V'

Agreement and Discrepancy in the Evaluation of Normal and Diabetic Oral Glucose Tolerance Test A-j. Valleron, Dr.Sc., E. Eschwège, M.D., L. Papoz, and G. E. Ro;selin, M.D., Paris, France

SUMMARY Systematic analysis with a five-hour OGTT of 340 subjects representative of people likely to be examined in a center specialized in diabetes detection was performed by multiple discriminant analysis, which provides indices of discrimination for different sets of blood glucose (BG) values. The relative sensitivity and the relative specificity of six different diagnostic methods: Fajans and r Conn, Wilkerson, WHO, British Diabetic Association, UGDP, and 1 European Study Group of Diabetes Epidemiology·were computed, i giving a quantitative determination for the degree of discrepancy in 1 the definition of diabetes: only 48 per cent of the subjects are cIassified in the same way by any of the diagnostic criteria. The time(s) of sampling and the index or indices of OGTT which are the most efficient in screening diabetes were estimated from homogeneous groups of subjects universally recognized as nondiabetic (URND) or as diabetic (URD) according to the different diagnostic methods. Better discriminating power (DP) between URD and URND compared with the maximum DP as measured by D2 of Mahalanobis from the seven BG values of the OGTT is given by the two·hour (70.2 per cent) than by the one-houl- (49.5 per cent) BG value when a single value is used; the one-two-hour BG value is the best set of two times (80.7 per cent). The different indices now in use for the classification of the OGTT have been found less effective than the weighted sum of one-two-hour BG values. The difficulty in obtaining highly specific diagnostic tests is discussed in relation to the consequences.on a partly automated screening in large poptllations. DIABETES 24:585-93, June, 1975.

! 1

1

The measurement of blood glucose after a glucose

1 load is the only generally accepted test to diagnose 1

ehemical diabetes in adults. 1- 2 However, from the frequency distributions of the blood glucose values ôbserved at different times during the oral glucose tolerance test (OGIT) it is impossible to deduce a eut off point between the normal and the diabetic subjects. As a consequence, sorne diagnostic methods that Use severaJ=3 or few 4 times of the OGTT have· been developed. In this study, the discrepaocy resulting

---------------------------------------------M ~rom the Institut National de la Santé et de la Recherche ed1cale Unité de Recherches Statistiques. I\ccepttJ for publication February 15, 1975.

I~ l

JUNE. 1975

from the appliCation of different criteria to the diagnosis of diabetes is quantitatively evaluated. The discriminating power of the OGTT is analyzed in order to locate the time(s) of sampling that are the most efficient in screening of diabetes. MATERIAL AND METHODS

Test and ana/ytica/ procedure. The subjects came for the first time to a center specialized in diabetes detection and did so either voluntarily or because they were suspected diabetics. Subjects with transient disease or inpatients were eliminated from the study. Most of them were examined in conditions very similar to those recommended by the World Health Organization 2 and the Committee on Statistics of the American Diabetes Association. 5 The 0-5-hr. OGTT was started between 8 and 8:30 a.m. after an overnight fast; the diet during the three days before the test included at least 200 ·gm. of carbohydrates. As glucose load is an important determinant of plasma glucose in normal Gand obese 7 people, most of the subjects were tested with a fixed dose of 75 gm. of glucose. Each subject ingested it in a volume of water not less than 150 ml. in less than five minutes. Subjects were resting and did not smoke. Samples were taken from the antecubital vein by an intravenous Braun catheter, thus avoiding repeated vc;:nopuncture. Samples were taken just before glucose ingestion (0 hr.) and exacdy l/z, 1, 2, 3, 4, and 5 hours later; blood glucose (BG) was measured on the same day by the glucose-oxidase method using Boehringer reactant (TC M2). Expmsion 01 the resll/ts. The different parameters analyzed were: the concentrations of blood glucose, the positiv~.area of blood glucose (BGa) levels circumscribed by the baseline (as extrapolated from the fasting values), the postglucose load curves, and either the intercept between both>these lines or the line of the five-hour abscissa, the peak value for glucose 585

.

EVALUATION OF NORMAL AND DlABETlC ORAL GLUCOSE TOLERANCE TEST

(BGp) (which is the maximum concentration less basal method are the following: multinormal distribution of level), the time for the peak value for glucose (BGpt), the variables, and equality of the variances and of the the recovery time required to return to the baseline correlation coefficients in the two groups; the use of logarithmic rather than arithmetic values fits these (BGrt) conventionally set ta five hours even if longer than five hours, and the sum of the blood glucose at assumptions better. different times (BGs). The indices proposed by KobRelative sensitivity-relative specificity: Specificity (Sp) berling et al. 3 and Danowski et al. 8 were also used. is the probability for a normal subject to be recognized For the purpose of multivariate statistical computa- as normal; sensitivity (Se) is the probabili~y for a pations, thirty-five subjects for whom data were missing tient with disease to be recognized as such. Each .of the were excluded. It was determined that there was no different criteria cannot be evaluated by an absolute definition, for there is no actual knowledge of who are bias in excluding these thictf-five subjects. The final sam pie after this selection comprised 340 subjects diabetics and who are not. So it is necessary to (204 men and 136 women). Six different diagnostic evaluate the different criteria by comparison with each methods were used ta classify the subjects as diabetics of the others. To do that, new quantities were defined: or nondiabetics: Fajans and Conn, H Wilkerson point relative specificity and relative sensitivity; Sp (AlB) is system, 1 0 World Health Organization (WHO) the relative specificity of the diagnosis method A with recommendations,2 British Diabetic Association reference to the diagnosis method B: it is the percen(BDA), I l University Group Diabetes Program tage of subjects classified as normal by method A in a (UGDP),12 and European Study Group of Diabetes population of subjects previously classified as normal by method B. Se (AlB) is the relative sensitivity of the Epidemiology (ESGDE).13 As these methods use different glucose loads and different methods of blood diagnosis method A with reference to the diagnosis method B; it is the percentage of subjects classified as glucose assay, ail the blood glucose values were transdiabetic by method A in a population of subjects preformed by using the equivalence described in the Appendix. The corpulence index 14 (CI) is defined by the viously classified as diabetic by method B. The two quantities Sp (AlB) and Se (AlB) are related to the relation: . CI = 100 (10 we.lg ht? (kg.) -1) blood glucose distribution in the population under g helght- (m.) study; the observed frequency distribution at faSting, The percentage of difference from the ideal weight one hour and two hours is indicated in table l'~ given by the table of the Metropolitan Life Insurance Company:15 -30 per cent, -10 per cent, 0 per cent, TABLE 1 10 per cent, and 30 per cent correspond to mean Distribution of blood glucose values corpulence indices of 21.5, 32.5, 37.0, 41.0, and at fasting, one hour, and two hours after 75-gm. oral glucose load 48.0, respectively. Multiple discriminant analysis: principle. Standard Values·of blood glucose levels (mg.! 100 mL) at: Fasting After oral glucose load statistical methods give the comparisons point by 2 hours 1 hour point between two groups (diabetic and nondiabetic Deciles subjects). These comparisons are not independent and 79 100 80 HJ% are difficult to interpret as a whole because the differ103 131 90 30% 122 em blood glucose values are linked as indicated by the 162 97 50% 148 197 107 70% correlations found in Results (table 6). The purpose of 258 263 136 90% multiple discriminant analysis is to replace these sepa48-600 56-568 68-400 Range rate point-by-point comparisons by a comparison between sets of values. Practically, this set-to-set comRESULTS parison is performed by using a weighted sum of the Discrepancy in the definition of diabetics. As indica~ed differenr blood glucose values. The weights assigned 16 in table 2, the frequency of diabetics in the populat/op to each blood glucose value are computed in order to is found ta be differenr by the various methods;'xhe take into account the correlations between the blood largest number of diabetics is observedwhen the BDA glucose values. The power of discrimination of a set of 2 c}lÎ:eria Il are used. The largest number of normals is BG values is measured by thë distance D o( observed with the Wilkerson criteria. 1 0 Forty-nine Mahalanobis, which is proportional to the squared per cent of subjects are c1assified neither as normal nor difference of the means of the weighted sum of these as diabetic by thèESGDE criteria. 13 variates; the statistical assumptions for using this 586

DIABETES, VOL.

! .

24, NO. 6

'A-J. VAllERON, DR, Sc., AND ASSOCIATES

TABLE 2

TABLE 4 Relativ~

Classification of the 340 subjects by six different methods of diabetes diagnosis B Diabetics %

Normals %

Suspects %

45

55 76 42 49 52 33

0 0 17 0 0 49

Fajans & Conn Wilkerson WHO' BOA' UGDP* ESGDE*

24 41 51 48 18

• For abbreviations see "expression of the resnlts."

Illtercomparison of diagnostic methods. We have compured the values Sp (AlB) and Se (AlB) for any couple A and B among six previously defined diagnostic methods. The results are shown in table 3 for relative specificities and in table 4 for relative sensitivities. If ail the methods gave the same results, these two tables would have aIl 100 per cent values. The discrepancies between the methods are measured by the deviation observed from this 100 per cent value. The Wilkerson point system classifies as nondiabetic'most of the subjects recognized as nondiabetic by the other methods (see line 2, table 3): 99.5 per cent of the normal subjects from Fajans and Conn, 100 per cent from WHO, UGDP, and ESGDE, and 99 per cent from BDA are normal for the Wilkerson point system. On the other hand, the Wilkerson point system (see line 2, table 4) recognizes as diabetics 54 per cent of the diabetics from Fajans and Conn, 5'( per cent of those from WHO,2 47 per cent of those from BDA, 51 per cent of those from- UGDP, and 88 per cent of those from ESGDE. Table 4 also indicates that Fajans and Conn generally declare as diabetics those so c1assified according to the other criteria. As seen in the vertical ESGDE, the subjects c1assifted as normal (table 3) or diabetic (table 4) are c1assified in the same way by three other methods (Fa jans and Conn, BDA, and UGDP) because their criteria of normality are low and , their diabetic criteria are high.

A Fajans & Conn Wilkerson WHO BDA

B

-

UGDp ESGDE

r

Fajans & Conn

Wilkerson WHO BDA UGDP ESGDE 99%

54% 80% 99% 92% 40%

95% 99% 100% 64%

88% 57% 89% 91% 43%

87% 47% 71% 87% 35%

86% 51% 77% 91%

100% 88% 100% 100% 100%

37%

Percentage of subjects cIassified as diabetic by methods listed in \ine A when these subjects are diabetic for methods Iisted in column B .

Definition of sltbjects "ltniversal/y recognized as nondiabetics" or "as diabetics." Because of the large discrepancies between diagnostic methods, two groups have been defined: subjects who are universally recognized as nondiabetic (URND) and subjects who are universally recognized as diabetic (URD). Among the 340 subjects of this study, 108 were c1assified URND and 53 URD. Only 48 per cent of the subjects are c1assified in the same way by any of these diagnostic criteria. Persons who are not recognized as URD or URND account for 179 subjects. They will be referred to 'as the conflicting diagnosis group (CD); they are partiy related to the group qualified as suspect by WHO and ESGDE. Among them, twenty-six are normal for five criteria and are c1assified as "suspect" either by ESGDE (n =22) or by WHO criteria (n =4). Thirty-two are diabetic for five criteria,. but are c1assified as normal by Wilkerson (n =7) or suspect by ESGDE (n=25). The characteristics of the three groups (URD, URND, CD) are shown in table 5; URD 'are, as a mean, significantly older and more obese than URND. This is a consequence of a general positive correlation between any blood glucose value from 0 to 4 hours and age or obesity (table 6, column total); however, when the well-defined URND and URD groups are considered, these correlations are no longer found (URND and URD columnsin table 6). Discrimination between URD and URND. The effec-

TABLE 3

TABLE 5

Relative Specificities: Sp (NB)

Characteristics of universally recognized as nondiabetic (URND), diabetic (URD), and conflicting diagnosis groups (CD)

Fajans & Conn

Wilkerson WHO BDA UGDP ESGDE 73%

99.5% 77 % 88 % 88 % 61 %

56% 65% 69% 44%

100% 100%

99% 99% 81%

94% 94% 92% 75% 67%

93% 100% 77% 87%

100% 100% 96% 98% 99%

64%

Percent age of subjects classified as normal by methods Iisted in Ine A when these subjects are normal for methods \isted in col umn B.

JUNE, 1975

A Fajans & Conn Wilkerson WHO BDA UGDP ESGDE

sensitivities Se(NB)

Group Total URND (n= 108) CD (n= 179) URD (n=53)

Age* (yr.) 42 :t 13.7 36 :t 12.6

Weight* (kg.) 73.7:t 16.4 70.0:t 13.3

Height* (cm.) 166:t 9 165:t 10

Obesity index* 41.3:t 8.5 39.8:t7.5

44~>:t

74.0:t 16.6

167:t 9

41

81.0±19

167± 8

45.2±8.9

13.2

50.5:t 11.7

:t8.7

*Mean ± S.D. 587

EVALUATION OF NORl-fAL AND DlABE1."IC ORAL GLUCOSE TOLERANCE TEST

TABLE 6 Correlations of blood glucose with age or obesity during the five-hour OGIT Age URND - 0.056

Total 0.266 xxx 0.319 xxx 0.375 xxx 0.320 xxx 0.232. xxx 0.172 xx 0.007

BGo Bm~

BG, BG. BG3 BG. BGs

- 0.231 x - 0.042

CD 0.199 xx 0.195 xx 0.159 x 0.187 x 0.078

- 0.027

- 0.041

0.038

- 0.114

- 0.005

0.073

0.085 0.217 x - 0.127 ~

URD - 0.087

Total 0.201 xxx 0.143 xx 0.171 xx 0.222 xxx 0.249 xxx 0.200 xxx 0.051

- 0.107 - 0.001 - 0.075

x, xx, xxx indicate significant correlation coefficients with p

= 0.05,

tiveness of discrimination between URD and URND as measured by D 2 of Mahalanobis (see Method) was calculated for different times and geometrical parameters and for different statistical indices. The discriminating power of each index was established as the ratio of its D 2 to the maximum D 2 obtained from the seven blood glucose values of the OGTT (D 2 max = 39.2). When a single value,is used (table 7), better discrimination is given by the two-hour (70.2 per cent) than by the one-hour blood glucose value (49.5 per cent). The discriminating power is indicated in table 7 for different OGTT inrices and table 8 for any set of two or three times of the OGTT. When two times are used, the best combination is the 1-2 hrs, closely foIlowed by the 0-2 hrs, which is more frequently used in screening diabetes. The corresponding weighted sum (WS) is WSl-2 = 2 log Gl + 3 log G2

P

= 0.01,

P

= 0.001,

Obesity URND 0.197 x 0.129

CD 0.015

URD 0.084

-0.116

0.057

0.074

-0.034

0.097

0.133

0.078

0.088

- 0.112

0.223

- 0.149

0.200 xx 0.082

0.243

- 0.039

-0.067

0.204

respectively.

for the 1-2 hrs and WSO-2 = 4 log BG2 - log BGo for the 0-2 hrs. We have computed the cumulative histograms for WSO-2 in the three groups URND, URD, and CD; the results are shown in figure 1. There is no overlap between URND and URD groups: 100 per cent of the URND subjectshave a WSO-2 :::;;; 6.2 indicated by the point a on the discriminating scale, while 100 per cent of the URD 'subjects have a WS larger than 6.6, as indicated by b. The CD group is largely scattered in these URND and URD regions: for instance,/25 per cent of the CD subjects have a WSo-2 :::;;; a and 25 per cent of the CD subjects have a WSO-2 > b; the remaining 50 per cent are in the undecided zone (a-b, figure 1) of the WSO-2. On the other hand,. if a subject has a WSO-2 larger than 7.2 (point c) he is surely diabetic because aIl the WSO-2 values of the CD group are

TABLE 7 Glucose tolerance test results for the universally recognized as nondiabetic (URND), as diabetic (URD), and conflicting-diagnosis (CD) groups. Group (N)

Geometrical parameters* mean ± S.D.

Blood glucose values (mg.liOO ml.) mean ± S.D. Oh

~h

1h

2h

3h

4h

5h

BGp

BGpt

BOrt

BGa

BGs

URND (108)

87.6 ±9.3

123.9 ±20.5

114.2 ±22.4

90.9 ± 15.3

76.8 ± 15.3

'74.7 ± 1l.7

79.5 ±7.7

42 ±17

39 ±14

133 ±60

2890 ± 1707

648 ±56

CD (179)

99.4 ± 13.1

157.2 ±28.5

176.6 ±36.4

135.3 ±31.0

95.0 ±30.1

79.8 ± 17.8

79.8 ± 12.4

85 ±28

61 . ±25

177 ±52

8001 ±3557

±~2

URD (53)

179.7 ±67.1

254.1 ±74.1

302.7 ±77.6

313.4 ± 100.4

246.9 ± 101.6

178.4 ± 102.6

153.0 ±89.0

145 ±40

103 ±39

246 ±54

21 154 ±7635

1628 ±'580

23.4

27.2

49.5

±70.2

38.7

6.2

18.8

18.1

7.5

20.5

47.8

Discrimi_ nating power

•.,!0,.4~ .

823

1

*For abbreviations: BOp, blood glucose peak; BOpt, time for blood glucose peak; BOrt, time of retum to the baseline; BOa, bJood glucose area; BOs, sum of blood glucose values. 588

DlABETES, VOL.

24, NO. 6

A-J. VALLERON, DR. Sc., AND ASSOCIATES

'l, 100

/

,

1

1

1

i

1

i

75

,

1

1 1 1 1 1

i

..... .......... -. __ (4)

100

1

i

- - - _ 100

--

1

1 1

i

1

1

1

i

25

--

, ,,

i i

50

5

,

/

1

i

150

1 1 1

.,,

i / / , ----/

~.

o

5

/

6

200 7

8

250

WEIGHTED SUM: 4IogBG2-logBGO

'" '~

Observed cumulative distributions of 0-2-hr weighted sums in URND, CD, and URD groups; a, b, c, are used to interpret lines drawn on figure 2.

than c. In order to use that quantity practically in the screening of diabetes we have drawn up nomograms shown on figure 2. For anygiven 2-hr and O-hr ~Iood glucose value, the two corresponding points on the scales are joined by a straight line. The intercept of that line with the discriminating scale gives the . classification of the test by reference co the URD, the l]RND, and the CD; four exam~es are shown in figure 2. The measurement of the discriminating power of each set of three times is given in table 8. When three TABLE 7 (continued) Glucose tolerance test results for the universally recognized as nondiabetic (URND) , as diabetic (URD), and conflicting-diagnosis (CD) groups. Group

OOTT Indicest mean ± S.D.

(N)

URND (108)

205.0 ±29.5

Danowski 1 416 ±46

CD (179)

312.0 ±53.2

568 ±75

506 ±71

616.0 ± 175.0

1,059 ±308

1,043 ±332

78.7

62.8

69.7

Kobberling

2 370 ±40

Kobberling 3 index is BO, + BO.· the Danowski 8 indices are BO, + BO. and BOo + + BO. + B03.

+ BO'n +

Bo,

400

8

Oh BG Values

2 h 0_2 h \1 1 Weighted Sum va ues

. BG

FIG. 2. Nomograms for computation of 0-2 hr. weighted sums. The intercept of the line through fasting and 2-hr. BG values with the weighted-sum line indicates to which groups (URND, URD, or CD) a subject may belong. a, b, c, are the limits shown in figure 1. Four examples are shown: line (1), subjed belonging to URD group; (:2) to URD or CD group; (3) to CD group; (4) to URND or CD group.

times are used, the best combination appears ta be the 1-2- 5 hr., with a discriminative power. of 91 per cent of the highest discrimination closely followed by the 0-1-2, which is frequently used in screening diabetes. The corresponding weighted sum is WSO-I-Z = 2 log BGo - 3 log BGl - 5 log BGz. The same analysis was carried out CO measure the effectiveness of each set of four, five, and six times_ The detailed results will not b'esh'6wn here, but the most striking facts are that the best four samples set are the 1-2-3-5 hrs (98.5 per cent of total discrimination) with corresponding weighted sums of WSI-Z':::3-:; = log BGl + 1.3 log BGz + 0.7 log BG3 - 0.85 log BG:;; the best [ive 589

!

,

EVALUATION OF NORMAL AND DIABETIC ORAL GLUCOSE TOLERANCE TEST

TABLE 8

Iikely ta be expected in centers that are partly specialized in diabetes detection. It is of interest to notice that the use of another methodology on another population in another European country gives similar Three Times Two Times results 3 for the comparison between the Fajans and Sel of Set of Sel of Conn fi and the Wilkerson methods. 1 0 The relative times (j1rs.) OP limes OP limes (hrs.) OP specificity and the relative sensitivity exhibit large o -lh - 1 51.2 V2-1-2 29,6 81.1 0- 'h V,-1-3 63.6 o -V2 - 2 75.1 51.0 differences whatever methods are compared. Thus, 0-1 o -'h - 3 46.5 'h-1-4 50.9 71.0 0-2 none of the methods now prevailing in diabetes diago -lh - 4 29,9 'h-1-5 39,3 51.8 0-3 nosis gives a classification between normals and diaV2-2-3 24,0 o -'h - 5 34.3 71.5 0-4 V2-2-4 ni 27.5 betics accepted by the other methods. Homogeneous 0-5 0- 1 - 2 85.4 'h-2-5 83.4 groups of subjects may be obtained only by the simul'h-3-4 0- 1 - 3 65.6 56.8 50,7 'h- 1 taneous use of the six methods for discriminating be71,3 0- 1 - 4 51.4 V2-3-5 63.5 v,- 2 V2-4-5 33.1 0- 1 - 5 53.7 v,- 3 45.3 tween diabetics and normals, and were qualified as 27,5 lh- 4 URD and URND. The exclusion of one of these 1 -2-3 80.9 0-2-3 72.6 v,- 5 27.8 methods does not modify any of the results demon1 -2-4 87.4 0- 2 - 4 74.0 91.4 1 -2-5 0-2-5 79.0 strated above. Furthermore, the inclusion of other 80,7 1- 2 75.9 1 -3-4 63.6 1- 3 criteria would not modify the classification because it 71,6 1 -3-5 0- 3 - 4 50.2 49.6 1- 4 would only give a smaUer number of people in the 1 -4-5 0- 3 - 5 55.3 51.1 49.6 1- 5 0- 4 - 5 29,5 URD and URND and would not bring new subjects 2 -3-4 79.9 70,7 2- 3 into these groups. Age- or obesity-adjusted limits 84.7 2 -3-5 2-4 73.9 were not used in the definition of URD and URND 78.7 2 -4-5 71,8 2- 5 46.8 3 -4-5 groups: such corrections are not indicated by the au44,5 3- 4 thors except by Fajans and Conn, fi who do not re3- 5 46.7 commend the use of their criteria for subjects over 50; moreover, the correlation between blood glucQse val4-5 11.6 ues, age, and obesity, which is found/in the The doser 10 100 per cenl is the discriminating power (OP) of each heterogeneous total group of subjects under study, is set of two or thrce blood gluco~e values, the better the corresponding no longer found in our URND and URD groups. discrimination by reference 10 Ihe discrimination obtained with the seven times of the five- hour OGTT. The obtaining of homogeneous normal and diabetic populations of this type is of interest in the definition sampI es are the Y'î-1-2-3-5 hrs. (99.3 per cent of total of the most simple procedure likely to diagnose dia-' discrimination); the best six samples are the betics in a large population. Without this definition O-Y'î-1-2-3-5 hrs. (99.9 per cent of total discriminathe effectiveness of the diagnosis would be dependent tion). on the initial method of classification, and on the absence of the relative specificity tbat already appears DISCUSSION when aU the values of one rriethod are determined and The methods employed were devised to diagnose wiU be even more amplified when a more simple prodiabetes by taking into account the blood glucose cedure related to the same method is used; the impotlevel during the fast and after glucose load. As these tance of these facts is substantiated by the theoretical methods defined diabetes without other indices such diagram shown in figure 3, where the joint variations as weight, age, or family history of diabetes, the inof specificity, sensitivity, and the percentage of diabettercomparison of the diagnostic methods was set up ics obtained after diagnosis are shown. The theoretiwith only the blciod glucose lev el of the population cal frequency of diabetes was taken as 5 per cent. sample. The relative specificity and the relative senWith a 99 per cent specific test, the so-caUed diabftic sitivity were chosen to quantify the comparison. AIsam pIe contains between 84 per cent and 71 per Cent though the results are dependent on the population of truly diabetic subjects when sensitivity ranges befrom which they have been computed, they can be "tween 100 per cent and 50 per cent. A decrease in extrapolated to other groups of subjects as far as these specificity downward to 95 per cent wiU reduce these percentages of true diabetics by 40 per cent. Inasmuch groups exhibit the same frequency distribution of the blood glucose level as in the population studied. This as the purpose orthe diagnosis is to determine those who are really diabetic, the choice of the most specific population is consistently representative of people Oiscriminating Power (OP)* when two or three times (logarithmic values) are used

590

DIABETES. VOL.

24,

NO.

6

A-J. VALLERON, DR. Sc., AND ASSOCIATES

% !3 of diab,tics .fter "".ning 100 Spocificity _ . _ . _ _ • • . 99

75

50

•_ _ .----------

--.----- . --' --' --,,- _ -.---'.---"-.

25

-,.

.95

..

.90

.

..

'.--.--

.80

5

.50

.60

.80

.90

Sensitivily

FIG. 3. Relations between percentage of diabetic subjects in the screened population, sensitivity and specificity of the diagnostic method used. A 5 per cent diabetes incidence was assumed for computations.

procedure has to be made. Thus, the effectiveness of different diagnostic procedures was studied in a population in which aIl subjects ;tre recognized as diabetic oc nondiabetic by al! the criteria. The sum of a pacticular set of glucose values as proposed by Danowski 8 and Kobberling,3 the blood .glucose area as proposed by Karp, 1 7 or another geometric index have been tound less effective than the index obtained from multiple discriminating analysis using only three times correctIy chosen (0-1-2 hr.). The best single blood glucose value for discriminating between URD and URND is the 2-hr. blood glucose value, and the best set of two times is the 0-2-hr. blood glucose value and the best set of two times is the 0-2-hr. test. This cesult does not reflect the initial criteria, because the same results are obtained even if the WH0 2 or the ESGDE,13 which give a special importance ta the 2-hr. blood glucose level, are excluded from the definition of the URD groups. Furthermore, the discriminating value of the 2-hr. blood glucose value is the same when the computations are done within age

ognized as such by the criteria of diabetes now in use There are to date no arguments other than genetic tc select future diabetics when looking at a single case with borderline or conflicting values of blood glucose However, these conflicting blood glucose value: within those of the conflicting-diagnosis group werc taken as risk factors in the epiderniological study 0 cardiovascular disease. 1 [1-24 Epidemiological studies 0: diabetes are fewer. 25-26 The isolation of groups of thù type provides an opportunity ta differentiare the rela· tionship of blood glucose level with insulin secretior and with the administration of drugs from that 0: groups not arbitrarily chosen but defined by the general acceptance of the different diabetes diagnosi~ criteria now ln use. ACKNOWLEDGMENT

This study was supported by the Institut Nationai de la Santé et de la Recherche Médicale and the Caissé Nationale de l'Assurance maladie des Travailleurs Salariés. We thank Mrs. Tholozan for technical assistance. REFERENCES lLabbe M., Labbe, H., and Nepveux, F.: Glycémie et hyperglycémie expérimentales chez I~s sujets normaux/C.R. Soc. Biol. 85:387-91, 1921. 2McDonald, G. W., Hoet, J. P;., and Butterfield, W.J.H.: Diaberes mellitus: Report of a WHO expert committee. WHO Rep. Series No. 310, 1965. 3Kobberling, J., and Creutzfeldt; W.: Comparison of different methods for the evaluation of the orjtl glucose colerance test. Diabeees 19:870-77, 1970. i 4Jackson,'W.P.U., Goldberg, M.D., Marine, N., and Vinik, A.!.: Effectiveness, reproducibili ty and weight relaeion of screening-teses for diabetes. Lancee 2: 110 1-05, 1968. 5Commicree on Seatistics of the American Diabetes Association: Seandardization of the oral glucose colerançe test. Diabetes

18:299-310, 1969. 6Casero, A., Scoee, J.P., Greeeie, D.P., Macfarlane, D., and Bailey, R.E.: Plasma insulin and glucose responses of healchy and weight classes. subjeces co varying glucose loads during ehree ho ur oral glucose The choice of methods which are stricter for the colerance eests. Diaberes 19:842-51, 1970. diagnosis of the disease conùibutes to the definition of 7Eschwège, E., Valleron, A.J., Papoz, L., and Rosselin, E.G.: a large group of people who are neither diabetic nor Influence de la charge orale de glucose sur les réponses normal. The glucose tblerance test in these people has glycémiques et hormonales au cours de l'hyperglycémie provoquée 8 per os sur 5 heures chez des sujees obèses non diabétiques. i!-nn. sometimes been called equivocal; this group is here Biol. Clin. 30:225-32, 1972. ! . termed as of conflicting diagnosis', the definition of a MD anows k'1, T .S. ,arons, A J. H ., H Yd OVlez, . ) . Q ., an d .:~;. W lnggroup as borderline by the WH0 2 and ESGDE 13 ere, ).P.: Utility of equivocal glucose colerance. Diabetes method gives criteria of diabetes that are more specific ",),}}:524-26, 1970. than the other methods. As demonstrated above, the "Fajans, S.S., and Conn, ).5.: Prediabetes, subclinical diabetes 1 and latent c1inical diabetes. Interpretaeion, diagnosis and treatg ucose values sit'uated in this range cannot be taken k ment. 1n 0 n Th e Nature an dT reatment 0 f D'la b etes, N ew Y or, for the screening of diabetes without including as Excerpta Medica Found. Int. Congr. Series No. 84, 1965, pp.

diabetics a majority of people who would be not recJUNE.

1975

641~56.

591

EVALUATION OF NORMAL AND DIABETIC ORAL GLUCOSE TOLERANCE TEST

l('Wilkerson, H.L.C: Diagnosis, oral glucose colerance tests. III Diabetes MeIIitus, Diagnosis and Treatment, New York, American Diaberes Association, 1964, pp. 31-34. IIFitzgerald, M.G., and Keen, H.: Diagnostic classification of diabetes. Br. Med. J. 1:1568, 1964. 12Klimt, C.R., Meinerr, c.L., Miller, CL., and Knowles,H.C: University Gcoup Diabetes Program (U.G.D.P.) A study of the relationship of therapy to vascular and other complications of diabetes. ln Tolbutamide Afcer Ten Years, Bcook Lodge Symposium, August 1967. New York, Excerpta Medica Found. Int. Congr. Series No. 149, 1967, pp. 261-69. 13European Association for the Study of Diabetes: A brief account of the European Diabetes Epidemiology Study Group Recommendations and Research. Dial:fetologia 6:453-54, 1970. 14Schwarrz, D., Lellouch, J., Anguera, G., Beaumont, J.L., Lenegre, J.: Tobacco and other -faccors in the aetiology of ischemic heart disease in man. Results of a retrospective survey. ]. Chrono Dis. 19:35-56, 1966. IODocumenta Geigy: Tables Scientifiques, sixth edit., Geigy J.R., A.G., Base!, 1963, p. 634. 16Anderson, LW.: On an introduction co multivariate statÏstical analysis. New York: John Wiley and Sons, Inc., 1958. l 'Karp, M., Brown, M., and Laron, Z.: A contribution co the imerpretation of the Oral Glucose Tolerance Test (OGTT). Diabecologia 8:381-84, 1972. 18 Andres, R.: Relation of physiologie changes in aging co medical changes of disease in the aged. Mayo Clin. Proc. 42 :674-84, 1967. I!lRandle, P.J., GarIand, P.B., Hales, CN., Newsholme, E.A.: The glucose farcy acid cycle. Lancet 1 :785-89, 1963. 2°Albrink, M.J., and Davidson P.C: Impaired glucose toi erance in patients with hypertriglyceridemia. J. Lab. Clin. Med.

67:573-84, 1966. 2JOstrander, L.D., Francis, T., Jr., Hayner, N.S., Kjeksberg, M.O., and Epstein, F. H.: The relationships of cardiovascular disease co hyperglycemia. Ann. lm. Med. 62: 1188-98, 1965. 22Epstein, F.H.: Hyperglycemia. A risk faccor in coronar,y heart disease. Circulation 36 :609- 19, 1967. 23Rosselin G., Claude, J.R., Eschwège, E., Parois, E., Warnet, ]. M., and Richard, J.L.: Diabetes survey-Plasma insulin during 0-2 h oral glucose rolerance test systematically carried out in a professional group. Diabecologia 7:34-45, 1971. 24Eschwège, E., Claude., J. R., Pacois, E., Richard, J.L., Rosselin, G., and Warnet, J.M.: Epidemiological study of glycemia during the oral glucose load test 0-2 h. 7th Int. Congr. Clin. Chem. Geneva/Evian 1969; Vol. 3, 47-57, Kayer, BaseI/MünchenlParis/New York, 1970. 25Marks, H.H., Krall, L.P., and White, P.: Epidemiology and detection of diabetes. Chap. 2 pp. 10-34. ln Joslin·s Diabetes Mellitus. A. Marble, P. White, R.F. Bradley, L.P. KraIl, Eds. Philadelphia, Lea & Febiger, 1971. 26McDonaJd, G. W.: The epidemiology of diabetes. ln Diaberes Mellitus, Theory and Practice 23 :582-93. M. Eilenberg, H. Rifkin, Eds. New York, McGraw-Hill Book Company, 1970. 27Danowski T.S., Khurana, R.C, Nolan, S., Stephan, T., Gegick, CG., Chae, S., and Vidalon, C: Insulin patterns in equivocal glucose colerance test (chemical diabetes) Diabetes 22:808-13, 1973. 2sEschwège, E., Warnet ].M., Pàcois, E., Rosselin, G.E., Claude, J. R.: Comparaison des glycémies mesurées par la méthode automatique réductimérrique et à la glucose oxydase chez 211 sujets à jeun et 2 heures après charge orale glucosée. Pathologie-Biologie 20:(3-4), 187-90, 1972.

APPIlNDIX

The different glucose criteria for diagnosis of diabetes that have been used are summarized in table A. In

the first three columns the table gives the name of the method, the characteristic~ of the glucose load, and 1"

TABLE A Glucose criteria for diagnosis of diabetes

,f Definition of

Procedure of OGIT Authors {Ref)

1 Fajans & Conn"

1. 75 gm./kg. IWt

II WiIkerson lo

100 gm.

Diabetic Subjects analysis BGI* > 160 and BGI1f.z> 140 and BG2~ 120 Ferricyanide (155) or Somogyi-Nelson Somogyi-Nelson BGo > 110 ~ 1 point; BGI> 170 ~ 1f.z pt (115) (165) 120 ~ 1f.z point; BGa> 110 -> 1 pt . (115) Sum of points > 2

III WH02

50 gm. or 100 gm.

Ferricyanide

IV BDN'

50gm.

Glucose-oxidase (BGo or BG'h or BG" or BG,'h > 160) and BG2 >110

others

V UGDpl2

30 gm.!m. 2 BS.t

others

VI ESGDEla

50 gm.

BGo + BG, + BG2 + BGa > 500 (475) Glucose-oxidase BG1f.z or BG,-,''i'f BG,'h > 220 and BG2 > 150

glucose load

BG2> 130 (135)

Normal Subjects others others

BG2> 110 (115)

Ferricyanide

BG1f.z or BG , or BG, 'h < 160 and BG2 < 120

*BGs indicate the blood glucose values observed at time; jf necessary, corrected values are shown·below, in parentheses. tlW = ideal weight; BS = body surface. 592

DlABETES. VOL. 24, NO. 6

A-J. VAllERON, DR. Sc., AND ASSOCIATES

the glucose as say method. The last two columns give the transformed limits used in the intercomparison. These transformed limits were computed in the following way: methods l, II, and V were restandardized to a unique 40-gm./m 2 glucose load following the indications of the Committee on Statistics. 5 The assay methods 1 to VI being different, we standardized the glucose limits as if al! the samples were tested by the

glucose-oxidase methods (GO) on whole venous blood glucose (VBG). Table B gives the equivalences between whole venous blood glucose values (VBG) and plasma glucose' values (BG) calculated ftom the indications of the Committee on Statistics 5 (table 2, left); conversion of ferricyanide plasma glucose values (FC) to glucose-oxidase values (GO) was computed fol!owing a method previously described. 28

TABLE B Conversio'h of whole venous blood glucose value 75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160 165 170 175

to plasma venous blood • glucose value 92 98 104 110 115 121 127 133 138 144 150 156 161 167 173 179 184 190 195 201 207

Conversion of ferricyanide plasma glucose value 90 100

to glucose oxidaset whole venous blood glucose

110 120

97 105

130 140

113 121

150 160

128 136

170

144

180 190

152 159

200 210

167 175

82 89

*Multiplying the whole venous blood values by 1. 15 and adding a constlmt of 6 mg. per 100 ml. converts the whole venous blood to plasma glucose. 5 ; tMultiplying the ferricyanide plasma glucose by 0.77 and adding a constant of 12 mg. ~er 100 ml. converts the ferricyanide plasma to glucose oxidase whole venous blood glucose. 28

i

JUNE, 1975

593

120

Diabetes and Coronary Heart Disease Epidemiological Study E. Eschwege, A.J. Valleron, and J.L. Richard

G.t.

Rosselin, J.A. Claude, J.M. Warnet

In a prospective survey to measùre the frequency of coronary heart disease in relation to difjerent clinical and biological parameters in a presumably healthy population, a simplified glucose tolerance test was carried out, together with a c1inical cardiovascular examination, obesity measurement and information obtained on the personal and family diabetes history. The simplified oGTT consisted of measuring, in each blood sample at fasting and two'hours after a 75 g oral glucose load, plasma glucose (pG), free fatty acids CFF A), triglycerides (TG), cholesterol (CT), insulin (lRI), cortisol (CORT) and human 'grdwth hormone CHGH). TIlis work was carried out in the "Groupe d'Etudes pour l'Epidemiologie de l'Atherosclerose" (GREA)*, and concerns 18~8 men in their fifties, employed in a large section of the Parisian Civil Service. Among the 1828 subjects examined, 88 diagnosed diabetics (57

',"