Prospective cohort study of body mass index in adolescence and death from stomach cancer in Japan

Abstract

We investigated the association between body mass index (BMI) values and the risk of death from stomach cancer later in life. From 1992 to 2000, 28 443 participants (13 211 men and 15 232 women) were followed, and 129 deaths from stomach cancer (86 men and 43 women) occurred. BMI (BMI at the baseline) and BMI_20y (BMI at age 20 years) were obtained by questionnaire at the beginning of the study. Men in the middle and highest tertiles of BMI_20y had a higher risk of death from stomach cancer than those in the lowest tertile (after controlling for covariates, hazard ratio [HR] = 2.73 with 95% confidence interval [CI] 1.26–5.89 for the middle tertile; HR = 2.15 with 95% CI 0.97–4.73 for the highest tertile), although there was no significant trend (P = 0.30). Likewise, women in the middle and highest tertiles of BMI_20y had an increased risk of death from stomach cancer (HR = 5.17 with 95% CI 1.50–17.87 for the middle tertile; HR = 4.22 with 95% CI 1.18–15.05 for the highest tertile; P for trend = 0.05). The analysis of BMI at the baseline did not show a significant relationship with the risk of death from stomach cancer. These results suggest that a high BMI_20y is associated with an increased risk of death from stomach cancer. (Cancer Sci 2007; 98: 1785–1789)

It has been indicated that body size,^{( 1} , ^{2 )} especially body mass index (BMI),^{( 3} , ⁴ , ^{5 )} has an influence on carcinogenesis. Previous studies have demonstrated that high BMI in adulthood is a risk factor for several cancers, such as breast, ovarian, colorectal and prostate cancers.^{( 6} , ⁷ , ⁸ , ⁹ , ¹⁰ , ^{11 )} In contrast, results from studies on BMI in adulthood and stomach cancer have generally shown no relationship.^{( 12 )}

Stomach cancer is the second leading cause of cancer death in men and the leading cause of cancer death in women in Japan.^{( 13 )} For this serious health threat, many epidemiological studies on stomach cancer risk have been conducted in Japan. These studies have paid attention to the association between stomach cancer risk and diet,^{( 14} , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ^{19 )} drinking^{( 20} , ^{21 )} or smoking.^{( 22} , ²³ , ²⁴ , ²⁵ , ^{26 )} To our knowledge, only two studies in Japan have examined the direct association between BMI and stomach cancer risk. Kuriyama et al. prospectively assessed the relationship between BMI in adulthood and morbidity from various cancers in Japanese men and women, and observed statistically no significant association with risk of stomach cancer.^{( 27 )} Inoue et al. observed no significant association between BMI in adulthood and risk of stomach cancer in a case‐control study of Japanese postmenopausal women.^{( 28 )} However, they found that a high BMI at age 20 years increased the risk of stomach cancer.

Body mass index in adolescence but not in adulthood may influence the risk of stomach cancer. A significant positive association between BMI at age 20 years and the risk of stomach cancer was also reported in a study conducted in a Western population.^{( 29 )} In the present study, we examined the association between BMI at age 20 years and in adulthood, and the risk of death from stomach cancer later in life in a community‐based cohort of Japanese men and women.

Subjects and Methods

In September 1992, we established a study cohort with 36 990 men and women who were 35 years old or older and resided in Takayama, Japan.^{( 30 )} At the baseline, a self‐administered questionnaire was conducted within the cohort. The questionnaire contained information regarding smoking history, alcohol intake, education, physical activity and marital status. Thirty‐four thousand and eighteen subjects reported to the questionnaire, equaling a response rate of 92.0%. Some of the individuals (n = 595, 1.7%) completed 45% or less of the questionnaire, and others (n = 1871, 5.5%) provided unreliable or inconsistent responses; they were all excluded from the cohort. Further details of the exclusion criteria of the self‐administered questionnaire have been reported previously.^{( 30 )} The final fixed cohort consisted of 31 552 subjects. The present study was approved by the Ethical Committee of Gifu University School of Medicine.

Information on self‐reported height and weight at the baseline and recalled weight when participants were approximately 20 years old was collected from the questionnaire. Based on the information, BMI at the baseline (BMI) and BMI at age 20 years (BMI_20y) were estimated by dividing the baseline weight and weight at age 20 years in kg by height in m². Information about the validity and reproducibility of the self‐administered questionnaire concerning an individual's current height and weight has been reported elsewhere.^{( 31 )} The Spearman correlation coefficients, which compare the self‐reported values in the questionnaire and the values actually measured, were 0.93 for height and 0.97 for weight in both sexes.

Years of smoking and the number of cigarettes smoked each day were reported in the questionnaire. Individuals who had smoked a total of 20 or more packets of cigarettes in their life were defined as smokers. Diet, including alcohol intake, was assessed by a semiquantitative food‐frequency questionnaire. The food‐frequency questionnaire was designed to obtain information about the average frequency and serving size of 169 food items consumed during the previous year. The amount of food and nutrient intake were estimated from the Japanese Standard Tables of Food Composition, 5th edition. ⁽³²⁾ The questions on alcohol consumption included six types of alcohol (sake, beer, light beer, shochu, wine and hard liquor). For each item, information about nine frequency categories and the number of cups, glasses and bottles consumed was asked. The total amount of alcohol consumed in g was estimated based on responses to the questionnaire. Detailed information on the food‐frequency questionnaire, including that on validity and reproducibility, has been reported elsewhere.^{( 33 )} The amount of physical activity was measured in h per week. Participants were asked in the questionnaire if they had engaged in various types of activities within the previous week. Details of the physical activity questionnaire, including the results of the validity tests, have been reported elsewhere.^{( 34} , ^{35 )}

The study cohort was followed between 1 September 1992 and 31 December 1999. We excluded 410 men and 199 women who reported the history of partial or total gastrectomy at the baseline of this study. During 7 years of follow‐up, any death occurring in Takayama City as well as its cause was confirmed with data from the office of the National Vital Statistics. The Statistics and Information Department of the Japanese Ministry of Health and Welfare obtained information on all deaths and coded the causes of death using the International Classification of Diseases (ICD). Permission to review the data regarding dates and causes of deaths was obtained from the Ministry of Internal Affairs and Communications, Japan. Stomach cancer was considered to be the underlying cause of death when the ICD‐9 code was 151 or the ICD‐10 code was C16. Study subjects who moved away from Takayama during the study period were identified from the city residential registers.

To measure the association between BMI at baseline and at age 20 years and the risk of stomach cancer death over the 7‐year study period, we utilized the Cox proportional hazard model. From the model, the hazard ratios (HR) of death from stomach cancer by the level of BMI at baseline as well as at age 20 years were calculated with 95% confidence intervals (CI). The length of follow‐up of each subject was the time interval from the beginning of the study period to the earliest occurrence of one of the following events: the date of death from stomach cancer; the date of death due to any other cause; the date on which the person moved out of Takayama; or the end of the study period. Individuals were categorized into tertiles according to the distribution of baseline BMI as well as BMI_20y. The lowest tertile was always used as a reference category. Initially the association between baseline BMI and the risk of death from stomach cancer was analyzed after controlling for age. The association between BMI_20y and the risk of death from stomach cancer was analyzed after controlling for age and baseline BMI. Further, several factors covering the spectrum of the likely measured confounders were used for adjustments. These variables included smoking history (pack‐years), alcohol intake (g/day), education (>8 years, 9–11 years or <12 years), physical activity score (METs) and marital status (married, never married or others). Tests for trends were carried out with continuous variables using the median value of the categories. All statistical analyses were carried out using the SAS program (SAS Institute, Cary, NC, USA).

We excluded the subjects whose BMI_20y values were missing to assess the relationship between BMI_20y and the risk of death from stomach cancer. That left 13 211 men and 15 232 women in the analysis. For the analysis concerning BMI at the baseline and the risk of death from stomach cancer, we excluded the subjects who had died from stomach cancer within the first 2 years of follow‐up. Thus, the dataset consisted of 13 148 men and 15 387 women in the analysis.

Results

Table 1 summarizes the baseline characteristics of the study participants by tertiles of BMI_20y. In both sexes, age and smoking status increased across the strata of BMI_20y. In contrast, the percentage of people who had undertaken more than 12 years of education decreased across the strata of BMI_20y in both sexes. The physical activity score increased across the strata of BMI_20y in men. Alcohol intake decreased across the strata of BMI_20y in women. BMI at baseline was significantly associated with marital status and years of education (data not shown). The percentage of married men as well as women increased across the strata of BMI at baseline. A longer period of education was associated with an increase in BMI at baseline in men and a decrease in women. The other variables listed in Table 1 showed no significant associations with BMI at baseline.

Table 1.

Baseline characteristics by tertile of body mass index (BMI) at age 20 years in the Takayama study, 1992–1999

BMI20y (kg/m2)	Men			Women
	Low <20.3	Medium 20.3–22.2	High >22.2	Low <20.0	Medium 20.0–22.0	High >22.0
Age (mean years)	51.5 (11.2)	53.0 (11.4)	57.7 (12.8)	51.8 (11.4)	53.7 (12.2)	59.1 (13.9)
Alcohol intake (mean g/day)	40.1 (40.3)	42.3 (41.4)	41.5 (41.5)	8.6 (18.8)	7.6 (15.6)	7.1 (15.7)
Smoking (mean pack‐years)	21.8 (13.7)	23.4 (13.9)	24.9 (14.8)	10.1 (9.0)	10.8 (10.5)	11.1 (9.8)
Education of >12 years (% yes)	49.7	45.3	34.3	41.7	36.3	25.0
Physical activity score (METs; mean)	26.1 (40.5)	28.3 (41.5)	29.2 (43.5)	20.1 (29.8)	21.2 (31.2)	18.6 (29.9)
Marital status (% married)	91.7	92.2	90.7	79.2	78.1	69.6

Values in parentheses are standard deviations. BMI_20y, body mass index at age 20 years; METs, metabolic equivalents.

During the follow‐up over 7 years, 129 deaths (86 men and 43 women) occurred due to stomach cancer. The associations between BMI_20y and risk of death from stomach cancer are shown in Table 2. The Pearson correlation coefficients between BMI at baseline and BMI_20y were 0.38 (P < 0.001) in men and 0.31 (P < 0.001) in women. Additional adjustment for baseline BMI was carried out in the calculation of stomach cancer risk according to tertile of BMI_20y. The men in the middle tertile and the subjects in the highest tertile of BMI_20y were at significantly increased risk of death from stomach cancer compared with those in the lowest tertile after controlling for age or age plus baseline BMI. However, the linear trend of the association between BMI_20y and the risk of death from stomach cancer was not significant in men. Additional adjustment for smoking history, alcohol intake, physical activity score, education and marital status somewhat attenuated the risk for the highest tertile of BMI_20y. Significantly increased risk of death from stomach cancer was observed among women in the middle and highest tertiles of BMI_20y compared with women in the lowest tertile after controlling for age or age plus baseline BMI. The trend was statistically significant. Additional adjustment for other covariates somewhat increased the HR for the middle and highest tertile of BMI_20y. Adjustment for BMI at the baseline somewhat strengthened the association between BMI_20y and the risk of death from stomach cancer in men as well as women.

Table 2.

Hazard ratios (HR) of death from stomach cancer by tertile of body mass index at age 20 years (BMI_20y) in the Takayama study, 1992–1999

BMI20y	Men				Women
	Tertile of BMI20y			P for trend	Tertile of BMI20y			P for trend
	Low	Medium (95% CI)	High (95% CI)		Low	Medium (95% CI)	High (95% CI)
No. of cases	12	33	41		4	17	22
No. at baseline	4427	4394	4390		5104	5096	5032
Age‐adjusted HR	1.00	2.46 (1.27–4.76)	2.03 (1.06–3.90)	0.23	1.00	3.64 (1.22–10.85)	3.26 (1.10–9.65)	0.03
Multivariate HR †	1.00	2.57 (1.32–5.01)	2.40 (1.23–4.67)	0.06	1.00	4.00 (1.34–11.95)	3.98 (1.32–11.99)	0.01
Multivariate HR ‡	1.00	2.53 (1.18–5.43)	1.72 (0.79–3.73)	0.76	1.00	4.64 (1.35–15.97)	3.35 (0.96–11.72)	0.18
Multivariate HR §	1.00	2.73 (1.26–5.89)	2.15 (0.97–4.73)	0.30	1.00	5.17 (1.50–17.87)	4.22 (1.18–15.05)	0.05

^†Adjusted for age and baseline body mass index (BMI).

^‡ Adjusted for age, smoking history, alcohol intake, education, physical activity score and marital status.

^{§ §} Adjusted for age, smoking history, alcohol intake, education, physical activity score, marital status and baseline BMI. CI, confidence interval.

Table 3 shows the risk of death from stomach cancer by the level of BMI at the baseline for both sexes. The HR in the highest tertile of the BMI tended to be lower than that in the lowest tertile in men and women, but the associations were not statistically significant.

Table 3.

Hazard ratios (HR) of death from stomach cancer by tertile of body mass index (BMI) at baseline in the Takayama study, 1992–1999

BMI	Men				Women
	Tertile of baseline BMI			P for trend	Tertile of baseline BMI			P for trend
	Low	Medium (95% CI)	High (95% CI)		Low	Medium (95% CI)	High (95% CI)
No. of cases	29	20	16		16	12	8
No. at baseline	4393	4404	4351		5138	5131	5118
Age‐adjusted HR	1.00	0.84 (0.47–1.48)	0.82 (0.44–1.52)	0.47	1.00	0.90 (0.42–1.91)	0.59 (0.25–1.38)	0.34
Multivariate HR †	1.00	0.68 (0.34–1.33)	0.53 (0.24–1.20)	0.12	1.00	0.96 (0.43–2.13)	0.50 (0.19–1.32)	0.28

^† Adjusted for age, smoking history, alcohol intake, education, physical activity score and marital status. CI, confidence interval.

We repeated the analysis among men and women combined. After controlling for sex and the other covariates, the HR were 2.87 (95% CI 1.57–5.25) and 2.63 (95% CI 1.42–4.86) for the middle and highest tertiles of BMI_20y, respectively (P for trend = 0.02). The corresponding figures for BMI at baseline were 0.66 (95% CI 0.39–1.10) and 0.71 (95% CI 0.43–1.18), respectively (P for trend = 0.15).

The associations between height and risk of death from stomach cancer are shown in Table 4. The inverse association between height and the risk of death from stomach cancer in men was of borderline significance after controlling for covariates (P for trend = 0.06). However, height was unrelated to the risk of death from stomach cancer in women.

Table 4.

Hazard ratios (HR) of death from stomach cancer by tertile of body height at baseline in the Takayama Study, 1992–1999

Body height	Men				Women
	Tertile of baseline body height			P for trend	Tertile of baseline body height			P for trend
	Low	Medium (95% CI)	High (95% CI)		Low	Medium (95% CI)	High (95% CI)
No. of cases	54	29	6		34	6	6
No. at baseline	4990	4924	3753		6488	4830	4545
Age‐adjusted HR	1.00	0.95 (0.59–1.51)	0.43 (0.18–1.04)	0.66	1.00	0.43 (0.18–1.06)	0.66 (0.26–1.70)	0.36
Multivariate HR †	1.00	0.83 (0.46–1.50)	0.30 (0.09–1.02)	0.06	1.00	0.88 (0.26–3.01)	1.34 (0.35–5.11)	0.77

^† Adjusted for age, smoking history, alcohol intake, education, physical activity score and marital status. CI, confidence interval.

Additional adjustment for height did not alter the association between BMI_20y and risk of death from stomach cancer substantially; HR were 2.63 (95% CI 1.22–5.70) and 2.03 (95% CI 0.92–4.50) for the middle and highest tertiles of BMI_20y, respectively (P for trend = 0.36). The corresponding figures for women were 5.06 (95% CI 1.50–17.57) and 4.03 (95% CI 1.12–14.48), respectively (P for trend = 0.07). Similarly, the association between height and risk of death from stomach cancer was not altered substantially after controlling for BMI_20y; HR for men were 0.89 (95% CI 0.50–1.58) and 0.43 (95% CI 0.15–1.26) for the middle and highest tertiles, respectively (P for trend = 0.27), and those for women were 0.52 (95% CI 0.19–1.43) and 0.99 (95% CI 0.36–2.68), respectively (P for trend = 0.57 in women).

An additional adjustment for intake of foods and nutrients, such as total energy, salt, vegetables and fruits, did not alter the associations between BMI at the baseline and the risk of death from stomach cancer. The adjustment for these variables also did not alter the association between BMI_20y and the risk of death from stomach cancer.

We also examined the association between BMI_20y and the risk of death from all types of cancer. The associations between BMI_20y and the risk of death from all types of cancer showed no statistical significance. Among men, HR of death from all types of cancer for the highest and middle tertiles were 0.89 (95% CI 0.66–1.19) and 0.86 (95% CI 0.65–1.14), respectively, after adjusting for the covariates, compared to the lowest tertile. The corresponding figures for women were 0.84 (95% CI 0.54–1.31) and 1.24 (95% CI 0.84–1.83), respectively.

Discussion

We found an elevated risk of death from stomach cancer among subjects who were in the middle and highest tertiles of BMI at age 20 years compared with that among the subjects in the lowest tertile. We could not evaluate the association of BMI_20y with each type of cancer besides stomach cancer well because of the small number of deaths. However, we observed no association between BMI_20y and the risk of death from all types of cancer. Our results indicate that BMI at age 20 years is an important factor that is associated with the risk of death from stomach cancer. The BMI at baseline was inversely associated with death from stomach cancer in men and women, but these associations were not statistically significant. This inverse association may be explained by the effect of underlying illness due to stomach cancer on BMI at the baseline, although our analyses excluded the subjects whose deaths occurred during the first 2 years of the study period.

To our knowledge, only two studies have addressed the association between the risk of stomach cancer and BMI in adolescence.^{( 28} , ^{29 )} In a population‐based case‐control study, Hansson et al. found that the highest BMI quartile at age 20 years was associated with an increased risk of gastric cancer (odds ratio [OR] = 2.27 in men, OR = 2.39 in women).^{( 29 )} In another case in Japan, Inoue et al. observed no significant association between BMI in adulthood and risk of stomach cancer (OR = 1.6 in women at age 20 years) in a case‐control study of postmenopausal women.^{( 28 )} However, they found that a high BMI at age 20 years increased the risk of stomach cancer. These findings were similar to the findings that we obtained. The lack of association between BMI in middle age and the risk of stomach cancer has been reported in several studies.^{( 2} , ¹² , ^{36 )}

High BMI in adolescence has been associated with some other cancers including ovarian cancer, breast cancer and colorectal cancer.^{( 5} , ³⁷ , ³⁸ , ^{39 )} Okasha et al. speculated that over‐nutrition in early life may be associated with high BMI and insulin‐like growth factor‐1 levels, and thereby could lead to the increase in breast cancer risk.^{( 37 )} Similar mechanisms may explain the observed association between BMI at age 20 years and the risk of death from stomach cancer, although the implication of IGF‐1 in this association is equivocal. We observed no dose–response relationship between the values of BMI_20y and death from stomach cancer. There may be a threshold for the effect on the risk of stomach cancer.

In the present study, body height was inversely associated with the risk of death from stomach cancer in men, although the association was not statistically significant. Shorter body height has been shown previously to be a risk factor for stomach cancer with some consistency,^{( 11} , ²⁹ , ⁴⁰ , ⁴¹ , ^{42 )} but not all.^{( 2} , ²⁸ , ^{43 )} Height as well as BMI at age 20 may reflect childhood nutrition and health. However, we failed to observe an inverse association between height and risk of death from stomach cancer in women.

This study has several limitations. Due to the size of the cohort and the length of the follow‐up, the number of deaths from stomach cancer that occurred during the study period was relatively small. In our cohort, the distribution of the BMI values was moderately homogeneous, and subjects who had extremely high or low BMI were scarce. The results of the validity test showed a high correlation between the self‐reported baseline BMI values and actual measured values in our study (correlation coefficients were 0.93 for height and 0.97 for weight in both sexes) as well as other studies (e.g. Kawada et al.,^{( 44 )} Klipstein‐Grobusch et al.^{( 45 )} and Spencer et al.^{( 46 )}). BMI_20y was calculated based on the current height and past weight of 15 or more years ago. The accuracy and reliability of reports of a person's weight in the past could be questioned in our study. However, Klipstein‐Grobusch et al. evaluated the recalled information on past bodyweight.^{( 45 )} In their study, 75.8% of middle‐aged subjects showed reproducibility of their weight at age 25 years within an error stratum range of ±3 kg. Therefore, the measurement error regarding recall of past bodyweight should not be great. In addition, it is unlikely that subjects who had over‐reported their past bodyweight tended to have a higher risk of stomach cancer death.

We should use body height at 20 years of age rather than that at the baseline for calculating BMI at age 20 years. Fujiwara reported that body height decreases approximately 1.2 cm in men and 1.3 cm in women at age 50–59 years and 1.5 cm in men and 2.1 cm in women at age 60–69 years in comparison with the highest height.^{( 47 )} Therefore, the calculated BMI_20y in our study might be overestimated. However, it is unlikely that the observed positive association between BMI_20y and the risk of death from stomach cancer is due to the greater decrease in height among those with the greater BMI_20y.

In the present study, we were not able to obtain information on stomach cancer incidence within the cohort. However, it is unlikely that BMI in youth influences stomach cancer mortality exclusively and not the incidence of stomach cancer. A similar pattern of relationship that we observed between BMI in youth and stomach cancer mortality should exist with regard to stomach cancer incidence.