Serum cholesterol response to changes in the diet

Serum Cholesterol Response to Changes in the Diet III. Differences Among Individuals By ANCEL KEYS, JOSEPH T. ANDERSONAND FRANCISCO GRANDE Data from 227 men in 10 sets of controlled dietary experiments in 5 institutions gave the least-squares solution:

positions of the diets are known. Comparison of predicted with observed average values in recently published data on samples of free-living people changing diets on prescription designed to lower serum cholesterol gave, predicted versus observed a: -30.0 vs. -28.5 and -27.2 vs. -30.1 mg./lOO ml. in men and women in caloric balance. In a sample of men who were also losing weight on a low-fat, lowcholesterol diet a change of -33.5 was predicted versus - 39.8 observed.

AX/AX = 1.84 X/X - 0.84, with S.E. of slope = 10.44, where X is the serum cholesterol level of an individual, X is the average for all men on the same diet, and n is the response to a given dietary change. Equations, a chart and a table are provided for the prediction of the serum cholesterol response when

change is made from one diet to another

when cholesterol

and fatty

acid

com-

I

DIFFER in their serum cholesterol levels when they are all on an identical diet and they also differ in their responsiveness to specified diet changes. However, these differences do not necessarily prevent useful estimation of individual responses. In general, responsiveness to change in the diet tends to be related to the intrinsic characteristic of the individual as indicated by his average serum cholesterol level on a specific diet. For various dietary changes in our own experiments, as well as data from independent experiments reported in the literature, we investigated the relationship: NDIVIDUALS

(XB

(III, 1)

=a+b(S*)/iq)

-X,)/(x,-X*)

where X, and X, are the cholesterol

values in the serum of the individual on

diets A and B, respectively and X,% and X,, are the average values for “normal” men on the corresponding diets1 Least-squares solutions for equation (III, 1) were obtained with the findings from 5 of our own experiments and the only 3 sets of adequate data in the literature. In that material of 8 experiments, averaging 21 men per experiment, the weighted mean estimates for the coefficients were a = -0.91, b = 1.91. In the separate experiments the values of the slope, b, ranged from b = 1.59 + 0.28 in our experiment IX to b = 2.80 * 0.74 for the data of Hatch et al.,’ where the -+ value is the standard error of the slope. From

the

Laboratory

of Physiological

Hygiene,

Unioersity

Minnesota. Aided by research grants from the U. S. Public Health no. HE-04401 to F. G. and J. T. A.) Received for publication Feb. 19, 1965.

of Minnesota,

Service

Minneapolis,

(‘no. HE-04997

to A. K.,

766 METABOLISM,VOL.

14, No.7

(JULY),

1965

SERUMCHOLESTEROL RESPONSETO CHANGES IN DIET

767

NEW RESULTS This question of individual responsiveness is of both theoretical and practical importance so it seemed desirable to re-examine it with data from new dietary experiments. The new material was obtained in fully controlled metabolic units in the Hastings State Hospital and in the Faribault State School and Hospital, using the same methods as in previous studies reported from this laboratory. All subjects were physically healthy and were judged to be free of metabolic abnormality. The patients at Hastings were stabilized schizophrenic men; those at Faribault were mentally defective men. The new data from the Hastings State Hospital concerned 22 men who were maintained on diet CHD for 4 weeks, then for 3 weeks on each of diets LCE, HCE, LCF and HCF and again on Diet CHD for 4 weeks, blood samples being drawn from each man twice at the end of each dietary period. The mean values for cholesterol, mg./lOO ml. of serum, were 232.7 s.d. = 38.6, on diet CHD; 193.5, s.d. = 28.1, on diet LCE; 224.5, s.d. = 36.5, on diet HCE; 228.0, s.d. = 35.7 on diet LCF; and 256.7, s.d. = 49.4, on diet HCF. The least-squares solution to equation (III, 1) with these data gave a = -0.78 and b = 1.78, coefficients almost identical with the average of our own 5 experiments reported previously, and noted above, which also concerned schizophrenic men at the Hastings State Hospital. Perhaps more interesting is the study at the Faribault State School and Hospital because it concerned more men ( N = 36)) of a different type (mental defective but not psychotic), in a different institution. All men were fed diet B for 4 weeks, then diet S for 6 weeks and again diet B for 4 weeks, blood samples being taken from each man twice at the end of each period on diet B and twice in the third and again in the sixth week on diet S. The data are summarized in table 1. On diet B, these 36 men had an average of 266.4 mg. of cholesterol/I00 ml. and on diet S the average decrease was 43.5. Table 1 gives the averages for each man on diet B, those averages as percentage of the grand average on diet B (106 per cent = 266.4), and the averages of the individual decreases on diet S, expressed as percentage of the grand average decrease (i.e., per cent of 43.5), and the least-squares solution of equation (III, 1) gave a = -0.51, b = 1.51. The standard error of the slope, b, is 5 0.211 and the 95 per cent confidence limits are 1.093 and 1.922. Using these values for a and b, the regression equation (III, 1) properly describes the relationship between the intrinsic cholesterol characteristics of the subjects on diet B and the cholesterol responses to diet S and would be applicable, for example, to the problem of predicting the cholesterol response of other subjects whose serum cholesterol values on diet B are known and who are to be changed to diet S. But the true correlations between the intrinsic cholesterol characteristic and the response cannot be ascertained simply by calculating the coefficient of correlation between X, and (X, - X, ) ; a spurious correlation necessarily exists between X, and (X, - X,). Sp urious correlation can be avoided” by calculating the correlation X, - X, and X, + X,. The results of this latter computation with the data on the 36 men at Faribault is r = 0.71; i.e. there is

1 4 5 6 7 8 9 11 12 1 2 3 4 F; 7 8 9 10

116 108 88 70 130 121 98 124 84 92 79 91 96 111 116 116 100 75

Xl%

the tlifferencra

S,

X2) %

FBX FBY FBY FBY FBY FBY FBY FBY FBY FBY FBZ FBZ FBZ FBZ FBZ FBZ FBZ FBZ

11 2 3 4 5 6 7 9 10 11 2 4 5 7 9 10 11 12

Subject NO.

FB

“31 221

326 380 216 295 236 304 231 297 275 225 276 200 222 283 239 300

+ N,.

--S,)/(N,-~S,):

~ S., and thr s~nns S,

143 81 111 89 114 87 111 103 84 104 75 83 106 90 113 87 83

1””

Xl%,

(Xl

69 99 83 99 96 131 119 99 92 57 83 14” 71 124 106 55

(Xl

of Mean

X2) %

92 179

-

Percentage

-c X2) %>

111 83 83

125 139 82 111 89 115 90 110 106 83 104 76 83 103 91

lS~‘S.‘)~~=lnO~S1-C?;.‘)/~.\;I+S~~.

of thy average for all fawn irl tlw cyxrimcnt,

X1 mg. c;

nd., on t1ic.t S. and. c~~pwssed as percrntag:ra

115 106 90 73 127 118 100 124 85 91 78 90 97 108 117 112 105 75

+

Table 1 .-Experiment (Xl

of Mean

X2,%

122 129 62 37 161 147 64 126 74 110 90 96 78 142 96 152 41 76

-

in my./100

(Xl

s,~~(100s,)/s,:(s,-~.\;,~~~100i~,

diet S,.

values for sernm cholesterol

310 288 235 186 346 322 260 330 224 246 210 “32 255 296 ,308 308 267 201

X1 mg. $L

the, values on diet S,,

Indivithd

FBW FBW FBW FBW FBW FBW FBW FBW FBW FBN FBS FBS FBS FBS FBS FBS FBS FRS

Subject NO.

Percentage

E

1 m

: u

g

$ g g

7

SERUMCHOLESTEROL RESPONSETO CHANGES IN DIET

769

Table 2 Reference

Minnesota, Hastings, present Minnesota, Faribault, present Minnesota, Hastings, 5 old series1 Hatch et al.2 Best et al.4 Farquhar and Sokolow”

No. Men

b

22 36 116 24

1.78 1.51 1.75

kO.58 20.21 to.37

2.80 2.10 1.69

kO.74 to.22 i0.56

14 15

SEb

Summary of results from the analysis of the relationship between the individual relative cholesterol value and the response to dietary change as indicated by equation (III, I) :

CXB - XA)/(XB group average coefficient, b.

- xA)

= a + bXA/zA,

and the subscripts

refer

where

X refers

to diets. SE,

indeed a true correlation between the intrinsic the responsiveness to dietary change.

to the individual,

is the standard

cholesterol

error

.% is the

of the slope

characteristic

and

PRACTICALSIGNIFICANCEOF THE REGRESSION

It will be noted that we now have various solutions to equation (III, 1)) above, with slope values ranging from b = I.51 for the study at Faribault to b = 2.80 for the data of Hatch et al.” These are summarized in table 2. The value from the data of Hatch et al.’ is the most divergent but it also has a very large standard error. It would seem that the average of all estimates, weighted according to the number of men, would be the most reasonable choice to recommend for general application. This weighted average of the least-squares solutions is a = -0.84 and b = 1.84, the standard error of the slope, being SE., = to.44 for the combined data on 227 men in 10 sets of experiments. A value of r = 0.71, though highly significant, does not mean a high degree of accuracy in predicting individual responses; more than half of the variance remains unexplained when r = 0.71. On the other hand, allowance for this relationship will much improve the accuracy of predicting the averages of groups of men who differ in their intrinsic cholesterol characteristic. This is illustrated, for example, by classification of the 36 men in the Faribault study into the upper and lower halves of the distribution of their values on diet B. On diet B the average for the 18 men in the upper half, U, was 115.6 per cent of the average for all 36 men; the average for the other 18 men, L, was 84.4 per cent. With no knowledge of the relationship of response to intrinsic characteristic the predicted average changes for the 2 groups would be identical, i.e., 43.5 mg./IOO ml.; the actual averages were 53.6 for the U group and 33.5 for the L group; using equation (III, Id), the predictions are 56.0 for the U group and 31.0 for the L group. THE INTRINSICCHARACTERISTIC,7, ANDREFERENCE STANDARDS The intrinsic, or personal, serum cholesterol characteristic, indicated by the serum value of the person, compared with the average of reference men on the same diet, may be designated as x, calculated as x = X/x. It is necessary, then, to specify reference men and a reference diet as standards.

770

KEYS, ANDERSON

AND GRANDE

The obvious reference men are those from whose data were derived the average relationships between the diet and the serum cholesterol level presented in Parts I and II of the present series of communications. The average results for these men studied over the years in Minnesota provide the equation, for cholesterol in mg./lOO ml.: ’ Chol. = 164 + 1.35(2s

(III, 2)

- P) + 1.52.

where S and P are percentages of total calories provided by glycerides of saturated and polyunsaturated fatty acids, respectively, and Z is the square root of the dietary cholesterol, in mg./lOOO Cal. of diet. The reference relationship carries no necessary implication that it represents the most desirable characteristic or center of “normality;” it is simply a base from which calculations can be made. The expression, 1.35(2S - P) + 1.52, may be termed the Diet Factor and designated as $. The reference diet proposed is one with a value of + = 60, which corresponds with the average “house” diet in the hospitals at the time when the controlled experiments were conducted on the reference men in those hospitals. As equation (III, 2) shows, the reference men have an average of 224 mg. of cholesterol/IO0 ml. when + = 60. The expected cholesterol value, X A, of other men on that diet with $ = 60 may be estimated from a rearrangement of equation (III, 1 ), incorporating the values for the coefficients obtained from the least-squares analysis, i.e., a = -0.84 and b = 1.84: 0.84X, x,

(III, la)

1 ~

(ftB

184(FB

or, since x;

-

:y;A ) + x*xs

-

xA)

____+ FA

= 224 when 4 = 60 for diet A, 188x,

(III, lb)

X,

+ 224X,

-

42112

= 1.84x,

-

188

example, if, with + = 40, X, = 233, and, since xB = 204, substituting in (III, lb), g ives X, = 258. Further, for the man who has X, = 233 at + = 40, his 7rvalue is x = 258/224 =1.15. If the value of x is known for a person, the expected serum cholesterol value X, on any diet, Q, is easily predicted from another rearrangement of equation (III, 1)) noting that ?r = X*/224 or X,, = 224 7: For

(III, lc)

X,

= 224x + (1.84~

-

0.84)

(Kg

-

224),

using the value for X, from equation (III. 2). corresponding to the dietary factor, +, of diet Q. For easy estimation, fig. 1, and table 3, have been provided. Figure 1, allows prediction of the serum cholesterol value, X,, on a diet characterized by + = 60 when the serum cholesterol level is known for some other diet specified as to the value of 4. The value XB meaning the average for our reference

men, is

simply 4 + 164, and the chart gives the family of lines for various values of X,

771

SERUM CHOLESTEROL RESPONSE TO CHANGES IN DIET

Fig. l.-Chart

for estimating

serum cholesterol,

X,

on reference diet, + =

60,

on which reference men have 224 mg./lOO ml. Observed values, X13, are on the abscissa; values of reference men, xi3, on diet B, are on the diagonals; values predicted for XA are on the ordinate. for corresponding values of 4. In using the chart, the observed value Xg, noted on the abscissa, is projected vertically upward to the intersection with the diagonal line for the corresponding value of xB; from horizontal projection to the ordinate gives the value of X,. Figure 1, may also be used for the reverse calculation and it is desired to predict the cholesterol value for diet B. The value of X, on the ordinate is projected

that intersection

the

when X, is known the same man on horizontally to the

intersection with the diagonal for the reference men on diet B, i.e., for xE and from the point of intersection the vertical projection downward to the abscissa gives the prediction for X,. Table 3, may be used for the same purpose utilizing A = XJ224, the personal characteristic, to find the value, X0, on diet Q.* STUDIES ON FREE-LIVING

PEOPLE

Dietary intakes are readily measured if only a liquid-formula

diet is used, as

‘A nomogram, devised by Dr. E. S. Fetcher, facilitates these calculations; a copy of this will be supplied on request addressed to the Laboratory of Physiological Hygiene.

77.2

KEYS. ANDERSON

Table

AND GRANDE

3

%

0.7

0.X

0.9

1 .n

1.1

1 .:i

1.1

1 .s

150 160

124 128 133

132 1.39 14.5 151 158

141 14Q

150 160

159 171

1.58 166 174

170 180 190

182 194 206

i76 19” “07 “23

18s 202 220 237

lQ4 21:3 “32 252

182 190

200 210

21X “30

-738 0.53 369

2*55 “7” 289

“71 “QO 309

198 206

220 230

248 254

“H.5 301

:307 324

:32H 345

215 223

240 250

265 277

3 16 X3”

:3-I1 :3riH

367 :386

231

260

289

2.39 247

270 280

301 313

:347 :36:;

376 393

40.~ 424

,378

411

434

170 180 190 200 210

137 142 146 151 155 159 164

164 170 177 183 189

260 270

168 173 177

196 202 208

280

182

215

220 230 240 250

Serum cholesterol teristic and the in the left-hand

X-9, mg./lOO

cholesterol value column. Column

gives values of the dietary of the table.

factor

ml.,

on tlic>t Q predicted

from

r.

the

personal

for reference men, %
( = xQ - 164);

vdws

of S,

art* given

i’,

-14 -3 6 I6 26 :3fi 46 56 66 76 86 96

106 116 characQ given column

in the

botl)

in the system of Ahrens et al.,” but the use of more natural foods and menus requires great care and labor, including periodic chemical analyses of aliquots of diet composites, to assure accuracy, even in a locked and self-contained metabolic unit. It might seem, then, that little of quantitative value can be secured in studies on free-living people, with estimates based onlv on food “diaries,” shopping lists, interviews with dietitians, recollections of ‘what was eaten and about how big were the portions, and final computation from tables of “average food composition” that necessarilv ignore the large natural variations from place-to-place, time-to-time and silmple-t”-sample. Such crude methods cannot provide reliable details for individuals but when applied with care it is hoped that they may !richldreasonable approximations for group averages of cooperative people who have frequent, repeated contact with the supervising dietitians. It is instructive, therefore, to examine data on free-living people who were unusually motivated and whose dietary records were examined frequently and with care. Green et al.’ studied married medical students in a program in which special provisions were made to standardize food items and their sources by provision of a subsidized commissary and the dietary experiment was a part of the professional education and experience of the subjects. The results are summarized in table 4. From the estimates of fat in the diet it appears that, with substantially constant total calories (body weight was almost constant) in the experimental period, the subjects tended to decrease their total fat intake slightly and to change marked& the proportions of saturated and polyunsaturated fatty acids in the diet. The chanqe in cholesterol content in the diet was not stated but since no special emphasis was placed on this factor in the prescription, the estimates in table 4, in line (4)) should be

SERUMCHOLESTEROL RESPONSETO CHANGES IN DIET

773

reasonable; if the change in dietary cholesterol was actually twice as great, i.e. from Z” = 250 down to Z2 = 150 instead of being as indicated in line (4), the effect predicted would be only about 2 mg./lOO ml. of serum more than estimated here. Line (7) in table 4, gives the serum cholesterol value that would be predicted on the particular diet used if the subjects were intrinsically the same as our reference men, i.e., if they had the personal characteristic x = 1.00; the entry is $ + 164. But if these men should subsist on a diet with + = 60, calculation from equation (III, lb) or with figure 1 gives the prediction as entered in line (8). Note that the value in line (8) divided by 224 is the value of rr for these subjects. Thus the average for the 2.5 men is x = 216.3/224 = 0.966 and it is predicted that these men would be slightly less responsive to dietary change than our reference men. The 14 women have an average value of x = 206.0/224 = 0.920 and they would be expected to be still less responsive. Line (10) in table 4, gives final predictions calculated from equation (III, lc). For th e men, the prediction is a decrease of 30.0 mg./lOO ml. as compared with the observed value of 28.5; for the women the average predicted change is -27.2 versus -30.1 observed. The data in the study by Stamler et al.,” also summarized in table 4, presents greater problems. Many of these men were obese and there was an average gradual loss of weight on the experimental diet, averaging a total loss of 13.9 pounds at the end of the prolonged experimental period. But the outstanding peculiarity of these men was their marked hypercholesterolemia with an average of 271.5 mg./lOO ml. on the control diet. For our reference men on that diet, with + = 54.8, the expected average would be only 218.8. The average personal characteristic of these men is 7r = 278.8/224 = 1.245. The final predicted change of these men on the experimental diet is -33.5 versus an observed change of -39.8 mg. of cholesterol/100 ml. of serum. The small discrepancy could easily be the result of the weight loss. If account were not taken to the value of T for these men, the prediction would be that the change would be only -23.0. DISCUSSION The considerations and relationships developed in this paper are based on a large mass of experimental data and would seem to be applicable to many situations in which it is desirable to characterize people in regard to serum cholesterol level or to predict their response to dietary change. But it is necessary to point out that the predictions are for calorie balance; if the subject is on a low-calorie reducing diet the serum changes will be greater than predicted here. Further, individuals do not only tend to differ from each other in regard to serum cholesterol level on a given diet. While interindividual difference is estimated by the value of T, this does not take care of intraindividual differences. Under metabolic ward conditions on a rigidly constant diet the average intraindividual standard deviation is about 12 mg./lOO ml., and in free-living

L'6Z L'Z8

L’9 T’OT

6'0

Z’LCXTC’9-

9’CT

O'P8T

O’OL-

O’CZO’CZ-

C’Z-

8'961 8'16

8'66-

-

0’8% L’TCZ -

E’9S’BC-

-

C’LT O’CP 0’68

6'SI O'PSZ 8'S

8'8TZ 8'PS

S'TLZ S'TLZ 8'8LZ Z‘Z:6Z”ZC-

6’ZZ’LZT’OC-

T'TT 6'66 Z'SS

9’LT

(T’PT) (ooz)

C'P8T E'OZ

C'69T -

Z’ZLT

-

6'6s Z'9S

L’ST

(8'ST) (osz) Z'P

S’9TZ S’ZS

P'66T P'66T 0.9oz

Z'T-

L’SSP-

0'08

L’S‘1 COT L’SC

(T’PT) (ooz)

L’OZ

(L'T) (OS-_) C'TT

L'P8T

C'ZC-

S6LT T’T8T

S’ZC-

5'88-

O’OC-

ST

(8'ST) (osz) P'P 0'9T Z'OP Z"T8

O’LTZ O’ES

9'6OZ 9'602 C'QTZ

-

Z (s) 'I"3OOOTP" ‘zz (P) 'I"38‘6 (cl 'I"3%‘S (a) 'I"3%‘.!I (1) '%)I‘@!aMlCpOg

(OT - (6= v (TT) ~~uow!pa~dIw~ COT) PaJasqO (6) 509 = +Jo3 “a'! ‘vX UO!JXNJd (8) $00.1 = JL 3!uO!P?pa~d(L) 'IuIOOT/'iku“Ioy3 muas + ~op~?j~a!a(9)

SERUM CHOLESTEROL RESPONSE TO CHANGES IN DIET

775

men on an ostensibly constant diet this value is 20 or m0re.O Hence grossly discrepant predictions for individuals will be common unless many blood samples are analyzed to characterize the person on a given diet. Much better results can be expected when predictions are made for the averages of groups. Finally, it goes without saying that accurate values for the detaiIed composition of the diet are needed. Though tables of food composition are now available for cholesterol and the saturated and polyunsaturated fatty acids in foods lo these are only averages and there is much variation among samples of the various foods, including the percentage of total fat and the fatty acid composition

of the fat. REFERENCES Jr., Tsaltas,

1. Keys, A., Anderson, J. T., and Grande, F.: Serum cholesterol in man: diet fat and intrinsic responsiveness. Circulation 19:201, 1959. 2. Hatch, F. T., Abell, L. L., and Kendall, F. E.: Effect of restriction of dietary

1955. 3. Oldham,

P. D.:

Amer.

J.

Med.

19:48,

A note on the analysis

man. Lancet

5. Farquhar, J. W., and Sokolow, M.: Response of serum lipids and lipoproteins of men to beta-sitosterol and safflower oil. Circulation 6. Ahrens, E. H., Jr., Hirsch,

17:890, 1958. J., Insull, W.,

1:943.

of in

1957.

modified foods for serum cholesterol reduction. J. A. M. A. 183:5, 1963. 8. Stamler, J., Berkson, D. M., Young, Q. D., Lindberg, H. A., Hall, Y., Mojonnier, L., and Andelman, S. L.: Diet and serum lipids in atherosclerotic coronary heart disease: etiologic and

of repeated measurements of the same subjects. J. Chron. Dis. 15:969, 1962. 4. Best, M. M., Duncan, C. H., Van Loon, E. J., and Wathen, J. D.: The effect of sitosterol on serum lipids. Amer. J. Med. 19:61, 1955.

R., and

7. Green. J. G., Brown. H. B., Meredith, A. P., and Page, I. H.: Use of fat-

fat and cholesterol upon serum lipids and lipoproteins in patients with hypertension.

T. T., Blomstrand,

Peterson, M. L.: The influence dietary fats on serum-lipid levels

preventive considerations. Med. North America 47:3, 1963.

Clin.

9. Keys, A., Anderson, J. T., and Grande. F.: Prediction of serum-cholesterol responses of man to changes in fats in the diet. Lancet 2:959, 1957. 10. -,

and

Keys,

Stay Well, Doubleday,

M.

H.:

Eat

Well

and

2nd, rev. ed. New York, 1963, pp. 349-368.

Ancel Keys, Ph.D., Professor, School of Public Health, Director, Laboratory of Physiological Hygiene, University of Minnesota, Minneapolis, Minn. Joseph T. Anderson, Ph.D., Professor, School of Public Health, University of Minnesota, Minneapolis, Minn. Francisco Grande, M.D., Professor, School of Public Health, Un,iversity of Minnesota, Director, Jay Phillips Research Laboratory, Mount Sinai Hospital, Minneapolis, Minn.