Prospective study of screening for stomach cancer in Japan
Although screening for stomach cancer is a widespread community service in Japan, the benefits of the screening program remain unclear. Our study investigated prospectively the relation between participation in stomach-cancer screening during the past 12 months and subsequent deaths. Data was derived from the Japan Collaborative Cohort Study, in which 480 stomach-cancer deaths were identified during an 8-year follow-up period. Cox proportional hazard regression was used to estimate the relative risk of death from stomach cancer and that from other causes while adjusting for potential confounding factors. In men, screening participation was associated significantly with a reduced risk of death from stomach cancer (relative risk [RR] = 0.54; 95% confidence interval [CI] = 0.41–0.70). The extent of the risk reduction was greater than potential health selection (for deaths other than stomach, RR = 0.71). The adjustment for potential confounding variables, however, attenuated the difference in risk of death (stomach cancer, RR = 0.65; other causes, RR = 0.71). In women, the magnitude of the association between screening participation and death from stomach cancer (RR = 0.74; 95% CI = 0.52–1.07) was equal to that for deaths from non-stomach cancers (RR = 0.74). Subgroup analysis showed that women with a parental history of stomach cancer had a reduced risk of death from stomach cancer associated with screening (RR = 0.32; 95% CI = 0.12–0.87). The present results underline the potential for selection bias in observational studies, and thus it remains an open question whether screening for stomach cancer is effective. © 2003 Wiley-Liss, Inc.
Stomach-cancer screening using photofluorography is widespread in Japan, a nation afflicted by high stomach cancer mortality.1 The national health policy for the prevention of chronic diseases, including stomach cancer, has promoted such screenings by providing financial support for cancer screenings. In 1999, 4.2 million residents (13% of the target population) participated in stomach-cancer screenings that were organized by the local governments.2
Ideally, the effectiveness of a screening program is evaluated by an intervention study in which subjects are randomly allocated to the screened and unscreened groups. Due to the wide implementation of stomach-cancer screening in Japan, however, no such intervention study has been carried out. Instead, the effectiveness of the program has been assessed using other epidemiological methods, including geographical comparisons of mortality trends,3 case-control studies4, 5, 6 and cohort studies.7, 8, 9 Case-control studies have consistently shown a 40–60% reduction in risk of death from stomach cancer among screened subjects.4, 5, 6 In contrast, prospective studies give inconsistent results. One study found a sizable reduction in stomach cancer mortality among screened subjects,7 whereas 2 other studies found that screening had less or no effect.8, 9 Observational studies of the effects of screening suffer from selection bias. The individuals taking part in screening programs are generally healthier than those who do not. With the exception of one preliminary report,9 however, previous studies have not adjusted for lifestyle factors that may be associated with stomach cancer. Furthermore, findings from case-control studies, which may yield biased results through other mechanisms,10 need to be confirmed by other methods.
The goal of our present study was to investigate prospectively the association between participation in stomach-cancer screening and death from stomach cancer, while controlling for potential confounding factors, using data from the Japan Collaborative Cohort (JACC) Study, a nationwide population-based study.
MATERIAL AND METHODS
Screening for stomach cancer in Japan
In 1960, mass screening for stomach cancer using mobile X-ray system was launched in Miyagi Prefecture in Japan. In 1983, a national law was enacted that stipulated that each municipality provides annual screenings for inhabitants aged 40 or older. Several diagnostic methods were proposed, but the double contrast X-ray examination remained the standard procedure for mass screening. In 1984, the Japanese Society of Gastroenterological Mass Survey standardized the diagnostic method.11 The method requires that examinees swallow effervescent granules before examination, drink 200–300 ml barium, and have 7 X-ray films of the stomach taken in different positions. This method has been widely used in mass screenings. The films taken are then examined independently by 2 specialists. Screenees showing abnormal findings are instructed to undergo detailed examinations, such as full-size film X-ray examinations, gastrofiberscope examinations and biopsy. Cancer cases receive the suitable treatment, including surgery, if appropriate. National statistics2 indicate that in 1983, 2.2 million inhabitants (7% of those 40 years of age or over) participated in community-based screenings for stomach cancer. The number of participants has grown steadily, reaching 4.2 million (13% of those 40 years of age or over) in 1992, a figure that has remained stable over the last decade. Of all examinees in 1999, 12% were required to undergo intensive examination and 0.14% were diagnosed with stomach cancer.
Data was derived from the JACC Study. The primary goal of the study was to identify etiologic factors of cancer in the Japanese population. Details of the study procedure have been described elsewhere.12 The subject group consisted of over 125,000 individuals from a total of 45 regions who completed a self-administered questionnaire from 1988–90. The questionnaire included queries on their medical history and health-related lifestyles, such as smoking, drinking, diet, as well as participation in screening programs. There were some variations in queries in the questionnaire across study regions; 3 regions omitted the query on screening participation. In most study regions, participants were recruited at the time of their general health check-ups, which are periodically provided for inhabitants of each municipality. Informed consent was obtained by having subjects sign the cover page of the questionnaire. In a few study regions, different methods were used to recruit subjects and to obtain informed consent. As an example, all residents of specified districts were invited to the baseline survey. Informed consent was obtained at the group level after explaining the study objective and the confidentiality of the data to the community leaders. With the cooperation of the respective municipalities, study participants were followed up annually to determine whether they had died or had relocated. For the deceased, cause of death was ascertained with the permission of the Management and Coordination Agency of the Japanese Government, then coded according to the International Classification of Disease Revision 9 (after 1995, Revision 10). Censoring was conducted on either date of death, relocation, or the end of the follow-up period (December 1997), whichever came first. The research protocol of the present study was approved by the Ethics Committee for Medical Care and Research, University of Occupational and Environmental Health, Japan.
We defined members of the screened group as individuals who had participated in stomach-cancer screening or dock screening, generally including stomach examination, during the 12-month period preceding the baseline survey, according to response to the questionnaire. Participation in the screening program was not monitored during the follow-up period. With regard to exposure status, it was assumed that individuals who had participated in a screening during the 12-month period preceding the baseline had also participated in screening during the follow-up, whereas those who had not participated had not. In addition to stomach cancer deaths (ICD-9, 151; ICD-10, C16) as the main outcome, all cancer-related deaths (ICD-9, 140-208; ICD-10, C00-C97) except stomach cancer deaths and all causes of death except stomach cancer were used as outcome variables to evaluate potential selection bias. Cox proportional hazard regression was used to estimate relative risk (RR) and the corresponding 95% confidence interval (CI). RR was estimated for the following models: 1) age- (10-year interval) and study region-adjusted model; 2) age-, study region-, and smoking (never smoked, former smokers, current smokers)-adjusted model; 3) model with additional adjustment for level of education (left school before 16 years of age, attended school up to 16–18 years of age, attended school over 18 years of age), Western-style breakfast, rice (at least 4 bowls per day), miso soup (at least 3 bowls daily), salty foods (at least 3 times per week of dried/salted fish or tsukudani [food boiled in soy sauce], and pickled vegetables daily), yellow-green vegetables (at least 3 times per week of green-leaf-vegetables or carrot/pumpkin), and parental history of stomach cancer. These variables were either risk factors for stomach cancer in previous studies, including those involving Japanese subjects,13, 14, 15, 16, 17, 18, 19 or indicators of adherence to the traditional Japanese diet (style of breakfast, rice, miso soup). In women, adjustments were also made for screenings for uterine or breast cancer for deaths from causes other than stomach cancer. Observing an increased risk of stomach cancer associated with familial history of stomach cancer among women in our study group (RR = 2.5);20 we analyzed the subgroup of women who reported a parental history of stomach cancer. We repeated the above analyses while excluding deaths in the first 3 years of follow-up. Calculations were carried out using the PHREG procedure of the Statistical Analysis System.21
The present study involved 100,562 subjects who were aged 40–79 years old at the time of the baseline survey, were asked for their screening experience. Of these individuals, 219 subjects with a history of stomach cancer, including those diagnosed during 12-month period preceding the baseline survey, were excluded. In assessing general health selection associated with screening, we believed it necessary to exclude subjects with serious conditions capable of affecting both screening participation and prognosis. Thus, 4,645 subjects with a history of other forms of cancer, myocardial infarction, or cerebrovascular diseases were also excluded. Of these individuals, 8,386 subjects who did not provide information on participation in stomach-cancer screening were also excluded. This left 87,312 subjects for analysis.
Among the subjects for the present study, 36% of the men and 35% of the women had participated in stomach-cancer screenings during the 12 months preceding the baseline survey. Subjects aged 50–69 years old reported a higher rate of participation in stomach-cancer screenings than either male or female subjects from younger or older age groups. Table I shows potential confounding variables according to stomach-cancer screening status. Compared to the unscreened group, screened group was more likely to report parental history of stomach cancer, to consume yellow-green vegetables, rice, and miso soup, and was less likely to be a smoker. The female subjects also showed similar differences. But screened women were less likely to eat Western-style breakfasts than unscreened women.
Table I. Potential Confounding Factors to Stomach-Cancer Screening in 87,312 Subjects in the JACC Study, Japan 1988–19971
|Mean age (years)||57.1||57.4||57.5||57.5|
|Education (> 18 years)||17.3||20.0||10.3||10.1|
|Parental history of stomach cancer||7.8||11.3||8.4||11.5|
|Rice (> 3 bowls/day)||42.9||45.8||19.3||21.4|
|Miso soup (>2 bowls/day)||24.5||29.3||14.3||19.1|
|Uterus-cancer screened1|| || ||36.7||68.0|
|Breast-cancer screened1|| || ||22.6||44.6|
During a mean follow-up period of 8 years, a total of 480 stomach-cancer deaths (men, 322; women, 158) were identified among the study subjects. In age groups 50 years or older, men exhibited a mortality rate from stomach cancer approximately 3 times that of women.
As shown in Table II, in age- and study region-adjusted model, the screened men showed a 46% reduction in risk of death from stomach cancer (RR = 0.54; 95% CI = 0.41–0.70), compared to the unscreened men. Screening participation was also associated with a 20% reduced risk for cancers other than stomach and a 29% reduced risk of deaths from causes other than stomach cancer. Additional adjustment for potential confounding factors reduced the discrepancy in death risk associated with screening participation between stomach cancer and other causes. In women, the screened group had a 26% reduced risk of dying from stomach cancer (RR = 0.74; 95% CI = 0.52–1.07), while showing risk reductions of similar magnitude for non-stomach-cancer deaths (RR = 0.70) and for deaths from causes other than stomach cancer (RR = 0.74) (Table III).
Table II. COX Proportional Hazard Regression Results of the Association Between Stomach-Cancer Screening and Subsequent Death from Stomach Cancer and Other Causes in 36,155 Men in the JACC Study, Japan, 1988–19971
|Stomach cancer|| || || || || |
| Screened||78||76||0.54 (0.41–0.70)||0.55 (0.40–0.76)||0.65 (0.45–0.95)|
|Cancers, excluding stomach cancer|| || || || || |
| Screened||400||388||0.80 (0.70–0.90)||0.79 (0.68–0.91)||0.73 (0.61–0.89)|
|All causes of death, excluding stomach cancer|| || || || || |
| Screened||1026||996||0.71 (0.64–0.77)||0.73 (0.66–0.82)||0.71 (0.63–0.80)|
Table III. COX Proportional Hazard Regression Results of the Association Between Stomach-Cancer Screening and Subsequent Death from Stomach Cancer and Other Causes in 51,157 Women in the JACC Study, Japan, 1988–19971
|Stomach cancer|| || || || || |
| Screened||45||32||0.74 (0.52–1.07)||0.78 (0.49–1.25)||0.75 (0.42–1.34)|
|Cancers, excluding stomach cancer|| || || || || |
| Screened||203||143||0.70 (0.59–0.83)||0.67 (0.54–0.84)||0.84 (0.62–1.12)|
|All causes of death, excluding stomach cancer|| || || || || |
| Screened||627||442||0.74 (0.66–0.83)||0.74 (0.64–0.85)||0.77 (0.64–0.91)|
Table IV shows the results for 3,660 women with a parental history of stomach cancer. The screened group had a 68% reduced risk of death from stomach cancer (RR = 0.32; 95% CI = 0.12–0.87), while showing no reductions in risk of death from causes other than stomach cancer.
Table IV. Relative Risk and 95% Confidence Interval for Stomach-Cancer Screening in 3,660 Women Who Have a Parental History of Stomach Cancer in the JACC Study, Japan, 1988–19971
|0.32 (0.12–0.87)2||1.00 (0.52–1.90)2||0.84 (0.56–1.25)2|
|0.40 (0.14–1.15)3||1.18 (0.55–2.53)3||0.93 (0.58–1.48)3|
|0.34 (0.11–1.05)4||1.88 (0.77–4.62)4||0.97 (0.55–1.72)4|
Analyses excluding deaths during the first 3 years of the follow-up period gave similar results; age- and study region-adjusted RRs for men, women overall, women with a parental history of stomach cancer were 0.57 (95% CI = 0.42–0.77), 0.79 (95% CI = 0.52–1.21), and 0.37 (95% CI = 0.13–1.03), respectively.
The present study investigated prospectively the effects of stomach-cancer screening, using data from a nationwide cohort in Japan. In men, the risk of death from stomach cancer among screened subjects was nearly half that of unscreened subjects. The extent of risk reduction was somewhat greater than for deaths from other causes. In women, reduced risk of death associated with screening participation was not specific for stomach cancer.
Selection bias has been discussed as a major limitation of case-control studies,4, 5, 6 in which information on confounding factors were not obtained. Among the subjects, the screened group was more likely to report risk factors for stomach cancer (e.g., parental history of stomach cancer and a traditional Japanese diet), while also tending to have healthier lifestyles (e.g., nonsmokers, and frequent consumption of yellow-green vegetables). Similar differences in lifestyle according to screening history have been observed in a previous study.22 In men, we found that the adjustment for these variables shifted estimates toward a null value, indicating that the reduced risk of death for stomach cancer among screened subjects was partly attributable to differences in known risk factors.
We also assessed the magnitude of potential selection bias by comparing the risk of death for stomach cancer with that of deaths from causes other than stomach cancer. This approach may be useful in examining the effects of a bias that cannot be controlled using available information on known confounding factors. Among both men and women, screened individuals showed a 20–30% reduced risk of death from causes other than stomach cancer. Among men, the risk of death from stomach cancer was lower than that from other causes by 10–25 percentage points, leaving open the possibility that screenings may have provided some benefit. In women, the scale of risk reduction for death from stomach cancer was no greater than for other causes of death. The reduced risk of death from stomach cancer among the screened women could be attributable entirely to the effects of health selection. This kind of bias may be reduced if subjects are selected from among individuals who had participated in screening at least once during the last several years. As indicated by a finding made in a cohort study (a lower all-cause mortality among individuals who had participated in screening regularly than those who had participated irregularly23), it is likely that the effects of selection bias differs in scale even among screened subjects showing various degrees of adherence to screening.
Selection through other mechanisms may also result in the present association. Past participation in screenings, which may prevent diagnosed cases from future participation, would reduce the probability of having stomach cancer at the baseline, leading to a spuriously reduced risk of death from stomach cancer among the screened group defined in our study. Two case-control studies have shown that the effect of stomach-cancer screening persists at least over a 3-year period.5, 6 According to this data, we repeated the analysis by excluding deaths during the first 3 years of follow-up to eliminate the potential effects of past screenings. The minor change in estimates after the exclusions indicates that such bias is negligible in the present study; although it remains possible that the effects of screenings before the baseline survey have persisted for a longer period.
The validity of the exposure assessment should be addressed. The present study determined participation in the screening program based on information from self-administered questionnaires, assuming that individuals who had taken part in stomach-cancer screenings during the 12-month period preceding the baseline also participated in screening during the follow-up, whereas those who had not participated did not. We evaluated this assumption using data from the follow-up survey conducted approximately 5 years after the baseline. Among the present subjects, 28,757 (33%) provided information on their screening experiences for the 12-month period preceding the follow-up survey. Because participants in most study regions were recruited at the time of health checkups, however, those responding to the follow-up survey were more likely to have participated in screening programs. They do not constitute a representative sample of subjects in the baseline survey. We analyzed data for Region A (n = 3,771), where participants were recruited on a population-basis for both surveys, and for Region B (n = 2,555), where 87% of the those responding to the baseline survey again participated in the second survey. In Region A, 68% of the individuals who had attended screenings during the year preceding the baseline survey participated again 5 years later, whereas 67% of the individuals who had not been screened for the year preceding the baseline survey did not attend screening 5 years later. In Region B, the corresponding figures were 71% and 67%. The similarity of figures obtained in different regions suggests that recent participation in screening is a reasonable predictor of future participation. A slightly stronger agreement regarding screening status was reported in a previous study, in which 2 surveys were conducted at a 1-year interval.9 Non-differential misclassification in screening status should lead to a conservative estimate of the preventive effect of screening.
Case-control studies have consistently demonstrated the significant effects of stomach-cancer screening, with a 40–60% reduction in risk of death from stomach cancer.4, 5, 6 In contrast, cohort studies give inconsistent results. A prospective study of an 18-year follow-up period from 1960–77 revealed a 50% reduction in risk of death from stomach cancer among screened subjects,7 whereas 2 other studies showed less significant effects, similar to that for deaths due to causes other than stomach cancer.8, 9 In our present study, analysis without adjusting for confounders showed a 45% reduced risk of death from stomach cancer in men, a finding comparable to results from the positive studies. If all potential health selection is taken into consideration, however, the benefits of screening are considerably reduced: a mere 8% reduction in death risk, calculated by dividing 0.6 (0.71 − 0.65) by 0.71 (Table II, using fully-adjusted estimates).
Two case-control studies with positive findings have reported gender differences in the magnitude of the effect of stomach-cancer screening, but these results do not agree. One study showed a larger preventive effect among men,5 whereas the other study found the contrary result.4 In our present study, male subjects showed a greater reduction in risk of death from stomach cancer associated with screening than female subjects. Random variation may be one explanation for such inconsistent findings, but it is possible that the effects of selection bias differ in scale by gender, as well as from study to study.
In an analysis of women with parental history of stomach cancer, a subgroup among whom stomach cancer mortality is as high as that for men, the risk of death from stomach cancer in the screened group was markedly reduced (RR = 0.32), whereas the risk of death from other cancers (RR = 1.00) and risk of death from other causes (RR = 0.84) were not reduced. This suggests that the screening program is effective specifically for high-risk groups. A confirming study is required, however, for the finding based on subgroup analysis with small numbers of cases (n = 21).
In summary, when no consideration was taken for potential health selection, we observed a large reduction in risk of death from stomach cancer associated with screening participation in men. The reduction may be accounted for in part by known confounding factors. Most of the remaining effects may be explained by general health selection, leaving a mere 8% margin of benefit associated with screenings. In women, reductions in risks associated with screening participation may be explained entirely by general health selection. The present results underline the potential for selection bias in observational studies, and thus it remains an open question whether screening for stomach cancer is effective.
The JACC study has been supported by Grants-in-Aid for Scientific Research from the Ministry of Education, Science, Sports and Culture of Japan (Monbusho) (No. 61010076 , 63010074 [1987, 1988], 1010068 , 2151065 , 3151064 , 4151063 , 5151069 , 6279102 [1994–1999] and 11181101 ).