Developing an estimated HbA1c by multivariable factors. 

 

P. Sueyanyongsiri, M.D.¹

¹Department of internal medicine, Surin hospital, Thailand.

               

Abstract:

Objective: In endemic area of hemoglobinopathy, patients exhibit significant clinical variation, particularly in level of hemoglobin. A prediction model of generalized linear regression analysis is used to predict HbA1c values from more than one variable.

Material and Methods: A cross-sectional, generalized prediction model is used to predict HbA1c values from more than one variables, employing the slope of the population with normal hemoglobin group to forecast HbA1c values. For the validation, the same dataset was used, calculating the percentage error compared to the actual measured values at HbA1c. A t-test was then performed, dividing patients into subgroups based on their blood sugar (Dtx) levels.

Results: Of all the 578 patients enrolled in this study. Patients with high hematocrit (Hct) also had higher levels of HbA1c. Hct was found to be weakly associated with HbA1c levels, at Y=5.5+0.055Hct. and R-square =0.0243, which is highly significant at p-value<0.001. Hence the equation derived from multivariable regression analysis is Y=3.2+0.018 Dtx+0.05 Hct. Validation testing revealed that the lowest error rate was 8.0, close to 0. In the 6.0-10.0 range, the error was approximately -20 to 20%. Testing the equation with normal hemoglobin group showed a relatively good relationship.

Conclusion: Suggesting that HbA1c levels are affected by Hct levels, in addition to Dtx. The estimate HbA1c is used to estimate the level of Dtx for individual self-control by patients with altered blood concentration.

Keywords: HbA1c, eHbA1c, estimate HbA1c

Introduction

The average blood sugar level, known as “HbA1c” is the percentage measurement of glycated hemoglobin in the blood, which is a type of protein in the red blood cell. In screening, diagnosis and monitoring treatment response of diabetes mellitus (DM), HbA1c levels are measured along with pre and post prandial blood sugar levels over three months to provide a better overall assessment of blood glucose over time than isolated fasting blood sugar levels during patient visits to the physician^1-4. HbA1c is measured every 3-4 months or at least twice a year in patients who retain stable long-term diabetic control to assess the risk of complications.

Contemporary lifestyles and diets that are stressful, lacking in exercise, rest, together with obesity and genetic dispositions all contribute to the increased prevalence of DM^5-6. Patients with poor control of DM subsequently may suffer from comorbidities such as amputations, chronic kidney disease, ischemic heart disease and cerebrovascular disease which can lead to mortality. Hence the importance of proper monitoring of blood sugar levels and HbA1c for screening, diagnosis and risk assessment of complications in DM for evaluating and improving the patient’s ability to control the disease over time.

The American Diabetes Association suggested an arbitrary cut-off point of ≥6.5% for HbA1c for the diagnosis of DM due to its increased associated risk of developing diabetic retinopathy⁷. Furthermore, a goal of HbA1c <7% or average blood sugar levels of <120mg/dL is desirable for good diabetic control to avoid complications. Studies have found that strict diabetic control can prevent microvascular problems of the retina, kidney and foot^8-9. However, the average proportion of patients that can retain good diabetic control is only about 50% in various studies worldwide^10-13. The reality is it is difficult for patients to control DM let alone strict control, which can also lead to hypoglycemia, a complication that can range in severity from mild dizziness to hospitalization¹⁴. Data from the National Health Security Office of Thailand suggest that merely 35% of patients are able to retain good diabetic control.

The clinical use of HbA1c is somewhat limited due to inaccuracies of test results in patients with certain conditions such as thalassemia, anemias, hemoglobin E disorder, hemolytic conditions and chronic kidney disease. In regions where such conditions are endemic^15-18, the use of HbA1c can become very challenging as the patient’s test result may appear normal while their fasting blood sugar may be well above 120mg%. For example, the condition hemoglobin E disease is endemic and highly prevalent at an estimated 30-50% of the population in the southern Thai province of Surin, many of which have coexisting DM^15,17-18. Hence, this study aims to determine the relationship between capillary blood sugar levels (Dtx) and HbA1c levels to assess the need for control of Dtx to achieve HbA1c targets in populations with hemoglobin E disorder.

Material and methods

This cross-sectional study was conducted in an endemic area of hemoglobinopathy to evaluate variability in hemoglobin type and hemoglobin concentration among patients. The data are not publicly available due to privacy restrictions but are available from the corresponding author upon a declaration that the patient's information to use in publish articles in medical technology will not disclose the personal information of the patient in any way, without identifying the person who owns the data. A generalized linear regression model was developed to predict HbA1c values using multiple variables. The slope derived from the DCIP-negative population (control group) was applied to estimate HbA1c levels.

The study protocol was approved by the institutional review board and conducted at the diabetes clinic of Surin Hospital between January and December 2017. A total of 578 patients were enrolled and categorized into three groups based on hemoglobin status: homozygous hemoglobin E (HbEE), hemoglobin E trait (HbEA), and DCIP-negative (control group). Hemoglobin classification was determined from blood samples, which were also used for annual HbA1c measurement. Written informed consent was obtained from all participants.

Sample size estimation was performed using G*Power software, based on correlation coefficients observed in the control group, particularly between hematocrit (Hct) and fasting capillary blood glucose (Dtx). The primary population (DCIP-negative group) demonstrated a left-skewed distribution, with a substantial proportion of patients exhibiting relatively high Hct and HbA1c levels. Therefore, patients with hemoglobin E disorders were included to improve data distribution and enable a more robust assessment of the relationship between Hct and HbA1c.

Although HbA1c is considered most reliable in individuals with normal erythrocytes, its accuracy in patients with hemoglobinopathies remains uncertain. Accordingly, the model fixed the intercept and slope for the Dtx variable. For Hct, only the intercept was fixed, based on the assumption that hemoglobin type does not directly affect HbA1c. The slope derived from the control group was then applied to predict HbA1c across all three groups. The HbEE subgroup was included primarily to improve distributional balance rather than to contribute to slope estimation. All data from 2017 were analyzed to assess associations among hemoglobin type, hematocrit (Hct), fasting capillary blood glucose (Dtx), and HbA1c. Diabetes mellitus was diagnosed in all patients using a fasting plasma glucose threshold >126 mg/dL. All participants received standard-of-care treatment, including insulin therapy, oral hypoglycemic agents, or physician-directed dietary management.

Laboratory Measurements

Fasting blood samples were collected in the morning after an overnight fast for measurement of fasting plasma glucose, dichlorophenolindophenol (DCIP), and HbA1c. Participants were classified into three groups: DCIP-negative (control), hemoglobin E trait (HbEA), and homozygous hemoglobin E (HbEE). Samples with positive DCIP results underwent further hemoglobin typing using an Hb Gold analyzer (Drew Scientific Ltd., England) based on low-pressure liquid chromatography (LPLC). The DCIP test was performed using KKU-DCIP-clear reagent. HbA1c levels were measured using a turbidimetric inhibition immunoassay (TINIA) method on hemolyzed whole blood (Cobas®, Roche Diagnostics, USA). Additional clinical and laboratory parameters, including fasting capillary blood glucose (Dtx), complete blood count, hematocrit (Hct), and serum creatinine (Cr), were collected. HbA1c levels were compared across the three groups and correlated with preprandial capillary blood glucose levels.

Statistical Analysis

Descriptive statistics are presented as mean ± standard deviation or percentiles, as appropriate. One-way analysis of variance (ANOVA) was used to compare mean values across groups stratified by age, sex, Hct, Dtx, Cr, and HbA1c. Generalized linear regression analysis was performed for both univariable and multivariable models. A p-value <0.05 was considered statistically significant. Robust variance estimation was applied to account for potential correlations among dependent variables. In multivariable analysis, collinear variables were excluded, and representative variables were selected. Identified correlated pairs included fasting capillary blood glucose and fasting plasma glucose, hemoglobin and hematocrit, and hemoglobin type and hematocrit. Variables selected for the final model were those readily accessible in primary care settings or amenable to self-monitoring, specifically Dtx and Hct. To assess the relationship between hemoglobin type and hematocrit, hematocrit was modeled as a proxy variable representing different hemoglobin types. If significant correlation was identified, hemoglobin type was excluded from the model. Internal validation was performed using the same dataset by calculating percentage error between predicted and observed HbA1c values within the range of 6.0–10.0%. A t-test was conducted after stratifying patients into 15 groups based on Dtx levels (80–300 mg/dL). Validation analysis was restricted to patients with normal hematologic parameters, representing the majority of the diabetic population.

Results

The correlation coefficient between HbA1c and fasting capillary blood glucose (Dtx) was 0.5481, whereas the correlation between HbA1c and hematocrit (Hct) was weaker (r = 0.1279). The analyzed sample included 347 individuals, which exceeded the calculated sample size of 107 by approximately 3.5-fold, yielding a statistical power of 99.99%. Among the 578 patients enrolled, 178 were male and 410 were female. Of these, 347 patients were DCIP-negative (control group), while 231 had hemoglobin E disorders, including 194 with hemoglobin E trait (HbEA) and 37 with homozygous hemoglobin E (HbEE). Although females were more prevalent across all groups, the difference was not statistically significant. Mean Dtx, HbA1c, and serum creatinine levels did not differ significantly between groups. However, age and hematocrit levels differed significantly. The HbEE group demonstrated lower Hct values compared with the other groups. Overall, patients across all groups exhibited a wide distribution of Hct and HbA1c values, ranging from low to normal levels (Figures 1 and 2). Linear regression analysis demonstrated a positive association between Hct and HbA1c, with higher Hct corresponding to higher HbA1c levels. No significant association was observed between Hct and Dtx (Figure 3). In multivariable exploratory analysis, after adjusting for covariates and separating correlated variables, hemoglobin type was not independently associated with HbA1c. In contrast, Hct remained significantly associated with HbA1c (Figure 4 and Table 2). Univariable regression analysis showed comparable relationships between HbA1c and Dtx in the HbEA and control groups (HbEA: Y = 5.14 + 0.017Dtx, R² = 0.1970; control: Y = 4.50 + 0.020Dtx, R² = 0.3251). However, the slope of the relationship was attenuated in the HbEE group (Y = 5.44 + 0.011Dtx, R² = 0.3251), indicating a weaker association (Figure 5). Hematocrit was weakly but significantly associated with HbA1c (Y = 5.5 + 0.055Hct, R² = 0.0243, p < 0.001), explaining approximately 2.43% of the variance (Figure 6). Based on multivariable regression analysis, the predictive equation for HbA1c was:

Y=3.2+0.018(Dtx)+0.05(Hct)Y = 3.2 + 0.018(\text {Dtx}) + 0.05(\text {Hct}) Y=3.2+0.018(Dtx)+0.05(Hct).  After adjustment for missing data and implementation of a model incorporating both Dtx and Hct, the derived equation provided individualized HbA1c estimates adjusted for hematocrit (Table 3). The predicted HbA1c values did not differ significantly from observed values in terms of slope or intercept of the regression line. Internal validation demonstrated higher prediction error at extreme HbA1c levels (very low and very high values). The lowest error approached 0% at HbA1c ≈ 8.0%. Within the clinically relevant range of 6.0–10.0%, prediction error ranged approximately from −20% to +20%. Application of the model in diabetic patients within the DCIP-negative group showed good agreement between predicted and observed HbA1c values (Figures 7 and 8).

Table 1- Characteristics of the patients at baseline.

Variable	Control group	HbEA	HbEE	p-value*
Cases (n; %)	347(60.0%)	194(33.6%)	37(6.4%)
Age (years; mean, S.D.)	60(10.9)	57.9(10.4)	63.6(9.2)	0.006
Sex (Male; n, %)	105(30.3%)	62(32.0%)	9(24.3%)	0.664
Dtx (mg/dl; mean, S.D.)	146.1(47.6)	150.7(50.2)	145.4(45.9)	0.558
HbA1c (%; mean, S.D.)	7.67(1.9)	7.59(1.8)	6.95(1.3)	0.073
Hematocrit (%; mean, S.D.)	38.5(5.1)	38.1(4.2)	32.3(3.8)	<0.001
Cr (mg/dl; mean, S.D.)	1.07(0.5)	1.04(0.4)	1.03(0.3)	0.733

Dtx= Capillary blood glucose; control group= negative dichlorophenol-Indolephenol; Hb EA= hemoglobin E trait; Hb EE= homozygous hemoglobin E; P-value<0.05 significant

 

Table 2. Effect of fasting capillary blood sugar, hematocrit level and hemoglobinopathy group to HbA1c by multivariable regression.

Variable	Constant	Co-efficient	95% Confidence interval	p-value
Dtx	3.2	0.018	0.014-0.022	<0.001
Hct	3.2	0.048	0.020-0.076	0.001
Groups	3.2	-0.06	(-0.274) - (0.157)	0.592

Hct=Hematocrit; Dtx= Capillary blood glucose

 

Table 3. Effect of fasting capillary blood sugar, hematocrit to estimated HbA1c(eHbA1c) by multivariable regression.

Hct.	HbA1c targets  (%)	Estimate Dtx. (mg/dl)	Estimate Dtx. (mmole/dl)
30	7.0	127.8	7.10
32	7.0	122.2	6.79
34	7.0	116.7	6.48
36	7.0	111.1	6.17
38	7.0	105.6	5.87
40	7.0	100.0	5.56
42	7.0	94.4	5.24
44	7.0	88.9	4.94

Hct=Hematocrit; Dtx= Capillary blood glucose

 

Figure 1. Histogram of HbA1c level in HbEE, HbEA and negative DCIP cases.

Figure 2. Histogram of hematocrit level in HbEE, HbEA and negative DCIP cases.

Figure 3. Non-significant correlation between hematocrit level and fasting capillary blood sugar level, but significant correlation with HbA1c level.

 

 Figure 4. Coefficient plots estimates and confidence intervals.

Dtx= Capillary blood glucose; Hct=Hematocrit; Hb group=hemoglobin group; CHO=Cholesterol; TG=Triglyceride; HDL=high-density lipoprotein cholesterol; LDL=Low-density lipoprotein cholesterol

 Figure 5. Fitted line between fasting capillary blood sugar(mg%) versus HbA1c (%) in control group, hemoglobin EA and hemoglobin EE group.

 

 Figure 6. Fitted line between hematocrit level (%) versus HbA1c (%) in control group, hemoglobin EA and hemoglobin EE group.

 

Figure 7. Internal validation of estimated HbA1c from 6.0 to 10.0%.

Figure 8. Standard error bar charts for comparison of measured HbA1c and estimated HbA1c in stratified fasting capillary blood sugar groups.

Discussion

The present study demonstrated a modest correlation between fasting capillary blood glucose (Dtx) and HbA1c (R² = 0.2322), which is lower than previously reported findings. This discrepancy may be attributable to the high prevalence of hemoglobinopathies in the study population from Surin Province, as well as other confounding factors discussed above. In contrast, hematocrit (Hct) showed a statistically significant, albeit weak, association with HbA1c (R² = 0.0243, p < 0.001), indicating that Hct contributes modestly to HbA1c variability.    Analysis of regression models suggests that hemoglobin type (HbEE, HbEA, or DCIP-negative) primarily influences the intercept of the HbA1c prediction equation, but not the slope. This indicates that baseline HbA1c levels differ across hemoglobin subgroups, whereas the relationship between Dtx and HbA1c remains relatively consistent. Specifically, each 1 mg/dL increase in Dtx was associated with an approximate 0.017% increase in HbA1c. When Hct was incorporated into the model, it demonstrated a statistically significant independent effect on HbA1c, supporting its inclusion in the predictive equation. Given the presence of multicollinearity, only selected variables were retained in the multivariable model. Hemoglobin and hematocrit were highly correlated (r > 0.895), as were fasting capillary blood glucose and fasting plasma glucose (r = 0.921). Including these variables simultaneously would compromise model stability. Therefore, Dtx and Hct were selected as representative variables, given their clinical accessibility and lower collinearity (r = 0.198). This approach ensured model parsimony and practical applicability in resource-limited settings.

A substantial proportion of patients with hemoglobin E disorders exhibited low hematocrit levels, reflecting chronic anemia. In this context, hematocrit was preferred over hemoglobin concentration as a predictor variable due to its strong correlation with hemoglobin and its suitability for modeling. Notably, some patients with relatively low Hct values did not exhibit clinical symptoms such as hyperviscosity, suggesting that hematologic adaptation may occur in this population. The high prevalence of diabetes mellitus (DM) in the study cohort necessitated subgroup analyses. The study setting is unique in that patients with hemoglobin E disorders are routinely monitored as a distinct subgroup, allowing for a more detailed evaluation of their impact on HbA1c interpretation. Although only a minority of patients (37/578) had clinically significant anemia, the findings remain relevant for populations with similar genetic backgrounds.

Further exploratory analysis focused on patients with Hct <38%, in whom HbA1c interpretation may be particularly affected. Using a modeling approach that incorporated Dtx and Hct, HbA1c values were estimated in the absence of direct measurement. The threshold of 38% was empirically derived based on iterative adjustment of regression coefficients. For patients with Hct ≥38%, reverse estimation was performed by assuming a target HbA1c of 7.0% to approximate corresponding Dtx levels for self-monitoring. Model validation demonstrated good agreement between predicted and observed HbA1c values in the DCIP-negative group. However, predictive performance was less robust in patients with hemoglobin E disorders. Nevertheless, applying the slope derived from individuals with normal hematologic parameters to those with hemoglobinopathies appears to provide reasonable estimates. Further validation in larger and more diverse populations is warranted.

---

Conclusion

The derived equation for estimated HbA1c (eHbA1c) integrates fasting capillary blood glucose (Dtx) and hematocrit (Hct) to provide an individualized assessment of glycemic control in patients with diabetes and non-diabetes to self-control blood sugar. This approach may be particularly useful in settings where HbA1c measurement is unreliable due to underlying hemoglobinopathies. Despite its potential utility, estimated HbA1c should not replace standard laboratory measurements. Patients are advised to undergo regular HbA1c testing at certified laboratories every 3–4 months to ensure accurate long-term monitoring and to reduce the risk of diabetes-related complications

Competing interests

There are no potential conflicts of interest to declare.

References

1.             Little RR, Roberts WL. A Review of Variant Hemoglobins Interfering with Hemoglobin A1c Measurement. Journal of Diabetes Science and Technology. 2009;3(3):446-451. doi:10.1177/193229680900300307NGSP.

2.             Sacks DB. A1C versus glucose testing: A comparison. Diabetes Care. 2011;34(2):518-523. doi:10.2337/dc10-1546   

3.             Selvin E, Crainiceanu CM, Brancati FL, Coresh J.  Short-term variability in measures of glycemia and implications for the classification of diabetes.  Arch Intern Med. 2007;167(14):1545-1551. doi:10.1001/archinte.167.14.1545

4.             Gillett MJ. International expert committee report on the role of the A1c assay in the diagnosis of diabetes: diabetes care 2009; 32 (7): 1327–1334. The Clinical Biochemist Reviews. 2009 Nov;30(4):197.

5.             Roglic G. WHO Global report on diabetes: A summary. International Journal of Noncommunicable Diseases. 2016;1(1):3. doi:10.4103/2468-8827.184853

6.             Shaw JE, Sicree RA, Zimmet PZ. Global estimates of the prevalence of diabetes for 2010 and 2030. Diabetes Research and Clinical Practice. 2009;87(1):4-14. doi:10.1016/j.diabres.2009.10.007

7.             Association AD. 13. Children and Adolescents: Standards of Medical Care in Diabetes—2021. Diabetes Care. 2020;44(Supplement_1): S180-S199. doi:10.2337/dc21-s013

8.             Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33). The Lancet. 1998;352(9131):837-853. doi:10.1016/s0140-6736(98)07019-6

9.             Nathan DM, Genuth S, Lachin J, et al. The effect of intensive treatment of diabetes on the development and progression of Long-Term complications in Insulin-Dependent diabetes mellitus. New England Journal of Medicine. 1993;329(14):977-986. doi:10.1056/nejm199309303291401

10.          Winocour PH. Effective diabetes care: a need for realistic targets. BMJ. 2002;324(7353):1577-1580. doi:10.1136/bmj.324.7353.1577

11.          Goudswaard AN, Lam K, Stolk RP, Rutten GE. Quality of recording of data from patients with type 2 diabetes is not a valid indicator of quality of care. A cross-sectional study. Family Practice. 2003;20(2):173-177. doi:10.1093/fampra/20.2.173

12.          Eliasson B, Cederholm J, Nilsson P, Gudbjörnsdóttir S. The gap between guidelines and reality: Type 2 diabetes in a national diabetes register 1996–2003. Diabetic Medicine. 2005;22(10):1420-1426. doi:10.1111/j.1464-5491.2005.01648.x

13.          Bryant W, Greenfield JR, Chisholm DJ, Campbell LV. Diabetes guidelines: easier to preach than to practise? The Medical Journal of Australia. 2006;185(6):305-309. doi:10.5694/j.1326-5377.2006.tb00583.x

14.          Gerstein HC, Miller ME, Byington RP, et al. Effects of intensive glucose lowering in type 2 diabetes. New England Journal of Medicine. 2008;358(24):2545-2559. doi:10.1056/nejmoa0802743

15.          Sattarattanamai C, Thongsuk S, Sutjaritchep P et al: Prevalence of thalassemia and hemoglobinopathies in pregnant women at Surin Hospital. Med J Srisaket Surin Buriram Hosp 2000; 15: 1-12.

16.          Mccurdy PR. 32DFP and 51CR for measurement of red cell life span in abnormal hemoglobin syndromes. Blood. 1969;33(2):214-224. doi:10.1182/blood.v33.2.214.214

17.          Sueyunyongsiri P: Effect of Hemoglobin E disorder on Hemoglobin A1c in Diabetic patients. Med J Srisaket Surin Buriram Hosp 2008; 23: 637-643.

18.          Sueyanyongsiri P, Tangbundit P, Sueyanyongsiri S, et al. Artifactually Low Hemoglobin A1C in Diabetic Patients with Hemoglobin E Disorder: Surin Hospital, Thailand. Allergy drugs clin immunol 3: 117. Allergy drugs clin immunol. 2019;3(1):91-3.

19.          Prayongratana K, Polprasert C, Raungrongmorakot K et al. Low cost combination of DCIP and MCV was better than that of DCIP and OF in the screening for hemoglobin E. Medical journal of the Medical Association of Thailand. 2008 Oct 1;91(10):1499.

20.          Van Dongen S, Molenberghs N, Matthysen N. The statistical analysis of fluctuating asymmetry: REML estimation of a mixed regression model. Journal of Evolutionary Biology. 1999;12(1):94-102. doi:10.1046/j.1420-9101.1999.00012.x

21.          Nathan DM, Kuenen J, Borg R, Zheng H, Schoenfeld D, Heine RJ. Translating the A1C assay into estimated average glucose values. Diabetes Care. 2008;31(8):1473-1478. doi:10.2337/dc08-0545 doi: 10.2337/dc08-0545

22.          Musa IR, Omar SM, Sharif ME, Ahmed ABA, Adam I. The calculated versus the measured glycosylated haemoglobin (HbA1c) levels in patients with type 2 diabetes mellitus. Journal of Clinical Laboratory Analysis. 2021;35(8): e23873. doi:10.1002/jcla.23873

23.          Sueyanyongsiri P. Effect of Hematocrit levels on HbA1c Values in the Endemic Area of Hemoglobinopathy. J ASEAN Fed Endocr Soc [Internet]. 2023 Nov. 9 [cited 2025 Jul. 20];38(S3):62.

24.          Sueyanyongsiri P. Translating HbE1c from Fasting Capillary Blood Sugar and Hematocrit Level in Surin Hemoglobin E Homozygote Diabetic Patients, Thailand. J ASEAN Fed Endocr Soc [Internet]. 2023 Nov. 9 [cited 2025 Jul. 20];38(S3):61.