Participants

All participants were employees of the Tokyo Gas company (Tokyo, Japan) that supplies natural gas to the Tokyo area. The number of personnel in the Tokyo Gas company was 12,899 (male: 11,978 and female: 921) at the end of March 1982 when the study was instituted. Most of the workers were sedentary during a typical work day. Female workers were excluded from the present analysis because of the small number. Participants completed a submaximal exercise test, annual health examinations, and self-administered questionnaires regarding their health habits between1982 and 1988. People who were diagnosed as having cancer, cardiovascular disease, diabetes mellitus, gastrointestinal disease, and tuberculosis based on the baseline health examination between 1982 and 1988 (n = 1162) were excluded from the present study to prevent the influence of these diseases and medication use on the assessment of CRF. In addition, the participants were given an interview at the time of their submaximal exercise test between 1982 and 1988 to check for the presence or absence of asthma, liver disease, kidney disease, bone fractures, torsional deformity of the lumbar spine, lower back pain, and neuralgia. If the participants had any one of these conditions, they were also excluded from the CRF test (n = 1777). Finally, to minimize potential bias from undiagnosed cancer that may have led to weight loss, participants were excluded if they had a BMI that was less than 18.5 kg · m^-2 (n = 279). After these exclusions, the number of participants included in the present study was 8760 men.

This study was approved by the ethics committee of the National Institute of Health and Nutrition of Japan.

Health examination and fitness test

In Japan, the Industrial Health and Safety Law and related regulations require the employer to conduct annual health examinations of all employees. Employees are required by law to participate. Height and weight were measured on a standard physician’s scale with light clothing but no shoes. BMI was calculated from measured weight and height. Resting blood pressure was obtained with standard auscultatory methods. A questionnaire was administered before the start of the fitness tests to assess the individual’s smoking and alcohol habits. Participants underwent a submaximal exercise test on a cycle ergometer. The exercise test consisted of two or three four-minute, progressively increasing exercise stages. The initial loads for the participants in the 19–29, 30–39, 40–49, and 50–59 years of age brackets were 600 kpm, 525 kpm, 450 kpm, and 375 kpm, respectively. Heart rate was measured using an electrocardiogram monitor (Cardiosuper 2E32, Sanei Sottki Co. Ltd., Tokyo, Japan), and 85% of their age-predicted maximal heart rate (220 minus age in years), was set as the target heart rate. The load was increased by 225 kpm for each stage among all age groups, until heart rates during the course of the exercise reached the target heart rate, or until completion of the third stage. Based on the work rate obtained from the last one minute of the final stage and heart rate obtained from the last 10 seconds, maximal oxygen uptake (VO2max) was estimated using the Åstrand-Ryhming Nomogram [21] and Åstrand’s Nomogram correction factors [22].

Mortality surveillance

We followed the participants for cancer mortality until June 30, 2003. The mean follow-up period was 20.2 years (range: 15–253 months). We ascertained mortality in two ways. The data for death occurring during employment at the Tokyo Gas company was available from the company (n = 167). For those who died after retirement, death was identified based on a members list kept by the Tokyo Gas Retiree Office (n = 195). Individuals retired from the company have access to a variety of welfare services for a minimal membership fee, paying this membership fee to the secretariat every year. Because the Office collects this annual fee from members every year, the secretariat has contact annually with the member. This allows annual follow-up of every participant. The majority of retirees subscribe to the Tokyo Gas Retiree Office; only 311 people (3.4%) did not. Thus, these 311 had their follow up time censored at the time they retired from Tokyo Gas Company. Additional details of the mortality surveillance have been previously described [20].

Data analysis

Man-years was calculated from the year of the baseline examination (1982–1988) to the year of death, loss to follow-up, or June 30, 2003. We categorized men into tertiles based on distributions of estimated VO2max according to several age groups (19–29, 30–39, 40–49, and 50–59 yr of age) as we were concerned about residual confounding by age, even with statistical adjustment. We also categorized men into tertiles (18.5-21.5, 21.6-23.6, 23.7-37.4 kg · m^-2) based on the distribution of BMI. We used Cox proportional hazards models to evaluate the relationship between cancer mortality and CRF as well as BMI, and estimated p for trend across categories of CRF and BMI. Hazard ratios for cancer mortality and 95% confidence intervals were obtained by using the lowest tertile of CRF and second tertile (21.6-23.6 kg · m^-2) of BMI as reference. The second tertile of BMI was used as the referent group because data from a large study of East Asians showed lowest mortality in this BMI range [23]. We calculated age- and multivariable-adjusted hazard ratios that adjusted for age (continuous variable), systolic blood pressure (continuous variable), cigarette smoking (never-smokers, past smokers, current smokers of 1–19 cigarettes per day, 20 or more cigarettes per day), alcohol intake (none, 1–45 g per day, 46 g or more per day), and BMI (continuous variable) or CRF (continuous variable). These covariates were selected because they are potential confounders of the associations of CRF or BMI with cancer mortality. Because we were concerned about possible residual confounding by smoking, even with statistical adjustment, we examined associations separately for never smokers and smokers (past smokers, current smokers of 1–19 cigarettes per day, 20 or more cigarettes per day). Further, a statistically significant interaction between smoking status and BMI or CRF was also observed. To assess interaction, multiplicative interaction terms (smoking status × BMI or smoking status × CRF) were included in the multivariable models and statistical interaction was evaluated using the Wald test comparing models with or without these interaction terms. We also studied the relationship between joint categories of CRF and BMI with cancer mortality. Hazard ratios and 95% confidence intervals for cancer mortality were obtained by using the joint category of the second tertile of BMI and the lowest tertile of CRF as reference. In the multivariable model, we adjusted for age, systolic blood pressure, cigarette smoking, and alcohol intake. The multiplicative interaction term (BMI × CRF) was included in a model to examine statistical interaction. To test proportional hazards assumptions, we examined log-minus-log plots and no evidence was found of violation of proportionality assumptions. All analyses were performed using SPSS, version 15.0 J for Windows (SPSS, Inc., Chicago, IL). All P values provided are from two-sided tests.

Body mass index and cancer mortality

Previous large cohort studies have shown that high BMI is a risk factor for all cancer mortality, as well as site-specific cancer mortality [4–8]. Calle et al. investigated the relationship between BMI and cancer mortality among approximately 900,000 American people [5]. The results of their study showed that high BMI is a risk factor for overall cancer mortality, but they did not have measures of CRF for adjustment. Recently, Zheng et al. investigated the relation between BMI and mortality in a pooled analysis of more then 1.1 million Asians [23]. They reported a J-shaped association between BMI and the risk of cancer death in East Asians. Our results showed no clear relationship between BMI and cancer mortality. These results may reflect the narrow range of BMI among participants in our study.

Fitness and cancer mortality

Previous analyses from the Aerobics Center Longitudinal Study (ACLS) have examined the relationship between CRF and cancer mortality in a North American population. These studies showed inverse relationships between CRF and cancer incidence or mortality [9, 12, 13, 19]; smoking related cancer [14]; and digestive [17], breast [16], prostate [10, 15], and lung cancer [18]. In addition, Laukkanen et al. also reported an inverse relationship between CRF and cancer mortality among Finnish men [24]. We previously reported, using the same cohort, that Japanese male workers with high CRF had lower cancer mortality rates [20], consistent with the American and Finnish data. The present study extends these findings by examining cancer mortality rates among men jointly classified by BMI and CRF.

Fitness, body mass index, and cancer mortality

Two papers from the ACLS have published the combined effects of CRF and BMI on cancer mortality. Farrell et al. investigated the joint effects of CRF and several adiposity factors, such as BMI, percent body fat, and waist circumference, on cancer mortality among men [12]. Peel et al. also investigated the joint effects on mortality from digestive cancer [17]. Data from another study, the Lipid Research Clinics Study (LRCS), were used to investigated the joint effects of CRF and BMI on total cancer ( apart from skin cancer) mortality [11]. These studies showed that men with higher CRF have lower cancer mortality rates compared to men having lower CRF, regardless of BMI levels. Participants of the ACLS and LRCS were from a North American population and their average BMI was approximately 26 kg · m^-2. It is unclear whether these findings are applicable to other populations where the average BMI may be lower, such as in Asian populations. Therefore, this study extends the results of previous investigations to a Japanese population, where the average BMI is low.

Possible mechanisms

The biological mechanisms for the joint effect of CRF and BMI on cancer mortality are not precisely known. However, several mechanisms may explain why men who have high CRF are protected against cancer mortality, even in a lean population.

First, excess secretion of insulin and insulin-like growth factor, a cancer cell promoter [25], may be inhibited as a result of improved insulin sensitivity resulting from physical activity [26]. An increase in GLUT4 has been attributed to physical activity, with the possible effect of enhancing insulin sensitivity [27]. It also has been reported that muscle contraction itself may be a stimulus for glucose intake into skeletal muscles [28], which then suggests the possibility that physical activity may correct hyperglycemia, thereby preventing cancer cell proliferation. Second, cancer cell promotion may be inhibited by immune functions enhanced by physical activity [26]. Moreover, physical activity increases skeletal muscle mass [29]. Muscle mass conserves large amounts of protein, which plays an important role in the functioning of the immune system [30]. A major cause of cancer is increased and/or chronic inflammation that stimulates cell proliferation. Physical activity reduces adipocyte infiltration into tissues that generates adipokines such as tumor necrosis factor and interleukin-6 that increase inflammation [31]. Thus, systemic effects of these cytokines and hormones may contribute to reduced risk of cancer in physically fit people. Third, aerobic exercise may inhibit DNA oxidation and prevent cancer cell initiation [32, 33], despite the increase in free radicals that accompanies an increase in energy consumption caused by physical activity.

Strengths

The strength of the present study is the use of a unique population with a low average BMI. Previous reports on this issue have investigated North American populations who have relatively high BMI. We measured CRF, a more objective measure compared with self-reported physical activity [34]. Although, our measurements were indirect, Teräslinna et al. investigated the relationship between measured VO2max and estimated VO2max using the Åstrand-Ryhming Nomogram and correction factors utilized in the present study, obtaining a correlation coefficient of 0.92 [35].

Limitations

There also are several limitations in this study. First, because of small number of cancer deaths, we could not analyze site-specific cancers. Cancer is a collection of heterogeneous diseases with different aetiological pathways. Thus, site-specific cancer analyses are needed in future. In addition, low statistical power may lead to non-statistically significant findings. Future studies with large sample size are needed to confirm the present findings. Second, cause of death was reported in an interview with the next of kin, leading to a possible misclassification of the cause of death. However, it is assumed that in most cases, attending physicians directly informed the next of kin of the cause of death. Therefore, this is unlikely to be a major limitation. Third, certain dietary factors, such as an anti-oxidants, dietary fiber, omega-3 fatty acids, etc. may reduce cancer incidence [36]. We did not have information on diet and so residual confounding by diet may be present. Finally, we included Japanese men: it is unclear how these findings may apply to other Asian populations who have relatively low BMI levels in general. Furthermore, participants were employees of a single, Japanese urban company and may not be representative of other Japanese. This limits the generalizability of this study but not its validity.