Low Cholesterol is Associated With Mortality From Stroke, Heart Disease, and Cancer: The Jichi Medical School Cohort Study

Background We investigated the relationship between low cholesterol and mortality and examined whether that relationship differs with respect to cause of death. Methods A community-based prospective cohort study was conducted in 12 rural areas in Japan. The study subjects were 12 334 healthy adults aged 40 to 69 years who underwent a mass screening examination. Serum total cholesterol was measured by an enzymatic method. The outcome was total mortality, by sex and cause of death. Information regarding cause of death was obtained from death certificates, and the average follow-up period was 11.9 years. Results As compared with a moderate cholesterol level (4.14–5.17 mmol/L), the age-adjusted hazard ratio (HR) of low cholesterol (<4.14 mmol/L) for mortality was 1.49 (95% confidence interval [CI]: 1.23–1.79) in men and 1.50 (1.10–2.04) in women. High cholesterol (≥6.21 mmol/L) was not a risk factor. This association was unchanged in analyses that excluded deaths due to liver disease, which yielded age-adjusted HRs of 1.38 (95% CI, 1.13–1.67) in men and 1.49 (1.09–2.04) in women. The multivariate-adjusted HRs and 95% CIs of the lowest cholesterol group for hemorrhagic stroke, heart failure (excluding myocardial infarction), and cancer mortality significantly higher than those of the moderate cholesterol group, for each cause of death. Conclusions Low cholesterol was related to high mortality even after excluding deaths due to liver disease from the analysis. High cholesterol was not a risk factor for mortality.


INTRODUCTION
Both medical professionals and patients are well aware of the dangers of high cholesterol, but most know little about the risks of low cholesterol, despite the many studies that have examined the issue. [1][2][3] The first report to show a relationship between low cholesterol and cerebral hemorrhage was a Japanese cohort study, 4 and many subsequent observational studies have shown that low cholesterol is associated with cerebral hemorrhage, 5 cancer, suicide, injury, and noncoronary mortality. [6][7][8] However, there is no explicit evidence that these relationships are causal.
A meta-analysis of interventional trials showed that cholesterol-lowering therapy was associated with high mortality in a population with low cardiovascular risk. 9 Although this meta-analysis focused on interventions other than statins, studies of statins have also shown that statin administration is associated with increases in cancer incidence among elderly adults, 10 breast cancer incidence during the secondary prevention phase, 11 and total cancer incidence. 12 Japanese researchers reported that the relationship between low cholesterol and mortality disappeared when deaths due to liver disease were excluded. 13 To clarify this issue, we investigated the relationship between cholesterol and mortality with respect to cause of death (deaths due to stroke, heart disease, and cancer). In addition, the relationship between cholesterol and mortality was examined after excluding deaths due to liver disease.

METHODS Participants
This study was conducted as part of the Jichi Medical School (JMS) Cohort Study 13 ; 12 490 men and women aged 40 to 69 years participated. The JMS Cohort Study is a prospective cohort study that began in 1992 with the aim of investigating risk factors for stroke and cardiovascular diseases. We collected baseline data from April 1992 through July 1995 in 12 rural areas of Japan and completed a follow-up in December 2005, in which 65% of the subjects from mass screening examinations participated. Of these, 96% of participants completed follow-up; incomplete data were obtained for 409 participants who were not followed until the last day of our study because they had left the study area. There were no follow-up data for 97 participants. However, data from the abovementioned 409 subjects were included in the analyses, and the day they left the area was defined as the endpoint. Informed consent for follow-up was not obtained from 95 participants and from 2 additional participants who had already left the area at the beginning of follow-up. These 97 individuals were excluded from the analyses. Furthermore, we were unable to obtain total cholesterol data for 156 participants, including 4 who did not provide informed consent. In total, we analyzed data from 12 241 participants, ie, 98% of the total number of participants. The average follow-up period was 11.9 years. We observed participants for a total of 145 312 person-years. Details regarding the JMS Cohort Study are available elsewhere. 14 Exposure Total cholesterol was measured by an enzymatic method (Wako, Osaka, Japan; interassay coefficient of variation: 1.5%). All samples were measured at the same laboratory (SRL, Tokyo, Japan).

Confounding factors
We obtained information on confounding factors (smoking and drinking habits, blood pressure, height, weight, and highdensity lipoprotein [HDL] cholesterol) from the baseline data of the JMS Cohort Study.

Outcome
Information regarding cause of death was collected using data from death certificates and national vital statistics with the permission of the Agency of General Affairs.
Written informed consent was obtained from all participants. The Institutional Review Board of JMS was responsible for ethical review of this research and approved the study.

Statistical analysis
Because the relationship between cholesterol and mortality is not linear, subjects were divided into 4 groups according to total cholesterol level (<4.14 mmol/L, 4.14 mmol/L to <5.17 mmol/L, 5.17 mmol/L to <6.21 mmol/L, and ≥6.21 mmol/L). The second lowest group (4.14 mmol/L to <5.17 mmol/L) was used as the reference group. However, in the analyses of ischemic stroke in men and myocardial infarction in women, the highest group was defined as ≥5.70 mmol/L because there was no ischemic stroke or myocardial infarction in subjects with a total cholesterol level ≥6.21 mmol/L. The cutoff value used for the lowest group was selected on the basis of past research, 1 and the cutoff value for the highest group was based on criteria from the mass screening examination.
A Cox proportional hazards model was used to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for mortality. Both age-adjusted HRs and multivariateadjusted HRs were calculated. The multivariate-adjusted model adjusted for age, smoking status (current/former/ never), drinking status (current/former/never), systolic blood pressure, HDL cholesterol, and body mass index (BMI: <18, 18 to <22, 22 to <26, ≥26). Age, blood pressure, and HDL cholesterol were analyzed as continuous variables. All analyses were performed separately for each sex using STATA/SE for Windows (STATACORP, release 10, TX, USA). A P value of less than 0.05 was considered to indicate statistical significance.

RESULTS
All subjects were included in the age-adjusted analysis. Because of missing values, multivariate-adjusted analyses included data from only 11 869 subjects. Tables 1 and 2 show the baseline characteristics of subjects grouped by sex and cholesterol category. The background characteristics of subjects included and excluded from the multivariateadjusted analysis did not differ. In total, 635 men and 423 women died during the study period; 34 male deaths and 15 female deaths were due to liver diseases. Table 3a shows the number of deaths according to cause of death, including deaths due to liver disease, and incidence in each cholesterol group by sex. Liver cancer and hepatic failure were the main causes of death due to liver disease. Table 4a shows HRs and 95% CIs for mortality by cholesterol category. Crude, age-adjusted, and multivariateadjusted HRs are grouped by sex. Smoking status, drinking status, and BMI were entered into the models as categorical dummy variables. The multivariate-adjusted HR of the lowest cholesterol group (<4.14 mmol/L) was 1.38 (95% CI, 1.13-1.66) for men and 1.42 (1.02-2.00) for women. The HRs of men and women in the    highest cholesterol group (≥6.21 mmol/L) were not significant; the HR was less than 1 for women (HR, 0.93; 95% CI, 0.67-1.30).
Among women, there was an inverse association between total cholesterol and mortality age-adjusted analysis. The multifactor-adjusted HR in the lowest cholesterol group was 0.50 (95% CI, 0.16-1.55) in premenopausal women and 1.42 (0.98-2.06) in postmenopausal women. Table 4a shows the same results, after excluding deaths due to liver disease. In these analyses, the age-adjusted HRs of the lowest cholesterol group were statistically significant: 1.38 (1.13-1.67) for men and 1.49 (1.09-2.04) for women. There was no difference between these results and those for all causes of mortality.
The results of analyses that excluded deaths within the first 5 years of follow-up were similar to those that included all deaths (Table 4b). In addition, the results of analysis that excluded 320 participants with a history of stroke (113), myocardial infarction (65), or cancer (142) were similar (Table 4c). Table 5 shows HRs and 95% CIs for stroke, heart disease, and cancer mortality according to cholesterol category. The multivariate-adjusted HR of the lowest cholesterol group was higher than 1 for each cause of death and was statistically significant for cancer mortality in men.
We separately analyzed participants with and without ischemia for stroke and heart disease. Among subjects with the highest level of total cholesterol, the multivariate-adjusted HRs for ischemic stroke and myocardial infarction were not significant. The corresponding HRs for ischemic stroke in men and myocardial infarction in women were less than 1; however, the HR of the lowest group was 3.86 (95% CI, 1.18-12.68) for hemorrhagic stroke in women and 5.79 (1.07-31.27) for heart failure excluding myocardial infarction in women.

DISCUSSION
We noted a clear relationship between low cholesterol and increased mortality. Okamura et al 13 reported that occult liver diseases are associated with mortality; however, in the present study, the relationship between low cholesterol and increased mortality was unchanged in analyses that excluded deaths due to liver disease. Our results suggest that hemorrhagic stroke and heart failure excluding myocardial infarction, contribute to the relationship between low cholesterol and high mortality. Studies have shown a relationship between low cholesterol and non-cardiovascular mortality; however, in addition to cancer mortality, stroke mortality and heart disease mortality were also related to low cholesterol in our analyses. The relationship between low cholesterol and hemorrhagic stroke was similar to previously reported results. 3,5 Although the relationship between high cholesterol and ischemic stroke is not constant, it may be that the risk of high cholesterol disappears due to medical interventions for ischemic stroke and that the risk of low cholesterol is thus emphasized because of a lack of such interventions for low cholesterol.
It is difficult to interpret the relationship between low cholesterol and heart disease mortality. Although a relationship between cholesterol and heart failure was reported, high cholesterol, too, was a risk factor for non-ischemic heart failure. 16 We found no report of an association between low cholesterol and heart failure. The effects of malnutrition should be considered, as should the possible presence of beriberi heart disease and alcoholism. The relationship between low cholesterol and heart disease mortality was stronger in women than in men, so a disease like hyperthyroidism, which is more common in women, may be the culprit. A meta-analysis found an increase in cardiovascular mortality associated with subclinical hyperthyroidism, 17 but further investigations are necessary to confirm this hypothesis.
Because low cholesterol is associated with high cancer mortality, low cholesterol is a key finding in cancer. Previous studies reported an increase in liver cancer 13,15 ; however, in the present study, increased cancer mortality in the lowest cholesterol group was unchanged after excluding cases of liver disease from the analyses. This suggests a need for screening of cancers other than liver cancer in individuals with low cholesterol levels.
Our results differ from those of previous studies, 1,3 in that high cholesterol was not identified as a risk factor for mortality in the present study. The HR was 0.93 (0.67-1.30) for women, which indicates that the focus should be on adults with low cholesterol rather than those with high cholesterol.
Our analyses constitute a primary use of existing data. An important advantage of this study was that the follow-up rate was very high because the study was conducted in rural areas, where migration is far less than in urban areas.
Cause of death was ascertained using death certificates, so there were potential limitations in accuracy regarding cause of death. The setting was a periodical health examination to screen participants with high cholesterol. Thus, it is also necessary to consider the possibility that risk was underestimated due to medical therapy for high cholesterol.
Our results are specific to people living in rural areas of Japan and their lifestyle, and may not be applicable to urban Japanese or other ethnic groups. Because Japanese in rural areas have less coronary disease than people in Western countries, their risks from low cholesterol may be higher.
There are many factors that might contribute to the relationship between low cholesterol and high mortality. A correlation between high mortality and low cholesterol clearly exists, especially in populations with a low risk of coronary heart disease. 9 Although the dangers of a high cholesterol level are widely known, they are less important in regions-such as rural Japan-where cardiovascular disease is less common. It may therefore be necessary to highlight the risks of low cholesterol. In Japan, only LDL cholesterol and HDL cholesterol are measured at present; however, total cholesterol remains an important measure in predicting mortality.
In conclusion, we observed that low cholesterol was associated with increased risks of cancer, hemorrhagic stroke, and heart failure excluding myocardial infarction.