Association between body mass index and risk of gastric cancer by anatomical and histological subtypes in over 500,000 East and Southeast Asian cohort participants

Abstract

Background:

This study was performed to investigate the association between body mass index (BMI) and gastric cancer (GC) in East and Southeast Asia where most of GC is non-cardia GC.

Methods:

Based on 8,997 GC cases among the Asia Cohort Consortium participants from China, Japan, Korea, and Singapore (N=538,835), we assessed GC risk according to BMI by calculating hazard ratios (HRs) and 95% confidence intervals (CIs) using the Cox proportional hazard regression model.

Results:

A U-shaped associations between BMI and GC risk were observed. GC risks in underweight group (<18.5 kg/m²) and in obesity group (≥27.5 kg/m²) were higher than reference BMI group (23–24.9 kg/m²) (HR=1.15, 95% CI 1.05–1.25 for underweight; HR=1.12, 95% CI 1.03–1.22 for obesity, respectively). The associations of underweight and obesity with GC risk were consistent in the analyses for non-cardia GC, intestinal type GC, and late onset GC. No significant association of underweight and obesity with the risk of cardia GC, diffuse type GC, and early onset GC was observed. Additionally, we found that the U-shaped association between BMI and GC risk remained in non-smokers, while only underweight was related to increased GC risk in smokers.

Conclusion:

BMI has a U-shaped association with GC risk in East and Southeast Asian population, especially for the non-cardia GC, intestinal type GC, and late onset GC.

Impact:

Future studies with consideration of anatomical location and histology of GC are needed to establish the association of underweight as well as obesity with GC risk.

Introduction

Gastric cancer (GC) is the fifth most frequently diagnosed cancer worldwide, with the estimated new cases in excess of 1,000,000 in 2018 (1). Its incidence rate is the highest in East Asia, including China, Japan and Korea (Age-standardized incidence rate of 32.1 and 13.2 per 100,000 for men and women, respectively in 2018) (1). Major risk factors for GC, such as Helicobacter pylori (H. pylori) infection and cigarette smoking account for over 85% of all GC cases (2). Furthermore, sodium intake over the recommendation of the World Health Organization (WHO) is attributable to more than 10% of GC incidence (3). However, there still remain GC incidence that could not be explained by known risk factors above.

Excess body fatness has been defined as a risk factor for various cancers, such as esophagus (adenocarcinoma), liver, pancreas, kidney, endometrium and breast (postmenopausal women) (4). However, there were inconsistencies across results on assessing body fatness as a risk factor of GC. A previous meta-analysis on the association between GC and body fatness reported a sequential increase in GC risk from normal body mass index (BMI) to overweight and obesity (5–6). In contrast, another meta-analysis study found that there is no significant association of overweight or obesity with GC (7). A recent large retrospective cohort analysis using the Clinical Practice Research Datalink (CPRD) in United Kingdom reported a hazard ratio (HR) of 1.00 based on a linear model to evaluate GC risk with BMI which indicates no difference in GC risk according to BMI level, yielding a result supporting the latter meta-analysis (8–9). Those findings above raise the fundamental question of whether there is a significant association between BMI and GC risk, and the GC risk is linearly related to BMI or not if GC risk is related to BMI. In a non-linear model using the same CPRD, in fact, an inverse J-shaped association of BMI with the GC risk was suggested. Elevated risks of GC were observed in underweight and BMI of 30–35 kg/m² (HR=1.42 and 1.18, respectively) (8) in this study.

The inconsistency on the association between GC risk and BMI may be because GC is not a single disease but consists of several sub-diseases with different clinical courses and causes, such as cardia and non-cardia GC according to anatomical location, diffuse and intestinal subtypes according to histological findings (10).

The existing meta-analyses for only cardia GC showed consistent patterns in that the GC risk gradually increased from normal BMI to overweight and obesity (5–7). However, most of the results based on cardia GC were derived from North American and Europe (5–7, 11).

Since cardia GC is less than 10% among total GC in Asians (12) and the prevalence of obesity, BMI more than equal to 30kg/m² according to WHO standards, is low compared to Western population (13–14), it is not easy to have sufficient statistical power to assess the association between obesity and cardia GC based on a single study in Asia. In order to overcome such limitations and to investigate the causal association between BMI and GC, a large-scale Asian prospective cohort study data is needed.

Therefore, we aimed to confirm a non-linear association between BMI and the risk of overall GC and to identify the association between BMI and the risk of GC subtypes by anatomical and histological classification in a large-scale prospective cohort study pooling project based on East and Southeast Asian population.

Materials and Methods

Study design and population

This study was based on the Asia Cohort Consortium (ACC), an international collaboration involving more than a million participants across Asia to investigate the etiology of various diseases. Details of ACC have been introduced in previous studies (15, 16). Thirteen cohorts from China, Japan, Korea, and Singapore recruiting 577,605 participants were included in this study. Relevant cohort investigators provided data on age, sex, country, year of recruitment, smoking status, alcohol drinking status, height, and weight at baseline along with GC incidence.

All covariates used in this study were harmonized by the ACC coordinating center. A dictionary containing details about data processing such as the contents of each variable, variable type, variable name, and precautions when handling variables is provided to each cohort. Each cohort send the data processed by the processing method written in the dictionary to the ACC coordinating center. Then, the ACC coordinating center confirm the content, format, and distribution of variables in the received data and integrates them into a unified format.

Height and weight were measured at baseline in all of the cohorts included in this study and participants without baseline height or weight data were excluded. We defined acceptable BMI range as 10.0–50.0 kg/m² and included participants within this range. We additionally excluded those lacking data on age at enrollment, GC incidence, or follow-up period, and total 554,037 subjects were left in this study. Because BMI in individual on the cusp of GC development, or in terminal illness, is usually low, there might be reverse causation between low BMI and GC. Hence, we excluded GC incidence diagnosed within 2 years from cohort enrollment and finally 538,835 study subjects including 8,997 GC cases were defined as study population. This study was approved by the institutional review board of Seoul National University Hospital (IRB No. H-0110–084-002 and H-0901–040-269) and followed the Declaration of Helsinki principles. All participants signed a written informed consent document.

Exposure and outcome

We used two classification criteria for BMI to find the association with GC risk in Asians. The first criterion is the BMI standard classification proposed by the WHO as follows: <18.5, 18.5–22.9, 23.0–24.9, 25.0–29.9, and ≥30.0 kg/m². However, there are a few obese Asians with BMI more than equal to 30.0 kg/m² (only 2.2% of the ACC participants) and there had been researches suggesting that the association between BMI, body fat, and health risk is different from that of Europeans. So we used the 2002’ modified BMI criterion for Asians suggested by WHO which can be applied to all ethnics in Asia (17), with criteria <18.5, 18.5–23, 23–25, 25–27.5, and ≥27.5 kg/m². The criteria had been used in previous ACC studies for association with mortality risk (16).

We defined GC incidence according to the International Classification of Diseases for Oncology 2^nd/3^rd version (C16.0-C16.9 for ICD-O-3 version; 151.0–151.9 for ICD-O-2 version). Follow-up period for GC incidence was defined as the interval between date of cohort enrollment and GC diagnosis date for cases or last follow-up date for non-cases. Of thirteen cohorts, eight cohorts included information on the specific GC location and pathological type. Cardia GC and non-cardia GC were classified according to the ICD-O codes (for non-cardia GC, C16.1-C16.9 for ICD-O-3 version; 151.1–151.9 for ICD-O-2 version; for cardia GC, C16.0 for ICD-O-3 version; 151.0 for ICD-O-2 version).

Lauren’s classification is one of the commonly used histologic classifications of GC and divides GC into mainly intestinal type and diffuse type according to tumor cells formation and behavior. Since two histologic types of GC (intestinal type and diffuse type) differ in pathology, epidemiology, and etiology, we chose this histologic classification in the analyses to assess the association between BMI and GC risk. Therefore, two major histologic types of GC, intestinal type and diffuse type GC, were considered in this study, and indeterminate type or mixed carcinoma were excluded from the subgroup analysis.

Statistical analysis

To evaluate risk for GC incidence associated with BMI, we calculated HR and corresponding 95% confidence intervals (CI), in the Cox proportional hazard regression models adjusted for potential confounders, including age at enrollment, sex, country (China, Japan, Korea, and Singapore), cohort, smoking status (never, past, and current), and alcohol drinking status (never and ever). All analyses were performed after excluding GC incidence diagnosed within 2 years from cohort enrollment. Additionally, we performed the sensitivity analysis by excluding GC case diagnosed within 4-year follow-up period from cohort enrollment in order to evaluate the association between BMI and GC risk while minimizing the possibility of reverse causation. We confirmed the Cox proportional hazard assumption based on the log–log survival curve for each category of BMI and confirmed that the assumption was not violated by with parallel curves. We performed meta-analysis using both fixed-effects model and random-effects model and evaluated heterogeneity among studies using the I-squared and the Cochran Q statistics. We additionally performed the restricted cubic spline analysis by using the SAS LGTPHCURV9 macro code to assess non-linear association between BMI and GC risk.

On eight cohorts included information on the specific GC location and pathological type, we performed subgroup analysis according to anatomical location (non-cardia GC vs. cardia GC), pathological type (intestinal type GC vs. diffuse type GC), and age onset in GC cases (late onset [onset age > 70 years old], mid-aged onset [45 years old ≤ onset age ≤ 70 years old], and early onset [onset age < 45 years old]). We included unspecified GC (C16.9 for ICD-O-3 version; 151.9 for ICD-O-2 version) (N=2,014) and GC without information on ICD-O code (N=2,182) within the non-cardia GC group because most of GC cases identified in Asian countries are non-cardia GC, especially Japan and Korea reporting a proportion of non-cardia GC higher than 90% (12).

To assess the impact of smoking status on the association between BMI and GC risk, we classified study subjects into groups according to a cigarette smoking status, and evaluated GC risk according to BMI levels in each stratified group. We additionally assessed the association between BMI and GC risk in each stratified population by sex.

All statistical analyses were performed using the SAS, version 9.4 (SAS Institute, Cary, NC).

Data availability

The data underlying this article were provided by Asia Cohort Consortium by permission. Data will be shared on request with the permission of the corresponding author and Asia Cohort Consortium.

Results

Participants in their respective cohorts were enrolled during 1984–2006 and followed for a median period of 14.9 years. Among 554,037 participants in the study, a total of 10,006 GC cases were recorded during the follow-up period (Table 1) and 8,997 GC cases were left after excluding those diagnosed within 2-year from the cohort enrollment. Overall, men, cigarette smokers, and alcohol drinkers comprised 45.3%, 39.8% and 40.5% of the subjects, respectively (data partially shown in Supplementary Table 1). Mean age at enrollment was 54.4 years, with variation across the cohorts. Mean BMI at baseline was 23.3 kg/m² (standard deviation, 3.2 kg/m²). Most of GC cases were non-cardia GC (including unspecified GC) (93.2%) in this study subjects. Intestinal type GC was more prevalent than diffuse type GC in this study (73.0% vs. 6.4%).

Table 1.

Baseline characteristics of the participating cohorts in the Asia Cohort Consortium from 1984 to 2015

	Person-year	Cohort	Enrollment	Follow-up year	Men	Age	GC Cases	BMI 1 (kg/m2)
								<18.5	18.5–22.9	23–24.9	25–29.9	≥30.0
			Year	Median (IQR)	%	Mean (SD)	N	%	%	%	%	%
ACC	7,991,847	554,037	1984–2006	14.9 (10.5–17.6)	45.3	54.4 (10.3)	10,006	5.4	43.4	24.7	23.7	2.8
Japan
RERF	1,106,385	49,730	1963–1993	22.5 (11.3–32.9)	39.7	52.2 (13.7)	2,355	13.3	53.9	17.1	13.8	1.9
3pref. Miyagi	340,460	29,457	1984	15.0 (7.8–15.0)	44.9	56.8 (11.1)	286	5.3	45.0	24.0	23.1	2.7
3pref. Aichi	371,202	32,178	1985	15.2 (7.5–15.3)	47.4	56.2 (11.3)	582	10.0	54.3	20.9	13.8	1.0
Miyagi	703,009	43,776	1990	17.6 (17.6–17.6)	48.2	51.9 (7.5)	799	2.4	41.4	26.7	26.8	2.7
Takayama	402,895	29,654	1992	15.6 (15.5–15.6)	45.7	55.4 (12.7)	628	8.6	54.6	21.8	14.2	0.9
JPHC1	890,403	42,641	1990–1992	22.6 (22.1–22.8)	47.8	49.6 (5.9)	1,016	2.6	42.2	26.5	26.0	2.7
JPHC2	978,828	55,571	1993–1995	19.7 (18.7–19.8)	47.4	54.2 (8.8)	1,359	3.6	43.2	25.4	25.0	2.8
Ohsaki	482,367	44,741	1995	13.2 (9.6–13.2)	48.1	59.8 (10.3)	899	3.7	41.4	25.8	26.3	2.9
Korea
KMCC	260,383	18,961	1993–2004	13.4 (10.9–17.4)	40.0	53.7 (14.4)	403	4.5	40.9	23.5	27.7	3.5
KNCC	37,562	7,747	2002–2015	4.7 (3.4–6.2)	36.0	52.7 (8.4)	23	2.2	41.7	26.2	27.0	2.9
Singapore
SCHS	723,842	63,240	1993–1999	12.4 (9.8–13.8)	44.2	56.5 (8.0)	641	6.5	41.6	29.8	19.2	3.0
China
SMHS	579,937	61,436	2001–2006	9.6 (9.0–10.7)	100	55.4 (9.7)	562	4.3	36.6	26.1	30.5	2.6
SWHS	1,114,574	74,905	1996–2000	15.3 (14.7–16.0)	0.0	52.6 (9.1)	453	3.4	37.5	23.7	30.2	5.1

Abbreviations: GC, Gastric cancer; BMI, Body mass index; IQR, Interquartile range; SD, Standard deviation; N, Number; ACC, Asia cohort consortium; RERF, Radiation Effects Research Foundation; 3pref. Miyagi, Three-Prefecture Cohort Study, Miyagi; 3pref. Aichi, Three-Prefecture Cohort Study, Aichi; JPHC1, Japan Public Health Center-based prospective Study1; JPHC2, Japan Public Health Center-based prospective Study2; KMCC, Korean Multi-center Cancer Cohort; KNCC, Korean National Cancer Screenee Cohort; SCHS, Singapore Chinese Health Study; SMHS, Shanghai Men’s Health Study; SWHS, Shanghai Women’s Health Study

^1.Mean of BMI=23.3 kg/m² (SD, 3.2 kg/m²)

No significantly increased GC risk was observed in the obese group defined as BMI ≥30 kg/m² according to BMI classification by previous WHO standards (HR=1.08, 95% CI 0.93–1.25) (Supplementary Table 1). However, significantly increased GC risks were found in the underweight group (BMI <18.5 kg/m²) and obese group defined as the new WHO standard for Asians (BMI ≥27.5 kg/m²) (HR=1.15, 95% CI 1.05–1.25 for underweight; HR=1.12, 95% CI 1.03–1.22 for obesity, respectively), compared to the reference BMI (23.0–24.9 kg/m²) (Table 2). The results based on Asian BMI standards were persistent even in a limited group not exposed to tobacco smoking (HR=1.23, 95% CI 1.06–1.43 for underweight; HR=1.23, 95% CI 1.08–1.39 for obesity, respectively). The U-shaped association between BMI and GC risk was re-confirmed in the results from the restricted cubic spline model (Supplementary Figure 1).

Table 2.

BMI and risk for GC in the ACC¹ from 1984 to 2015

	BMI (kg/m2)
	< 18.5	18.5–22.9	23.0–24.9	25.0–27.4	≥ 27.53
Cohort
HR (95% CI)2	1.15 (1.05–1.25)	1.07 (1.02–1.13)	1	1.01 (0.94–1.08)	1.12 (1.03–1.22)
Person-year	419,544	3,519,088	1,947,967	1,355,198	734,302
Cases, N	692	4,290	1,977	1,318	720
Cohort with non-smoking & non-drinking
HR (95% CI)2	1.23 (1.06–1.43)	1.10 (1.01–1.20)	1	1.12 (1.00–1.24)	1.23 (1.08–1.39)
Person-year	236,973	2,052,568	1,172,545	836,105	488,037
Cases, N	234	1,492	761	599	386
Meta-analysis
HR (95% CI)4	1.01 (0.85–1.20)	1.06 (0.98–1.14)	1	1.00 (0.92–1.09)	1.13 (1.02–1.25)
HR (95% CI)5	1.12 (1.03–1.22)	1.08 (0.95–1.23)	1	1.05 (0.86–1.27)	1.21 (1.01–1.48)
I2 (%)	58%	51%		25%	23%
P Cochran Q test	0.07	0.08		0.46	0.49

Abbreviation: BMI, Body mass index; GC, Gastric cancer; ACC, Asia Cohort Consortium

^1.Analyzed excluding cases diagnosed within 2-year of cohort enrollment

^2.Adjusted for age at cohort enrollment, sex, country, cigarette smoking status and alcohol drinking status

^3.Participants with BMI ≥ 27.5 kg/m² are 9.2% and 8.9% among ACC cohort population and incident GC cases, respectively

^4.Random effects model

^5.Fixed effect model

Moreover, the association between BMI level and GC risk was also consistent in the meta-analysis based on the fixed-effect model (HR=1.12, 95% CI 1.03–1.22; HR=1.21, 95% CI 1.01–1.48, respectively). Although there was a significant heterogeneity between cohorts in the association between BMI and GC risk, no individual study alone led to the results of the pooled analysis (Supplementary Figures 2–5). In subgroup analysis by ethnicity, the U-shaped associations of BMI with GC risk were found only in East Asians, not Southeast Asians (Supplementary Table 2). In sensitivity analysis excluding GC cases confirmed within four years from cohort enrollment, the U-shaped association between BMI and GC risk remained (Supplementary Table 3).

The U-shaped association with BMI also remained in non-cardia GC subtype, (HR=1.22, 95% CI 1.10–1.35 for underweight; HR=1.09, 95% CI 0.98–1.21 for obesity, respectively) (Table 3). The similar pattern was also found in intestinal type GC subtype (HR=1.14, 95% CI 1.01–1.28 for underweight; HR=1.11, 95% CI 0.98–1.25 for obesity, respectively). There were some differences in the pattern of association between intestinal type GC subtype and BMI according to sex. The associations of increased GC risk with underweight was observed only in women (HR=1.51, 95% CI 1.24–1.83), and obesity was associated with elevated risk of GC only in men (HR=1.13, 95% CI 0.97–1.32), though this association was not statistically significant (Table 4). In contrast, no significant association with the risk of cardia GC and diffuse type GC was observed in underweight and obesity groups (Tables 3 and 4). The U-shaped association was also found in late onset GC (HR=1.13, 95% CI 1.01–1.26 for underweight; HR=1.21, 95% CI 1.08–1.36 for obesity, respectively); whereas early onset GC risk was not associated with BMI levels (Table 5).

Table 3.

BMI and risk for GC subtype according to anatomical location in the ACC ¹

	BMI (kg/m2)
	< 18.5	18.5–22.9	23.0–24.9	25.0–27.4	≥ 27.5
CGC
HR (95% CI)2	0.89 (0.58–1.38)	0.99 (0.78–1.26)	1	1.16 (0.86–1.57)	0.94 (0.62–1.43)
Cases, N	27	227	100	76	28
NCGC 3
HR (95% CI)2	1.22 (1.10–1.35)	1.09 (1.02–1.16)	1	0.97 (0.89–1.05)	1.09 (0.98–1.21)
Cases, N	540	3,126	1,341	858	449
Female
CGC
HR (95% CI)4	1.46 (0.75–2.85)	0.72 (0.43–1.19)	1	0.92 (0.50–1.69)	0.46 (0.19–1.12)
Cases, N	15	41	25	19	8
NCGC 3
HR (95% CI)4	1.42 (1.20–1.68)	1.09 (0.97–1.22)	1	1.00 (0.86–1.16)	1.05 (0.89–1.24)
Cases, N	224	992	427	314	197
Male
CGC
HR (95% CI)4	0.68 (0.37–1.27)	1.17 (0.89–1.54)	1	1.20 (0.85–1.70)	1.06 (0.65–1.73)
Cases, N	12	186	75	57	20
NCGC 3
HR (95% CI)4	1.15 (1.01–1.31)	1.09 (1.01–1.18)	1	0.95 (0.86–1.06)	1.12 (0.97–1.29)
Cases, N	316	2134	914	544	252

Abbreviation: BMI, Body mass index; GC, Gastric cancer; ACC, Asia Cohort Consortium; CGC, Cardia gastric cancer; NCGC, Non-cardia gastric cancer

^1.Analyzed excluding cases diagnosed within 2-year from cohort enrollment

^2.Adjusted for age at cohort enrollment, sex, country, cigarette smoking status and alcohol drinking status

^3.Cases including unspecified GC

^4.Adjusted for age at cohort enrollment, country, cigarette smoking status and alcohol drinking status

Table 4.

BMI and risk for GC subtype according to histological classification in the ACC ¹

	BMI (kg/m2)
	< 18.5	18.5–22.9	23.0–24.9	25.0–27.4	≥ 27.5
Intestinal type GC
HR (95% CI)2	1.14 (1.01–1.28)	1.07 (0.99–1.15)	1	0.96 (0.87–1.06)	1.11 (0.98–1.25)
Cases, N	400	2,455	1,065	664	360
Diffuse type GC
HR (95% CI)2	0.99 (0.63–1.55)	1.22 (0.95–1.57)	1	1.17 (0.85–1.61)	1.27 (0.85–1.89)
Cases, N	26	224	83	67	35
Female
Intestinal type GC
HR (95% CI)3	1.51 (1.24–1.83)	1.10 (0.96–1.26)	1	1.00 (0.84–1.19)	1.07 (0.88–1.30)
Cases, N	167	718	304	226	148
Diffuse type GC
HR (95% CI)3	0.96 (0.53–1.74)	1.06 (0.75–1.49)	1	0.97 (0.62–1.52)	1.14 (0.69–1.89)
Cases, N	15	111	48	33	22
Male
Intestinal type GC
HR (95% CI)3	1.02 (0.88–1.19)	1.06 (0.97–1.16)	1	0.94 (0.83–1.05)	1.13 (0.97–1.32)
Cases, N	233	1,737	761	438	212
Diffuse type GC
HR (95% CI)3	1.09 (0.55–2.18)	1.47 (1.00–2.15)	1	1.50 (0.93–2.41)	1.49 (0.79–2.83)
Cases, N	11	113	35	34	13

Abbreviation: BMI, Body mass index; GC, Gastric cancer; ACC, Asia Cohort Consortium

^1.Analyzed excluding cases diagnosed within 2-year from cohort enrollment

^2.Adjusted for age at cohort enrollment, sex, country, cigarette smoking status and alcohol drinking status

^3.Adjusted for age at cohort enrollment, country, cigarette smoking status and alcohol drinking status

Table 5.

BMI and risk for GC subtype according to onset age in the ACC ¹

	BMI (kg/m2)
	< 18.5	18.5–22.9	23.0–24.9	25.0–27.4	≥ 27.5
Early onset GC 3
HR (95% CI)2	1.07 (0.50–2.31)	1.19 (0.83–1.72)	1	1.08 (0.67–1.74)	0.84 (0.43–1.64)
Cases, N	8	97	41	29	11
Late onset GC 3
HR (95% CI)2	1.13 (1.01–1.26)	1.03 (0.96–1.11)	1	1.03 (0.93–1.13)	1.21 (1.08–1.36)
Cases, N	523	2,467	1,046	683	391

Abbreviation: BMI, Body mass index; GC, Gastric cancer; ACC, Asia Cohort Consortium

^1.Analyzed excluding cases diagnosed within 2-year from cohort enrollment

^2.Adjusted for age at cohort enrollment, sex, country, cigarette smoking status and alcohol drinking status

^3.The onset age of GC: ‘Early onset’ and ‘Late onset’ were defined as ‘Onset age < 45-year’ and ‘Onset age > 70-year’, respectively.

The associations of underweight and obesity with increased risk of non-cardia GC, intestinal type GC, and late onset GC were persistent even when we excluded GC cases diagnosed within 4-year from cohort enrollment (Supplementary Tables 4–6).

Among non-smokers, a U-shaped association of BMI on GC risk was consistently observed (HR=1.23, 95% CI 1.06–1.43 for underweight; HR=1.23, 95% CI 1.08–1.39 for obesity, respectively) (Supplementary Table 7). In contrast, the impact of underweight on increasing GC risk lessened and the elevation of GC risk in obesity group was not observed in smokers (HR=1.12, 95% CI 0.99–1.25 for underweight; HR=1.00, 95% CI 0.88–1.14 for obesity, respectively).

Discussion

We found elevated GC risk in both underweight and obesity groups based on this prospective cohort study consisting of more than 500,000 Asians. According to the anatomical subtype, non-cardia GC, which accounts for the majority of GC in Asians, was also related to the BMI in the same manner of association between BMI and overall GC risk. Intestinal type and late-onset GCs also exhibited patterns of association with BMI similar to those observed for total GC.

Several epidemiologic studies suggested a U-shaped pattern on the association between BMI and GC risk, but none of them could confirm the GC risk related to BMI level due to lack of statistical power (8, 18–20). One of the reasons was that it may be difficult to ascertain an increased risk of GC with underweight in the U.S. and European populations because underweighted people are rare and GC is uncommon compared to Asian population. On the contrary, in East Asians, obesity prevalence (BMI ≥30 kg/m²) is only 1%−5% (12), so it may be difficult to observe a high risk of GC in group with obesity.

In particular, cardia GC is known to be strongly associated with obesity; but the positive association was reported in only American and European studies (5, 11). There were three Asian population studies for cardia GC in relation to overweight or obesity. We conducted meta-analysis using previous three Asian studies (21–23), however increased risk of cardia GC in accordance with obesity was not confirmed (summary relative risk=1.23, 95% CI 0.93–1.64) (Supplementary Figure 6). In this study, we found that being overweight and obese in Asians did not increase the risk of cardia GC. Rather, excess BMI levels were associated with elevated risk for non-cardia GC, which was similarly observed in nested case-control studies derived from multiple Asian cohort studies (HpBCC) (20). Additional studies are warranted to investigate the underlying reasons for the observed difference in the BMI-cardia GC risk association between Asian and American/European populations.

In the cohort subjects who don’t smoke with adjustment for alcohol drinking status, the risks of GC in underweight population and obese population were still higher than that in population with reference BMI, suggesting that BMI is an independent factor involving in GC development regardless of the confounding effects of smoking or alcohol drinking.

Our result for U-shaped association with BMI on the risk of GC could be interpreted under the following biological mechanism. In general population, underweight, one of the forms of malnutrition, is associated with micronutrient deficiency (24, 25) and consequently causes poor immunity (26, 27), and increase oxidative stress (28). These processes above can increase the GC risk by elevating the likelihood of H. pylori infection and diminishing prevention effect against cancer. Additionally, smoking has been suggested to be inversely associated with body weight (29), thus smoking and low BMI are characteristics that can be clustered. When both factors exist concurrently, the carcinogenic effect of tobacco smoking may be greater due to a series of underweight courses for bad health mentioned above.

In contrast, the biological mechanism between overweight or obesity and GC risk may be the direct induction and progression of carcinogenesis due to a series of inflammatory responses and increased chronicity (30, 31). However, this was the mechanism explaining the association between the existing cardia GC and obesity. It has not been elucidated whether the association between non-cardia GC and obesity is due to chronic inflammatory condition. Therefore, additional studies are needed to determine mechanisms that could explain the link between non-cardia GC and overweight/obesity, especially in Asians.

In the analyses considering pathology and age at onset of GC, an elevation of GC risk according to underweight was observed only in case of intestinal type GC and late onset GC as well as non-cardia GC. It has been known that cardia GC is positively related to obesity and diffuse type GC in North American and European studies, which is more commonly observed in young GC patients and linked to hereditary factors rather than environmental factors compared to intestinal type GC (32). In light of the link among cardia GC, diffuse type, and young age onset in Westerners, it is a reasonable observation that the pattern of association of intestinal GC and late-onset GC with BMI in Asians having non-cardia GC in most cases (over 90%) is similar to the association between total GC and BMI.

When obesity was defined as the general obesity standard (BMI ≥30 kg/m²) by the WHO, it could be interpreted that only underweight is linked to the increased risk of GC and obesity (BMI ≥30 kg/m²) was not associated with an elevated GC risk in this present study. That result was clearly different from the previous meta-analyses showing a link between overweight/obesity and GC risk, mostly derived from Westerners’ studies. Further research is needed to determine whether the difference is due to variation in race itself, or that in the composition of GC or the exposure to environmental factors.

This study has several limitations. First, we did not consider H. pylori infection as a confounder in the analysis, despite the fact that it might be the key link between BMI and GC risk. Second, as only eight cohorts from Japan and Korea of the consortium had information of anatomical location of GC, the associations of BMI with cardia GC and non-cardia GC risk are limited in generalization. Third, elevated GC risk related to underweight could arise from reversed causation. Therefore, we performed analyses after excluding GC cases identified within 2-year from enrollment to minimize the reverse causation.

Despite these limitations, this study has several strengths. There were over 8,997 GC cases in this study even after excluding GC cases confirmed within 2 years from cohort enrollment, meaning the sample had sufficient statistical power even after eliminating GC cases diagnosed within 2 years of enrollment. This large-scale Asian cohort study presented results on whether the risk of GC according to BMI depends on the biological characteristics of GC (anatomical location, histological type, and age onset), though information on characteristics of GC was available only in eight cohorts.

In conclusion, this large-scale Asian cohort study established that underweight and obesity based on the Asian obesity standard (BMI ≥27.5 kg/m²) are associated with the increased risk of GC, especially non-cardia GC. The pattern in the association between BMI and GC risk was persistent in case of intestinal type GC and late onset GC. In contrast, even though this study was the largest Asian cohort study to date, increased risk of cardia GC in overweight and obese people in Asians couldn’t be confirmed since the proportion of overweight and obese population is low and most of GC cases are non-cardia GC in Asians.