Quantification of Self Selection Bias in Mass-screening for Gastric Cancer

Self selection bias in mass-screening for gastric cancer was quantified using data from a group consisting of 66,386 workers. Total observed person-year was 182,779. Among males self selection bias was thought to exert some influence that made the incidence rate about 1. 13(all ages, 95%CI=0.65-1.97) or 1.23(over 40, 95%CI= 0.69-2.20)-fold higher and the mortality, rate about 1.22(all ages, 95%CI=0.48-3.13) or 1.42(over 40, 95%CI=0.54-3.69)-fold higher in the screening-attending group. Among females self selection bias was thought to exert some influence that made the incidence rate about 0.30(all ages and over 40, 95%CI=0.042.47 and 0.04-2.48)-fold lower and the mortality rate about 1.09(all ages, 95%CI=0.18-6.68) or 1.11(over 40, 95%CI=0.18-6.76)-fold higher in the screening-attending group. J Epidemiol, 1991 ; 1 : 7-10.


SUBJECTS AND METHODS
The subjects of the study were 66,386 workers belonging to three health insurance societies located in the Tokyo metropolitan area.Sex and age distribution of the subjects are shown in Table 1.These three health insurance societies consist of workers belonging to about a hundred small or middle scale companies involved in steel, textile or paper products.
The gastric cancer screening was intended to be carried out on those who were over 40, but in fact some between ages 35 to 39 also attended.
The incidence of gastric cancer was observed for all subjects.The observation of the incidence of gastric cancer was carried out for the years 2 to 4, as shown in Table 2, using the data from "receipts" held by the health insurance societies.All the incidence cases were followed up for 5 years, using the "receipts" or resident cards ("Jumin-Hyo") held by municipalities.The cause of death during the 5-year-follow-up was presumed to be gastric cancer.In this study, the term "mortality" rate of gastric cancer is defined as the incidence rate of gastric cancer that causes death within five years after diagnosis (represented as "mortality" to distinguish from the usual meaning of mortality) .For all subjects whether or not they attended the gastric cancer screening during the year when the incidence was observed and the year prior to then, was determined from the records kept by the health insurance societies which administer the screening program.
The analysis was done by the person-year method.The personyears shown in Table 3, were divided into two groups with the criterion being whether or not the subject attended the gastric cancer screening in the year before the incidence was observed (screening-attending group and screening-absent group, respectively).Each group was divided into twelve strata by sex (male, female), age (-39, 40-59, 60-) and attendance (yes, no) at the screening in the year of the incidence observation.
Incidence cases of gastric cancer and incidence cases of fatal gastric cancer were divided in the same way.The incidence and "mortality" rate per year of gastric cancer for each group, risk ratio and its 95% confidence interval were calculated by Miettinen's method3) from this stratified data.Separate calculations were done with respect to sex (male and female) and age (on all subjects and on those over 40).
To confirm the validity of this study, expected number of incidence and death were calculated from sex and age distributions shown in Table 3, using data from the Malignant Neoplasm Surveillance of 19794)

RESULTS
The expected numbers of incidence and death are shown in Table 4, with observed numbers.Observed incidence was 99 and observed death was 43, which were less than the expected value of 125 and 61 respectively.
Two cases of the observed 99 incidence cases were missed during a 5-year-follow-up, and "mortality" was calculated without the inclusion of these cases.In Table 5, observed person-year, incidence and death for gastric cancer for the attending group and the absent group are shown.
Among males the incidence rates of gastric cancer in the screening-attending groups were higher than those in the screening-absent group, and the risk ratios were 1.13(95%confidence interval= 0.65-1.97)for all ages and 1.23(95%CI=0.65-1.97)for over 40.But among females the incidence rates of gastric cancer in the screening-attending groups were lower than those in the screening-absent group, and the risk ratios were 0. 30(95%CI= 0.04-2.47 and 0.04-2.48)both for all ages and for over 40.Among both males and females the "mortality" rates in the screeningattending group were higher than those in the screening-absent group, and the risk ratios were 1.22(95%CI=0.48-3.13) for males all ages , 1.42(95% CI=0.65-1.97)for males over 40, 1.09(95%CI=0.18-6.68)for females all ages and 1.11(95%CI=0.18-6.76)for females over 40.These results are shown in Table 6.

DISCUSSION
When subjects are workers belonging to a few large companies, the healthy workers effect might exert a strong influence6),7).In this study however, subjects were workers from many small or middle scale companies, so the bias due to this factor is thought to be minimal.Another possible problem is entry and withdrawal of members.In this study, entry and withdrawal of members was less than 5% of the total members per year, and therefore was thought to exert little influence.Steelworkers have been reported to have a high risk for gastric cancer8) while similar relationships have not been reported so far for textile or paperproducts workers.In this study, no difference in incidence rate or "mortality" rate was observed between the subjects belonging to the companies dealing with steel products and the other subjects.
While there is the probability that the subjects may attend other gastric cancer screening programs such as those administered by local government, since the health insurance societies pay most of the cost for the screening and the personal defrayal is very little, and furthermore since the gastric cancer screening program is scheduled conveniently for the subjects and the companies, this probability is very small.Observed numbers of incidence and of death were both smaller than expected.
One possible reason for this is the above-mentioned healthy workers effects6),7).Another reason is that the incidence and mortality rate is declining in Japan9) and the date of this study is later than 1980 or 1983.Taking these factors into account, the results of this study are thought to be valid.
Among males, the incidence rate of gastric cancer in the screening-attend group was higher than that in the screening-absent group and the incidence rate of fatal gastric cancer in the screening-attending group was also higher than that in the screening-absent group.
Among females, the incidence rate of gastric cancer in the screening-attend group was lower than that in the screening-absent group but the incidence rate of fatal gastric cancer in the screening-attending group was higher than that in the screening-absent group.The 95% confidence interval for risk ratio included 1.0 in all calculations.
Several reports10)-12) have evaluated self selection bias in mass-screening programs for gastric cancer using the data of cancer registration, and have concluded that the incidence rate of gastric cancer in the screening-attending group was higher than that in the screening-absent group (risk ratio =0.80-1.57)and the mortality rate in the former was lower than that in the latter (risk ratio =0.53-0.72): most of which were not statistically significant.Incidence rate for fatal gastric cancer is equal to mortality rate, if death of gastric cancer occurs randomly.Therefore the result for "mortalit y" in this study is comparable to the results for mortality in these other reports.
The results in this study were similar to the other reports with respect to incidence among males.But with respect to mortality among both males and females, the results in this study were different from the other reports.In Japan, persons who have symptoms in the stomach, especially males, have been observed to tend to attend gastric cancer screening instead of consulting a physician.This tendency may be the reason for higher incidence rate and "mortality" in the screening-attend group among males.
In the quantification of self selection bias, the influence of the actual effectiveness of gastric cancer screening program is a problem.
In this study this influence was removed by stratification based on attendance at the screening in the year of the incidence observation.But in the other reports the influence may not have been removed, because those who attended gastric cancer screening were thought to have attended gastric cancer screening more frequently after that than those who didn't.
This may be a reason for the difference between the results in this study and the other reports.
For incidence rate the results were inverse between male and female.Many factors are supposed to influence behavior with regard to attendance at screenings.In Japan, the attitude of management in a company toward the health of workers and the working condition in the company are also important factors13).Adding to these socioeconomic factors, an individual's or his/her family's perception regarding health, his/her present health condition and past experience at examination may influence behavior.The occurrence of gastric cancer in one's circle may also have influence.Thus self selection bias may change depending on age, sex, occupation and years.
It was concluded that among males self selection bias exerted some influence that made the incidence rate about 1.13 (all ages, 95%CI=0.65-1.97)or 1.23 (over 40, 95%CI=0.69-2.20)-foldhigher and the mortality rate about 1.22 (all ages, 95%CI=0.48-3.13)or 1.42 (over 40, 95%CI=0.54-3.69)-foldhigher in the screening-attending group.Although the reliability of the results regarding females in this study may be low due to the cases of female gastric cancer being only 14, it was also concluded that among females self selection bias exerted some influence that made the incidence rate about 0.30 (all ages and over 40, 95%

Table 1 .
Sex and age distribution of subjects .

Table 5 .
Observed person-year , incidence and death of gastric cancer regarding attending group and absent group.

Table 6 .
Age adjusted incidence and mortality rate of gastric cancer/ 100 ,000 person-year,