434 (0.533–11.112)	0.251
IIIA	2.918 (0.688–12.367)	0.146
IIIB	4.753 (1.126–20.065)	0.034
IIIC	7.412 (1.713–32.063)	0.007
Adjuvant chemotherapy		0.898
No	Ref
Yes	0.973 (0.635–1.491)
Postoperative complication		0.083
No	Ref
Yes	1.558 (0.943–2.575)

Table 2.

Univariate and multivariate Cox analyses of risk factors associated with RFS in patients with HAS.

Variables	Univariate analysis	P	Multivariate analysis	P
	HR (95% CI)		HR (95% CI)
Age (years)		0.307
<60	Ref
≥60	1.251 (0.814–1.922)
Sex		0.786
Male	Ref
Female	0.936 (0.580–1.511)
BMI		0.550
<25	Ref
≥25	0.864 (0.535–1.395)
ASA		0.243
1	Ref
2	0.705 (0.441–1.126)	0.144
3	1.021 (0.514–2.025)	0.953
AFP		0.883
<20	Ref
≥20	1.036 (0.651–1.648)
CEA		0.001		<0.001
<5.0	Ref		Ref
≥5.0	2.083 (1.375–3.156)		2.349 (1.529–3.608)
Tumor location		0.315
Upper
Middle	0.649 (0.310–1.359)	0.252
Lower	0.849 (0.523–1.378)	0.507
Mixed	1.308 (0.727–2.352)	0.371
Tumor size (mm)		0.474
<50	Ref
≥50	1.164 (0.768–1.763)
Lymphovascular invasion		<0.001
Negative	Ref
Positive	2.293 (1.455–3.614)
Perineural invasion		<0.001		0.002
Negative	Ref		Ref
Positive	2.584 (1.703–3.921)		1.989 (1.292–3.061)
Histologic type		0.517
MHAS	Ref
SHAS	0.872 (0.577–1.319)
pT		0.051
1	Ref
2	0.946 (0.318–2.815)	0.920
3	1.659 (0.680–4.048)	0.266
4a	2.265 (2.265–5.289)	0.059
4b	0.733 (0.148–3.634)	0.704
pN		<0.001		<0.001
0	Ref
1	2.223 (0.982–5.030)	0.055	1.764 (0.772–4.032)	0.178
2	2.198 (1.000–4.831)	0.050	1.343 (0.590–3.053)	0.482
3a	3.220 (1.508–6.877)	0.003	1.853 (0.834–4.120)	0.130
3b	7.267 (3.315–15.931)	<0.001	4.499 (1.987–10.188)	<0.001
pTNM Stage		0.001
IA	Ref
IB	0.227 (0.024–2.183)	0.199
IIA	1.601 (0.382–6.708)	0.520
IIB	2.004 (0.571–7.036)	0.278
IIIA	1.785 (0.536–5.942)	0.345
IIIB	2.846 (0.860–9.421)	0.087
IIIC	4.858 (1.439–16.404)	0.011
Adjuvant chemotherapy		0.756
No	Ref
Yes	1.067 (0.708–1.608)
Postoperative complication		0.233
No	Ref
Yes	1.361 (0.821–2.256)

To further determine the impact of pT staging on the prognosis of patients with HAS, the survival of patients with different pT stages was analyzed. The results showed that there were no statistically significant differences in OS and RFS between adjacent pT stages (3-year OS: pT1, 77.0%; pT2, 76.8%; pT3, 65.7%; pT4a, 53.4%; pT4b, 57.3%. 3-year RFS: pT1, 69.4%; pT2, 75.4%; pT3, 61.3%; pT4a, 49.0%; pT4b, 72.0%; all P>0.05, Supplementary eFig. 5, Supplemental Digital Content 2, http://links.lww.com/JS9/D483). Supplementary eFigure 6 (Supplemental Digital Content 2, http://links.lww.com/JS9/D483) (Supplemental Digital Content 2, http://links.lww.com/JS9/D483) showed the relationship between pT stage and the number of lymph node metastases. The results demonstrated that with increasing tumor infiltration, the number of lymph node metastases increased. Similar findings were observed for the relationship between PNI and pN staging, as the proportion of PNI-positive patients increased with more advanced pN stages (pT1, 20.8%; pT2, 31.3%; T3, 38.1%; pT4a, 53.4%; pT4b, 69.0%; P<0.05; Supplementary eFig. 7A, Supplemental Digital Content 2, http://links.lww.com/JS9/D483). Although the proportion of PNI-positive patients increased with tumor infiltration (pT1, 13.6%; pT2, 12.5%; T3, 52.1%; pT4a, 45.2%; pT4b, 45.5%; P<0.05; Supplementary eFig. 7B, Supplemental Digital Content 2, http://links.lww.com/JS9/D483), no incremental relationships were observed. These results suggest that PNI is significantly correlated with pT staging and pN staging and serves as an independent prognostic factor in patients with HAS. PNI may act synergistically with pT staging and potentially serve as a complementary factor to pT staging.

Establishment of modified pT stage

The modified pT (mpT) staging system was developed by integrating pT staging with the independent prognostic pathological variable PNI using RPA (Fig. 1A): mpT1 for pT1-3 and PNI-negative, mpT2 for pT4a-b and PNI-negative, and mpT3 for PNI-positive (Fig. 1B). The 3-year OS and RFS rates of the different subgroups formed by a combination of PNI and pT staging are shown in Supplementary eTables 9 and 10 (Supplemental Digital Content 2, http://links.lww.com/JS9/D483). The prognostic significance of mpT staging was validated through K–M analysis, and the results showed significant differences in the 3-year OS and RFS among the different mpT stages (mpT1 vs. mpT2 vs. mpT3: 3-year OS: 85.2 vs. 63.5% vs. 38.3%; 3-year RFS: 80.7% vs. 56.6% vs. 38.1%; all P<0.05; Supplementary eFig. 8, Supplemental Digital Content 2, http://links.lww.com/JS9/D483). Supplementary eFigure 9 (Supplemental Digital Content 2, http://links.lww.com/JS9/D483) demonstrates an increasing trend in the number of lymph node metastases with higher mpT stages (mpT1: 3.0±3.6 nodes, mpT2: 6.9±8.9 nodes, mpT3: 9.2±9.2 nodes, P<0.05), suggesting that mpT staging can differentiate the number of lymph node metastases.

Figure 1.

Modified pathological T stage (mpT stage). (A) Proposed mpT stage through RPA; (B) definition for mpT stage.

Prognosis analysis of HAS with different modified pTNM stage

A modified pTNM staging system (mpTNM) was constructed based on the combination of mpT and pN staging according to the 8th AJCC staging system (Table 3). The 3-year OS and RFS rates of the different subgroups formed by the combination of mpT and pN staging are shown in Supplementary eTables 11 and 12 (Supplemental Digital Content 2, http://links.lww.com/JS9/D483). K–M analysis showed the survival curves of different mpTNM stages (3-year OS: IA 95.8%, IB 79.1%, IIA 67.0%, IIB 61.3%, IIIA 54.4%, IIIB 35.2%, IIIC 31.6%; 3-year RFS: IA 92.1%, IB 73.7%, IIA 66.5%, IIB 56.5%, IIIA 56.3%, IIIB 34.5%, IIIC 21.1%, Supplementary eFig. 10A, B, Supplemental Digital Content 2, http://links.lww.com/JS9/D483) and eighth AJCC stages (3-year OS: IA 82.5%, IB 93.8%, IIA 68.4%, IIB 69.3%, IIIA 62.3%, IIIB 54.5%, IIIC 33.1%; 3-year RFS: IA 75.0%, IB 95.0%, IIA 64.6%, IIB 61.1%, IIIA 62.3%, IIIB 50.0%, IIIC 28.4%, Supplementary eFig. 10C, D, Supplemental Digital Content 2, http://links.lww.com/JS9/D483). Cox multivariate analysis showed that mpT staging, pN staging, and CEA level were independent predictors of OS in the first-step regression analysis, whereas PNI, pN staging, and CEA level were independent predictors of RFS (Tables 4–5). In the second-step multivariate analysis, mpTNM staging replaced the other pathological factors (PNI, mpT, and pN staging) as independent predictors of OS and RFS in patients with HAS (Tables 4–5).

Table 3.

The modified pTNM staging definitions for HAS.

mpTNM	pN0	pN1	pN2	pN3a	pN3b
mpT1	IA	IB	IIA	IIB	IIIB
mpT2	IB	IIA	IIB	IIIA	IIIB
mpT3	IIA	IIB	IIIA	IIIB	IIIC

Table 4.

Two-step multivariate analysis of the prognostic factors for OS in patients with HAS.

Variables	Multivariate analysisa	P	Multivariate analysisb	P
	HR (95% CI)		HR (95% CI)
CEA		0.006		0.006
<5.0	Ref		Ref
≥5.0	1.882 (1.203–2.944)		1.867 (1.197–2.913)
pN		0.021
0	Ref
1	1.373 (0.560–3.365)	0.489
2	1.562 (0.668–3.651)	0.304
3a	1.917 (0.828–4.437)	0.129
3b	3.582 (1.487–8.627)	0.004
mpT		0.001
1	Ref
2	2.359 (1.167–4.771)	0.017
3	3.595 (1.830–7.065)	<0.001
mpTNM		/		<0.001
IA			Ref
IB			6.178 (0.759–50.291)	0.089
IIA			9.944 (1.291–76.606)	0.027
IIB			9.566 (1.256–72.881)	0.029
IIIA			13.193 (1.712–101.662)	0.013
IIIB			21.868 (2.922–163.668)	0.003
IIIC			39.277 (5.052–305.328)	<0.001

^a Step 1, with consideration of all significantly important prognostic factors in the univariate analysis except for the modified pTNM stage.

^b Step 2, with consideration of all significantly important prognostic factors in the univariate analysis, including the modified pTNM stage.

Table 5.

Two-step multivariate analysis of the prognostic factors for RFS in patients with HAS.

Variables	Multivariate analysisa	P	Multivariate analysisb	P
	HR (95% CI)		HR (95% CI)
CEA		<0.001		<0.001
<5.0	Ref		Ref
≥5.0	2.349 (1.529–3.608)		2.240 (1.462–3.433)
Lymphovascular invasion
Negative
Positive
Perineural invasion		0.002
Negative	Ref
Positive	1.989 (1.292–3.061)
pN		<0.001
0	Ref
1	1.764 (0.772–4.032)	0.178
2	1.343 (0.590–3.053)	0.482
3a	1.853 (0.834–4.120)	0.130
3b	4.499 (1.987–10.188)	<0.001
mpTNM		/		<0.001
IA			Ref
IB			3.608 (0.777–16.753)	0.101
IIA			4.808 (1.073–21.537)	0.040
IIB			5.372 (1.231–23.443)	0.025
IIIA			5.039 (1.116–22.742)	0.035
IIIB			8.995 (2.085–38.809)	0.003
IIIC			23.071 (5.205–102.261)	<0.001

^a Step 1, with consideration of all significantly important prognostic factors in the univariate analysis except for the modified pTNM stage.

^b Step 2, with consideration of all significantly important prognostic factors in the univariate analysis, including the modified pTNM stage.

Prognostic performance of the modified pTNM staging system

The prognostic accuracy of the different staging systems and components was evaluated using the integrated area under the ROC curve (iAUC) with 1000 bootstrap resamplings (Fig. 2). The results showed that the predictive accuracy of mpT (0.664, 95% CI: 0.611–0.717) for the 3-year OS in patients with HAS was significantly better than that of pT (0.585, 95% CI: 0.531–0.639, P=0.020) and PNI (0.628, 95% CI: 0.575–0.681, P=0.002). The staging system combining mpT and pN demonstrated significantly better prediction accuracy (0.707, 95% CI: 0.650–0.763) for 3-year OS in patients with HAS than the eighth AJCC TNM staging system (0.667, 95% CI: 0.610–0.723, P=0.016). AIC analysis demonstrated that the mpTNM staging had superior goodness of fit compared with the eighth edition AJCC staging, and mpTNM staging had better survival prediction homogeneity (likelihood ratio χ ²: 41.51 vs. 27.10). Compared with the eighth AJCC staging system, mpTNM staging showed improved survival prediction performance (NRI, 47.7%; IDI, 0.083; P<0.05; Table 6). Time-dependent ROC curves demonstrated that mpTNM staging was superior to the eighth AJCC staging over time (Fig. 3).

Figure 2.

Predictive accuracy of pathological prognostic factors for prediction of OS in patients with HAS (the predictive accuracy for 3-year overall survival based on the iAUC with 1000× bootstrap resampling for each parameter is shown in a box plot. Median values of 1000× bootstrap resampling are shown with thick lines).

Table 6.

Comparison of the prognostic performance of the AJCC eighth edition pTNM staging system and the modified pTNM staging system.

Variables	Eighth pTNM	mpTNM	P
AIC	835.94	824.69
Likelihood ratio chi-square	27.10	41.51
NRI (95% CI)		47.7% (2.26–76.2%)
IDI (95% CI)		0.083 (0.026–0.141)	0.006

Figure 3.

Time-dependent receiver operating characteristics (ROC) curves for the AJCC eighth edition pTNM staging system and the modified pTNM staging system.

Discussion

HAS, a relatively rare subtype of GC, exhibits clinical and pathological characteristics distinct from those of CGC and is associated with poor prognosis²⁶. Huang et al. and Kang et al. showed that pT staging is not a significant prognostic factor in patients with HAS^27,28. Similarly, Zhou et al.²⁹ found that clinical or pathological TNM staging was not an independent risk factor for the prognosis of patients with HAS. Therefore, further studies with a large sample size are required to validate the applicability of the current pTNM staging system for HAS. In our study, clinicopathological data were collected from 14 centers in the China-HASSG. The analysis demonstrated that pT stage, pN stage, LVI, and tumor size were independent pathological prognostic factors in patients with CGC. However, the pT stage was not an independent prognostic factor in patients with HAS. Survival analysis revealed that the eighth AJCC pT staging system failed to effectively differentiate the prognosis of patients with HAS.

Owing to the limited sample size of available HAS patients, Zhou et al. categorized HAS patients into two groups based on T staging (T1-3/T4) in their study. While T staging showed a statistically significant predictive value for prognosis in HAS patients in that study, significant within-group heterogeneity may still exist due to the categorization. Moreover, several studies have suggested that pathological features, such as LVI, PNI, and tumor differentiation, may influence the prognostic value of T staging in malignant tumors. Incorporating the important prognostic pathological features may improve the prognostic predictive ability of T staging^13–15. Therefore, in this study, PNI, an independent prognostic pathological feature, was integrated with pT staging. PNI refers to the infiltration of tumor cells into the perineurium or nerve bundles, and is considered the fifth mode of tumor metastasis^21,30. Previous research has reported that PNI is not only the result of direct infiltration of tumor cells, but can also involve invasion into lymphatic vessels and veins surrounding nerves³¹. However, the perineural space is now recognized as an independent pathway for cancer spread, as it differs anatomically and ultra-structurally from the lymphatic vessels. By allowing direct interactions between tumor cells and nerves, PNI creates a favorable microenvironment that supports tumor growth and survival. Additionally, PNI can promote cancer-related pain and serves as a valuable pathological marker for early metastasis, providing important prognostic information³². Thus, PNI has been identified as a crucial invasion pathway in various malignant tumors, including pancreatic cancer, biliary tract cancer, prostate cancer, and head and neck cancer³³. PNI positivity is a precursor to decreased survival rates in patients with malignant tumors²¹. Reports have shown that in CGC, the rate of PNI positivity increases with tumor undifferentiation, infiltrative depth, and size³⁴. The present study observed similar characteristics in patients with HAS. Furthermore, mpT staging, developed by integrating pT staging with PNI, demonstrated good discriminative ability for survival in patients with HAS.

In the present study, the prognostic predictive ability of the mpTNM staging system was evaluated by internal validation, performed by resampling the dataset. Internal validation using bootstrap resampling is a useful method when the sample size is small³⁵. In this study, the predictive accuracy of the mpTNM and original pTNM staging systems, along with their components, was assessed using the iAUC calculated from 1000 bootstrap resamples. The results showed that mpT had significantly better predictive accuracy for the 3-year OS in patients with HAS compared to pT alone. The staging system combining mpT and pN demonstrated significantly superior predictive accuracy for the 3-year OS in patients with HAS compared to the staging system combining pT and pN. Additionally, the discriminative ability of different survival curves of mpTNM stages was superior to that of the eighth AJCC stages. This study represents the first investigation of the applicability of the eighth AJCC staging system in postoperative patients with rare-subtype HAS. By utilizing nationwide multicenter data and considering the unique characteristics of patients with HAS, a customized mpTNM staging system was newly developed. The current eighth edition of the AJCC staging system lacks a separate classification for HAS, therefore, the results of this study are expected to serve as a supplement to the eighth edition of the AJCC staging system for this rare disease. The mpTNM staging system established for HAS in this study could more accurately predict the prognosis of patients with HAS, which may help clinicians develop personalized treatment plans, potentially improving treatment outcomes and reducing unnecessary interventions. Furthermore, the mpTNM staging system could also guide the development of follow-up plans, assisting clinicians in determining appropriate follow-up frequencies and strategies. Timely detection of early recurrence and prompt intervention could help slow tumor progression, with the aim of improving patient survival rates and quality of life. Regarding the adjuvant therapy, another study by the China Hepatoid Adenocarcinoma of the Stomach Study Group found that receiving current adjuvant chemotherapy regimens did not significantly improve the survival of patients with HAS after radical surgery¹⁸. Similar to the AJCC-TNM staging system, whether the mpTNM staging system could modify the existing regimens of adjuvant therapy in clinical practice needs to be further confirmed by multicenter prospective randomized controlled clinical trials.

Although we hope to develop a modified pTNM staging system applicable to HAS, the staging system itself has certain limitations. The staging system can only evaluate the pathological stage of the tumor from the depth of tumor invasion, PNI, lymph node, and distant metastasis, and does not consider other factors that may affect the prognosis of the patient, such as molecular biological characteristics and treatment response. The data for this model were derived from Asian countries, and there may be differences in the biological behavior of hepatoid adenocarcinoma between Asian and Western countries. We are attempting to seek external validation using large Western databases, such as the SEER database. Due to the limited sample size, external validation of this newly developed staging system using multicentre data was not conducted. Only after verifying its effectiveness and reliability in large-scale clinical practice can clinical application and promotion of the staging system be ensured. Moreover, as this was a retrospective study, potential selection bias and variations in treatment and follow-up approaches among different institutions should be considered. Although each center strictly followed surgical procedures and perioperative management according to the guidelines during that period, standardizing and normalizing the diagnosis and treatment of GC bridged the gap between individuals and medical centers to some extent. However, variations still exist in the implementation of laparoscopic surgery and adjuvant chemotherapy across different centers. The K–M analysis showed no significant differences in the 3-year OS and RFS between the 2004–2010, 2011–2015, and 2015–2019 groups (Supplementary eFig. 11, Supplemental Digital Content 2, http://links.lww.com/JS9/D483), indicating that prognosis was not significantly affected by the different time periods. Nevertheless, given the long timespan of patient inclusion in this study, during 15-year period, changes in TNM staging, surgical techniques, and technologies may have introduced some biases, such as selection bias. Additionally, the lower proportion of patients receiving adjuvant chemotherapy in the HAS group than in the CGC group could be influenced by factors including surgical outcomes, complications, physician and patient preferences, and variations in clinical practice. More precise results require confirmation through prospective studies.

In conclusion, we developed an mpTNM staging system suitable for patients with HAS based on nationwide multicenter data, which demonstrated superior predictive performance compared with the eighth edition of the AJCC pTNM staging system. The mpTNM staging system allows for a more accurate assessment of the postoperative prognosis of patients with HAS.

Ethical approval

Ethical approval for this study (IRB number 2021KY159) was provided by the institutional review boards of Fujian Medical University Union Hospitals, Fuzhou, China, on 02 November 2021.

Consent

Each participant signed informed consent before participating in this study following the Declaration of Helsinki. Informed consent or a substitute for it was obtained from all patients for being included in the study.

Source of funding

Supported by Fujian Provincial Medical Construction Fund of ‘Creating double High Levels’ ([2021] No. 76), Fujian Research and Training Grants for Young and Middle-aged Leaders in Healthcare (No. [2022] 954) and Fujian third batch of ‘Innovation star’ talent project (No. [2022] 22).

Author contribution

Y.-Q.H., Z.-N.H., Q.-Q.H., P.Z., Z.-Z.Z., and L.H.: conception and design; Y.-Q.H., Z.-N.H., Q.-Q.H., P.Z., Z.-Z.Z., and L.H.: drafting of the manuscript; Y.-Q.H. and Z.-N.H.: statistical analysis; C.-M.H.: obtained funding; P.L., C.-H.Z., C.-M.H., and Q.-Y.C.: administrative, technical, or material support; C.-H.Z., C.-M.H., and Q.-Y.C.: supervision. All authors contributed in acquisition, analysis, or interpretation of data, critical revision of the manuscript for important intellectual content, and have read and approved this version of the manuscript, and due care has been taken to ensure the integrity of the work.

Conflicts of interest disclosure

The authors declare no conflicts of interest.

Research registration unique identifying number (UIN)

This multicenter retrospective cohort study (ClinicalTrials.gov Identifier NCT05274464) was performed in accordance with the Declaration of Helsinki and Ethical Guidelines for Clinical Studies.

Guarantor

Ping Li.

Data availability statement

The dataset analyzed for this study is available from the corresponding author upon reasonable request.

Provenance and peer review

Not commissioned, externally peer-reviewed.

Supplementary Material

js9-111-0718-s001.doc ^{(115.1KB, doc)}

js9-111-0718-s002.pdf ^{(1.7MB, pdf)}