Jump to
Abstract
Background Many studies have reported an association between a low or lowered blood total cholesterol (TC) level and subsequent nonatherosclerotic disease incidence or death. The question of whether low TC is a true risk factor or alternatively a consequence of occult disease at the time of TC measurement remains unsettled. To shed new light onto this problem, we analyzed TC change over a 6- year period (from exam 1 in 1965 through 1968 to exam 3 in 1971 through 1974) in relation to subsequent 16-year mortality in a cohort of Japanese American men.
Methods and Results The study was based on 5941 men 45 to 68 years of age without prior history of coronary heart disease, stroke, cancer, or gastrointestinal-liver disease at exam 1 who also participated in exam 3 of the Honolulu Heart Program. The association of TC change with mortality end points was investigated with two different approaches (continuous and categorical TC change) with standard survival analysis techniques. Falling TC level was accompanied by a subsequent increased risk of death caused by some cancers (hemopoietic, esophageal, and prostate), noncardiovascular noncancer causes (particularly liver disease), and all causes. The risk-factor–adjusted rate of all-cause mortality was 30% higher (relative risk, 1.30; 95% CI, 1.06 to 1.59) among persons with a decline from middle (180 to 239 mg/dL) to low (<180 mg/dL) TC than in persons remaining at a stable middle level. By contrast, there was no significant increase in all-cause mortality risk among cohort men with stable low TC levels. Nonillness mortality (deaths caused by trauma and suicide) was not related to either TC change or the average of TC levels in exams 1 and 3.
Conclusions These results add strength to the reverse-causality proposition that catabolic diseases cause TC to decrease.
A J- or U-shaped relation between blood total cholesterol (TC) level and all-cause mortality has been reported in several1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 but not in other21 22 23 24 25 26 27 28 29 30 31 32 33 studies. The reports of inverse relations have caused some alarm and controversy about the implications of low TC in the general population.34 35 36 37
Three interpretations of this association are possible. One view is that a low TC level plays an etiologic role in a variety of nonatherosclerotic diseases (direct causality). Another possibility, however, is that the low TC–mortality relation is attributable to a hypocholesterolemic effect of disease in a preclinical stage (reverse causality). Finally, the association between low TC and mortality may be due to other unmeasured or unknown factors related to both low TC and disease (confounding).38 39 40 41 With a few exceptions,4 8 earlier studies on this topic have not adequately addressed the direct versus reverse causality problem, largely because most statistical analyses were based on a single measurement of TC. Consequently, inferences about temporality could not be made.
Previous studies in the Honolulu Heart Program (HHP) relating baseline TC with subsequent 9-year,7 13-year,8 and 17-year37 mortality demonstrated inverse relations of TC with hemorrhagic stroke, liver disease, chronic obstructive lung disease, and cancers of the esophagus, colon, liver, and hemopoietic system. The relations for cancer and benign liver disease were stronger in the first 5 years and showed flattening when deaths in the first 5 years of follow-up were deleted.
The aim of this study was to extend these preceding reports by assessing the relation of changes in TC level from 1965 through 1968 (exam 1) to 1971 through 1974 (exam 3) with subsequent 16-year mortality. The hypothesis of reverse causality, that low TC is a consequence of disease, would be compatible with a situation in which a declining TC level (resulting from catabolic diseases such as cancer or liver disease), not stable low level, would be related to subsequent mortality. Stated differently, a declining TC level would probably be caused by disease, whereas a stable low TC level would be the result of dietary, lifestyle, or genetic factors.
Methods
Study Population and Procedures
The HHP is a long-term, prospective study of cardiovascular disease among middle-aged men of Japanese ancestry who were residing on Oahu Island, Hawaii, in 1965. Details of the cohort recruitment and study design and procedures were given elsewhere.42 43 In brief, a baseline comprehensive examination was carried out during 1965 through 1968 among 8006 study participants. Of those, 6860 men (85% of the original sample) completed the third examination cycle during 1971 through 1974. At both the baseline examination (exam 1) and the reexamination 6 years later (exam 3), detailed information was collected on sociodemographic characteristics, medical history, anthropometric measures, and smoking and alcohol consumption habits. For each subject, blood pressure was recorded three times (twice by a nurse and once by a physician) with a standard sphygmomanometer applied to the left arm of a seated subject, and the average reading was used for the analysis. Body mass index was computed as weight (in kilograms) over height (in meters squared). Laboratory tests included determination of TC, blood glucose 1 hour after a 50-g glucose oral challenge, and uric acid. Nonfasting TC was measured by the Auto Analyzer N24a method at both examinations,44 but the analyses were performed by different laboratories. During 1965 through 1968, frozen sera (−20°C) were shipped in dry ice to the US Public Health Service Heart Disease Control Program Laboratory in San Francisco, Calif; during 1971 through 1974, samples were processed at the Kuakini Medical Center in Honolulu, Hawaii. A physical activity index was calculated as a weighted sum of hours per day spent in five levels of activity, as used in the Framingham Study.45 A dietician made dietary assessments using the 24-hour recall method.46 Measures of diet and physical activity were available only at baseline.
For the present report, mortality follow-up through hospital surveillance, newspaper obituaries, and state health department records covered a 16-year period beginning in 1973 and continuing through the end of 1988 and is believed to be essentially complete. For each fatal event, a medical review panel assigned an underlying cause of death in light of relevant clinical records and coded it according to the International Classification of Diseases, eighth revision. Attrition in this cohort is known to be very small; a follow-up survey in 1984 found that only 1.3% of the men could not be located.8 To avoid the possible effect of disease at the initial examination on risk factor levels or blood lipids, the present analysis excluded men with documented history of coronary heart disease (n=325), stroke (n=113), cancer (n=81), gastrectomy (n=234), colectomy (n=36), hepatic cirrhosis (n=3), or premalignant intestinal disease (n=36) at baseline. Of the initial 8006 men, a total of 758 had one or more of the above medical conditions. Of the 7248 men free of documented prevalent disease, 256 were excluded from analysis because they died before 1973. Of the remaining 6992 men, 739 did not attend exam 3 and thus also were excluded. This left 6253 men, 312 of which had missing values on TC and study confounders. Thus, the final sample for multivariate analysis comprised 5941 men.
Data Analysis
Multiple linear regression was used to identify independent correlates of TC change between exams 1 and 3. Next, the association of changes in TC during a 6-year interval with subsequent 16-year mortality was examined by two different statistical approaches. The first method consisted of modeling mortality rate ratios as a function of continuous TC change adjusted for average TC level. We computed within-individual change in TC as TC in exam 3 minus TC in exam 1; then we used this continuous change as the main covariate of interest. To avoid misleading results that may have arisen by using the TC change alone or by entering initial TC as a covariable, we chose to include the average TC between exams 1 and 3 (but not the initial value). The problem with using the initial value is that it is typically correlated strongly and negatively with the change variable (as a result of regression to the mean), a circumstance that may lead to spurious results. Conversely, the mean TC value, which represents the value half-way between the two examinations, usually shows a very small correlation with the change.47 48 The Cox proportional-hazards model49 50 was used to estimate age-adjusted associations of categorical and continuous TC change with cause-specific mortality end points. The continuous TC change approach, less constrained by sample size, allowed us to examine individual causes of death. Multivariate analysis was carried out, controlling for potential confounding by baseline body mass index, blood pressure, blood glucose, uric acid, physical activity, percent calories from fat, changes in smoking and alcohol consumption status, and body mass index change between exams 1 and 3. Changes in smoking status between examinations were represented by dummy (dichotomous) variables for past smokers at baseline, quitters between examinations, and continuing smokers relative to men who never smoked. Similarly, changes in alcohol consumption status encompassed dummy variables for continuing abstainers, continuing heavy drinkers (>40 mL ethanol per day, top quintile, at both exams 1 and 3), and heavy drinkers becoming abstainers at exam 3 relative to other drinking groups combined.
To further elucidate the TC change–mortality associations, we also carried out survival analysis with categorical TC change. First, the study population was subdivided into groups according to patterns of TC change from baseline to the second cholesterol measurement 6 years later. At each examination, cholesterol levels in individuals were ranked as low (<180 mg/dL), middle (180 to 239 mg/dL), or high (≥240 mg/dL). Then we compared rates of broad mortality categories (cardiovascular disease, cancer, other causes, and all-cause mortality) across subgroups of the population defined by all possible patterns of TC change. There were three groups of stable TC level (high-high, middle-middle, and low-low), three groups of declining TC level (high-middle, high-low, and middle-low), and three groups of rising TC level (low-middle, low-high, and middle-high). The low cutoff point of 180 mg/dL for the low level (which represented the 13th percentile point of the baseline TC distribution in this population) was used for consistency with earlier investigations of low TC in relation to mortality in this same cohort.7 8 37 The middle range of exam 1 TC (180 to 239 mg/dL) encompassed the zone of minimum death rate in this population,37 whereas the high cholesterol cutoff point of 240 mg/dL for the high level was selected so as to agree with the National Cholesterol Education Program “high-risk” category.51 This method allowed us to assess the two competing hypotheses (direct versus reverse causality) by comparing the associations of declining versus consistently low TC level with mortality outcomes. Relative risks were computed relative to the subgroup maintaining the middle TC level.
Results
Death was documented in 1370 of the 5941 eligible men during 16 years of follow-up (23% crude mortality rate). The number and causes of death were distributed as follows: 557 (41%) to cancer, 385 (28%) to cardiovascular causes, and 428 (31%) to noncardiovascular noncancer causes (see Table 3⇓ for a detailed breakdown of causes of death).
Correlates of Serum Cholesterol Change
TC level fell substantially from exam 1 to exam 3 among participants developing coronary heart disease, gastrointestinal disease, and liver cirrhosis between 1969 and 1975 (Table 1⇓). No important decline in TC was observed for incident cancer and lung disease during the same years. On the other hand, men with incident cerebrovascular disease experienced a modest rise in TC (Table 1⇓).
Mean Serum Cholesterol Levels at Exam 1 and Exam 3 (1971 Through 1974) and Mean Cholesterol Differences by Incident Medical Conditions Between 1969 and 1975: The Honolulu Heart Program
Seven variables were found to relate significantly and independently to TC change: age (β=−0.23; SEM=0.06), initial TC (β=−0.39; SEM=0.009), body mass index change (β=3.31; SEM=0.26), initial uric acid level (β=−0.08; SEM=0.02), past smoking at exam 1 (β=−1.80; SEM=0.83), liver cirrhosis diagnosed between 1969 and 1975 (β=−14.30; SEM=6.3), and cerebrovascular disease diagnosed between 1969 and 1975 (β=6.65; SEM=3.50). The adjusted R2 of the model was .24, indicating that these seven variables explained about 24% of the variation of TC change. The direction of these associations demonstrated that TC tended to diminish from exam 1 to exam 3 with a decrease in body mass index, older age, higher uric acid concentrations, history of past smoking, and newly diagnosed liver cirrhosis and that it was strongly influenced by initial TC level.
Analysis Using Continuous TC Change
As expected, very little correlation existed between the TC change and average TC (r=−.025; P=.06). Computation of mean TC differences between exams 1 and 3 according to causes of death revealed that almost all causes were associated with a generally modest decline in TC (Table 2⇓), except for large differences in nonmalignant liver disease (−17.5 mg/dL), hemopoietic cancer (−11.6 mg/dL), prostate cancer (−14.6 mg/dL), esophageal cancer (−22.3 mg/dL), rectal cancer (−10.5 mg/dL), and other circulatory deaths (−11.7 mg/dL). For decedents from nonillness mortality (deaths resulting from trauma and suicide), TC change was similar (−1.8 mg/dL) to that for survivors (−1.3 mg/dL). This also was the finding for nonhemorrhagic stroke (−1.7 mg/dL), with this value only slightly larger for decedents from hemorrhagic stroke (−3.0 mg/dL) and for lung cancer (−3.4 mg/dL).
Mean Serum Cholesterol Levels at Exam 1 and Exam 3 and Mean Cholesterol Differences by Causes of Death: The Honolulu Heart Program, 16 Years of Follow-up
Table 3⇓ gives the multivariate-adjusted relative risks of mortality derived from the analysis with continuous TC change. A reduction of 32 mg/dL in TC (1 SD of TC change) was significantly associated with significant excess risk of hemopoietic, prostate, esophageal, and total cancer and with mortality resulting from nonmalignant liver disease and all causes combined. To compensate for potential confounding as a result of the inclusion of early mortality in the analysis (through TC-lowering effect of occult disease, leading to early death), we also estimated the associations of TC change and mean TC with late mortality (excluding deaths through year 5). After early deaths were excluded, the associations were generally maintained, except for some weakening of the relation between TC change and hemopoietic malignancies.
Relative Risks or Cause-Specific Mortality Associated With a Decline in Serum Cholesterol of 32 mg/dL (0.83 mmol/L) by Cause of Death: The Honolulu Heart Program, 16 Years of Follow-up
It should also be pointed out that after TC change and the other risk factors were taken into account and early deaths were excluded, the average TC between exams 1 and 3 (data not shown) was a direct determinant of fatal coronary heart disease (P=.0001) and an inverse correlate of total cancer (P=.02) and fatal hemorrhagic stroke (P=.01). Nonillness mortality (deaths resulting from trauma and suicide) was not related to either TC change or the average of TC levels in HHP exams 1 and 3.
Analysis Using Patterns of TC Change
Table 4⇓ shows the distribution of the cohort by patterns of TC change. Men maintaining a stable middle level constituted by far the largest group (46% of the cohort). The two groups of primary interest—men with stable low TC levels and men with TC changes from middle to low—each represented about 6% of the cohort. Not surprisingly, very few subjects experienced TC changes of a large magnitude, eg, high to low or low to high.
Mean Serum Cholesterol Level at Exam 1 and Exam 3 and Mean Cholesterol Differences by Patterns of Cholesterol Change: The Honolulu Heart Program, 16 Years of Follow-up
Table 5⇓ gives the results from the multivariate analysis performed with categorical variables for patterns of TC change. Relative to men remaining at a middle TC level, those with consistently high TC levels and those experiencing a rise from a low to a high level had a significantly elevated risk of cardiovascular mortality. However, only one cardiovascular death was observed in the low-to-high TC change category. For cancer mortality, subjects whose TC levels changed from middle to low showed a significantly increased risk. Likewise, a significant elevation of noncardiovascular noncancer mortality and all-cause mortality risk was seen among study subjects displaying a TC decline from a middle to a low level. Other patterns of TC change did not have significant relations with the risk of mortality. Notably, no significant relation was observed between all-cause mortality and a sustained low TC level, although there was some marginal excess risk of cancer and “other” deaths in this group. As Table 5⇓ indicates, the middle-high subgroup had a slightly (nonsignificant) lower all-cause mortality than the middle-middle subgroup.
Relative Risks of Cause-Specific Mortality by Patterns of Serum Cholesterol Change Between Exam 1 and Exam 3: The Honolulu Heart Program, 16 Years of Follow-up
Discussion
The present report addresses the association of changes in TC during a 6-year period (from examinations done during 1965 through 1968 to examinations done during 1971 through 1974) with subsequent 16-year cause-specific mortality in a cohort of middle-aged Japanese American men. The fundamental question was whether and how strongly stable low TC or declining TC levels were associated with specific mortality outcomes. If only declining levels are associated with death by catabolic diseases (eg, cancer or liver disease), then the low TC–mortality association may be explained by reverse causality: disease-lowering TC. If, on the other hand, a stable low TC level is a significant risk factor for mortality, then the data would support the directionality of low TC to disease.
Observed changes in TC in the HHP cohort are considered to be spontaneous because only 1.7% of the decedents and 1.4% of the cohort survivors in 1988 were receiving lipid-lowering treatment at entry into the study (G. Wergowske, MD, personal communication, January 1995). No attempt was made to exclude these individuals.
Our results suggested that age, initial TC level, change in body mass index, former smoking at exam 1, baseline uric acid levels, and newly diagnosed liver cirrhosis were crucial factors related to TC change. A decrease in TC with increasing age has been reported in several investigations of older populations52 53 and may be due to decreased energy expenditure with concomitant reduced energy intake. For this cohort, the average decline was about 3.2 mg/dL per decade after adjustment for lifestyle and biological factors. The observed relation between initial levels and subsequent change is probably the result of the presence of regression to the mean, with initial extreme values converging toward middle values at exam 3. A direct relation between TC and weight was reported in cohort12 54 and case-control studies.55 56 Moreover, in one case-control study, the authors concluded that a decrease in weight and TC were related to the progression of malignant disease.55
We observed a significant risk of cancer, noncardiovascular noncancer, and all-cause mortality in men whose TC levels changed to low (<180 mg/dL) from the middle (180 to 239 mg/dL) level at baseline. The above finding for cancer mortality agrees with results from the Multiple Risk Factor Intervention Trial,4 in which participants who died of cancer experienced within 1 year of death a fall from baseline in serum cholesterol level that was 22.7 mg/dL (0.59 mmol/L) greater than that of surviving participants of the same randomization group and smoking status. These results also support longitudinal data published by Pekkanen et al57 from the Seven Countries Study in which a decline in TC also tended to be associated with death caused by cancer among the older cohort. A decreasing TC also was associated with elevated overall mortality and cardiovascular mortality in both men and women in the 30-year follow-up of the Framingham Study.58
The current results with TC change as a continuous variable noted a relation between TC decline and all-cause mortality that was largely the result of associations of falling TC with cancers of the hemopoietic system, esophagus, and prostate and with nonmalignant liver disease. The association between hemopoietic malignancies and low TC was reported in several large prospective studies.4 9 13 19 22 28 Alterations in plasma lipids and lipoprotein fractions were demonstrated in patients with acute leukemia and non-Hodgkin’s lymphoma that were related to the degree of underlying tumor burden and to the presence of bone marrow involvement.59 Several studies clearly demonstrated that high receptor-mediated uptake and degradation of LDL by the leukemic cells may cause hypocholesterolemia in acute myelogenous leukemia.60 61 62 Moreover, Knekt et al17 noted a significant inverse relation between TC and prostate cancer that persisted after exclusion of the first 4 years of follow-up. An association between low TC and prostate cancer was found in three other cohorts.28 63 64 Recent work in patients with metastatic carcinoma of the prostate showed an enhanced fractional elimination rate of plasma LDL associated with the tumor, which explained the low LDL levels, a situation similar to that in patients with myeloproliferative disorders.65
Deaths caused by nonmalignant liver disease in the HHP cohort consisted almost exclusively of alcoholic cirrhosis of the liver, a condition known to alter lipid metabolism, causing a decrease in lipoprotein production.66
Interestingly, coronary heart disease mortality was unrelated to cholesterol change but had a direct significant association with the mean TC level. This is in agreement with significant associations of serum cholesterol with the 10-year incidence of coronary heart disease,67 sudden cardiac death within 18 years,68 and autopsy-determined atherosclerosis in the coronaries and aortas in this same cohort.69
Mean cholesterol was an inverse determinant of fatal hemorrhagic stroke. Very low serum cholesterol concentrations have been hypothesized to be a causative factor in the development of intracranial aneurysms, lesions that are liable to rupture and cause intracerebral bleeding. The proposed mechanism would involve a weakening effect of the vascular architecture by inadequate cholesterol content in the membranes of the endothelial cells of small intracranial vessels.70 71 Contrary to this notion, however, persons with hypobetalipoproteinemia and children (who sustain very low TC levels for years) do not have hemorrhagic strokes. In any case, the absolute number of hemorrhagic stroke deaths in persons with very low TC (<160 mg/dL) was very small (n=5), so this association may not be important from a public health perspective.
The inverse relation between TC change and hemopoietic cancer may be explained by the cholesterol-lowering activity of neoplastic cells in the asymptomatic stage. Further evidence for reverse causality is provided by studies showing that low TC reverts to usual concentrations with successful chemotherapy and remission of disease.72 Moreover, the association between cancers of the hemopoietic system and TC decrease as a continuous variable diminished substantially when early deaths were eliminated from the analysis. On the other hand, the association of a decrease in TC and cancers of the prostate, esophagus, and all sites combined did not weaken when the first 5 years were excluded. It has been proposed that this long-term association may be due to prolongation of survival by treatment.73
The observed slope of TC was fitted across only two measures, either of which could be atypical for the subject. Thus, it is important to remember that the observed associations between TC change and mortality could be biased by measurement error. In the case of multiple exposure groups (as is the case in the analysis with categorical TC change), the direction of bias is less predictable and depends on not only the misclassification rates but also the distribution of subjects across exposure levels.74 In the case of continuous TC change with adjustment for multiple confounders, measurement error in TC change could attenuate, inflate, or leave the coefficient magnitude unaltered.75
The key finding of this report is that spontaneously falling TC levels were associated with increased risk of nonmalignant liver disease, total cancer, and most noticeably, cancers of the esophagus, prostate, and hemopoietic systems. By contrast, after confounders were controlled for, a stable low TC level was not associated with significantly increased mortality risk, although some marginal risk existed owing to an association of very low TC with fatal hemorrhagic stroke.
In conclusion, the results of this longitudinal analysis, together with existing coherent biological plausibility of reverse causality, support the hypothesis that low TC, independent of the risk factors considered in this study, appears to be a manifestation of tumor activity or the consequence of chronic liver disease. Future research should concentrate on the temporal association of serial TC measurements with disease and the potential etiologic role of chronically low TC in the pathogenesis of intracerebral hemorrhage.
Acknowledgments
The Honolulu Heart Program is supported by research contracts (NO1-HC-02901 and NO1-HV-02901) from the NHLBI, NIH, Bethesda, Md. This work was part of Dr Iribarren’s doctoral dissertation at the University of Southern California School of Medicine, Los Angeles.
Footnotes
-
Reprint requests to James H. Dwyer, PhD, Institute for Prevention Research, USC School of Medicine, 1540 Alcazar St, CHP 205, Los Angeles, CA 90033.
- Received April 27, 1995.
- Revision received July 17, 1995.
- Accepted August 16, 1995.
- Copyright © 1995 by American Heart Association
References
This Issue
Jump to
Article Tools
Share this Article
- Low Serum Cholesterol and MortalityPermalink:
