Relation Between Serum Total Cholesterol Level and Cardiovascular Disease Stratified by Sex and Age Group: A Pooled Analysis of 65 594 Individuals From 10 Cohort Studies in Japan

Background The relation between serum total cholesterol (TC) and cardiovascular disease in women and in the elderly is unclear, especially in Asian populations. Methods and Results We examined this relation in the largest-scale pooled analysis of the Japanese population, the Evidence for Cardiovascular Prevention from Observational Cohorts in Japan (EPOCH-JAPAN) study. A total of 65 594 participants who were 40 to 89 years of age and did not have a past history of cardiovascular disease were examined. Cox proportional-hazards models were used to estimate hazard ratios for death from total stroke, cerebral infarction, intracranial cerebral hemorrhage, or coronary heart disease. The mean follow-up period was 10.1 years, with the number of deaths from total stroke, cerebral infarction, cerebral hemorrhage, and coronary heart disease being 875, 457, 212, and 374, respectively. The participants were divided into 2 age groups: middle-aged (40 to 69 years; mean age 55 years) and elderly (70 to 89 years; mean age 75 years). In men, the multivariate-adjusted hazard ratios for coronary heart disease in the highest TC category (≥6.21 mmol/L) compared with the lowest category (<4.14 mmol/L) were 2.52 (95% confidence interval: 1.15–5.07) in middle-aged participants and 2.77 (1.09–7.03) in elderly participants. In women, the hazard ratios of the highest TC category (≥6.72 mmol/L) compared with the lowest category (<4.66 mmol/L) were 3.20 (1.44–7.09) in middle-aged participants and 1.02 (0.42–2.49) in elderly participants. TC levels were not associated with cerebral infarction in any age or sex group and were associated negatively with total stroke and cerebral hemorrhage. Conclusion High serum TC levels are associated with coronary heart disease in middle-aged Japanese men and women, but evidence in elderly Japanese individuals is still limited.

H ypercholesterolemia is a well-documented and established risk factor for coronary heart disease (CHD). 1,2 However, evidence of this association is mainly from middle-aged or relatively young elderly men <70 years of age, whereas evidence from women or the elderly, especially in Asian populations, is scarce. One large metaanalysis based on observational studies found that high levels of serum total cholesterol (TC) were associated with an increased CHD mortality rate in both men and women. 3 However, this study was stratified only by age and sex and was not adjusted for other confounders. Another metaanalysis based on observational studies showed a weaker association between TC and CHD in women and in participants ≥75 years of age. 4 To our knowledge, no observational study has demonstrated a clear positive relation between serum TC levels and death from CHD in women and the elderly specific to Asian populations. 1 Furthermore, less evidence is available on the influence of serum TC on stroke than on CHD.
We therefore investigated the associations between serum TC level and death due to cardiovascular disease (CVD), such as CHD and stroke, after stratification by sex and age in the largest-scale pooled analysis carried out in the Japanese population. Our a priori hypothesis is that a high serum TC level is a risk factor for CVD in Japanese after stratification by both sex and age.

Study Design
This study was part of a pooled project called EPOCH-JAPAN (Evidence for Cardiovascular Prevention from Observational Cohorts in Japan), which incorporates a meta-analysis of individual participant data from 13 cohorts across Japan. The project was designed to conduct pooled analyses and examine the relation between cause-specific mortality rate and various exposures, including laboratory measures and lifestyle factors. The guidelines for a cohort recruitment of EPOCH-JAPAN were as follows: collection of health examination measures, >1000 participants, and >10 years of followup (although each cohort's collaborator had discretion to choose the size of his or her data set for pooling). Both nationwide and regional cohort studies were included. We have reported the detailed characteristics of each cohort previously. 5

Study Population
Of the 13 cohorts, 10 provided data on the cause of death (n=90 528). 5 The people who were (1) <40 or ≥90 years of age (n=10 528), (2) had past history of CVD (n=7422), and (3) lacked data on TC level (n=2122) at the baseline survey were removed. Moreover, 4862 participants were removed because of missing data for at least one of the following covariates: sex, age, body mass index, blood pressure, and smoking and drinking status. Finally, a total of 65 594 participants were included in the analysis. The levels of serum TC were measured enzymatically in all the cohorts, with the exception of the NIPPON DATA80 cohort, in which TC was measured by the Lieberman-Burchard direct method.

Ascertainment of Death
The causes of death were sought in great detail from the available sources in each cohort study. In most studies, death certificates were reviewed or the National Vital Statistics were used after permission had been obtained. Other sources used in some studies included autopsy reports, medical records, health examinations, and questionnaires. The underlying cause of death was coded according to the International Classification of Diseases (ICD) for National Vital Statistics based on the criteria proposed by the World Health Organization. 6 These classifications were based on the ICD-9 until the end of 1994 and on the ICD-10 from the beginning of 1995. The respective classification codes for ICD-9 and ICD-10 used in the study were as follows: death from CVD (390 to 459; I00 to I99), total stroke (TS) (410 to 414 or 430 to 438; I20 to I25 or I60 to I69), cerebral infarction (433 or 434 or 437.8; I63 or I69.3), intracranial cerebral hemorrhage (431 to 432; I61 or I69.1), and CHD (410 to 414; I20 to I25).

Statistical Methods
Sex-specific analysis was performed. TC was categorized into 7 categories (<4.14, 4.14 to 4.65, 4.66 to 5.16, 5.17 to 5.68, 5.69 to 6.20, 6.21 to 6.71, and ≥6.72 mmol/L) in accordance with a previous Japanese cohort study, 7 which had provided key evidence for the guidelines of the Japan Atherosclerosis Society for diagnosis and prevention of atherosclerotic CVD for Japanese. 8 However, because only a small number of participants had TC levels ≥6.72 mmol/L in men and <4.14 mmol/L in women, with the number of events in these participants being limited, we decided to combine these TC levels into the adjacent category (6.21 to 6.71 mmol/L in men and 4.14 to 4.65 mmol/L in women). The lowest level in both sexes (men, <4.14 mmol/L; women, <4.65 mmol/L) served as the reference group.
The study population was divided into 2 age groups in both men and women: middle-aged (40 to 69 years; mean age 55 years) and elderly (70 to 89 years; mean age 75 years). Age group-and sex-specific analyses were performed. Cox proportional-hazards models stratified by cohorts 9 were used to estimate the hazard ratios (HRs) for cardiovascular outcomes according to baseline TC. Deaths from CHD and from TS and its subtypes (cerebral infarction and cerebral hemorrhage) were used in the analysis. In the Cox model, age, body mass index, systolic blood pressure, smoking status (current smoker, ex-smoker, never-smoker), and drinking status (current drinker, ex-drinker, never-drinker) were used as confounding variables.
All confidence intervals were estimated at the 95% level, and the significance level was set at P=0.05. All the statistical analyses were performed in Statistical Analysis System release 9.13 (SAS Institute, Inc., Cary, NC).

Results
The baseline characteristics of the participants in the 10 cohorts are shown in Table 1. Each baseline survey was performed between 1977 and 1990, with the number of participants ranging from 1608 in the Tanno-Sobetsu cohort †In the studies of Tanno-Sobetsu, Ohasama, and Oyabe, ex-drinkers were classified as nondrinkers.
‡In this nationwide cohort study, the participants were from all areas of Japan.
to 24 940 in the Japan Collaborative Cohort (JACC) study. Mean age ranged from 47 years in the YKK cohort to 61 years in the Osaki cohort. The number of total participants was 65 594 (27 054 men and 38 540 women), and the mean age was 57 years. The mean ± standard deviation (SD) serum TC level of the total participants was 4.95±0.91 mmol/L for men and 5.33±0.96 mmol/L for women. The levels were lowest in the Oyabe cohort for men (4.71 mmol/L) and in the Tanno-Sobetsu cohort for women (5.02 mmol/L) and were highest in the YKK cohort for men (5.22 mmol/L) and in the Radiation Effects Research Foundation (RERF) cohort for women (5.56 mmol/ L). The mean follow-up period was 10.1 years, with the number of deaths from TS, cerebral infarction, cerebral hemorrhage, and CHD being 875, 457, 212, and 374, respectively.
In the Cox regression models, the relation between serum TC levels and CHD death was continuous and positive overall, with the exception of elderly women. However, when TC level was treated as a continuous variable, the relation was not significant in elderly men ( Table 2). In middle-aged men, the multivariate-adjusted HR of the highest TC category (≥6.21 mmol/L) for CHD was 2.52 (95% confidence interval [CI]: 1.15-5.07) compared with the lowest TC category (<4.14 mmol/L), and the multivariate-adjusted HR for a 1-SD increment in serum TC level (0.98 mmol/L) was 1.26 (95% CI: 1.11-1.42). In elderly men, the multivariate-adjusted HR of the highest TC category for CHD was 2.77 (95% CI: 1.09-7.03) compared with the lowest TC category. The multivariateadjusted HR for a 1-SD increment in serum TC level in these participants was 1.23 (95% CI: 0.96-1.56).
On the other hand, serum TC levels were not associated with cerebral infarction in any age or sex group and were associated negatively with cerebral hemorrhage and TS. The multivariateadjusted HRs for cerebral infarction for a 1-SD increment in serum TC were 0.92 (95% CI: 0.74-1.14) in middle-aged men, 1.04 (95% CI: 0.87-1.24) in elderly men, 1.08 (95% CI: 0.83-1.39) in middle-aged women, and 0.97 (95% CI: 0.80-1.16) in elderly women. The multivariate-adjusted HR for cerebral hemorrhage for a 1-SD increment in serum TC was 0.84 (95% CI: 0.72-0.97) in the combined participants. The multivariateadjusted HR for TS for a 1-SD increment in serum TC was 0.93 (95% CI: 0.862-0.997) in the combined participants. The Figure 1 summarizes the associations with CHD and cerebral infarction by sex and age groups.
Although 52% (n=34 379) of the participants in the present study had information on antihypertensive medication, the relations between TC and death from CVD (HRs) did not change substantially when use of hypertension medication (16%) was added as a covariate. In the subgroup analysis of participants with information on self-reported diabetes (n=28 793) or casual blood glucose (n=32 384), the relations between TC and death from CVD also were not altered when diabetes was added as a covariate. Because it is likely that there were time period differences in CVD event rates, we reanalyzed the data excluding 2 cohorts with earlier baseline surveys (Tanno-Sobetsu [1977] and NIPPON DATA80 [1980] cohorts). This analysis showed that the relations between TC and death from CVD (HRs) did not change substantially (data not shown).

Discussion
In this large cohort study in Japan, we found a positive relation between serum TC level and CHD death in both middle-aged women and middle-aged men. We also observed that the highest TC group (≥6.21 mmol/L) had an increased risk for CHD in elderly men, although the multivariate-adjusted HR did not reach a statistically significant level when TC level was treated as a continuous variable. There was also no relation between TC and CHD in elderly women. Except for elderly women, our results are similar to those reported in 2 previous large-scale studies. 3,4 To our knowledge, this is the first finding from a large-scale study specific to an Asian population that demonstrates a positive relation between hypercholesterolemia and CHD in middle-aged women with a mean age of 55 years. On the other hand, serum TC levels were not associated with cerebral infarction in any age or sex group and were associated negatively with TS and cerebral hemorrhage death.
The Prospective Study of Pravastatin in the Elderly at Risk (PROSPER) study, which included participants between 70 and 82 years of age with prior vascular disease (mean age 75 years), showed that pravastatin reduced the risk of CHD events. 10 In addition, the Management of Elevated Cholesterol in the Primary Prevention Group of Adult Japanese (MEGA) study showed that pravastatin reduced the risk of CHD in Japanese participants ≥60 years of age without a history of CHD or stroke. 11 The Justification for the Use of Statins in Primary Prevention: An Intervention Trial Evaluating Rosuvastatin (JUPITER) study also showed that rosuvastatin reduced the incidence of major cardiovascular events, including CHD, in people ≥70 years of age without prior vascular disease who had an elevated high-sensitivity C-reactive protein without apparent hyperlipidemia. 12 However, observational studies on HR indicates hazard ratio; CVD, cardiovascular disease. *One SD of total cholesterol was 0.98 mmol/L. †Crude mortality rate was expressed as per 1000 person-years. ‡HR was adjusted for age, systolic blood pressure, body mass index, smoking categories, and drinking categories. All analyses were stratified by cohort. elderly people at low risk or in the primary care setting are very rare, and there is little evidence of a sex difference. Moreover, elderly people, especially elderly women, are the dominant population in the currently aging societies of developed countries. Therefore, the sex-specific findings we observed in elderly community dwellers are important and suggest an increased risk for CHD in the highest TC group (≥6.21 mmol/L) in elderly men. However, this relation was not significant when TC level was treated as a continuous variable. The Seven Countries Study showed that Japan had the lowest CHD mortality rate among developed countries, which was attributed largely to remarkably low serum TC  Figure 1. Multivariate-adjusted hazard ratios for death from (A) coronary heart disease and (B) cerebral infarction according to total cholesterol levels. Hazard ratio was adjusted for cohort, age, systolic blood pressure, body mass index, and smoking and drinking categories. CI indicates confidence interval. levels in the 1950s. 13 However, changes in lifestyle toward a Westernized pattern in Japan have resulted in a continuous increase in dietary fat intake and serum TC levels. Moreover, it was reported recently that Japanese born after World War II had serum TC levels similar to those of white men 14 and that the incident rate of CHD in Japan is increasing in some areas. 15,16 In the Western population, a positive relation between TC and CHD was observed in both sexes and in all age groups, although its association was attenuated in the elderly. 3 The weak association or lack of association between serum TC and CHD in elderly Japanese participants could be due to relatively less exposure to hypercholesterolemia in their young and middle-aged periods before the baseline, a time when mean serum TC level was known to be very low. 17,18 Although we do not have information on serum TC levels before the baseline measurements in the present study, elderly Japanese, even those with hypercholesterolemia, might not have had higher serum TC levels throughout their entire lives. Especially for elderly women, the exposure period to high serum TC could be shorter than for men, as serum TC levels in women before menopause are considerably lower than in men. This could be one reason for the lack of relation between serum TC levels and CHD in elderly women. Furthermore, our findings could be explained by survivor bias. 19 In other words, elderly participants in the present study might have some beneficial characteristics that helped them avoid CHD due to hypercholesterolemia.
In the present study, serum TC levels were not associated with cerebral infarction in any age or sex group. This finding is different from several other large-scale cohort studies in Western populations, which showed a weak but positive association between serum TC and cerebral infarction, 3,4,20,21 but is similar to previous cohort studies in Japan and some Western countries. 2,[22][23][24][25] These discrepancies could be due to differences in the prevalence of subtypes of cerebral infarction. Cerebral infarction consists of 3 major pathological subtypes-namely lacunar, atherothrombotic, and cardioembolic infarctions. In some Western populations, atherothrombotic infarctions account for approximately one half of cerebral infarctions, 26 whereas in some Japanese populations, it accounts for only approximately one quarter of cerebral infarctions. Furthermore, cardioembolic infarction is more common than atherothrombotic infarction, accounting for 23% to 38% of cerebral infarctions. [27][28][29][30] The Hisayama study in a Japanese community showed that serum low-density lipoprotein cholesterol was associated positively with only atherothrombotic infarctions, whereas it was associated negatively with cardioembolic infarction and showed no association with lacunar infarction. 27 One possible mechanism for the inverse association between low TC and cardioembolic infarction is that low TC increases the occur-rence of atrial fibrillation, 31 which is the predominant risk factor for cardioembolic infarction. The aforementioned heterogeneity in pathological background of cerebral infarction might be a major reason for the variation in findings observed among cohort studies.
In the present study, serum TC levels were associated negatively with risk of cerebral hemorrhage death. This result is similar to previous studies in Japan and the United States. 20,32,33 Low serum TC can induce angionecrosis, possibly in coexistence with hypertension. Experimental evidence from one study showed that a hypercholesterolemic diet given to spontaneously hypertensive rats reduced angionecrosis of smooth muscle cells in intracerebral arteries, leading to the occurrence of hemorrhagic stroke. 34 Low serum TC also can reflect nutritional status, which is known to be related to death after onset. Further basic, clinical, and epidemiological studies on these associations are required. Consequently, because TC can be associated inversely with the cardioembolic type of cerebral infarction in addition to cerebral hemorrhage, it is not unexpected that TC was associated inversely with TS death in the present study, similar to another recent report from Japan. 35 Several limitations need to be considered when these results are interpreted. First, we did not take into account the use or nonuse of cholesterol-lowering therapy, including statins, the main drug used to treat hypercholesterolemia. However, baseline surveys in 7 cohorts of EPOCH-JAPAN were performed before the introduction of the first statin in Japan (1989). 36 Another 3 cohorts were also started around 1990, and therefore it is likely that only a few participants were taking statins at baseline. Consequently, it is not likely that this limitation would have changed our inferences substantially. Second, the results were based on single health examinations and were likely to have underestimated the true association because of regression dilution bias. Third, the participants in the study volunteered to receive their health examinations, and for that reason their characteristics could be somewhat different from those of nonparticipants or the general population. This would influence the absolute measure of effect (mortality rate) and could therefore underestimate the risk. However, these differences have little effect on relative measures of effect, such as HRs. Fourth, the number of deaths from CHD in the study might not be sufficient in elderly women to estimate the association with serum TC.
In conclusion, this largest-scale pooled analysis specific to Asians found a significant positive relation between serum TC and CHD in both middle-aged men and middle-aged women who were 40 to 69 years of age and did not have a past history of CVD, although a similar relation in elderly participants was not confirmed. Further research is therefore warranted in elderly men and women.