Methods

Top Methods Results Author & Article Info References

We searched the medical literature to find animal experiments, descriptive and observational studies, and clinical trials related to cholesterol and injuries. A computer MEDLINE search of English-language articles was done for the years 1966 to 1993 using the Medical Subject Heading terms "cholesterol," "accidents," and "mortality". We examined the bibliographies of articles and reviews to identify additional related literature. During the spring of 1993, Current Contents was searched by computer to identify relevant new articles.

Relative risks for death from injury in primary prevention trials of cholesterol reduction were calculated using Epi-Info software (USD; Stone Mountain, Georgia) [4], and summary estimates across trials were obtained with a random-effects Bayesian model [5, 6]. To ensure that important associations were not overlooked because of the assumptions involved in a random-effects model, we also calculated summary estimates for groups of trials using a fixed-effects model [7].

Results

Top Methods Results Author & Article Info References

Animal Studies

Experimental studies in animals have focused on behaviors rather than injuries. Monkeys that ate a low-fat diet and had low serum cholesterol levels (average, 3.76 mmol/L) were more aggressive than monkeys that ate a high-fat diet and had elevated serum cholesterol levels (average, 12.1 mmol/L) [8]. In another study by the same group, monkeys with low cholesterol levels had an altered release of prolactin in response to a stimulus, a finding that the authors interpreted as suggesting altered brain serotonin activity [9]. Two studies from a Canadian group reported that dietary fat can influence some measures of behavior in rats; for example, albino rats (but not hooded rats) fed a diet high in soybean fat took longer to lick their paws when placed on a hot plate [10, 11]. These two studies did not measure lipid levels, and the behavioral changes had no clear relation to the potential for injury. Overall, the few animal studies do not support a relation between cholesterol and injury.

Cross-sectional and Ecologic Studies

Cross-sectional surveys simultaneously measure exposure information and disease prevalence in a population. These studies have a common weakness in design in that serum cholesterol is measured after the injury has already occurred, thus making conclusions regarding cause and effect difficult.

Most cross-sectional studies have looked for associations between serum cholesterol and personality traits that might be proxy measures for violent behavior and deaths from injury. Jenkins and coworkers [12] administered psychological tests to California firemen and Georgia supermarket employees. They found higher cholesterol levels in men who valued social norms, dependability, and conscientiousness. Virkkunen [13-15] has published a series of articles on violent personality traits. In one study, prisoners with "antisocial personality" were found to have lower cholesterol levels (mean, 5.05 mmol/L) compared with other prisoners (mean, 6.0 mmol/L) [13]. A second study reported that murderers with a history of violence while consuming alcohol had lower cholesterol levels than other murderers [14]. A third study of Finnish boys with attention-deficit disorder found that an aggressive subgroup had lower cholesterol levels than the others [15].

Stewart and Stewart [16] attempted to replicate Virkkunen's findings on the association between cholesterol and antisocial personality. They studied psychiatric inpatients but found no difference in cholesterol levels between those with and those without "antisocial personality". In a cohort of London civil servants, researchers found no important correlation between serum cholesterol and hostility as measured by the Cook-Medley scale: r = 0.016 for men (n = 4246, P > 0.2) and r = –0.005 for women (n = 1742, P > 0.2) [17]. The investigators of the Coronary Artery Development in Young Adults (CARDIA) study, which included 5115 young adults in four states, used the same hostility scale and reported no statistically significant differences in serum cholesterol by hostility score quartile [18]. Among elderly men and women (n = 1592) in Edinburgh, Scotland, who were evaluated with the Bedford Foulds Personality Deviance Scales, investigators noted little correlation between total cholesterol and measures of hostility [19]. For example, the Spearman rank-correlation coefficients between total cholesterol and hostile thoughts and between total cholesterol and hostile acts in men were 0.06 and –0.02,respectively.

Morgan and colleagues [20] used the Beck depression inventory in 1020 men aged 50 to 89 years in Rancho Bernardo, California. They reported that among men older than 70 years, those who were in the lowest quartile of serum cholesterol were more likely to have high scores for depression compared with men whose cholesterol levels were higher (P = 0.03). They speculated that low cholesterol could result in depression and suicide. In men younger than 70 years, the age group more commonly followed in cohort studies and entered in clinical trials, no relation was found between depression and cholesterol level; no depressed patients were identified in the lowest cholesterol quartile.

One cross-sectional study actually examined injuries themselves rather than psychological measures [21]. Cholesterol levels from workers at a British nuclear plant were drawn at a screening examination that was offered to middle-aged men. The researcher then examined accident records for the previous 2 years and classified the 410 men as having one of the following: no injury, an injury that did not involve time lost from work, or an injury that caused absence from work. Mean cholesterol levels for the three groups were 5.73 mmol/L, 5.70 mmol/L, and 5.66 mmol/L, respectively. The differences among the groups were not statistically significant, and the author concluded that no relation existed between cholesterol level and history of injury.

Ecologic studies adopt groups of people as the unit of study. Investigators from the Seven Countries Study reported a negative correlation between the average initial serum cholesterol levels of 16 cohorts and the 25-year injury mortality rates (r = –0.27,P > 0.2) [22]. In summary, cross-sectional and ecologic studies have not supported an association between cholesterol and death from injury. The assumption in some of the studies that certain personality measures are related to death from injury is unproven.

Cohort Studies

The Conference on Low Blood Cholesterol summarized mortality data from 19 cohort studies and specifically looked for associations between low cholesterol and traumatic death [23]. Using a proportional hazards model, the authors pooled 14 cohort studies to obtain a relative risk of 1.4 for death from injury in men with serum cholesterol levels lower than 4.14 mmol/L compared with men whose levels were 4.14 to 5.15 mmol/L (P = 0.003). Twelve of the 14 studies found that in men, an increased risk for death from injury was associated with low cholesterol levels. The data from patients screened in the Multiple Risk Factor Intervention Trial (MRFIT) were examined separately, and the relative risk was 1.27 (P = 0.06). For women, data from six studies yielded a risk ratio of 1.26 for the low-cholesterol group (P > 0.2). These estimates were adjusted for age, diastolic blood pressure, cigarette smoking, and body mass index.

The investigators of some studies included in the Conference on Low Blood Cholesterol summary have recently reported their individual results in detail. The Whitehall study of London male civil servants [24] found no statistically significant association between cholesterol and death from injury; the age-adjusted relative risk for a 1.21 mmol/L decrease in cholesterol was 1.02 (95% CI, 0.83 to 1.26). For deaths from suicide, the relative risk for a 1.21 mmol/L decrease in cholesterol level was 1.19 (CI, 0.9 to 1.57), whereas for other deaths from injury the relative risk was 0.93 (CI, 0.72 to 1.21). In the Honolulu Heart Program [25], which followed men of Japanese ancestry, the highest cholesterol group had the highest age-adjusted death rate from injury, and the lowest cholesterol group had the next highest rate. Among the 350 977 men screened for the MRFIT study and followed for an average of 12 years, crude rates of death from injury were highest in the lowest cholesterol group (P < 0.05) [26]. When deaths from injury in patients in the MRFIT study were classified as nonintentional, homicide, and suicide, the highest crude mortality rate was always in the lowest cholesterol group; however, after adjusting for age, race, income, smoking, blood pressure, and season, a statistically significant inverse relation was seen only for deaths from suicide. The relative risk for suicide among men whose cholesterol levels were lower than 4.14 mmol/L was 1.61 (CI, 1.19 to 2.17) compared with men whose cholesterol levels were greater than that level.

Several studies were not included in the Conference on Low Blood Cholesterol meta-analysis [27-31]. In one study, researchers who followed a cohort of 1537 Italian men from the Seven Countries Study found a small, insignificant positive relation between cholesterol and death from injury [27]. In an analysis of the cohort of 1580 Finnish men of the Seven Countries Study, Pekkanen and colleagues [28] reported that in western Finland, the age-adjusted mortality rate from injury increased as cholesterol levels increased (P = 0.25), whereas in eastern Finland it decreased (P = 0.21). This study defined the lowest cholesterol group as patients with cholesterol levels lower than 6.05 mmol/L, which is much higher than a level lower than 4.14 mmol/L, used by the Conference on Low Blood Cholesterol study to define this group. The Low Blood Cholesterol study found that the risk for death from injury was almost flat for cholesterol levels greater than 4.14 mmol/L [23]; the Finnish data therefore appear to cover a cholesterol range that shows no variation in death from injury in most cohorts.

A study of Chinese men and women (n = 9021) [29], which was also not included in the Conference on Low Blood Cholesterol article, had many participants who fell into the "low" cholesterol range (<4.14 mmol/L). The rate of death from injury increased with lower levels of serum cholesterol; the ratios of observed to expected deaths for the lowest to highest quartiles of cholesterol were 1.55, 1.44, 0.85, and 0.23; a chi-square test for trend was significant (P = 0.04). After using a proportional hazards model to adjust for age, sex, diastolic blood pressure, smoking, and alcohol use, the association was marginally significant (P = 0.05).

Researchers studying a large Swedish cohort (n = 54 385) have reported their findings after 20.5 years of follow-up [30]. In men, low serum cholesterol was related to death from injury, especially from suicides, during the first 7 years of follow-up; age-adjusted relative risk in the lowest cholesterol quartile compared with the highest was 2.75 (CI, 1.52 to 4.96). However, this relation was no longer found 7 to 21 years after cholesterol levels were first measured. For women in this study, no relation was found between cholesterol and death from injury. A fifth study from the Netherlands of 2954 men and women concluded that low cholesterol was associated with death from injury, and the effect was present for 28 years; relative risk in the lowest tertile of cholesterol compared with the highest was 2.4 (CI, 1.0 to 5.4), after adjustment for age, sex, alcohol use, and smoking [31]. In summary, five cohort studies that were not reviewed by the Conference on Low Blood Cholesterol have reported some information on deaths from injury; the three largest studies reported evidence that the relative risk for death from injury was highest in the group with the lowest cholesterol.

The Bogalusa Heart Study is limited to children and young adults 26 years of age or younger [32]. From 1978 to 1989, 66 deaths occurred in study participants between ages 3 and 26 years; these deaths represent 85% of all Bogalusa-area deaths for that age range; 90% of the deaths were caused by injuries. In those who died, the average cholesterol level before death was 3.85 mmol/L, compared with an average of 4.09 mmol/L in a reference population of Bogalusa children. Thus, low cholesterol levels and death from injury may be associated even in children.

Possible Confounding in Cohort Studies

Male sex and age are strongly related to death from injury [33]. However, sex did not confound the relation between injury and cholesterol in most cohort studies because nearly all of these studies were either limited to men or stratified the analysis on sex. Investigators controlled for age in the analysis [23-31].

Deaths from injury are known to be associated with alcohol use, and this could falsely create an association between low cholesterol and death from injury in cohort studies if alcohol use is more common among those with low serum cholesterol [34]. Although the Conference on Low Blood Cholesterol did not present the trauma data adjusted for alcohol intake, the investigators did present the risk ratio for noncardiovascular, noncancer death stratified by alcohol intake, and almost no effect of alcohol intake was found in pooled results for men [23]. A few of the individual cohort studies were adjusted for alcohol intake [25, 29, 31]. Several studies have reported either no association or a direct association between total cholesterol and alcohol intake [28, 35-38]. It appears that alcohol use is unlikely to account for any association between low cholesterol and death from injury.

Another potential confounder is poverty, which is associated with both intentional and unintentional death from injury [33]. In a Yugoslavian cohort study (included in the Conference on Low Blood Cholesterol report), low cholesterol levels were associated with both injury mortality and low social class [39]. After adjustment for years of education and other variables, the association between cholesterol and deaths from injury changed from negative to positive. The MRFIT study found that the crude injury mortality rate was highest in the low-cholesterol group, but after adjusting for income, race, and other covariates, the relative risk estimate for each mg/dL increase in serum cholesterol was actually positive (1.0001; CI, 0.9989 to 1.0022) [26]. The Whitehall study [24] of London civil servants reported that lower employment grade was associated with lower serum cholesterol, and another British study [40] noted lower cholesterol levels in manual laborers. It is possible that in cohort studies, confounding by socioeconomic status could explain at least part of the association between low cholesterol and death from injury.

Primary Prevention Trials

Muldoon and coworkers [3] published a meta-analysis that summarized the results of the six published trials [1, 2, 41-44] of cholesterol reduction for the primary prevention of coronary disease. They limited the analysis to men in randomized trials that intervened only on cholesterol. They reported that the odds ratio for death from injury in the intervention groups compared with controls was 1.76 (CI, 1.19 to 2.58). This analysis has been criticized for improper data extraction [45] and for including trials that were not analyzed by the intention-to-treat method [29]. We reviewed the studies used by Muldoon and colleagues and concluded that the Veterans Administration diet trial should not be included in a meta-analysis because it reported deaths from injury only during the last 2 years of the study (four deaths from injury in the treatment group, none in controls), an interval examined specifically because the death rate was higher in the treated group during those years. The World Health Organization (WHO) clofibrate trial recently reported mortality results by intention to treat, and we used those figures in our meta-analysis [42, 46] (Table 1). With these revisions to eligible trials, the summary relative risk for death from injury among treated men compared with controls was 1.41 (CI, 0.95 to 2.10). A test for heterogeneity was not significant (P > 0.2), and with a fixed-effects model the summary odds ratio and CI were almost identical (odds ratio, 1.40; CI, 0.96 to 2.04). (The results for random and fixed models are so similar for all summary estimates and confidence limits that only the results of random-effects summaries of relative risk are presented.)

Although women have a lower incidence of homicide, suicide, and nonintentional death from injury, no reason exists to believe a priori that their risk for death from injury would be affected differently by attempts to reduce cholesterol. Including data on women from the Minnesota diet study [44] with data from the other primary prevention trials in Table 1 yields a summary relative risk estimate of 1.28 (CI, 0.90 to 1.83) for patients receiving treatment compared with controls.

An argument has been made for including the Expanded Clinical Evaluation of Lovastatin (EXCEL) trial as a primary prevention study because only 28.7% of those patients had a history of coronary disease [47]. Although this view of the trial has been disputed [48], it seems reasonable to us. Investigators in that trial administered lovastatin to 6582 patients and placebo to 1663 controls for 48 weeks [49-51]. No deaths from injury were reported, and if the EXCEL study is included in the meta-analysis, the results are almost unchanged: The relative risk was 1.3 (CI, 0.88 to 1.91) for all patients and 1.42 (CI, 0.94 to 2.15) for men (Table 2).

Another study of 52 patients may also be considered a primary prevention trial, but so many patients were lost to follow-up that we did not include this study in our meta-analysis [52]. No deaths from injury were reported, and including the study would have no effect on the results.

One concern about meta-analysis of deaths from injury in clinical trials of cardiac disease is that the definition of death from injury may differ among studies. For example, the Minnesota Coronary Survey included deaths caused by dental extraction and drug reactions with deaths caused by trauma and suicide [44].

Muldoon and coworkers [3] excluded from their analysis three primary prevention trials that used multiple interventions: the Oslo Study, which used two interventions, diet and antismoking efforts [53]; the MRFIT study, which treated hypertension, cholesterol, and smoking [54]; and a Finnish trial [55] similar to the MRFIT study. Muldoon and associates [3] argued that they wished to assess the unconfounded effect of efforts to lower cholesterol and that including multiple-intervention trials might bias their estimate of effect. If the three multiple-intervention trials are combined with the other primary prevention trials, the summary relative risk for death from injury is reduced to 1.17 (CI, 0.87 to 1.58) (Table 2). The European Collaborative Trial of Multifactorial Prevention is a community- and work-site-based study that encouraged a low-cholesterol diet, smoking reduction, weight control, and exercise, but information on deaths from injury has not been reported [56, 57].

Muldoon and colleagues [3] concluded that the association between lowering cholesterol and death from injury is causal but that the reason for this association is unknown. They have pointed out that the association may be with efforts to lower cholesterol rather than with a low cholesterol level itself: "Our paper concerned cholesterol reduction rather than low cholesterol concentrations. The process of lowering cholesterol concentration may perturb behavior, mood, or nervous system function and may do so only in certain people" [58]. Wysowski and Gross [59] reviewed the deaths from injury that occurred in the Lipid Research Clinics trial [2] and the Helsinki Heart Study [43]. They made it clear that the people who died of injuries in the treated groups did not have "low" cholesterol levels as defined by the cohort studies reviewed above; most had levels of 6.46 mmol/L or greater.

Smith and Pekkanen [60] divided primary intervention studies into those that used diet alone and those that used medication. They concluded that the relative risk for death from injury among men was increased by drug therapy (odds ratio, 1.75; CI, 1.07 to 2.85) but not by diet (odds ratio, 1.20; CI, 0.75 to 1.93). Their analysis used the incomplete Veterans Administration trial data [1], the old data on mortality from the WHO clofibrate trial [42], and a Finnish mental hospital diet study that used a different design from the randomized trials [61]. We believe that the first six trials listed in Table 1 are the primary prevention trials suitable for analysis. Using data only for men, the pooled relative risk estimate for death from injury in the treated patients of the five medication studies [2, 41-43, 46, 49-51] was 1.39 (CI, 0.84 to 2.29) compared with controls. Among the men in the single-diet study [44], the relative risk for the treated group was 1.50 (CI, 0.76 to 2.94). Thus, we found no evidence to support the view that risk for death from injury is associated only with drug therapy.

Secondary Prevention Trials

Secondary prevention trials of coronary disease are those in which most of the patients already had some clinically apparent evidence of coronary disease when they were enrolled in the study—usually a recent myocardial infarction, but sometimes evidence of angina or angiographic findings. If efforts to lower cholesterol affect the risk for injury, in principle this association could be seen in secondary prevention trials as well. A recent meta-analysis by Smith and colleagues [47] listed 25 secondary prevention trials that used diet or drugs to lower cholesterol [62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89], plus two additional trials that attempted to lower cholesterol in patients with diabetic retinopathy [90] and cerebrovascular disease [91]. These studies are small compared with the primary prevention trials; the six major primary prevention trials enrolled 38 094 patients, whereas the 27 secondary prevention trials included 14 802 patients. Unfortunately, most secondary prevention studies do not provide information about possible deaths from injury, and the investigators of one study that had two deaths from injury did not report the deaths by treatment group [75]. Table 1 shows the results for eight studies with published information about deaths from injury. When these results are added to the six single-intervention primary prevention trials, the summary relative risk for death from injury with treatment is 1.24 (CI, 0.88 to 1.77) (see Table 2). The relative risk estimates for individual studies and the most important pooled estimates are presented in Figure 1. Several secondary prevention trials have been done, and more are planned [47, 92, 93]. It is hoped that the new studies will assess and report on deaths from injury.

View larger version (22K): [in this window] [in a new window]   Figure 1. Relative risk estimates and 95% confidence intervals for death from injury of treated patients compared with controls in cholesterol-lowering trials. Relative risk greater than 1.0 indicates that cholesterol reduction increases the risk for death from injury. The reference number is given for individual studies; the number of studies used is given for summary estimates. Studies marked with an asterisk had relative risk estimates of infinity. Studies with no deaths from injury are not illustrated (references 49, 62, 70, 72, 78, 85), although they are included in summary estimates. WHO = World Health Organization; LRC = Lipid Research Clinics; MRFIT = Multiple Risk Factor Intervention Trial; CDP = Coronary Drug Project; POSCH = Program on the Surgical Control of Hyperlipidemia.

Potential Mechanisms for a Causal Association

If a relation exists between serum cholesterol or efforts to lower cholesterol and deaths from injury, the biological mechanism for this relation is unknown. Engelberg [94] has proposed that low cholesterol may result in low levels of cerebral serotonin, and this may in turn increase impulsive behavior that could lead to suicide or violence. Muldoon and colleagues [3, 58] speculated that patients who change their diets to lower their cholesterol intake might suffer mood changes sufficient to lead some to suicide or put them at increased risk for nonintentional injury. Smith and Pekkanen [60] have advanced the theory that an increased number of deaths from injury in primary prevention trials may be caused by adverse effects of the medications used in some trials rather than to actually lowering cholesterol. Jacobs [34] recently reviewed the role of cholesterol in cellular function, steroid synthesis, and the immune system, and pointed out that numerous ways exist in which low cholesterol might cause disease, including injuries. However, he concluded, and we concur, that few data are currently available to support such a mechanism.

Conclusions

Currently, animal, cross-sectional, and ecologic studies have not provided much evidence for an association between serum cholesterol and injuries. In most cohort studies, low cholesterol levels (<4.14 mmol/L) are associated with an increased risk for death from injury. But the strength of the relation is weak and may be caused by confounding from factors such as socioeconomic status.

The best evidence that cholesterol and death from injury are connected comes from data on men treated in single-intervention cardiac primary-prevention trials. Here, the association may be with the efforts or drugs used to lower cholesterol rather than with low levels of cholesterol. The association has only modest strength and borderline statistical significance. Because the strongest indication for an association between cholesterol lowering and injuries comes from cardiac prevention trials, it would be helpful if researchers involved in the primary and secondary prevention trials whose results have been published continue to report details of deaths from injury. To assist future studies of this question, new trials should prospectively collect data on deaths from injury with as much precision as possible.