Serum Cholesterol and Cancer Mortality in Japanese Civil Service Workers : Findings from a Nested Case-control Study

To examine an association between serum cholesterol level and cancer mortality, a nested case-control study was conducted among 5, 796 civil service workers in Japan who underwent periodic health examination. One-hundred and thirty-one deaths (cases), including 73 cancer deaths, were identified in the study period from 1980/1981 to September 1, 1991. Two controls were randomly selected for each case, matched for age, sex, year of examination and job status. As a major result, an increase in serum cholesterol of 10 mg/dl significantly reduced cancer risk by 0.91 times in men, but not in women. This reduction of cancer risk in men was found not to be confounded by body mass index (BMI), smoking habits, and drinking habits. No significant association could be found between serum cholesterol level and specific sites of cancer. Separate analysis by follow-up period significantly revealed an inverse association between serum cholesterol and cancer deaths in men in 6 years or later after serum measurement. This inverse association was believed not to be ascribable to an effect of a preclinical cancer. J Epidemiol, 1993; 3 : 99-107.

. In the present study, therefore, we attempted to clarify an association between serum cholesterol level and cancer mortality among civil service workers in Japan, and to add a further evidence of a long-term inverse association between them.

MATERIALS AND METHODS
We conducted a nested case-control study among civil service workers who underwent periodic health examination in 1980 and 1981. Study population, identification of cases and control selection are as follows.

Study population
In 1980 and 1981, baseline and periodic health examinations were carried on a total of 5,796 (4,153 men and 1,643 women) civil service workers aged 25,30,35,40,45,50, and 55 years old each year (response rate : 92%). The examination included self-administered questionnaire, anthropometric measurement and physical examination, electrocardiogram (ECG), urinalysis, and biochemical blood test. Blood specimens were drawn under fasting condition. Information collected by self-administered questionnaire included past history of major diseases, current health status, family history, smoking and drinking habits, and preference for foodstuffs and taste. Smoking (drinking) habits were inquired to classify daily-smoker (drinker), occasional smoker (drinker), and exor non-smoker (drinker).
For daily drinkers, alcohol consumption per day was estimated by assuming alcoholic volume of 27 g in 180ml of "sake", 633 ml of beer, or 1 double glass of whisky. Serum cholesterol was measured enzymatically48).

Identification of cases
Study population was followed up to September 1, 1991 after examination.
All dead subjects were identified among those who are continuously on active service and also among those retired, but subsequently supported by retirement pension. Causes of death, classified according to International Classification of Diseases (ICD-9), were routinely checked by reviewing documents sent to the pension office to notify the death of pensioners.
Those who retired before September 1, 1991 and were not subsequently supported by pension, accounting for 11.8% (683 in number) of a total study population, were excluded from the present analysis, since their vital status could not be traced.
One-hundred and thirty-one deaths (94 deaths during active service and 37 deaths after retirement) were identified between the examination and September 1, 1991. Table I summarizes causes of death by sex and subsequent follow-up year. Seventy three deaths were due to cancers (60 men and 13 women), being accounted for 55.7% of 131 total deaths. Control selection Controls were selected from the same population matching to cases for age (within 1 year), sex, year of examination, and job status (blue or white color worker). First, we constructed potential sources of controls for each case, and then two controls were randomly selected from each source by identifying those whose year of employment was the nearest and who are confirmed as alive on September 1, 1991. There was one case who succumbed to liver cancer, and for whom we could not set up two eligible controls. This pair was also included in the present analysis.

Statistical methods
Analyses were performed separately by sex to explore the relationship between serum cholesterol level and total mortality, cancer mortality from all sites, and mortality from major primary sites of cancer (stomach, lung, digestive organ cancers, smokingrelated cancers), when death occurred in 2 years or later after examination, i.e. excluding those who died in less than 2 years after examination, probably due to diseases already suffered. We included cancers of the esophagus, stomach, colon and rectum in cancer of the digestive organ. For men, smoking-related cancer was defined as cancers of the lung, mouth, larynx, esophagus, pancreas, bladder, and leukemia.
For women, cancer of the cervix was also included19) in smokingrelated cancer. The entire follow-up period was divided into three periods : (1) 2 to 5 years, (2) 6 to 8 years, and (3) 9 to 11 years. All statistical analyses were conducted using the SAS programs at Nagoya University Computation Center. For multivariate analysis, we used a conditional logistic regression model for matched data49). In multivariate analyses, body mass index (BMI : weight(kg)/height(m)2), smoking habits, and drinking habits at baseline (1980 or 1981) were used as covariates.

RESULTS
Some selected characteristics of the study subjects are shown in Table 2. The cases were more likely to be non-drinkers and current smokers, but heavy drinkers defined as consumers of 81g or more alcohol per day, appeared to be more prevalent among the cases. They also more frequently suffered from hypertension or hepatitis ; more often complaining of dizziness, weight loss or thirst at entry.
As shown in Table 3, there was no significant difference in mean serum cholesterol for "all causes" and "all causes except accident and suicide" , though the mean values were consistently lower in the cases. For Crude and two kinds of adjusted relative risks were also summarized in Table 3. An increase in serum cholesterol of 10 mg/dl significantly reduced cancer risk by 0.91 times in men, but no reduction was noted in women. This reduction of cancer risk in men was found not to be confounded by BMI, smoking habits and drinking habits, since two adjusted relative risks were 0.89 (95% confidence interval 0.81-0.98) and 0.87 (0.78-0.97). Table 4 gives the relative risks of death from some sites of cancer in men. In women, numbers of death from specific sites of cancer are too small to calculate the relative risks. Lung cancer death tended to show a positive, and stomach cancer death a negative association with serum cholesterol.
Risk of dying from cancers of the digestive organ and from smokingrelated cancers appeared to be lower, when serum cholesterol is higher. Risk of cancer death with serum cholesterol was separately evaluated by follow-up period in men (Table 5). For the first period (follow-up 2-5 years), no association was obvious. For the second (followup 6-8 years) and the last period (follow-up 9-11 years), mean serum cholesterol were significantly lower in the cases, and an inverse association was statistically significant, when adjusted for BMI, and smoking and drinking habits. In women, however, no significant associations were noted in any follow-up periods between serum cholesterol level and cancer mortality (not shown in the text).

DISCUSSION
At first, we will discuss a few methodological shortcomings in our study and how we overcame them. First, the present study was based on relatively small numbers of cancer death. When case-control study is conducted based on small numbers of cases and controls, significant associations are not likely to be obtained.
When a significant association is found, however, it may not be a distorted relationship between a given disease and a risk factor, unless cases and controls were selected with biases. Second, we could not trace all individuals in the study population up to September 1, 1991 to identify all dead subjects.

A . Tamakoshi et al
The missed individuals were more likely to be women by sex, younger in men and older in women by age, compared to those who could be successfully traced. Thus, non-respondent bias could possibly exist in the present study, but we could not exclude this bias. Since some deaths were likely to be missed, we adopted the study design of a nested case-control study, in which cases were limitedly identified, and controls were definitely confirmed as survivors . This means, accordingly, that missclassification of cases and controls is not likely at all. Third, some studies have revealed an U-shape association of total mortality with serum cholesterol1-9). According to our separate analysis by categorizing serum cholesterol into four levels, we found no U-shape association, but dose-response relationship with total mortality in men. We then handled serum cholesterol as a continuous variable , but not as a categorical variable, in our analysis. As many other studies2, 10-13.17-20) we found an inverse association between serum cholesterol and cancer mortality in men, but not in women. No apparent association in women might possibly be due to small numbers of cancer death in women, and partly to sex differentials in predominant sites of cancer18) .
In several studies, low serum cholesterol was found to be inversely associated with colon cancer 1,2,14,15,18,23,[50][51][52], lung cancer 5, 6,12,14,15,223,25,32,33) or smoking-related cancer 6,18-20). Our study, however, failed to demonstrate any significant associations between serum cholesterol level and specific sites of cancer, possibly because of small numbers of death from individual sites of cancer. A positive association with lung cancer deaths and negative association with deaths from smoking related cancers in our study, could not be readily explainable due to small number of deaths in these sites.
An association of low serum cholesterol level with cancer deaths was found in most reports3, 5,11,[20][21][22][23][25][26][27][28][29][30][31][32]. when cancer deaths which occurred within a few years after cholesterol measurement were analyzed. This short-term association is believed to reflect an effect of preclinical cancer5) which lowered serum cholesterol. In our study all deaths which occurred in less than 2 years after examination were intentionally excluded. But in male patients who died in less than 2 years after cholesterol measurement, mean serum cholesterol was apparently lower by about 20 mg/dl when compared to their controls. This short-term effect was probably ascribable to increased catabolism resulting from enhanced low density lipoprotein (LDL)-receptor activity in malignant cells53). Henriksson and his colleagues-"), who compared prostatic cancer patients with and without metastasis and their controls, found that the patients with metastasis cleared LDL faster than those without metastasis and controls.
According to this finding that indicates faster catabolism of LDL, a mechanism by which cancer reduces serum cholesterol is theoretically plausible.
In several studies 4,10-18), a long-term inverse association of serum cholesterol level with cancer was also obtained likewise our study. This long-term inverse association could not be explained by preclinical cancer, though implications of this association have not been fully understood.
There are three possible explanations for the long-term inverse relationship between low serum cholesterol level and cancer. The first , it is just a chance observation ; the second, a third factor will exist which lowers serum cholesterol and simultaneously increases the probability of cancer development ; and the third, low serum cholesterol itself actually causes cancer10,12,55-57 First, long-term association was found in 104,10-18) reports out of 291-5,10-33) cohort studies in which an inverse association of serum cholesterol level with cancer was reported.
Law and Thompson45), who reviewed 33 published prospective studies by metaanalysis, confirmed a long-term association of low serum cholesterol with cancer. It may be, therefore, unlikely to be observed only by chance.
Second, a possible third factor may include smoking habits or beta-carotene.
Since serum cholesterol is known to be pharmacologically lowered by smok-ing45), serum cholesterol level could possibly be confounded by smoking habits. It is , however, found in most studies 1,4,[11][12][13][14][15]18,19,[22][23][24][25][26]31,32) including ours that an inverse association between serum cholesterol and cancer persisted even after statistical adjustment for smoking habits. Smoking is also known to lower high density lipoprotein (HDL) cholesterol58-61). Most prospective studies have examined an association of cancers with total serum cholesterol except a few studies 18,62,631 in which an inverse association of serum HDL cholesterol with cancer was not consistently demonstrated.
In our study, the relation with HDL cholesterol was examined as well, but no association with cancer mortality was found (unpublished data) . Another possibility is that beta-carotene is a third factor to influence both cancer mortality and serum cholesterol level. Shekelle and his colleagues64) reported that the association between low serum cholesterol level and increased risk of lung cancer did appear to be stronger in strata with low intake of beta-carotene than in the total cohort. And they hypothesized that betacarotene mediated the association between low serum cholesterol and increased risk of lung cancer. Though beta-carotene may possibly play an important role in a serum cholesterol-cancer relationship, there have been few prospective studies64), which examined the interrelationship between serum cholesterol level and serum beta-carotene level or dietary intake of beta-carotene. There were also potential confounding factors which have been taken into account in previous studies; systolic blood pressure1,12-14,16,23) relative weight 14,15,24,25,32), educational level 11,24,32) alcohol consumption 4,11,14,18,19,23), hematocrit24), physical activity23), or social class1,12. When adjusted for these variables, significant inverse associations between serum cholesterol and cancer were not altered65) Third, if an association is real, a mechanism by which low serum cholesterol causes cancer should be rational. Isles and his colleagues 12) indicated the theoretical possibility that low cholesterol concentration predisposes directly to cancer, as cholesterol is an essential component of cell membranes. Kark and his colleagues15) also discussed that serum cholesterol levels may be associated predictively with a risk of developing certain cancers, possibly through an effect of serum cholesterol concentration on the characteristics of cell membranes, or by serum cholesterol levels reflecting membrane cholesterol levels. It is also possible that "malnutrition" expressed as "low serum cholesterol" may promote cancer development in some populations.
According to Lowell and his col-leagues66), laboratory studies have documented detrimental effects of malnutrition on host defenses and immune function by decreasing functions of lymphocytes, granulocytes, and macrophages.
Since immune effector cells are known to mediate antitumor effects in vitro67), malnutrition may possibly be implicated in oncogenesis and/or the body's ability to control the development and spread of cancer by immune depression, though the overall mechanisms remain unknown.
Increasing dietary intake of saturated fat and cholesterol is known to increase serum cholesterol level 61-71), and international correlation was positive between per capita consumption of diets high in saturated fat and animal fat and cancer risk 72,73). This positive correlation is quite contradictory to low serum cholesterolcancer association.
In 1992, Kritchevsky 17), who examined the role of dietary fat and cholesterol, found that diet cannot explain the low serum cholesterolcancer association.
Based on these discussions, it may probably be hypothesized that men with naturally occurring low levels of cholesterol are at an increased risk of cancer 11,74). To resolve this hypothesis, such researches as intervention trials of increasing serum cholesterol in low-serum-cholesterol populations may be needed.
In conclusion, we found a statistically significant inverse association between serum cholesterol level and cancer mortality in men. This inverse association was most apparent in 6 years or later after serum measurement. This finding was believed not to be ascribable to an effect of preclinical cancer.