- Email: [email protected]
Longitudinal trends in total serum cholesterol levels in a Japanese cohort, 1958–1986
J Clin Epidemiol Vol. 50, No. 4, pp. 425-434, Copyright 0 1997 Elsevier Science Inc.
0895.4356/97/$17.00 PII SO8954356(96)00423-4
1997
ELSEVIER
Longitudinal Trends in Total Serum Cholesterol Levels in a Japanese Cohort, 1958-1986 Michiko Yama&,‘~* F. Lennie Wang,’ Kazunori Kodama, * Hideo Sasaki,’ Katsutaro Shimaoka,’ and Michio Yamakido3 DEPARTMENTS 3T~~ SECOND
OF ‘CLINICAL DEPARTMENT
STUDIES AND ‘STATISTICS, RADIATION EFFECTS RESEARCH FOUNDATION, OF INTERNAL MEDICINE, HIROSHIMA UNIVERSITY SCHOLL OF MEDICINE,
HIROSHIMA HIROSHIMA
AND JAPAN
ABSTRACT. The 28-year follow-up of a Japanese cohort, having collected vast amounts of data collected on total serum cholesterol (TC), provided an exceptional opportunity to examine TC temporal trends. The longitudinal statistical method of growth-curve analysis was used to elucidate the age-related changes in TC levels and to characterize these trends in relation to sex, birth cohort, time period, place of residence, and body mass index (BMI). Japanese TC levels at initial examination were remarkably lower than those in western countries. During the study period from 1958 to 1986, TC levels increased dramatically with age in both sexes. The slope of the cholesterol growth curve was steeper for women than for men, with the difference growing larger after age 40 years. Drastic changes in Japanese behavior and lifestyle, especially westernization of the diet, are thought to have affected the TC values as time-period effects. As a result of this temporal change, which affected different cohorts at different ages, TC values were higher in members of the younger cohort. The increase of the TC values as time-period effects were larger in earlier period than in later period. These time-period effects appeared to be almost similar in men and women. The TC growth curves also varied by city of residence. Subjects in urban areas had higher TC values than subjects in rural areas. Changes associated with BMI from 1958 to 1986 were only partially responsible for the increased steepness of the TC growth curve. J CLIN EPIDEMIOL 50;4:425434, 1997. 0 1997 Elsevier Science Inc. KEY WORDS. Japanese cohort, growth-curve
analysis, longitudinal
INTRODUCTION Total serum cholesterol (TC) is a recognized major risk factor for coronary heart disease (CHD) [l-3], whereas an inverse relationship between TC and cerebral hemorrhage has been reported [4,5]. From a public health standpoint, the possible relationship between TC and cancer is of great interest [6-81. Epidemiological studies have been conducted to better characterize the secular trends of TC [3]. Many studies examined the relationships between TC and basic demographic factors such as sex, age, and obesity [9-201. Most of these studies were cross-sectional in design. The few studies that have examined longitudinal changes in TC were based on a small number of measurements per person at intervals of less than 15 years and/or were of limited age range [14-201. Moreover, such studies have been conducted only in western or North American populations in which the level of TC is high.
‘Address Studies,
for correspondence: Radiation Effects
Michiko Research
Yamada, Foundation,
Department of Clinical 5-2 Hijiyama Park,
Minami-ku, Hiroshima, Japan. Current address for Dr. Katsutaro Center, 1000 Carson St., Torrance,
Shimaoka CA 90509.
is Harbor-UCLA
Medical
studies, total serum cholesterol
The Adult Health Study (AHS) was begun in 1958 at the Atomic Bomb Casualty Commission to study the occurrence of illnesses and the changes in physiological and biochemical function resulting from exposure to atomic-bomb radiation among residents of Hiroshima and Nagasaki. In the AHS, blood cholesterol has been measured biennially since 1958 in about 10,000 persons including both irradiated and nonirradiated subjects ranging from adolescents to the elderly. This provided an exceptional opportunity to examine the TC temporal trends in the Japanese over a 30year period. Although TC levels in Japan in the 1960s were reported to be remarkably lower than TC levels in western countries [21], the rapid westernization of the Japanese lifestyle [22,23] has caused drastic change in the TC profile [5]. The present analysis was conducted to elucidate these changes, both quantitatively and qualitatively. TC growth curves were estimated using data from both the irradiated and nonirradiated subjects, taking into account any effects of radiation exposure. An appropriate model for the nonirradiated subjects results by substituting zero for radiation dose in the model. The findings are presented in the context of age, birth cohort, and time period, along with sex, obesity, and place of residence.
M. Yamada et al.
426
MATERIALS AND METHODS Subjects The subjects are members of the Atomic Bomb Casualty Commission-Radiation Effects Research Foundation (ABCC-RERF) AHS cohort [24]. The AHS cohort at its inception was composed of 19,961 subjects, about half of whom had been within 2,000 m of the hypocenter, a quarter beyond 3,000 m, and the remainder not in the cities (NIC) at the time of the bombings (ATB). Since 1958, subjects have been invited to participate in biennial health examinations conducted by the ABCC-RERF clinical staff. These health examinations consist of clinical evaluation and routine laboratory determinations that include TC and blood pressure, and various anthropometric measurements such as of height and weight. Because of high attrition and the fear that the sample would eventually become too small for the detection of radiation effects, 2436 subjects were added to the study group in 1977. Attrition due to death was onethird of the original total population by the end of 1986, and attrition due to migration out of the cities has ranged from 9% to 13% throughout follow up. The level of participation in the AHS has been high, ranging from 71 to 86% among residents of both cities since the biennial examinations started [25-271. This study excludes the 5000 persons who were not in either city at the time of the bombings, because the socioeconomic status of this group differed from the other participants. Also excluded were 3283 persons for whom Dosimetry System 1986 (DS86) estimates were not available, 3529 who did not participate in the examinations, 383 completely lacking TC data, and 246 lacking all data on either height or weight. Since the accuracy of the recorded height and weight data could not be verified directly by reviewing the medical charts, we compared each measurement of an individual against his/her mean height or weight after age 20. A height that deviated by more than 55 cm and a weight by more than 25% were recorded as missing because they were considered to have been input errors. One hundred twenty-nine people lacking either height or weight data from every attended exam were excluded. Subjects diagnosed as having certain medical conditions that might adversely affect the TC levels were also excluded as follows: 11 with nephrotic syndrome, 92 with diseases of the gallbladder and biliary ducts, 14 with disorders of the adrenal glands, 61 with acute hepatitis, and 16 with nutritional marasmus. These medical conditions were identified by means of the three-digit International Classification of Diseases (ICD) codes that were entered into a computer database whenever diagnoses were made during the AHS examinations. Familial hypercholesterolemic cases were not excluded because their ascertainment based on only the ICD codes was considered unreliable; such cases were expected to constitute only a small fraction of the data. The remaining 9633 people, consisting of 3437 males (Hiroshima, 2416; Nagasaki, 1021) and 6196 females (Hiroshima, 4602; Nagasaki, 1594), were the subjects of our analyses.
Laboratory
Determinations
Nonfasting TC levels were determined by the KendallAbel1 method from 1958 to 1965. Later, TC levels were measured using three types of automated methods using a Technicon Autoanalyzer: the N24 method from 1965 to 1966; the N37A method from 1966 to 1967; and the N24A method from 1967 to 1986. A Hitachi model 7050 autoanalyzer has been used since 1986. Because autoanalyzers were not available at the Nagasaki Laboratory, blood specimens collected there were frozen and sent to the Hiroshima Laboratory for analysis. To maintain accuracy and reproducibility in each laboratory, quality control was conducted internally as well as externally, the latter with the Centers for Disease Control (Atlanta, GA). Standardization was performed to ensure comparability of TC measurements obtained from various cholesterol-determination methods. From each of the five systems, 45 to 105 TC measurements were obtained for estimating the equations that relate measurements from the four earlier methods to those of the autoanalyzer 7050. The relationship assumed was ln(Y,,,,) = a + b X ln(YnonP7&, where Ynon-7050is a TC measurement from one of the four earlier methods, YTosOis that from the autoanalyzer 7050, and a and b are parameters to be estimated. When b = 1, the above equation simplifies to Yad, = c X Y,,,,.-iOsO[28]. The correction factors c for measurements from Kendall-Abell, N24, N37A, and N24A methods were estimated as 0.9513, 0.7991, 0.8842, and 0.9468, respectively. The adequacy of the conversion factors was assessed graphically by examining whether an overt departure from linearity was exhibited by pairs of points in the calibration data set. Visual inspection showed the conversions to be satisfactory. Statistical
Analysis
We used the growth-curve technique for the analysis of serially measured data, using the mixed-effects model of Laird and Ware [29]. Let ys denote the natural log (In) of the jth TC measurement for person i, where j ranges from 1 to 14, depending on the number of TC determinations. The following growth function was used to relate the level of the natural log (In) of TC, yl,, in an individual with the time-varying covariable, age, and body mass index [BMI = (weight/height*) X 1001. Y,, = PO, + A,(ase,) + PdBMI,,
+ /Mat&)
+ MBMI,,)
. age,) + PdBMI,,
. ag$
(1) + cd.
The growth curve coefficients p = [PO1 , . . . , Wi] are specific to each subject, describing the cholesterol profile for individuals. We let p be a linear function of the fixed effects covariates, which include city (coded 0;l for Hiroshima;Nagasaki), sex (0;l for male;female), and the actual year of birth minus 1945 (YOB”) for distinguishing the birth cohorts. Each individual’s growth curve parameters were assumed
Longitudinal
Trends
427
Total Serum Cholesterol
in
to be linearly related to the effects of city (C), YOB* (Z), in the following way:
sex (S),
Pot = a0 + a,(C), + ads), + a3(Z), + ab(C . 3, + a,(C
Z), + a&S
PIL= aa + adc), + alo( Z), + al&S
+ a2,(C
Z), + azz(S . Z), + a&C
a32
+
a33(c),
+ a3dC
+
a34(s)L
.S
Z), + boz
+ all(Z), + a12(C . 9,
+ al,(C
+ a19(C . Z), + a&S
p41=
Z), + a,(C
. Z), + aJC
.S
. S . 2$ + bz,
. Z), + a,,(C +
a3dz)i
Z), + b,,
. S . Z), + b3, +
. Z), + a3&S . 23, + adC
a36(c
. s),
S . Z), + bqi
P5,= a4o+ a41(C),+ a42(S),+ a4343(Z), + a4,(C . S), + a&C
. Z), + a&
23, + a&C
. S Z), + b,,
(2) The restricted maximum likelihood estimates of the parameters were estimated using the program REML [30]. After nonsignificant effects were removed as sets based on the multivariate chi-square tests of zero simultaneous effects of the covariates, followed by step-wise removal of individual nonsignificant terms, a parsimonious growth model of ln(TC) was obtained. This step-wise algorithm is described in detail elsewhere [3 I]. Although our goal is to present the temporal trends of TC apart from radiation exposure aspects, restricting the analysis to only the nonirradiated subjects excludes 57% of the available data. This reduces the accuracy of the nonradiation-related parameter estimates, which was circumvented by adjusting for the effects of radiation dose in the model estimated by using all eligible subjects. To evaluate the significance of radiation dose effects, radiation dose (DOSE) and its interactions with city, sex, and YOB* were added in the model as fixed effects. Analyses showed that the resultant growth curve of ln(TC) included terms involving radiation dose, but coefficients of DOSE not involving sex were all zero, indicating that radiation effects were exclusive to females. At this stage, the fit of the observed to predicted was not
satisfactory,
particularly
among
the
older
cohorts.
It was
suspected that the convolution of the post-war social modernization process that took place in Japan might have affected TC in a nonlinear manner among different birth cohorts. This hypothesis was assessed by adding (YOB*)* and SEX.(YOB*)* terms in the model, which were found to be significant.
RESULTS The distribution of the subjects, their mean age in 1958, and the median number of cholesterol determinations per person are shown in Table 1 for each city and sex by 10 birth-year intervals. Overall, the median number of TC measurements was 7 per subject, with a range of 1 to 14. TC measurements were available for ages 13 to 98 years in males (birth years 1875 to 1945) and for 13 to 96 years in females (birth years 1877 to 1945). As an initial descriptive step before embarking on a more formal longitudinal analysis, especially in the absence of efficient statistical programs for analyzing a large amount of serially measured data, one might examine the data as depicted in Fig. 1. There, the mean TC values of the AHS participants grouped by birth cohorts are plotted against mean age in each examination cycle. It is clear that a remarkable change took place in the levels of TC over the course of 28 years. However, Figure 1 does not accurately depict the longitudinal changes encountered in an individual’s TC profile over time, because the trends are based on averages from examinations in which the composition of the study population varied. In the present analysis, the population growth curve was estimated using information obtained from each individual’s growth curves. The parameters of the estimated population growth curve of ln(TC) are shown in Table 2. The first column contains the terms for individual growth function and the second row contains the fixed-effects covariates included in the estimated population model. The estimated function may be reconstructed by going across each row by column to read off the elements of a, substituting 1 for the intercepts. That is, ln(TC) = 3.63 + 0.49(SEX) O.O17(YOB*) + . . . -1.6 x 1O-5 (CITY. SEX. YOB* . age* BMI) + 6.3 X 10m5 (DOSE . SEX . age’ BMI). We present the temporal trends of TC apart from radiation effects by substituting zero for DOSE in the above equation to calculate the expected values of ln(TC). Also used were the mean BMIs and the corresponding mean attained ages calculated from the nonirradiated subjects grouped into 10 birth-year and 5 year-attained age intervals. The model corresponding to Table 2 is difficult to interpret because of the many interaction terms. It can be said, in general, that age-related variations in TC levels are modified by gender, place of residence, and year of birth. Moredetailed descriptions of the differences are better presented graphically.
Sex Difference Time Period,
and Combined Effects of Age, and Birth Year on the TC Cjmwth
Curves
Figure 2 shows the observed and predicted TC levels and the 95% confidence intervals (CI) for men and women of Hiroshima born in 1910, 1920, 1930, and 1940. A relatively good fit of the predicted growth curves to the observed data is indicated.
Female
Male
All subjects No. with DS86 = 0 Mean age in 1958 Median no. TC determinations All subjects No. with DS86 = 0 Mean age in 1958 Median no. TC determinations All subjects No. with DS86 = 0 Mean age in 1958 Median no. TC determinations
“DS86 = Dosimetry System 1986 estimated dose. Atomic-bomb survivors no radiation exposure. bMedian number of cholesterol determinations per subject during AHS
Total
Nagasaki
to the year
All subjects No. with DS86 = 0 Mean age in 1958 Median no. TC determinations
Item
according
Female
subjects
All subjects No. with DS86” = 0 Mean age (years) in 1958 Median no. TC determinationsb
of the
Male
Sex
1. Distribution
Hiroshima
City
TABLE
136 52 63.9 4 200 71 67.0 4 24 16 66.8 4 28 8 66.0 4 388 147 66.9 4
examination
cycles 1-14.
available cohorts
1875 763 48.6 8
227 94 48.5 9
1024 362 48.5 8 157 112 48.5 8
467 195 48.7 8
1905-1914
Birth
data
to this system. If a survivor
1538 637 57.9 6
157 71 58.1 6
128 81 57.7 6
794 304 57.3 6
58.0 6
181
459
1895-1904
of the
doses according
description
1885-1894
and
are assigned radiation
::.6 3
30
0 0 -
-
0 0
4 76.5 3 19 6 76.7 3
11
51884
of birth
557 332 28.7 8 2707 1269 28.8 8
346 221 28.2 6
1104 428 29.1 8
700 288 28.6 8
1925-1934
has a dose of 0, he or she is considered
1748 730 38.3 10
10
353 168 37.9
113 79 39.4 10
10
982 370 38.0
300 113 39.3 9
1915-1924
to have received
1347 563 17.6 5
272 126 18.3 6
253 118 17.9 5
479 193 17.4 5
343 126 17.1 5
1935-1945
Longitudinal
Trends
in Total
Serum
429
Cholesterol
Men 210,
I
1906-15
<1906
1916-25 ?i I-
I 30
130 --20
I 40
I 50 Age
I
I
I
60
70
80
90
(years)
Women
210
shima women born in 1920, 1930, and 1940 are 158.3 mg/ dl, 176.3 mg/dl, and 184.7 mg/dl, respectively. The expected TC level for 53 year old Hiroshima women born in 1910, 1920, and 1930 are 169.4 mg/dl, 188.7 mg/dl, and 195.7 mg/dl, respectively. Changes in TC values between two individuals of the same age but 10 years apart in time were very similar for any given age. Moreover, the increase in TC level with age was greater in earlier than in later periods. For example, the TC levels of Hiroshima women aged 33, 43, and 53 years in 1963 and 1973 increased by 19.5 mg/dl, 18.0 mg/dl, and 19.3 mg/dl, respectively, over the lo-year period. On the other hand, the increase between 1973 and 1983 for Hiroshima women aged 43, 53, and 63 years was only 8.4 mg/ dl, 7.0 mg/dl, and 10.0 mg/dl, respectively. Similar patterns of change were seen in men as well.
.".
cc
The Effects of Place of Residence on the TC Cjrowth
'---\
1916-25
l1301 20
I 30
I 40
I 50 Age
I 60 (years)
I
I
I
70
80
90
TC growth curves for women of Hiroshima Nagasaki born in 1910, 1920, 1930, and 1940 are shown in Fig. 3. TC levels were consistently higher for residents of Hiroshima than for residents of Nagasaki. Nagasaki residents experienced somewhat smaller age-related changes in TC. Similar city differences were also seen in men. Sex differences and combined effects of age, time period, and year of birth on the TC growth curves were similar between the two cities.
The Effects of Varying FIGURE 1. Means of total ages for men and women Study (AHS) examination
serum cholesterol plotted against of Hiroshima during Adult Health cycles 1-14 by birth cohort.
TC levels increased with age. TC levels for subjects aged 18 to 48 years in 1958 ranged from 124 mg/dl to 146 mg/ dl in men and from 123 mg/dl to 153 mg/dl in women. TC levels 28 years later in 1986 for the corresponding subjects ranged from 173 mg/dl to 184 mg/dl in men and 185 mg/ dl to 203 mg/dl in women. The greatest change occurred for women born in 1930, for whom the average increase over the 28-year period was 66.9 mg/dl (95% CI [64.7 mg/ dl, 69.1 mg/dl]). The smallest change occurred for men born in 1910, for whom the average increase during the same period was 27.6 mg/dl (95% CI [25.6 mg/dl, 29.6 mg/dl]). The rate of increase in TC values declined gradually with age. Values for men began to taper off at younger ages than for women. At under 40 years of age, TC levels were higher in men than in women, although sex difference was small. At over 40 years of age, however, TC levels were lower in men than in women, and the difference increased with further aging. TC levels at a given age were higher for those born later. For example, the expected TC level at age 43 years for Hiro-
Curves
BMI
on the TC Qrowth
Cumes
As shown in Table 3, BMI generally increased with age, up to about 60 years in men and to mid-60s in women, and decreased with further aging. The actual magnitude of the BMI effect on TC at a given age depended on sex, city of residence, and YOB*. Figure 4 shows the expected TC growth curves for Hiroshima women assuming that their BMI values to be fixed at their mean levels in 1958 throughout the follow-up period and observed BMI. For example, between 1958 and 1986, the mean BMI of Hiroshima women born during 1925-1934 increased from 0.2112 to 0.2297 (or 4.4. kg, at a constant height of 154 cm) and TC increased from 128.8 mg/dl to 195.7 mg/dl. The magnitude of change in TC that can be attributed to this increase in BMI was less than 10 mg/dl. Thus, an increase in TC level can be explained only in part by changes in BMI. DISCUSSION Past longitudinal studies of North American populations and those leading the so-called “western” lifestyles displayed a wide-ranging trends in cholesterol levels that are largely associated with the age structure and environmental conditions inherent to those populations [14-201. In industrialized countries, TC levels tended to increase with age
M. Yamada et al.
Men 2101 T
18
23
28
33
YOB = I-930
38
43
48
53
58
63
68
73
’ 38
’ 43
’ 48
’ 53
’ 58
’ 63
’ 68
I 73
Women
‘f 0zl:b
YOB = 1930
1101 18
’ 23
’ 28
’ 33
Age (years)
FIGURE 2. Predicted growth curves for total serum cholesterol and the 95% confidence intervals by attained age for the u&radiated subjects of Hiroshima, plotted for men and women whose year of birth (YOB) was 1940, 1930, 1920, or 1910.
T
190
-
9 E p .2 I B 5s
170-
150
-
$0 Tii 5 130c
110""""' 18
23
28
33
38
43
48
53
58
-----Nagasaki 1 ' 63 68
Age(years)
FIGURE 3. Predicted growth curves of total serum cholesterol by attained age for the u&radiated women of Hiroshima and Nagasaki, whose year of birth (YOB) was 1940, 1930, 1920, or 1910.
1 73
Longitudinal
TABLE
3. Means 1958-1986,
women,
of body mass index by birth cohort
by
Birth Sex
18 23 28 33 38 43 48 53 58 63 68 73 18 23 28 33 38 43 48 53 58 63 68 73
Female
0.1977 0.2056 0.2134 0.2190 0.2227 0.2255
for
men
1920
0.2061 0.2088 0.2140 0.2181 0.2218 0.2237
0.2098 0.2145 0.2204 0.2246 0.2268 0.2295
age
and
cohort
1930
1940
As
Male
431
Trends in Total Serum Cholesterol
1910
0.2080 0.2096 0.2130 0.2170 0.2180 0.2176
0.2112 0.2156 0.2213 0.2264 0.2292 0.2297
0.2190 0.2188 0.2191 0.2183 0.2178 0.2171
0.2188 0.2248 0.2306 0.2338 0.2346 0.2334
0.2268 0.2236 0.2243 0.2262 0.2272 0.2256
[15-201, except during adolescence [14]. Only a few longitudinal studies were conducted of populations in non-industrialized countries in which mean population levels of TC are low [32,33]; no increase or only a small increase in serum cholesterol with age was found. For most studies, followup times spanned less than 15 years and TC was measured in-
210,
s ?
190
2 T
170-
D d '0
150
s i m B t-
1 -
-
130-
110""""""
16
23
26
33
36
43
48
53
58
63
68
73
Age (years)
FIGURE 4. Growth
curves of total serum cholesterol for the unirradiated women of Hiroshima, whose year of birth (YOB) was 1940, 1930, 1920, or 1910, assuming the body mass index (BMI) to be fixed at the mean value in 1958 (dotted curves BMI = 0.210, 0.211, 0.219, and 0.227, respectively) and observed BMI (solid curves).
collected TC data frequently [15-191. Th us, our biennially spanning a 28-year period offered a unique opportunity for characterizing the longitudinal trends of TC in an Asian cohort of wide-ranging ages. Our results show that temporal changes in TC levels differed among men and women. Before 40 years of age, sex difference in TC levels was small, although TC levels tended to be higher in men than in women. However, after 40 years, not only did the female TC level rise dramatically, sex difference also increased significantly. Similar trends have been observed in past cross-sectional studies [9-111. The huge sex difference in TC levels occurring after age 40 can be explained in part by changes in lipid metabolism that occur exclusively in women. A decrease in estrogen causes an increase in the TC level [34]. Estrogen levels change rapidly during the perimenopausal period, with TC rising to a higher level in postmenopausal than in premenopausal women [34]. Cross-sectional and longitudinal studies have shown that higher TC levels in postmenopausal women result from changes in lipid metabolism following hormone deprivation [34-371. This association has been confirmed further by studies of estrogen-replacement therapy in postmenopausal women which lowered their TC levels [37,38]. Although women exposed to 1 Gy (Gray [Gy] is the international unit for absorbed dose. 1 Gy = 1 J/kg = 100 rad) of atomic bomb radiation in our data showed an average increase of 2 mg/dl in TC levels, the result of an increasing temporal trends in TC is independent of radiation effects. We found that the TC level at a given age was considerably higher in those born later, which initially gave an impression of a cohort effect, that is, a difference in TC profile among subjects born in different years. This “cohort effect” must be interpreted with care since the three time-related factors affecting TC levels, that is, calendar time period, age, and birth year (which may possibly impart physiological variation among birth cohorts), are confounded. Their individual effects could not be isolated by our analysis, which took into account the last two factors, but that in relation to time period could not be examined simultaneously. The unmodeled period effect would show up as part of age and cohort effects. Thus, it is important to note that the differences seen in the shapes of TC growth curves among the various birth cohorts in our study largely resulted from time period-related forces acting at different ages on birth cohorts as they passed through time. It is not inconceivable that an inherent physiological difference exists among humans born in different decades that may affect TC levels for a lifetime (this might be called the “true” cohort effect). Although we’re not able to quantify the magnitude, we suspect that such cohort effects are quite small relative to the effects of time period and age. It is known that over the years the Japanese have adopted lifestyles more closely resembling those of westernized populations, for example, in diet and physical activities. Similar changes
432
in TC values over a decade were observed at any given age, which were larger in earlier period than in later period. These changes appeared to be almost similar in men and women. Systematic bias in laboratory procedures is not thought to be a factor since internal as well as external quality controls were conducted. These results suggest to us that time period factors related to behavior and life style in Japan had a large effect on the Japanese TC profile. Studies have shown that changes in lipoprotein cholesterol levels in individuals are also influenced by diet, tobacco smoking, alcohol consumption, vasectomy, and menopause [19,20,34-451. Unfortunately, these factors could not be evaluated in the context of growth-curve analysis due to lack of data for some of the study participants. Although detailed information on nutrition intake was not available for our study, it is generally known that from 1950 to 1975 Japanese fat intake (especially animal fat) had been increasing, and carbohydrate intake had been decreasing [23]. Saturated fatty acid (SFA) intake increases the levels of total and low-density lipoprotein (LDL) cholesterol, whereas substitution of polyunsaturated fatty acids (PUFA) for SFAs causes a decrease in total as well as LDL and high-density lipoprotein (HDL) cholesterol in humans [40]. So it may be reasonable to attribute the increased TC levels in the Japanese to an increased fat intake and the related increase in the percent of total calories derived from saturated fatty acids. Although it is not clear whether the changes affected all generations equally; it is interesting to note that TC levels were observed to increase at similar magnitude during the same period in all generations. The relationship between BMI and TC level was investigated by Gillum et al. who followed the cholesterol levels in the same individuals from young adulthood to middle age [20]. They showed that BMI was significantly related to TC levels. Burns et al. investigated TC and BMI of students for about ten years and showed that changes in BMI partly explained the changes in TC in men but not in women [19]. The amount of increase in BMI with age observed in our AHS cohort of comparable age groups was smaller or equal to that of these two studies, but the increase in TC was larger than in these studies. Nevertheless, the difference among the expected TC values based on fixing the BMI value to its mean in 1958 and those based on the observed mean BMI values showed that only a small part in the rise in TC levels could be explained by changes in BMI. The effects of other factors were not evaluated quantitatively in this study. TC levels of Hiroshima residents were found to be higher than those of Nagasaki residents, for both men and women of all birth cohorts. A nutritional mail survey conducted in the AHS from 1983 to 1985 indicated that fat intake was higher in Hiroshima subjects, suggesting dietary differences to be one possible cause for the intercity variation in TC levels. A Japanese national survey on nutrition showed that fat intake in urban areas was higher than in rural areas [23].
M. Yamada et al.
The difference between Hiroshima and Nagasaki is likely due to this urban-rural difference in dietary habits, since Hiroshima is more urban than Nagasaki. Such difference in TC levels, attributed to variation in dietary habits and life style, was also found in a study of urban and rural residents of Shanghai, China [46]. In westernized countries, the efficacy of diet-control projects in systematically reducing TC levels for the prevention of coronary heart disease has been demonstrated in epidemiological studies [10,47,48]. The Framingham Offspring Study showed that the age-adjusted secular trends of TC decreased from 1971 to 1975 and from 1979 to 1983, although changes in some lipoprotein profiles were not necessarily desirable, such as higher levels of very-low-density lipoprotein (VLDL) cholesterol and lower levels of HDL cholesterol in women [15]. In Japan, however, the secular trend of TC levels showed a remarkable increase between 1960 and 1980. TC levels of approximately 1000 men and women aged 40-60 years residing in rural northeastern Japan were measured during 1963-1966, 1972-1975, and 1980-1983 [5]. Between the time periods 1963-1966 and 1980-1983, significant upward shifts occurred in the mean and distribution of TC, in which the age-adjusted mean TC levels rose 22 mg/dl in men and 29 mg/dl in women [5]. A similar increase was observed in the current AHS cohort. Moreover, the increase was greater for those born more recently. Our results may not be generalized to the Japanese population as a whole because not all age groups are represented uniformly in our cohort. Furthermore, the participation rate in the AHS drops off after age 70 and birth cohorts from more recent decades are not included in the atomic-bomb survivor population. Bias due to out-migration from Hiroshima and Nagasaki, and refusal to participate in examinations can distort the true trends in TC values, but these are expected to occur randomly in any strata of TC level. The participation rate in the AHS cohort has ranged from 71% to 86% during the 28-year study period. Bias due to death caused by coronary heart disease and stroke will be considered in future studies. Elevated TC is clearly a risk factor for coronary heart disease [l-3] and also for myocardial infarction in the AHS population [49]. Despite the extraordinary rise in TC levels in the Japanese between 1960 and 1980, no significant increase was noted in the incidence of myocardial infarction in either the AHS [50] or the rural population in northeastem Japan [5]. This may be explained in part by the relatively low TC levels of the older Japanese who are more at risk for these diseases. Serum cholesterol was shown to be inversely associated with the incidence of cerebral hemorrhage [5,5 11. The substantial decline in cerebral hemorrhage incidence [5] and mortality [52] in the Japanese may be attributed to increased TC levels as shown in this study. More recently the younger cohorts have shown TC val-
Longitudinal
Trends
in Total
Serum
433
Cholesterol
those of their western counterparts. If the age-related trends in the TC growth curves, as shown in our study, continue for the younger subjects, we may expect a rise in the incidence of myocardial infarction. Appropriate guidance and surveillance are believed to be necessary at this time to help prevent such an occurrence. ues
approximating
This publication is based on research performed at the Radiation Research Foundation (RERF), Hiroshima and Nagasaki, Japan. is a private, nonprofit found&on funded equally by the Japanese try of Health and Welfare and the United Stares Department of through the National Academy of Sciences.
Effects RERF MinisEnergy
References 1. Kannel WB, Dawber TR, Kagan A, et al. Factors of risk in the development of coronary heart disease six-year follow-up experience. The Framingham Study. Ann Intern Med 1961; 55: 33-50. 2. Kozarevic D, Mcgee D, Vojvodic N, et al. Serum cholesterol and mortality. The Yugoslavia cardiovascular disease study. Am J Epidemiol 1981; 114: 21-28. 3. National Center for Health Statistics-National Heart, Lung and Blood Institute Collaborative Lipid Group: Trends in serum cholesterol levels among US adults aged 20 to 74 years. Data from the National Health and Nutrition Examination surveys, 1960 to 1980. JAMA 1987; 257: 937-42. 4. Iso H, Jacobs DR, Wentworth D, et al. Serum cholesterol levels and six-year mortality from stroke in 350,977 men screened for the multiple risk factor intervention trial. N Engl J Med 1989; 320: 904-910. 5. Shimamoto T, Komachi Y, Inada H, et al. Trends for coronary heart disease and stroke and their risk factors in Japan. Circulation 1989; 79: 503-515. 6. Experience of an international collaborative group: Circulating cholesterol level and risk of death from cancer in men aged 40 to 69 years. JAMA 1982; 248: 2853-2859. 7. Sherwin RW, Wentworth DN, Cutler JA, et al. Serum cholesterol levels and cancer mortality in 361,662 men screened for the multiple risk factor intervention trial. JAMA 1987; 257: 943-948. 8. Epstein FH. Low serum cholesterol, cancer and other noncardiovascular disorders. Atherosclerosis 1992; 94: l-2. 9. Linder FE, Woolsey TD, Sagen OK, et al. Serum cholesterol levels of adults: United States, 1960-62. National Center for Health Statistics. Public Health Service Publication No. 1000, Series 11, No. 22. U.S. Public Health Service, Washington, DC; 1967. 10. Fulwood R, Kalsbeck W, Rifkind B, et al. Total serum cholesterol levels of adults 20-74 years of age: United States, 197680. National Center for Health Statistics. Vital and Health Statistics, Ser. II, No. 236. DHHS Pub. No. (PHS) 861686. U.S. Public Health Service, Washington, DC; 1986. 11. Mann JI, Lewis B, Shepherd J, et al. Blood lipid concentrations and other cardiovascular risk factors: Distribution, prevalence, and detection in Britain. Br Med J 1988; 296: 1702-1706. 12. Williams DP, Going SB, Lohman TG, et al. Body fatness and risk for elevated blood pressure, total cholesterol, and serum lipoprotein ratios in children and adolescents. Am J Public Health 1992; 82: 358-363. 13. Sprafka JM, Norsted SW, Folsom AR, et al. Life-style factors do not explain racial differences in high-density lipoprotein cholesterol, The Minnesota Heart Survey. Epidemiology 1992; 3: 156-163.
DS, Cresanta JL, Srinivasan SR, et al. Longitudinal 14. Freedman serum lipoprotein changes in white males during adolescence: The Bogalusa Heart Study. Metabolism 1985; 34: 396-403. 15. Anderson KM, Wilson PW, Garrison RJ, et al. Longitudinal and secular trends in lipoprotein cholesterol measurements in a general population sample, The Framingham Offspring Study. Atherosclerosis 1987; 68: 59-66. 16. Newschaffer CJ, Bush TL, Hale WE, et al. Aging and total cholesterol levels: Cohort, period, and survivorship effects. Am J Epidemiol 1992; 136: 23-34. 17. Clark DA, Allen MF, Wilson FH, et al. Longitudinal study of serum lipids, 12-year report. Am J Clin Nutr 1967; 20: 743752. 18. Garry PJ, Hunt WC, Koehler KM, et al. Longitudinal study of dietary intakes and plasma lipids in healthy elderly men and women. Am J Clin Nutr 1992; 55: 682-688. 19. Berns MAM, Vries JHMD, Katan MB. Determinants of the increase of serum cholesterol with age: A longitudinal study. Int J Epidemiol 1988; 17: 789-796. 20. Gillum RF, Taylor HL, Brozek J, et al. Blood lipids in young men followed 32 years. J Chron Dis 1982; 35: 635-641. 21. Kato H, Tillotson J, Nichaman MZ, et al. Epidemiological studies of coronary heart disease and stroke in Japanese men living in Japan, Hawaii, and California. Serum lipids and diet. ABCC TR 28-72. 22. Health and Welfare Statistics Association, Kokumin Eisei no Doko, Showa 62, Tokyo, Health and Welfare Statistics Association; 1987: pp. 92-94 (in Japanese). 23. Nutrition Section, Ministry of Health and Welfare: National Nutrition Survey, 1991. Tokyo: Daiichi Shuppan Publishers; 1991 (in Japanese). JW, Beebe GW. ABCC-JNIH Adult Health 24. Hollingsworth Study: Provisional research plan. ABCC TR 9-60. 25. Wong L, Yamada M, Sasaki H, et al. Noncancer disease incidence in the atomic bomb survivors: 1958-1986. Radiat Res 1993; 135: 418-430. 26. RERF, participation rate in cycle 1- 15 of Adult Health Study. RERF Update. Hiroshima: Radiation Effects Research Foundation; 1992; 4: 10. 27. Kodama K, Mabuchi K, Shigematsu I. A long-term cohort study of the atomic-bomb survivors. J Epidemiol 1996; 6(Suppl.): 95-105. 28. Brown GW, Mood AM. On median tests for linear hypothesis. Neyman J, Ed. Proceedings of the Second Berkley Symposium on Mathematical Statistics and Probability. Berkeley and Los Angeles: The University of California Press; 1951: pp. 159-166. 29. Laird NM, Ware JH. Random-effects models for longitudinal data. Biometrics 1982; 38: 963-974. 30. Cook N. A FORTRAN program for random-effects models, Technical Report, Cambridge, MA: Harvard School of Public Health, Dept. of Biostatistics; 1982. 31. Wong FL, Kodama K, Sasaki H, Yamada M, Hamilton H. Longitudinal study of the association between ABO blood phenotype and total serum cholesterol level in a Japanese cohort. Genet Epi 1992; 9: 405-418. 32. Werner GT, Sareen DK. Serum cholesterol levels in the population of Punjab in north west India. Am J Clin Nutr 1978; 31: 1479-1483. 33. Conner WE, Cerqueira MT, Connor RW, Wallace RB, Malinow MR, Casdorph HR. The plasma lipids, lipoproteins, and diet of Tarahumara Indians of Mexico. Am J Clin Nutr 1978; 31: 1131-1142. 34. Jensen J, Nilas L, Christiansen C. Influence of menopause on serum lipids and lipoproteins. Maturitas 1990; 12: 321-331. 35. Matthews KA, Meilahn E, Kuller LH, et al. Menopause and
434
36.
37.
38.
39.
40.
41.
42.
43.
M.
risk factors for coronary heart disease. N Engl J Med 1989; 321: 641-646. Lindquist 0. Intraindividual changes of blood pressure, serum lipids, and body weight in relation to menstrual status: Results from a prospective population study of women in Gotebtirg, Sweden. Prev Med 1982; 11: 162-172. Campos H, Wilson PWF, Jimenez D, et al. Differences in apolipoproteins and low-density lipoprotein subfractions in postmenopausal women on and off estrogen therapy: Results from the Framingham Offspring Study. Metabolism 1990; 39: 1033-1038. Egeland GM, Kuller LH, Matthews KA, et al. Hormone replacement therapy and lipoprotein changes during early menopause. Obst Gynecol 1990; 76: 776-782. Wilson PWF, Garrison RJ, Abbott RD, et al. Factors associated with lipoprotein cholesterol levels. The Framingham Study. Arteriosclerosis 1983; 3: 273-281. Stulb SC, Mcdonough JR, Greenberg BG, et al. The relationship of nutrient intake and exercise to serum cholesterol levels in white males in Evans County, Georgia. Am J Clin Nutr 1965; 16: 238-242. Montoye HJ, Block WD, Metzner HL, et al. Habitual physical activity and serum lipids: Males, age 16-64 in a total community. J Chronic Dis 1976; 29: 697-709. Wood PD, Stefanick ML, Dreon DM, et al. Changes in plasma lipids and lipoproteins in overweight men during weight loss through dieting as compared with exercise. N Engl J Med 1988; 319: 1173-1179. Dwyer JH, Rieger-Ndakorerwa GE, Semmer NK, et al. Lowlevel cigarette smoking and longitudinal change in serum cho-
lesterol
among
adolescents.
The
Berlin-Bremen
Yamada
Study.
et al.
JAMA
1988; 259: 2857-2862. 44. Keys A, Anderson 45. 46.
JT, Grande F. Serum cholesterol response to changes in the diet. Metabolism 1965; 14: 747-787. Grundy SM, Denke MA. Dietary influences on serum lipids and lipoproteins. J Lipid Res 1990; 31: 1149-1172. Zhuang H, Han Q, Chen H. Study of serum lipids and lipoproteins of healthy subjects in Shanghai. Chin Med J 1986; 99:
657-659. 47. Burke GL,
Sprafka JM, Folsom AR, et al. Trends in serum levels from 1980 to 1987. The Minnesota Heart Survey. N Engl J Med 1991; 324: 941-946. 48. Hallfrisch J, Muller D, Drinkwater D, et al. Continuing diet trends in men: The Baltimore Longitudinal Study of Aging (1961-1987). Gerontology: Medical Sciences 1990; 45: 186191. 49. Szatrowski TP, Peterson Jr, AV, Shimizu Y, et al. Serum cholesterol, other risk factors, and cardiovascular disease in a Japanese cohort. J Chron Dis 1984; 37: 569-584. 50. Kodama K, Shimizu Y, Sawada H, et al. Incidence of stroke and coronary heart disease in the Adult Health Study sample, 1958-78. Technical Report 22-84, Radiation Effects Research Foundation, Hiroshima, Japan, 1984. 51. Ueshima H, Iida M, Shimamoto T, et al. Multivariate analysis of risk factors stroke. Eight-year follow up study of farming villages in Akita, Japan. Prev Med 1980; 9: 722-740. 52. Health and Welfare Statistics Association, Kokumin Eisei no Doko, 1994, Tokyo: Health and Welfare Statistics Association: 1994; p. 47 (in Japanese). cholesterol
0895.4356/97/$17.00 PII SO8954356(96)00423-4
1997
ELSEVIER
Longitudinal Trends in Total Serum Cholesterol Levels in a Japanese Cohort, 1958-1986 Michiko Yama&,‘~* F. Lennie Wang,’ Kazunori Kodama, * Hideo Sasaki,’ Katsutaro Shimaoka,’ and Michio Yamakido3 DEPARTMENTS 3T~~ SECOND
OF ‘CLINICAL DEPARTMENT
STUDIES AND ‘STATISTICS, RADIATION EFFECTS RESEARCH FOUNDATION, OF INTERNAL MEDICINE, HIROSHIMA UNIVERSITY SCHOLL OF MEDICINE,
HIROSHIMA HIROSHIMA
AND JAPAN
ABSTRACT. The 28-year follow-up of a Japanese cohort, having collected vast amounts of data collected on total serum cholesterol (TC), provided an exceptional opportunity to examine TC temporal trends. The longitudinal statistical method of growth-curve analysis was used to elucidate the age-related changes in TC levels and to characterize these trends in relation to sex, birth cohort, time period, place of residence, and body mass index (BMI). Japanese TC levels at initial examination were remarkably lower than those in western countries. During the study period from 1958 to 1986, TC levels increased dramatically with age in both sexes. The slope of the cholesterol growth curve was steeper for women than for men, with the difference growing larger after age 40 years. Drastic changes in Japanese behavior and lifestyle, especially westernization of the diet, are thought to have affected the TC values as time-period effects. As a result of this temporal change, which affected different cohorts at different ages, TC values were higher in members of the younger cohort. The increase of the TC values as time-period effects were larger in earlier period than in later period. These time-period effects appeared to be almost similar in men and women. The TC growth curves also varied by city of residence. Subjects in urban areas had higher TC values than subjects in rural areas. Changes associated with BMI from 1958 to 1986 were only partially responsible for the increased steepness of the TC growth curve. J CLIN EPIDEMIOL 50;4:425434, 1997. 0 1997 Elsevier Science Inc. KEY WORDS. Japanese cohort, growth-curve
analysis, longitudinal
INTRODUCTION Total serum cholesterol (TC) is a recognized major risk factor for coronary heart disease (CHD) [l-3], whereas an inverse relationship between TC and cerebral hemorrhage has been reported [4,5]. From a public health standpoint, the possible relationship between TC and cancer is of great interest [6-81. Epidemiological studies have been conducted to better characterize the secular trends of TC [3]. Many studies examined the relationships between TC and basic demographic factors such as sex, age, and obesity [9-201. Most of these studies were cross-sectional in design. The few studies that have examined longitudinal changes in TC were based on a small number of measurements per person at intervals of less than 15 years and/or were of limited age range [14-201. Moreover, such studies have been conducted only in western or North American populations in which the level of TC is high.
‘Address Studies,
for correspondence: Radiation Effects
Michiko Research
Yamada, Foundation,
Department of Clinical 5-2 Hijiyama Park,
Minami-ku, Hiroshima, Japan. Current address for Dr. Katsutaro Center, 1000 Carson St., Torrance,
Shimaoka CA 90509.
is Harbor-UCLA
Medical
studies, total serum cholesterol
The Adult Health Study (AHS) was begun in 1958 at the Atomic Bomb Casualty Commission to study the occurrence of illnesses and the changes in physiological and biochemical function resulting from exposure to atomic-bomb radiation among residents of Hiroshima and Nagasaki. In the AHS, blood cholesterol has been measured biennially since 1958 in about 10,000 persons including both irradiated and nonirradiated subjects ranging from adolescents to the elderly. This provided an exceptional opportunity to examine the TC temporal trends in the Japanese over a 30year period. Although TC levels in Japan in the 1960s were reported to be remarkably lower than TC levels in western countries [21], the rapid westernization of the Japanese lifestyle [22,23] has caused drastic change in the TC profile [5]. The present analysis was conducted to elucidate these changes, both quantitatively and qualitatively. TC growth curves were estimated using data from both the irradiated and nonirradiated subjects, taking into account any effects of radiation exposure. An appropriate model for the nonirradiated subjects results by substituting zero for radiation dose in the model. The findings are presented in the context of age, birth cohort, and time period, along with sex, obesity, and place of residence.
M. Yamada et al.
426
MATERIALS AND METHODS Subjects The subjects are members of the Atomic Bomb Casualty Commission-Radiation Effects Research Foundation (ABCC-RERF) AHS cohort [24]. The AHS cohort at its inception was composed of 19,961 subjects, about half of whom had been within 2,000 m of the hypocenter, a quarter beyond 3,000 m, and the remainder not in the cities (NIC) at the time of the bombings (ATB). Since 1958, subjects have been invited to participate in biennial health examinations conducted by the ABCC-RERF clinical staff. These health examinations consist of clinical evaluation and routine laboratory determinations that include TC and blood pressure, and various anthropometric measurements such as of height and weight. Because of high attrition and the fear that the sample would eventually become too small for the detection of radiation effects, 2436 subjects were added to the study group in 1977. Attrition due to death was onethird of the original total population by the end of 1986, and attrition due to migration out of the cities has ranged from 9% to 13% throughout follow up. The level of participation in the AHS has been high, ranging from 71 to 86% among residents of both cities since the biennial examinations started [25-271. This study excludes the 5000 persons who were not in either city at the time of the bombings, because the socioeconomic status of this group differed from the other participants. Also excluded were 3283 persons for whom Dosimetry System 1986 (DS86) estimates were not available, 3529 who did not participate in the examinations, 383 completely lacking TC data, and 246 lacking all data on either height or weight. Since the accuracy of the recorded height and weight data could not be verified directly by reviewing the medical charts, we compared each measurement of an individual against his/her mean height or weight after age 20. A height that deviated by more than 55 cm and a weight by more than 25% were recorded as missing because they were considered to have been input errors. One hundred twenty-nine people lacking either height or weight data from every attended exam were excluded. Subjects diagnosed as having certain medical conditions that might adversely affect the TC levels were also excluded as follows: 11 with nephrotic syndrome, 92 with diseases of the gallbladder and biliary ducts, 14 with disorders of the adrenal glands, 61 with acute hepatitis, and 16 with nutritional marasmus. These medical conditions were identified by means of the three-digit International Classification of Diseases (ICD) codes that were entered into a computer database whenever diagnoses were made during the AHS examinations. Familial hypercholesterolemic cases were not excluded because their ascertainment based on only the ICD codes was considered unreliable; such cases were expected to constitute only a small fraction of the data. The remaining 9633 people, consisting of 3437 males (Hiroshima, 2416; Nagasaki, 1021) and 6196 females (Hiroshima, 4602; Nagasaki, 1594), were the subjects of our analyses.
Laboratory
Determinations
Nonfasting TC levels were determined by the KendallAbel1 method from 1958 to 1965. Later, TC levels were measured using three types of automated methods using a Technicon Autoanalyzer: the N24 method from 1965 to 1966; the N37A method from 1966 to 1967; and the N24A method from 1967 to 1986. A Hitachi model 7050 autoanalyzer has been used since 1986. Because autoanalyzers were not available at the Nagasaki Laboratory, blood specimens collected there were frozen and sent to the Hiroshima Laboratory for analysis. To maintain accuracy and reproducibility in each laboratory, quality control was conducted internally as well as externally, the latter with the Centers for Disease Control (Atlanta, GA). Standardization was performed to ensure comparability of TC measurements obtained from various cholesterol-determination methods. From each of the five systems, 45 to 105 TC measurements were obtained for estimating the equations that relate measurements from the four earlier methods to those of the autoanalyzer 7050. The relationship assumed was ln(Y,,,,) = a + b X ln(YnonP7&, where Ynon-7050is a TC measurement from one of the four earlier methods, YTosOis that from the autoanalyzer 7050, and a and b are parameters to be estimated. When b = 1, the above equation simplifies to Yad, = c X Y,,,,.-iOsO[28]. The correction factors c for measurements from Kendall-Abell, N24, N37A, and N24A methods were estimated as 0.9513, 0.7991, 0.8842, and 0.9468, respectively. The adequacy of the conversion factors was assessed graphically by examining whether an overt departure from linearity was exhibited by pairs of points in the calibration data set. Visual inspection showed the conversions to be satisfactory. Statistical
Analysis
We used the growth-curve technique for the analysis of serially measured data, using the mixed-effects model of Laird and Ware [29]. Let ys denote the natural log (In) of the jth TC measurement for person i, where j ranges from 1 to 14, depending on the number of TC determinations. The following growth function was used to relate the level of the natural log (In) of TC, yl,, in an individual with the time-varying covariable, age, and body mass index [BMI = (weight/height*) X 1001. Y,, = PO, + A,(ase,) + PdBMI,,
+ /Mat&)
+ MBMI,,)
. age,) + PdBMI,,
. ag$
(1) + cd.
The growth curve coefficients p = [PO1 , . . . , Wi] are specific to each subject, describing the cholesterol profile for individuals. We let p be a linear function of the fixed effects covariates, which include city (coded 0;l for Hiroshima;Nagasaki), sex (0;l for male;female), and the actual year of birth minus 1945 (YOB”) for distinguishing the birth cohorts. Each individual’s growth curve parameters were assumed
Longitudinal
Trends
427
Total Serum Cholesterol
in
to be linearly related to the effects of city (C), YOB* (Z), in the following way:
sex (S),
Pot = a0 + a,(C), + ads), + a3(Z), + ab(C . 3, + a,(C
Z), + a&S
PIL= aa + adc), + alo( Z), + al&S
+ a2,(C
Z), + azz(S . Z), + a&C
a32
+
a33(c),
+ a3dC
+
a34(s)L
.S
Z), + boz
+ all(Z), + a12(C . 9,
+ al,(C
+ a19(C . Z), + a&S
p41=
Z), + a,(C
. Z), + aJC
.S
. S . 2$ + bz,
. Z), + a,,(C +
a3dz)i
Z), + b,,
. S . Z), + b3, +
. Z), + a3&S . 23, + adC
a36(c
. s),
S . Z), + bqi
P5,= a4o+ a41(C),+ a42(S),+ a4343(Z), + a4,(C . S), + a&C
. Z), + a&
23, + a&C
. S Z), + b,,
(2) The restricted maximum likelihood estimates of the parameters were estimated using the program REML [30]. After nonsignificant effects were removed as sets based on the multivariate chi-square tests of zero simultaneous effects of the covariates, followed by step-wise removal of individual nonsignificant terms, a parsimonious growth model of ln(TC) was obtained. This step-wise algorithm is described in detail elsewhere [3 I]. Although our goal is to present the temporal trends of TC apart from radiation exposure aspects, restricting the analysis to only the nonirradiated subjects excludes 57% of the available data. This reduces the accuracy of the nonradiation-related parameter estimates, which was circumvented by adjusting for the effects of radiation dose in the model estimated by using all eligible subjects. To evaluate the significance of radiation dose effects, radiation dose (DOSE) and its interactions with city, sex, and YOB* were added in the model as fixed effects. Analyses showed that the resultant growth curve of ln(TC) included terms involving radiation dose, but coefficients of DOSE not involving sex were all zero, indicating that radiation effects were exclusive to females. At this stage, the fit of the observed to predicted was not
satisfactory,
particularly
among
the
older
cohorts.
It was
suspected that the convolution of the post-war social modernization process that took place in Japan might have affected TC in a nonlinear manner among different birth cohorts. This hypothesis was assessed by adding (YOB*)* and SEX.(YOB*)* terms in the model, which were found to be significant.
RESULTS The distribution of the subjects, their mean age in 1958, and the median number of cholesterol determinations per person are shown in Table 1 for each city and sex by 10 birth-year intervals. Overall, the median number of TC measurements was 7 per subject, with a range of 1 to 14. TC measurements were available for ages 13 to 98 years in males (birth years 1875 to 1945) and for 13 to 96 years in females (birth years 1877 to 1945). As an initial descriptive step before embarking on a more formal longitudinal analysis, especially in the absence of efficient statistical programs for analyzing a large amount of serially measured data, one might examine the data as depicted in Fig. 1. There, the mean TC values of the AHS participants grouped by birth cohorts are plotted against mean age in each examination cycle. It is clear that a remarkable change took place in the levels of TC over the course of 28 years. However, Figure 1 does not accurately depict the longitudinal changes encountered in an individual’s TC profile over time, because the trends are based on averages from examinations in which the composition of the study population varied. In the present analysis, the population growth curve was estimated using information obtained from each individual’s growth curves. The parameters of the estimated population growth curve of ln(TC) are shown in Table 2. The first column contains the terms for individual growth function and the second row contains the fixed-effects covariates included in the estimated population model. The estimated function may be reconstructed by going across each row by column to read off the elements of a, substituting 1 for the intercepts. That is, ln(TC) = 3.63 + 0.49(SEX) O.O17(YOB*) + . . . -1.6 x 1O-5 (CITY. SEX. YOB* . age* BMI) + 6.3 X 10m5 (DOSE . SEX . age’ BMI). We present the temporal trends of TC apart from radiation effects by substituting zero for DOSE in the above equation to calculate the expected values of ln(TC). Also used were the mean BMIs and the corresponding mean attained ages calculated from the nonirradiated subjects grouped into 10 birth-year and 5 year-attained age intervals. The model corresponding to Table 2 is difficult to interpret because of the many interaction terms. It can be said, in general, that age-related variations in TC levels are modified by gender, place of residence, and year of birth. Moredetailed descriptions of the differences are better presented graphically.
Sex Difference Time Period,
and Combined Effects of Age, and Birth Year on the TC Cjmwth
Curves
Figure 2 shows the observed and predicted TC levels and the 95% confidence intervals (CI) for men and women of Hiroshima born in 1910, 1920, 1930, and 1940. A relatively good fit of the predicted growth curves to the observed data is indicated.
Female
Male
All subjects No. with DS86 = 0 Mean age in 1958 Median no. TC determinations All subjects No. with DS86 = 0 Mean age in 1958 Median no. TC determinations All subjects No. with DS86 = 0 Mean age in 1958 Median no. TC determinations
“DS86 = Dosimetry System 1986 estimated dose. Atomic-bomb survivors no radiation exposure. bMedian number of cholesterol determinations per subject during AHS
Total
Nagasaki
to the year
All subjects No. with DS86 = 0 Mean age in 1958 Median no. TC determinations
Item
according
Female
subjects
All subjects No. with DS86” = 0 Mean age (years) in 1958 Median no. TC determinationsb
of the
Male
Sex
1. Distribution
Hiroshima
City
TABLE
136 52 63.9 4 200 71 67.0 4 24 16 66.8 4 28 8 66.0 4 388 147 66.9 4
examination
cycles 1-14.
available cohorts
1875 763 48.6 8
227 94 48.5 9
1024 362 48.5 8 157 112 48.5 8
467 195 48.7 8
1905-1914
Birth
data
to this system. If a survivor
1538 637 57.9 6
157 71 58.1 6
128 81 57.7 6
794 304 57.3 6
58.0 6
181
459
1895-1904
of the
doses according
description
1885-1894
and
are assigned radiation
::.6 3
30
0 0 -
-
0 0
4 76.5 3 19 6 76.7 3
11
51884
of birth
557 332 28.7 8 2707 1269 28.8 8
346 221 28.2 6
1104 428 29.1 8
700 288 28.6 8
1925-1934
has a dose of 0, he or she is considered
1748 730 38.3 10
10
353 168 37.9
113 79 39.4 10
10
982 370 38.0
300 113 39.3 9
1915-1924
to have received
1347 563 17.6 5
272 126 18.3 6
253 118 17.9 5
479 193 17.4 5
343 126 17.1 5
1935-1945
Longitudinal
Trends
in Total
Serum
429
Cholesterol
Men 210,
I
1906-15
<1906
1916-25 ?i I-
I 30
130 --20
I 40
I 50 Age
I
I
I
60
70
80
90
(years)
Women
210
shima women born in 1920, 1930, and 1940 are 158.3 mg/ dl, 176.3 mg/dl, and 184.7 mg/dl, respectively. The expected TC level for 53 year old Hiroshima women born in 1910, 1920, and 1930 are 169.4 mg/dl, 188.7 mg/dl, and 195.7 mg/dl, respectively. Changes in TC values between two individuals of the same age but 10 years apart in time were very similar for any given age. Moreover, the increase in TC level with age was greater in earlier than in later periods. For example, the TC levels of Hiroshima women aged 33, 43, and 53 years in 1963 and 1973 increased by 19.5 mg/dl, 18.0 mg/dl, and 19.3 mg/dl, respectively, over the lo-year period. On the other hand, the increase between 1973 and 1983 for Hiroshima women aged 43, 53, and 63 years was only 8.4 mg/ dl, 7.0 mg/dl, and 10.0 mg/dl, respectively. Similar patterns of change were seen in men as well.
.".
cc
The Effects of Place of Residence on the TC Cjrowth
'---\
1916-25
l1301 20
I 30
I 40
I 50 Age
I 60 (years)
I
I
I
70
80
90
TC growth curves for women of Hiroshima Nagasaki born in 1910, 1920, 1930, and 1940 are shown in Fig. 3. TC levels were consistently higher for residents of Hiroshima than for residents of Nagasaki. Nagasaki residents experienced somewhat smaller age-related changes in TC. Similar city differences were also seen in men. Sex differences and combined effects of age, time period, and year of birth on the TC growth curves were similar between the two cities.
The Effects of Varying FIGURE 1. Means of total ages for men and women Study (AHS) examination
serum cholesterol plotted against of Hiroshima during Adult Health cycles 1-14 by birth cohort.
TC levels increased with age. TC levels for subjects aged 18 to 48 years in 1958 ranged from 124 mg/dl to 146 mg/ dl in men and from 123 mg/dl to 153 mg/dl in women. TC levels 28 years later in 1986 for the corresponding subjects ranged from 173 mg/dl to 184 mg/dl in men and 185 mg/ dl to 203 mg/dl in women. The greatest change occurred for women born in 1930, for whom the average increase over the 28-year period was 66.9 mg/dl (95% CI [64.7 mg/ dl, 69.1 mg/dl]). The smallest change occurred for men born in 1910, for whom the average increase during the same period was 27.6 mg/dl (95% CI [25.6 mg/dl, 29.6 mg/dl]). The rate of increase in TC values declined gradually with age. Values for men began to taper off at younger ages than for women. At under 40 years of age, TC levels were higher in men than in women, although sex difference was small. At over 40 years of age, however, TC levels were lower in men than in women, and the difference increased with further aging. TC levels at a given age were higher for those born later. For example, the expected TC level at age 43 years for Hiro-
Curves
BMI
on the TC Qrowth
Cumes
As shown in Table 3, BMI generally increased with age, up to about 60 years in men and to mid-60s in women, and decreased with further aging. The actual magnitude of the BMI effect on TC at a given age depended on sex, city of residence, and YOB*. Figure 4 shows the expected TC growth curves for Hiroshima women assuming that their BMI values to be fixed at their mean levels in 1958 throughout the follow-up period and observed BMI. For example, between 1958 and 1986, the mean BMI of Hiroshima women born during 1925-1934 increased from 0.2112 to 0.2297 (or 4.4. kg, at a constant height of 154 cm) and TC increased from 128.8 mg/dl to 195.7 mg/dl. The magnitude of change in TC that can be attributed to this increase in BMI was less than 10 mg/dl. Thus, an increase in TC level can be explained only in part by changes in BMI. DISCUSSION Past longitudinal studies of North American populations and those leading the so-called “western” lifestyles displayed a wide-ranging trends in cholesterol levels that are largely associated with the age structure and environmental conditions inherent to those populations [14-201. In industrialized countries, TC levels tended to increase with age
M. Yamada et al.
Men 2101 T
18
23
28
33
YOB = I-930
38
43
48
53
58
63
68
73
’ 38
’ 43
’ 48
’ 53
’ 58
’ 63
’ 68
I 73
Women
‘f 0zl:b
YOB = 1930
1101 18
’ 23
’ 28
’ 33
Age (years)
FIGURE 2. Predicted growth curves for total serum cholesterol and the 95% confidence intervals by attained age for the u&radiated subjects of Hiroshima, plotted for men and women whose year of birth (YOB) was 1940, 1930, 1920, or 1910.
T
190
-
9 E p .2 I B 5s
170-
150
-
$0 Tii 5 130c
110""""' 18
23
28
33
38
43
48
53
58
-----Nagasaki 1 ' 63 68
Age(years)
FIGURE 3. Predicted growth curves of total serum cholesterol by attained age for the u&radiated women of Hiroshima and Nagasaki, whose year of birth (YOB) was 1940, 1930, 1920, or 1910.
1 73
Longitudinal
TABLE
3. Means 1958-1986,
women,
of body mass index by birth cohort
by
Birth Sex
18 23 28 33 38 43 48 53 58 63 68 73 18 23 28 33 38 43 48 53 58 63 68 73
Female
0.1977 0.2056 0.2134 0.2190 0.2227 0.2255
for
men
1920
0.2061 0.2088 0.2140 0.2181 0.2218 0.2237
0.2098 0.2145 0.2204 0.2246 0.2268 0.2295
age
and
cohort
1930
1940
As
Male
431
Trends in Total Serum Cholesterol
1910
0.2080 0.2096 0.2130 0.2170 0.2180 0.2176
0.2112 0.2156 0.2213 0.2264 0.2292 0.2297
0.2190 0.2188 0.2191 0.2183 0.2178 0.2171
0.2188 0.2248 0.2306 0.2338 0.2346 0.2334
0.2268 0.2236 0.2243 0.2262 0.2272 0.2256
[15-201, except during adolescence [14]. Only a few longitudinal studies were conducted of populations in non-industrialized countries in which mean population levels of TC are low [32,33]; no increase or only a small increase in serum cholesterol with age was found. For most studies, followup times spanned less than 15 years and TC was measured in-
210,
s ?
190
2 T
170-
D d '0
150
s i m B t-
1 -
-
130-
110""""""
16
23
26
33
36
43
48
53
58
63
68
73
Age (years)
FIGURE 4. Growth
curves of total serum cholesterol for the unirradiated women of Hiroshima, whose year of birth (YOB) was 1940, 1930, 1920, or 1910, assuming the body mass index (BMI) to be fixed at the mean value in 1958 (dotted curves BMI = 0.210, 0.211, 0.219, and 0.227, respectively) and observed BMI (solid curves).
collected TC data frequently [15-191. Th us, our biennially spanning a 28-year period offered a unique opportunity for characterizing the longitudinal trends of TC in an Asian cohort of wide-ranging ages. Our results show that temporal changes in TC levels differed among men and women. Before 40 years of age, sex difference in TC levels was small, although TC levels tended to be higher in men than in women. However, after 40 years, not only did the female TC level rise dramatically, sex difference also increased significantly. Similar trends have been observed in past cross-sectional studies [9-111. The huge sex difference in TC levels occurring after age 40 can be explained in part by changes in lipid metabolism that occur exclusively in women. A decrease in estrogen causes an increase in the TC level [34]. Estrogen levels change rapidly during the perimenopausal period, with TC rising to a higher level in postmenopausal than in premenopausal women [34]. Cross-sectional and longitudinal studies have shown that higher TC levels in postmenopausal women result from changes in lipid metabolism following hormone deprivation [34-371. This association has been confirmed further by studies of estrogen-replacement therapy in postmenopausal women which lowered their TC levels [37,38]. Although women exposed to 1 Gy (Gray [Gy] is the international unit for absorbed dose. 1 Gy = 1 J/kg = 100 rad) of atomic bomb radiation in our data showed an average increase of 2 mg/dl in TC levels, the result of an increasing temporal trends in TC is independent of radiation effects. We found that the TC level at a given age was considerably higher in those born later, which initially gave an impression of a cohort effect, that is, a difference in TC profile among subjects born in different years. This “cohort effect” must be interpreted with care since the three time-related factors affecting TC levels, that is, calendar time period, age, and birth year (which may possibly impart physiological variation among birth cohorts), are confounded. Their individual effects could not be isolated by our analysis, which took into account the last two factors, but that in relation to time period could not be examined simultaneously. The unmodeled period effect would show up as part of age and cohort effects. Thus, it is important to note that the differences seen in the shapes of TC growth curves among the various birth cohorts in our study largely resulted from time period-related forces acting at different ages on birth cohorts as they passed through time. It is not inconceivable that an inherent physiological difference exists among humans born in different decades that may affect TC levels for a lifetime (this might be called the “true” cohort effect). Although we’re not able to quantify the magnitude, we suspect that such cohort effects are quite small relative to the effects of time period and age. It is known that over the years the Japanese have adopted lifestyles more closely resembling those of westernized populations, for example, in diet and physical activities. Similar changes
432
in TC values over a decade were observed at any given age, which were larger in earlier period than in later period. These changes appeared to be almost similar in men and women. Systematic bias in laboratory procedures is not thought to be a factor since internal as well as external quality controls were conducted. These results suggest to us that time period factors related to behavior and life style in Japan had a large effect on the Japanese TC profile. Studies have shown that changes in lipoprotein cholesterol levels in individuals are also influenced by diet, tobacco smoking, alcohol consumption, vasectomy, and menopause [19,20,34-451. Unfortunately, these factors could not be evaluated in the context of growth-curve analysis due to lack of data for some of the study participants. Although detailed information on nutrition intake was not available for our study, it is generally known that from 1950 to 1975 Japanese fat intake (especially animal fat) had been increasing, and carbohydrate intake had been decreasing [23]. Saturated fatty acid (SFA) intake increases the levels of total and low-density lipoprotein (LDL) cholesterol, whereas substitution of polyunsaturated fatty acids (PUFA) for SFAs causes a decrease in total as well as LDL and high-density lipoprotein (HDL) cholesterol in humans [40]. So it may be reasonable to attribute the increased TC levels in the Japanese to an increased fat intake and the related increase in the percent of total calories derived from saturated fatty acids. Although it is not clear whether the changes affected all generations equally; it is interesting to note that TC levels were observed to increase at similar magnitude during the same period in all generations. The relationship between BMI and TC level was investigated by Gillum et al. who followed the cholesterol levels in the same individuals from young adulthood to middle age [20]. They showed that BMI was significantly related to TC levels. Burns et al. investigated TC and BMI of students for about ten years and showed that changes in BMI partly explained the changes in TC in men but not in women [19]. The amount of increase in BMI with age observed in our AHS cohort of comparable age groups was smaller or equal to that of these two studies, but the increase in TC was larger than in these studies. Nevertheless, the difference among the expected TC values based on fixing the BMI value to its mean in 1958 and those based on the observed mean BMI values showed that only a small part in the rise in TC levels could be explained by changes in BMI. The effects of other factors were not evaluated quantitatively in this study. TC levels of Hiroshima residents were found to be higher than those of Nagasaki residents, for both men and women of all birth cohorts. A nutritional mail survey conducted in the AHS from 1983 to 1985 indicated that fat intake was higher in Hiroshima subjects, suggesting dietary differences to be one possible cause for the intercity variation in TC levels. A Japanese national survey on nutrition showed that fat intake in urban areas was higher than in rural areas [23].
M. Yamada et al.
The difference between Hiroshima and Nagasaki is likely due to this urban-rural difference in dietary habits, since Hiroshima is more urban than Nagasaki. Such difference in TC levels, attributed to variation in dietary habits and life style, was also found in a study of urban and rural residents of Shanghai, China [46]. In westernized countries, the efficacy of diet-control projects in systematically reducing TC levels for the prevention of coronary heart disease has been demonstrated in epidemiological studies [10,47,48]. The Framingham Offspring Study showed that the age-adjusted secular trends of TC decreased from 1971 to 1975 and from 1979 to 1983, although changes in some lipoprotein profiles were not necessarily desirable, such as higher levels of very-low-density lipoprotein (VLDL) cholesterol and lower levels of HDL cholesterol in women [15]. In Japan, however, the secular trend of TC levels showed a remarkable increase between 1960 and 1980. TC levels of approximately 1000 men and women aged 40-60 years residing in rural northeastern Japan were measured during 1963-1966, 1972-1975, and 1980-1983 [5]. Between the time periods 1963-1966 and 1980-1983, significant upward shifts occurred in the mean and distribution of TC, in which the age-adjusted mean TC levels rose 22 mg/dl in men and 29 mg/dl in women [5]. A similar increase was observed in the current AHS cohort. Moreover, the increase was greater for those born more recently. Our results may not be generalized to the Japanese population as a whole because not all age groups are represented uniformly in our cohort. Furthermore, the participation rate in the AHS drops off after age 70 and birth cohorts from more recent decades are not included in the atomic-bomb survivor population. Bias due to out-migration from Hiroshima and Nagasaki, and refusal to participate in examinations can distort the true trends in TC values, but these are expected to occur randomly in any strata of TC level. The participation rate in the AHS cohort has ranged from 71% to 86% during the 28-year study period. Bias due to death caused by coronary heart disease and stroke will be considered in future studies. Elevated TC is clearly a risk factor for coronary heart disease [l-3] and also for myocardial infarction in the AHS population [49]. Despite the extraordinary rise in TC levels in the Japanese between 1960 and 1980, no significant increase was noted in the incidence of myocardial infarction in either the AHS [50] or the rural population in northeastem Japan [5]. This may be explained in part by the relatively low TC levels of the older Japanese who are more at risk for these diseases. Serum cholesterol was shown to be inversely associated with the incidence of cerebral hemorrhage [5,5 11. The substantial decline in cerebral hemorrhage incidence [5] and mortality [52] in the Japanese may be attributed to increased TC levels as shown in this study. More recently the younger cohorts have shown TC val-
Longitudinal
Trends
in Total
Serum
433
Cholesterol
those of their western counterparts. If the age-related trends in the TC growth curves, as shown in our study, continue for the younger subjects, we may expect a rise in the incidence of myocardial infarction. Appropriate guidance and surveillance are believed to be necessary at this time to help prevent such an occurrence. ues
approximating
This publication is based on research performed at the Radiation Research Foundation (RERF), Hiroshima and Nagasaki, Japan. is a private, nonprofit found&on funded equally by the Japanese try of Health and Welfare and the United Stares Department of through the National Academy of Sciences.
Effects RERF MinisEnergy
References 1. Kannel WB, Dawber TR, Kagan A, et al. Factors of risk in the development of coronary heart disease six-year follow-up experience. The Framingham Study. Ann Intern Med 1961; 55: 33-50. 2. Kozarevic D, Mcgee D, Vojvodic N, et al. Serum cholesterol and mortality. The Yugoslavia cardiovascular disease study. Am J Epidemiol 1981; 114: 21-28. 3. National Center for Health Statistics-National Heart, Lung and Blood Institute Collaborative Lipid Group: Trends in serum cholesterol levels among US adults aged 20 to 74 years. Data from the National Health and Nutrition Examination surveys, 1960 to 1980. JAMA 1987; 257: 937-42. 4. Iso H, Jacobs DR, Wentworth D, et al. Serum cholesterol levels and six-year mortality from stroke in 350,977 men screened for the multiple risk factor intervention trial. N Engl J Med 1989; 320: 904-910. 5. Shimamoto T, Komachi Y, Inada H, et al. Trends for coronary heart disease and stroke and their risk factors in Japan. Circulation 1989; 79: 503-515. 6. Experience of an international collaborative group: Circulating cholesterol level and risk of death from cancer in men aged 40 to 69 years. JAMA 1982; 248: 2853-2859. 7. Sherwin RW, Wentworth DN, Cutler JA, et al. Serum cholesterol levels and cancer mortality in 361,662 men screened for the multiple risk factor intervention trial. JAMA 1987; 257: 943-948. 8. Epstein FH. Low serum cholesterol, cancer and other noncardiovascular disorders. Atherosclerosis 1992; 94: l-2. 9. Linder FE, Woolsey TD, Sagen OK, et al. Serum cholesterol levels of adults: United States, 1960-62. National Center for Health Statistics. Public Health Service Publication No. 1000, Series 11, No. 22. U.S. Public Health Service, Washington, DC; 1967. 10. Fulwood R, Kalsbeck W, Rifkind B, et al. Total serum cholesterol levels of adults 20-74 years of age: United States, 197680. National Center for Health Statistics. Vital and Health Statistics, Ser. II, No. 236. DHHS Pub. No. (PHS) 861686. U.S. Public Health Service, Washington, DC; 1986. 11. Mann JI, Lewis B, Shepherd J, et al. Blood lipid concentrations and other cardiovascular risk factors: Distribution, prevalence, and detection in Britain. Br Med J 1988; 296: 1702-1706. 12. Williams DP, Going SB, Lohman TG, et al. Body fatness and risk for elevated blood pressure, total cholesterol, and serum lipoprotein ratios in children and adolescents. Am J Public Health 1992; 82: 358-363. 13. Sprafka JM, Norsted SW, Folsom AR, et al. Life-style factors do not explain racial differences in high-density lipoprotein cholesterol, The Minnesota Heart Survey. Epidemiology 1992; 3: 156-163.
DS, Cresanta JL, Srinivasan SR, et al. Longitudinal 14. Freedman serum lipoprotein changes in white males during adolescence: The Bogalusa Heart Study. Metabolism 1985; 34: 396-403. 15. Anderson KM, Wilson PW, Garrison RJ, et al. Longitudinal and secular trends in lipoprotein cholesterol measurements in a general population sample, The Framingham Offspring Study. Atherosclerosis 1987; 68: 59-66. 16. Newschaffer CJ, Bush TL, Hale WE, et al. Aging and total cholesterol levels: Cohort, period, and survivorship effects. Am J Epidemiol 1992; 136: 23-34. 17. Clark DA, Allen MF, Wilson FH, et al. Longitudinal study of serum lipids, 12-year report. Am J Clin Nutr 1967; 20: 743752. 18. Garry PJ, Hunt WC, Koehler KM, et al. Longitudinal study of dietary intakes and plasma lipids in healthy elderly men and women. Am J Clin Nutr 1992; 55: 682-688. 19. Berns MAM, Vries JHMD, Katan MB. Determinants of the increase of serum cholesterol with age: A longitudinal study. Int J Epidemiol 1988; 17: 789-796. 20. Gillum RF, Taylor HL, Brozek J, et al. Blood lipids in young men followed 32 years. J Chron Dis 1982; 35: 635-641. 21. Kato H, Tillotson J, Nichaman MZ, et al. Epidemiological studies of coronary heart disease and stroke in Japanese men living in Japan, Hawaii, and California. Serum lipids and diet. ABCC TR 28-72. 22. Health and Welfare Statistics Association, Kokumin Eisei no Doko, Showa 62, Tokyo, Health and Welfare Statistics Association; 1987: pp. 92-94 (in Japanese). 23. Nutrition Section, Ministry of Health and Welfare: National Nutrition Survey, 1991. Tokyo: Daiichi Shuppan Publishers; 1991 (in Japanese). JW, Beebe GW. ABCC-JNIH Adult Health 24. Hollingsworth Study: Provisional research plan. ABCC TR 9-60. 25. Wong L, Yamada M, Sasaki H, et al. Noncancer disease incidence in the atomic bomb survivors: 1958-1986. Radiat Res 1993; 135: 418-430. 26. RERF, participation rate in cycle 1- 15 of Adult Health Study. RERF Update. Hiroshima: Radiation Effects Research Foundation; 1992; 4: 10. 27. Kodama K, Mabuchi K, Shigematsu I. A long-term cohort study of the atomic-bomb survivors. J Epidemiol 1996; 6(Suppl.): 95-105. 28. Brown GW, Mood AM. On median tests for linear hypothesis. Neyman J, Ed. Proceedings of the Second Berkley Symposium on Mathematical Statistics and Probability. Berkeley and Los Angeles: The University of California Press; 1951: pp. 159-166. 29. Laird NM, Ware JH. Random-effects models for longitudinal data. Biometrics 1982; 38: 963-974. 30. Cook N. A FORTRAN program for random-effects models, Technical Report, Cambridge, MA: Harvard School of Public Health, Dept. of Biostatistics; 1982. 31. Wong FL, Kodama K, Sasaki H, Yamada M, Hamilton H. Longitudinal study of the association between ABO blood phenotype and total serum cholesterol level in a Japanese cohort. Genet Epi 1992; 9: 405-418. 32. Werner GT, Sareen DK. Serum cholesterol levels in the population of Punjab in north west India. Am J Clin Nutr 1978; 31: 1479-1483. 33. Conner WE, Cerqueira MT, Connor RW, Wallace RB, Malinow MR, Casdorph HR. The plasma lipids, lipoproteins, and diet of Tarahumara Indians of Mexico. Am J Clin Nutr 1978; 31: 1131-1142. 34. Jensen J, Nilas L, Christiansen C. Influence of menopause on serum lipids and lipoproteins. Maturitas 1990; 12: 321-331. 35. Matthews KA, Meilahn E, Kuller LH, et al. Menopause and
434
36.
37.
38.
39.
40.
41.
42.
43.
M.
risk factors for coronary heart disease. N Engl J Med 1989; 321: 641-646. Lindquist 0. Intraindividual changes of blood pressure, serum lipids, and body weight in relation to menstrual status: Results from a prospective population study of women in Gotebtirg, Sweden. Prev Med 1982; 11: 162-172. Campos H, Wilson PWF, Jimenez D, et al. Differences in apolipoproteins and low-density lipoprotein subfractions in postmenopausal women on and off estrogen therapy: Results from the Framingham Offspring Study. Metabolism 1990; 39: 1033-1038. Egeland GM, Kuller LH, Matthews KA, et al. Hormone replacement therapy and lipoprotein changes during early menopause. Obst Gynecol 1990; 76: 776-782. Wilson PWF, Garrison RJ, Abbott RD, et al. Factors associated with lipoprotein cholesterol levels. The Framingham Study. Arteriosclerosis 1983; 3: 273-281. Stulb SC, Mcdonough JR, Greenberg BG, et al. The relationship of nutrient intake and exercise to serum cholesterol levels in white males in Evans County, Georgia. Am J Clin Nutr 1965; 16: 238-242. Montoye HJ, Block WD, Metzner HL, et al. Habitual physical activity and serum lipids: Males, age 16-64 in a total community. J Chronic Dis 1976; 29: 697-709. Wood PD, Stefanick ML, Dreon DM, et al. Changes in plasma lipids and lipoproteins in overweight men during weight loss through dieting as compared with exercise. N Engl J Med 1988; 319: 1173-1179. Dwyer JH, Rieger-Ndakorerwa GE, Semmer NK, et al. Lowlevel cigarette smoking and longitudinal change in serum cho-
lesterol
among
adolescents.
The
Berlin-Bremen
Yamada
Study.
et al.
JAMA
1988; 259: 2857-2862. 44. Keys A, Anderson 45. 46.
JT, Grande F. Serum cholesterol response to changes in the diet. Metabolism 1965; 14: 747-787. Grundy SM, Denke MA. Dietary influences on serum lipids and lipoproteins. J Lipid Res 1990; 31: 1149-1172. Zhuang H, Han Q, Chen H. Study of serum lipids and lipoproteins of healthy subjects in Shanghai. Chin Med J 1986; 99:
657-659. 47. Burke GL,
Sprafka JM, Folsom AR, et al. Trends in serum levels from 1980 to 1987. The Minnesota Heart Survey. N Engl J Med 1991; 324: 941-946. 48. Hallfrisch J, Muller D, Drinkwater D, et al. Continuing diet trends in men: The Baltimore Longitudinal Study of Aging (1961-1987). Gerontology: Medical Sciences 1990; 45: 186191. 49. Szatrowski TP, Peterson Jr, AV, Shimizu Y, et al. Serum cholesterol, other risk factors, and cardiovascular disease in a Japanese cohort. J Chron Dis 1984; 37: 569-584. 50. Kodama K, Shimizu Y, Sawada H, et al. Incidence of stroke and coronary heart disease in the Adult Health Study sample, 1958-78. Technical Report 22-84, Radiation Effects Research Foundation, Hiroshima, Japan, 1984. 51. Ueshima H, Iida M, Shimamoto T, et al. Multivariate analysis of risk factors stroke. Eight-year follow up study of farming villages in Akita, Japan. Prev Med 1980; 9: 722-740. 52. Health and Welfare Statistics Association, Kokumin Eisei no Doko, 1994, Tokyo: Health and Welfare Statistics Association: 1994; p. 47 (in Japanese). cholesterol