University of Minnesota

J- and U-Shaped Associations

J- and U-shaped distributions of risk associated with physical and behavioral attributes are increasingly found as larger and longer-term epidemiologic studies are pursued systematically among cohorts. Clinicians have long been familiar with the obvious anomalies and diseases found at the extremes of genetic and personal characteristics. Early insurance actuarial experience, moreover, highlighted the excess risk of death associated with the extremes of relative underweight and overweight among standard insured lives. Excess deaths at the lower end were usually attributable to preexisting chronic or wasting diseases such as tuberculosis and cancer. In the current era, a train of epidemiologic observations has renewed scientific interest in J- and U-shaped distributions of mortality risk as found associated with body mass, blood pressure level, serum total cholesterol level (TC), and self-reported alcohol intake, among other characteristics (Blackburn and Jacobs 1994).

The tendency in the medical community over the years has been to reject causality of these associations under the reasonable concept that persons at the extremes are “rare birds” of strange and heterogeneous constitution or are already ill at the time of characterization. But this comfortable conclusion has been challenged by the increasing numbers and varieties of observations of these U shapes, by the strength of the associations found, and by the greater opportunity to explore confounding factors of risk as well as to account for preclinical disease or co-morbid states.

Specifically, Japanese investigators since the 1960s have remarked on the ecologic association between excess risk of hemorrhagic stroke and low average population level of total serum cholesterol (TC). They hypothesized that the traditional Japanese diet, low in saturated fatty acids and cholesterol and animal protein, and high in carbohydrates, leads to low serum cholesterol, which, in turn, is a co-cause with hypertension of hemorrhagic stroke (Komachi et al. 1971; Kimura et al. 1977). The therapeutic and public health implications of this hypothesis distressed their Western colleagues.

The Western investigators early pointed out the probable fallacy of this interpretation, particularly in the absence of any Japanese populations subsisting on low salt regimens, citing other cultures with low saturated fat diets and low serum cholesterol levels, but also having relatively low salt intake, and in which there was no excess of stroke; to the contrary. The crux of the Western argument was that low TC and hemorrhagic stroke were likely confounded by the Japanese traditional culture of high salt, high CHO, low protein, and high alcohol intake, along with a smoking tradition and highly prevalent hypertension (Blackburn and Jacobs 1989). The stroke risk should not, therefore, be blamed on their low-fat diet and low average TC.

Jacobs and Blackburn (1994) even proposed this truism in explanation of the presumed fallacy: “Medical people are often insensitive to the most obvious aspects of their own milieu, in which people are homogeneously and heavily “bathed” in a particular behavior or exposure, whether it be salt, fat, physical indolence, or tobacco. This phenomenon may account for the frequent ignoring by the colloquial medical community of “major and obvious ? ” population causes of mass disease [ed. For example: salt and hypertension, saturated fats and high serum cholesterol, tobacco and lung cancer, sloth, obesity, and heart attack.”

In the 1970s and 80s, a flurry of findings reencouraged the originally Japanese hypothesis, culminating in those among the 351,000 MRFIT screenees in the U.S. male population. Even this affluent population, with a high fat, moderate salt diet, and relatively high average TC, had a three-fold excess of hemorrhagic stroke in the lower part of the serum cholesterol distribution, importantly individually correlated also to elevated blood pressure (Iso et al. 1989).

These, along with other reports relating low serum cholesterol to colon cancer, precipitated a formal NHLBI Conference on the so-called risk of low TC. It concluded that “the findings do not represent a public health challenge, however, they do represent a scientific challenge” (Feinleib 1982). A subsequent analytical effort, as part of the US-Japan Health Treaty, included evidence from cholesterol-lowering trials and made meta-analyses of this relation in coronary heart disease-free populations, excluding early deaths, with individual and pooled data plus all the needed adjustments for risk factors and alcohol intake. The result was a symmetrical and strongly U-shaped distribution of all-cause deaths by serum cholesterol level, with a 15% excess male mortality at TC lower than 160 mg./dl. and over 240, but a flat relation for women, and with a steadily graded (monotonic) increased risk for all of CHD deaths versus TC. The causes of the excess deaths at low TC were non-cardiovascular, which relation held in all the individual studies (Jacobs et al 1992).

These several formal analyses pretty well eliminated the effect of preexisting disease as the main cause of excess deaths with low TC. But the evidence was insufficient to rule out confounding or some actual biologic cause. The meta-analyses of TC lowering trials similarly left uncertainty about confounding but the possibility of noxious drug effects.

The conclusion of all this flurry was that lower TC, or lowering elevated TC levels reduces both individual and population risk of CHD, that the colon cancer link to low TC was not confirmed, and that the excess mortality with low TC is attenuated by time and thus due in part to preexisting disease. The unexplained residual required more exploration of causes versus confounding.

Nevertheless, TC lowering remains the population goal in high-risk societies, whereas individuals therein having “low TC” are advised not to seek further TC lowering. The dramatic issue died with a whimper and a shrug. (Henry Blackburn)


Blackburn, H, DR Jacobs, Jr. 1989. The ongoing natural experiment of cardiovascular diseases in Japan. Circulation. 79: 718.

Blackburn, H, DR Jacobs, Jr. 1994. The implications of J- and U-shaped associations of risk with serum cholesterol level. Pages 63-72; In, Filer, LF, Jr., RM Lauer, and RV Luepker (Eds) Prevention of Atherosclerosis and Hypertension Beginning in Youth. Philadelphia: Lea and Febiger.

Feinleib, M: 1982. Summary of a workshop on cholesterol and non-cardiovascular disease mortality. Prev. Med. 11: 360.

Komachi, Y et al. 1971. Geographic and occupational comparisons of risk factors in cardiovascular disese in Japan. Jpn. Circ. J. 35: 189.

Kimura, N et al. 1977. Fifteen year follow-up population survey on cerebrovascular disease in rural Japan. Jpn. J.Med. 16: 142.

Iso, H. et al. 1989. Serum cholesterol levels and six-year mortality from stroke in 350,997 men screened for the Multiple Risk Factor Intervention Trial. N.Engl. J.Med. 320: 904.