J CIIo Epidemiol Vol. 41, No. 7, pp. 645-649, 1988 Printed in Great Britain. All rights reserved

0895-4356/88 $3.00 + 0.00 Copyright 0 1988 Pergamon Press plc

METHODOLOGIC CONSIDERATIONS IN MEASURING GLYCOSYLATED HEMOGLOBIN

IN EPIDEMIOLOGIC STUDIES

SCOT E. MOSS,’ RONALD KLEIN,‘,* BARBARA E. K. KLEIN,’ TERRY L. SPENNETTA~ and EARL S. SIIRAGO’

Departments of ‘Ophthalmology and *Medicine, University of Wisconsin Medical School, Madison, WI 53792, U.S.A.

(Received in revised form 18 January 1988)

Abstract-Multiple linear regression was used to predict an incubated glycosylated hemoglobin value from the unincubated value and blood glucose. Hemoglobin A, (HbA,) was measured by a disposable microcolumn technique in a large, geographically defined population of diabetic persons in southern Wisconsin. During the study, incubation of blood samples to remove pre-A,, was implemented. Multiple linear regression using data from a group of 788 patients yielded the equation: incubated HM, = 0.897 (non-incubated HbA,) - 0.00332 (blood glucose) + 0.388. This equation was “validated” by substituting the calculated value of incubated HM, for the actual value in a multinomial logistic regression with diabetic retinopathy as the dependent variable. Little change in the model resulted from the substitution. Further validation was obtained from an independent sample of diabetic persons. Calculated values of incubated HbA, were an average of 0.4% lower than the actual values.

Glycosylated hemoglobin Methodology Diabetes mellitus Epidemiology

INTRODUCTION

Glycosylated hemoglobin has recently been used in epidemiologic studies of diabetic per- sons when examining the relationships of meta- bolic control with the presence and severity of the chronic complications of diabetes [l]. Its advantage over measurements of blood glucose is that it provides an,index of glycemia over a period of 2-3 months rather than only at the specific time the blood is drawn.

For many laboratory determinations, tech- niques of measurement change as more reliable, valid or cost effective methods become avail- able. This has occurred for glycosylated hemo-

*Reprint requests should be addressed to: Dr Ronald Klein, Deoartment of Gphthalmoloav, University of Wisconsin M&Cal School, -600 Highland Avenue, -Madison, WI 53792, U.S.A.

This work was supported by Grant EYO3083 (Dr R. Klein) from the National Eye Institute. Glycosylated hemoglobin determinations were performed in the Core Laboratory of the Clinical Nutrition Center with support from USPHS NIH Grant AM AG 26659 (Dr E. Shrago).

globin. When analytic methods change or there is other variation in processing procedures, it is necessary to quantitate the comparability of different methods. This is important for consist- ency within a study when the change has been made during the period of observation. We have had the opportunity to evaluate the compar- ability of measures of glycosylated hemoglobin in a large study of diabetic persons in southern Wisconsin conducted from 1980 to 1982 when there were changes in the laboratory techniques.

The specific purpose of this paper is to exam- ine the relationships among blood glucose, glycosylated hemoglobin from unincubated samples, and glycosylated hemoglobin from in- cubated samples.

METHODS AND MATERIALS

Participants were recruited as part of a large population-based study, the Wisconsin Epi- demiologic Study of Diabetic Retinopathy. The

646 SCOT E. MOSS et al.

study population consisted of a sample of di- abetic persons receiving care in an 11 county area in south central Wisconsin who were exam- ined in 1980-1982. Case identification and de- scriptions of the population appear in detail in previous reports [2-4].

Blood glucose was determined in a semi- quantitative fashion from a fingerprick capillary blood sample using a reagent strip for visual estimation by a glucose oxidase/peroxidase re- action (Chemstrip bG@). Blood glucose values are expressed in terms of milligrams per deci- liter. For determination of glycosylated hemo- globin, a 100-200 ~1 sample of capillary blood was collected in a heparinized capillary tube and after transfer to a 1.5 ml microfuge tube kept refrigerated at 4°C until analyzed. Glycosylated hemoglobin was determined by a disposable microcolumn technique following manu- facturer’s instructions (Isolab) [5].

The hemoglobin fractions measured by the assay include hemoglobin A,, (HbA,,) as well as the minor components HbA,,, and HbA,,. These are referred to collectively as hemoglobin A, (HbA,). In addition, any other “fast” hemo- globin variant, such as HbF which elutes with HbA,, is also included and affects the assay value. For convenience we will refer to the assay value as HbA, with the understanding that it may include minor hemoglobin variants. HbA, values are expressed as a percent of total hemo- globin.

Initially, the samples were not processed to remove pre-A,, . When the existence of this unstable precursor to HbA,, became known during the course of the study, the samples were incubated in 1 ml of normal saline at 37°C for 4 hr to remove the pre-A,,. Both incubated and nonincubated samples were assayed through the remainder of the study. The term HbA, will be further modified by the adjectives “non- incubated” and “incubated” to indicate the presence or absence, respectively, of pre-A,,.

RESULTS

In order to determine the precision of the assay technique, split duplicate blood specimens were obtained and assayed. The -intra-assay coefficient of variation is 2.9% (N = 63) and 2.4% (N = 74) with non-incubated and incu- bated samples, respectively.

The effect of incubating samples before anal- ysis to remove the pre-A,, was determined by analysis of data from 788 participants for whom

blood glucose and both incubated and non- incubated HbA, were obtained. The correlation coefficient between incubated and non- incubated HbA, values is 0.95. The correlations between blood glucose and incubated and non- incubated HbA, values are 0.55 and 0.67, re- spectively. All correlations are highly significant statistically. Blood glucose is more highly cor- related with the non-incubated than the incu- bated HbA, which is consistent with previous reports that the non-incubated HbA,, which includes the pre-Ale fraction, is more dependent on short term variation in blood glucose P, 671.

A stepwise, multiple linear regression model was developed to further investigate the re- lationships among these variables. To predict the value of incubated HbA, based on the non-incubated value and the blood glucose one would use the resulting equation: incubated HbA, = 0.897 (non-incubated HbA,) - 0.00332 (blood glucose) + 0.388. Non-incubated HbA, is more important than blood glucose in ex- plaining the variability of the incubated HbA, . It explains 91.1% of the variation in incubated HbA,. Blood glucose explains an additional 1.4% which, although small, is still highly significant statistically. Thus, total R* is 92.5%.

The utility of the calculated value of the incubated HbA, as a substitute for the actual value was evaluated in a multinomial logistic regression with diabetic retinopathy as the de- pendent variable. This model is used to explore the association of several variables with a cate- gorical variable. Four categories of diabetic retinopathy were used: proliferative, severe non- proliferative, mild non-proliferative, and no retinopathy. For this analysis the study popu- lation was divided into two groups: younger onset (diagnosis of diabetes prior to 30 years of age (N = 396)) and older onset (diagnosis of diabetes at 30 years of age or older (N = 392)). The results of the analysis for the younger onset group are presented in Table 1. Three vectors of coefficients, denoted /I,, /I2 and /I3 are necessary to explain four categories of retinopathy. Only the statistically significant variables (p < 0.05) are included. Model 1 contains the actual incu- bated HbA, value and model 2 contains the computed value. Most coefficients change little, if at all, when the s,ubstitution is made. Entropy in a logistic regression is analogous to the R* in a multiple linear regression, i.e. it is the percent of the variation in the dependent variable which is explained by the independent variables. Thus,

Methodologic Considerations in Measuring Glycosylated Hemoglobin 647

Table I. Results of logistic regressions on diabetic retinopathy for younger onset diabetic persons (N = 396) Wisconsin, 1980-1982

Model 1 coefficients Model 2 coefficients

6, 6, 6, I?, L?, 6,

Constant - 12.0 - 9.8 - 3.6 - 12.0 - 10.0 - 3.6 Incubated HbA: (%) 0.23 0.30 0.20 0.24 0.33 0.20 Age at diagnosis (yr) 0.059 0.099 0.066 0.060 0.10 0.067 Duration of diabetes (yr) 0.51 0.42 0.36 0.51 0.42 0.36 Diastolic blood pressure 0.026 0.0016 - 0.034 0.025 0.00092 - 0.035

(mmHg) Urine protein (0 = no, 1 = yes) 1.7 0.090 - 0.69 1.8 0.14 - 0.65 Entropy 0.380 0.381

*Actual value in model 1, calculated value in model 2. See text for explanation of entropy.

38.0% of the variation in retinopathy is ex- plained by the independent variables. This was also little changed by the substitution of the calculated for the actual incubated HbA, . Sim- ilar results were obtained for the older onset group.

The optimum method for validating the cal- culated value of incubated HbA, involves use of an independent sample. We used 51 diabetic patients participating in another ongoing study for whom we had values for both incubated and non-incubated HbA, and a blood glucose (B.E.K. Klein, U. of Wisconsin Medical School, Madison, Wise. personal communication, 1986). The regression equation presented above was used to calculate an incubated HbA, for comparison with the actual value. The calcu- lated value underestimates the actual value in this population by 0.4%. A paired t-test shows that the calculated value is significantly lower, (p < 0.001) than the actual value.

DISCUSSION

Laboratory determinations are commonly used in epidemiologic studies of chronic disease. When the factor being studied is only recently under investigation the measurement may change as a result of the development of more reliable or more efficient laboratory procedures. In addition, laboratory error may result in the application of a different technique. Although it is reasonable to wait until there is consistency of laboratory technique and estimates of its vari- ability, it is neither reasonable, nor feasible, to wait until a technique is so standard that no further refinement of it is likely to occur. We, therefore, must proceed with the research effort, and with the occurrence of a change in technical procedure, assess the comparability of the old and new results and hope to be able to make

appropriate adjustments or corrections. This is a problem which has occurred in the past, for example, in the measurement of serum choles- terol. Moore and Gordon [8] describe difficulties in measuring serum cholesterol early in the Health Examination Survey. In that study, due to a change in laboratory technique, serum cholesterol values determined by the ferric-chloride method were converted to the Abell-Kendall method by a procedure similar to that of the present study. In this paper we have explored this problem with respect to measurement of glycosylated hemoglobin, a problem which is not unique to our study [9, lo]. Although our example is specific to our study situation, the problems have general applica- tion.

HbA,,, and to a lesser extent HbA,, is an indicator for long-term blood glucose control in diabetes [l]. HbA,, is formed continuously and irreversibly within erythrocytes with the rate dependent on the long-term blood glucose con- centration. However, the process of formation of HbA,, is a two step process in which hemo- globin A and glucose combine to form a precur- sor to HbA,,, pre-A],, which may then be converted to HbA,, . The first reaction is a rapid and reversible one in which the equilibrium state is affected by short-term changes in blood glu- cose. The second reaction is slow and irre- versible and little affected by short-term blood glucose change. As column chromatographic assay techniques cannot distinguish HbA,, from pre-A,,, HbA,, and HbA, determinations may be affected by short-term blood glucose fluctuations unless pre-A,, is removed prior to assay. This may be accomplished by dialysis of hemolysates against a glucose-free medium or by incubation of erythrocytes in saline.

More specific assays of glycosylated hemo- globin now exist such as isoelectric focusing.

648 SCOT E. Moss et al.

However, disposable microcolumns remain in use by a substantial number of laboratories [I 11. We feel that this method is a useful technique for measuring glycemia in a large population study such as ours. In addition, in our hands it is reproducible, as evidenced by intra-assay coefficients of variation of 2.9 and 2.4% for non-incubated and incubated samples, re- spectively.

In the present study, participants began to be examined when the existence of pre-A,, was only suspected. Thus, we performed what was con- sidered to be an established assay technique at that time without incubation or dialysis of samples before analysis. Approximately two- thirds of the way through the study, the exis- tence of pre-A,, and methods to remove it became known. The question in this situation is whether to continue with the old technique for consistency and comparability or to switch midstream to the new “state of the art” tech- nique and hope to be able to make a compara- bility correction to the old values. While there is no single answer to this question, it depends in part on the use of the results, costs, avail- ability of resources, and the ability to make a comparability correction. If the results are to be put into a small number of categories, such as above and below the median or good, fair, and poor control, then a switch to a new system can probably be accommodated by categorizing the old and new sets of values separately. If, how- ever, the results are to be used as a continuous variable, a switch to a new system is probably not advisable unless some adjustment can be made to make the two systems comparable. One approach is to calculate Z-scores for both meth- ods by subtracting the mean and dividing by the standard deviation, and then using the Z-scores as a continuous variable. The approach we took was to use a linear regression equation which also uses information contained in the blood glucose. We were able in the latter part of the study to obtain both incubated and non- incubated values of HbA, plus the blood glu- cose which is important in making a compara- bility correction. Thus, we had the option of using the old system throughout or correcting the old system to correspond to the new system. This we have done in both ways [3,4, 121. In 1443 patients we had a non-incubated HbA, and a blood glucose value. Using the regression equation developed above, we calculated an estimated, incubated HbA, value. There has been one other report of a linear regression

involving incubated and non-incubated HbA, . Jury et al. [13] developed a linear regression of pre-dialysis HbA, on post-dialysis HbA, in 21 non-diabetic and 35 diabetic samples. Their equation is: post-dialysis HbA, = 0.88 (pre- dialysis HbA,) + 0.335 with an R2 of 0.930. The data of the present study, leaving out the term for blood glucose, give a corresponding equa- tion of: incubated HbA, = 0.806 (non-incubated HbA,) + 0.724 with an R2 of 0.911. This com- pares favorably with the results of Jury et h1.[13]. -

Thus, in the current study it appears justifiable to use the glycosylated hemoglobin values as a study variable despite the alteration of laboratory techniques. In a large series such as ours where the primary purpose is a search for etiologic relationships, we feel the data are robust enough to overcome any error intro- duced by the change of laboratory technique and subsequent adjustment. However, as each research situation is unique, we cannot recom- mend the use of our adjustment procedure by others. Each investigator faced with a similar problem should determine if an adjustment is advisable and then develop his or her own equations. Indeed, in many situations such a procedure is not advisable.

Acknowledgemenrs-The authors are grateful to Drs Day- ton T. Miller. Vincent MaPe;io. David L. DeMets. Frank C. Larson, Polly Newcomb, a;> Matthew D. Davis ior consul- tation and criticism at different stages of the study; Moneen M. Meuer for project coordination; Stacy M. Meuer for data management; Anik Ganguly and Larry D. Hubbard for programming and data management advice; and Nancy L. Lange, Barbara B. Houser and Lori E. Shinstine for assistance with the manuscript.

I.

2.

3.

4.

5.

6.

REFERENCES Duncan BB, Heiss G. Nonenzymatic glycosylation of protein-A new tool for assessment of cumulative hyperglycemia in epidemiologic studies, past and future. Am J Epidemiol 1984; 120: 169-189. Klein R. Klein BEK, Moss SE et al. Prevalence of diabetes mellitus in southern Wisconsin. Am J Epi- demiol 1984; 119: 5461. Klein R, Klein BEK, Moss SE et al. The Wisconsin Epidemiologic Study of Diabetic Retinopathy II. Prev- alence and risk of diabetic retinopathy when age at diagnosis is less than 30 years. Arch Ophtbabnol 1984; 102: 520-526. Klein R, Klein BEK, Moss SE et al. The Wisconsin Epidemiologic Study of Diabetic Retinopathy III. Prevalence and risk of diabetic retinopathy when age at diagnosis is 30 or more years. Arch OphthaImol 1984; 102: 527-532. Isolab (Akron, OH). Qoik-step Fast Hemoglobin Test System. 1981; 1-8. Goldstein DE, Peth SB, England JD et al. Effects of acute changes in blood glucose on HbA,,. Diabetes 1980; 29: 623-628.

7.

8.

9.

10.

Methodologic Considerations in Measuring Glycosylated Hemoglobin 649

Compagnucci P, Cartechini MG, Bolli G et ai. The importance of determining irreversibly glycosylated hemoglobin in diabetes. Diabetes 1981; 30: 60742. 11. Moore FE, Gordon T. Serum ChoRaterol Levels of Adults: United States-1%&1962. Public Health Ser- vice. Vital and Health Statistics 1967: Public Health 12. Service Publication No. 1000, Series 11, No. 22. Sosenko JM, Gadia MT, Fournier AM et a/. Body stature as a risk factor for diabetic sensory neuropathy. 13. Am J Med 1986; 80: 1031-1034. Monson JP, Koios G, Toms GC et al. Relationship between retinopathy and glycaemic control in insulin-

dependent and non-insulin-dependent diabetes. J R Sot Mad 1986; 79: 274-276. College of American Pathologists. E4ectropboresis- Chromatogrnpby Survey. Skokie, Ill.: College of Amer- ican Pathologists; 1986. Klein R, Klein BEK, Moss SE er al. Glycosylated hemoglobin in a population-based study of diabetes. Am J-Epidamiol i98-7; 126: 415428. Jurv DR. Baker JR. Dunn PJ. Clinical imoortance of the-reversible fraction of haemoglobin A;, in type 2 (non-insulin dependent) diabetes. Dtnhetologia 1983; 25: 313-315.