Introduction
Haemoglobin A1c (HbA1c) is produced by non-enzymatic glycation of haemoglobin. It provides an estimate of glycaemia (mean glucose levels) over the preceding one to three months, proportionately weighted to more recent periods [1]. It is therefore used to diagnose diabetes mellitus and to monitor patients with diabetes. The American Diabetes Association (ADA) recommends a cut-off point of more than or equal to 6.5% (48mmol/mol) HbA1c level for a diagnosis of diabetes mellitus and 5.7–6.4% (39–47 mmol/mol) for a diagnosis of prediabetes [2]. It is recommended that HbA1c is measured every three to six months in newly diagnosed patients with type diabetes mellitus until stabilised, and thereafter every six months [3].
HbA1c levels can be reported using either International Federation of Clinical Chemistry and Laboratory Medicine (IFCC), expressed in mmol/mol, or National Glycohemoglobin Standardization Program (NGSP) reference systems, expressed as % [4]. Although a conversion formula between the two methods is available, they may not be directly interchangeable [5]. Further, within-subject variability in NGSP units is reported to be lower than in IFCC units [6]. Although the IFCC is considered to be the higher standard test, many countries continue to use NGSP units [7].
Reference ranges for diagnostic tests are usually set by comparison to a reference population, which allows assessment of between-subject variation. However, biological parameters also vary over time. This can be systematic and predictable, such as seasonal variation [8], or may be due to chance. This longitudinal within-subject variation is known as biological variation. Variation can also be introduced into a measurement from pre-analytical factors such as stress, exercise and food intake prior to the laboratory measurement and posture during the sampling procedure. Imperfect accuracy and precision of laboratory measurement mean that if measurement is repeated a number of times on the same sample, there will be a range of results around the true value. This is analytical variation.
The variability in measured HbA1c results encountered in real world clinical practice is because of a combination of biological, pre-analytical and analytical variability. Biological variability is the greatest component of total variability, since clinical laboratories usually set acceptable analytical variability at <0.5% of biological variability [9].
The greater the within-subject variability of a parameter, the lower the probability that a single measurement is an accurate reflection of the true mean of the parameter in that individual. This can lead to errors in diagnosis, prognosis and treatment. Specifically for HbA1c, an inaccurate result may lead to an incorrect, missed or delayed diagnosis of diabetes mellitus, causing inappropriate commencement of treatment or an inappropriate delay in commencing treatment, and an incorrect estimation of glycaemic control in patients with diabetes, leading to an inappropriate increase in treatment intensity or a delay in intensifying treatment.
Three previous systematic reviews on HbA1c variability have been published. Gonzalez-Lao et al [10] reported the results of a systematic review and meta-analysis of 17 papers. Based on three papers this gave a pooled estimate of within-subject coefficient of variation (CVi) for HbA1c in healthy adults of 0.013 (0.012–0.025) for results reported in IFCC units (mmol/mol) and based on four papers, a CVi of 0.013 (0.012–0.021) for results reported in non-IFCC units (%). CVi was slightly higher in patients with diabetes mellitus than subjects without diabetes. The European Federation of Clinical Chemistry and Laboratory Medicine (EFLM) database reports a CVi of 0.016 (0.013–0.024) for papers reporting IFCC units based on four papers and 0.012 (0.003–0.019) for papers reporting NGSP units (based on 7 papers) [11]. Braga et al performed a systematic review without meta-analysis of eight studies, with CVis ranging from <0.007 to 0.98 [12].
There is a need for an up to date and more comprehensive systematic review of within-subject variability of HbA1c.
Aim of review
To describe and evaluate the current literature on within-subject variability of measured HbA1c and from this to estimate variability in people with and without diabetes mellitus.
Methods
Searches were devised to identify cohort studies, clinical trials or any studies in which an HbA1c or glycosylated haemoglobin measurement was performed more than once in the same individual. See Appendix 1 in S1 File for sample Medline and Embase search strategy. The searches for this study were combined with a similar study into C-reactive protein (CRP) variability, and then CRP studies were excluded at the full text screening stage.
The international prospective register of systematic reviews, PROSPERO, was checked for ongoing reviews, and the protocol was registered with PROSPERO. On 5th August 2020, database searches were carried out. Medline, Embase, Cochrane Central, Epistemonikos and Open Grey were searched from inception to 5th August 2020. An update to the search was performed to include studies published up to November 2022 (Full details in Appendix 2 in S1 File). Subject experts were contacted for suggestions for further papers. Search terms were adapted for each database searched. The references of included papers were checked by hand for further relevant papers. Endnote reference management software was used to collate studies revealed by the search.
Studies were included if primary research data on the variability of at least two measurements of HbA1c or glycosylated haemoglobin within the same subject was recorded. Studies could be of any design. The population included adults and children, healthy or with any disease condition, in any setting. Outcome was variability which could be reported as coefficient of variation (CVI), standard deviation (SD), variability independent of the mean (VIM), index of individuality (II), Reference Change Value (RCV), index of heterogeneity, validity coefficient (VC), ICC agreement, ICC consistency, Cronbach’s alpha or Cohen’s kappa. There was no restriction on time of publication, language of publication, population, setting or sample size. Studies were excluded if participants were not in a steady state (measurements were before and after an intervention or had an acute or rapidly changing illness) or data were secondary (systematic and narrative reviews).
Studies were grouped for synthesis based on variability measure reported eg ICC, CV. Since this study was aimed to capture variability data wherever it was published, there was no limit on publication date, language or study design.
Titles and abstracts were screened independently by two reviewers (AG and EF) in Abstrackr systematic review software [13], and those identified of interest underwent full text searches. Full texts were screened by AG and exclusions confirmed by EF. Differences were resolved by discussion. Foreign language papers were translated by Google translate software. Data were extracted into Excel from the full text of the study papers where possible, and from abstracts where full texts were not available, by a single reviewer (AG). All data extraction was verified independently by a second reviewer (EF). Where the same study was reported in multiple papers, the full text paper was preferred over an abstract, English language preferred over non-English, and the earliest English version over the later if there was more than one.
Table 1 lists the outcome and other main variables extracted. All eligible outcomes were included where more than one outcome was reported within a paper. The primary outcome measure was variability of repeated measurements within the same subject. Where multiple measures of variability were given in a single study, the primary population only was analysed for the main meta-analysis. The primary population was the full study population (as opposed to subgroups), and if the full study population was not given, the primary outcome was identified using the following hierarchy: 1. Healthy population; 2. Most stable population (i.e. subjectively judged to be in the most steady-state such as disease course or treatment); 3. First outcome listed in the paper.
[Figure omitted. See PDF.]
Where information was missing or unclear, the information was not extracted with the following exception: where patient characteristic data was reported only for a whole study population, but variability data was only present for a sub-population, the patient characteristic data for the whole population was used.
A risk of bias tool was used adapted from the Consensus-based Standards for the selection of health Measurement Instruments (COSMIN) risk of bias tool for test reliability [14]. The template risk of bias form with associated explanatory notes is shown in Appendix 2 in S1 File. Risk of bias was assessed by AG and reviewed by EF. Disagreements between reviewers were resolved by discussion.
The main outcome measure of this review is coefficient of variation of repeated measurements within the same individual.
All studies were considered for synthesis provided they had a measure of variability that had a sufficient number of studies reported so they could be pooled. Coefficients of variation were converted from % to a decimal fraction.
Stata SE 17 was used to perform statistical calculations and generate forest plots. ICCs were transformed using Fisher’s Z transformation [15]. Z scores were then back-transformed (see Appendix 4 in S1 File) and 95% confidence intervals of the ICC were calculated. Coefficient of variations were considered for meta-analysis, but no well-described and validated methods currently exist for meta-analysis of coefficient of variations, so we describe the results identified here. Descriptive subgroup analyses were performed based on unit of measurement (IFCC or NGSP), health status, setting, number of measurements and whether short or long-term variability was measured. Sensitivity analyses were based on number of subjects and risk of bias.
Results
(Fig 1).
[Figure omitted. See PDF.]
After database searches and hand searching of reference lists of included studies, 2,675 non-duplicate studies were identified of which 2320 were excluded after title and abstract screening leaving 355 studies. After full text screening, 244 were excluded, and a further five during data extraction. Most full text exclusions were because no recognised measure of variability was reported.
One hundred and eleven studies met the inclusion criteria (31.3% of full text studies screened). Sample sizes ranged from 4 to 91,866 subjects, with a median of 378. One hundred-and-five studies (95%) reported longer term variability (arbitrarily defined for this study as seven or more days between measurements), two reported short-term variability (less than seven days between measurements), one reported both long and short term, and three did not report measurement intervals. One hundred and six studies (95.5% of included studies) were of a cohort design, three (2.7%) used data from the placebo arm of a randomised controlled trial and two (1.8%) were case-control studies Twenty studies (18.0%) were carried out in a primary care or community setting and 73 (65.8%) studies were carried out in secondary or tertiary setting such as universities, hospitals and laboratories. In eighteen (16.2%) studies the setting was not reported.
Study populations were diverse in terms of age, gender and health status. Ethnicity was recorded in 34 (30.6%) studies. Included studies and full study characteristics are presented in Appendix 4 in S1 File. With questions four and five of the risk of bias scoring excluded, 21 (18.9%) studies scored the best risk of bias score of one, 53 (47.7%) studies scored a risk of bias of two, and 37 (33.3%) studies scored a risk of bias of three. Risk of bias scores are presented in Appendix 5 in S1 File. Table 2 summarises the study characteristics of the included papers.
[Figure omitted. See PDF.]
Included studies consisted of ninety-four full texts and seventeen abstracts. The average study age of subjects in the primary population ranged from 8.8 to 76 years, with a median of 59.1. The percentage of subjects who were male ranged from 0% to 98% with a median of 52.3%. Ninety-three studies reported the setting. Twenty of the 93 (21.5%) studies that recorded setting were conducted in primary care or the community and 73/93 (78.5%) in secondary or tertiary care or in a laboratory/university situation. Health status of individuals was recorded in 110/111 studies of which 25 (22.5%) were healthy, 59 (53.2%) had type 2 diabetes, 19 (17.1%) had type 1 diabetes, 4 (3.6%) had both type 1 diabetes and type 2 diabetes, and 3 (2.7%) had other diseases.
The average number of measurements ranged from 2 to 36.8 with a median of 5. Eighty-four (75.7%) studies used NGSP (%) units to calculate the variability and four (3.6%) used IFCC (mmol/mol); 23 (20.7%) reported in both units or did not clearly specify the units.
Eighty-three (74.8%) studies reported total variation, 15 studies (13.5%) reported individual variation, and 1 (0.9%) study reported both. For 12 (10.8%) studies the type of variation reported was unclear). The CVI of the HbA1c primary population ranged from 0.0029 to 0.179, with a median of 0.070 (IQR 0.051). The within-subject standard deviation of the HbA1c of the primary population where the units were NGSP (%) ranged from 0.08 to 1.47 with a median of 0.62 (IQR 0.37). Only five papers reported ICC, with a median of 0.873 and a range of 0.59 to 0.97.
One paper [16] appears to report an incorrect value for the CV since the value appears to be implausibly low compared to other reports of this result implausible and is also inconsistent with the other variability measures reported in the paper. The value for CV from this paper used in this review was therefore calculated from the reported SD and mean. Two papers did not report a CV, ICC or SD. Segar [17] reported variability as average successive variability (average absolute difference between successive values), with a result of a mean of 0.6±0.3% (mean±SD). Sugawara [18] used an adjusted standard deviation to account for different numbers of HbA1c measurements, reporting an adjusted SD result of 0.79±0.6% (mean±SD). Three papers [19–21] reported variability measurements for units of both mmol/mol and % and the % unit is included in this review.
Median within-subject coefficient of variation (CVi) was 0.070 (IQR 0.034 to .09). Table 3 summarises the results of variability measures of the primary study population.
[Figure omitted. See PDF.]
(See Methods for definition).
With units reported as %, CVi in healthy subjects (median 0.017; IQ range: 0.013 to 0.022) was lower than in patients with type 2 diabetes (0.083; IQ range: 0.06 to 0.10), type 1 diabetes (0.084; IQ range 0.067 to 0.89) or other health conditions (0.079; IQ range 0.048 to 0.11) (Fig 2 and Table 4). Findings were similar for mmol/mol units (Fig 3).
[Figure omitted. See PDF.]
[Figure omitted. See PDF.]
[Figure omitted. See PDF.]
Figs 4 and 5 show results for studies measuring variability as SD or ICC.
[Figure omitted. See PDF.]
T2DM = type 2 diabetes mellitus, T1DM = type 1 diabetes mellitus.
[Figure omitted. See PDF.]
T2DM = type 2 diabetes mellitus.
Forest plots for CV for studies reporting units of %, stratified by risk of bias, setting, short- or long-term measurements and whether CVI or CVT was reported or whether this was uncertain can be found in Appendix 7 in S1 File.
Because of the variability was higher in patients with diabetes, for all studies where it was reported we plotted mean population HbA1c against CVi (Fig 6). These demonstrated a positive correlation (R2 0.48).
[Figure omitted. See PDF.]
Discussion
This review is the most comprehensive systematic review and meta-analysis of within-subject HbA1c variability. Of the three previous systematic reviews, one did not perform a meta-analysis and two only performed meta-analyses on healthy populations. The largest included only 17 studies in total, 3 the meta-analysis of variability measured using of mmol/mol and 4 in the meta-analysis of % units [10]. This review included 111 studies. Previous systematic reviews on HbA1c variability were designed to estimate a coefficient of variation under ideal conditions in order to inform analytical performance specifications, reference change values and population-based reference intervals. By contrast the current study was designed to be as broad as possible, increasing generalisability by including HbA1c variability in a variety of settings and health conditions.
There is no well-described, validated method for meta-analysis of CVis therefore it was not undertaken in this study.
Our median estimates of CVi (NGSP %) were 0.017 (IQ range: 0.013 to 0.022) in healthy individuals, 0.083 (IQ range: 0.06 to 0.10) in type 2 diabetes and 0.084 (IQ range 0.067 to 0.89) in type 1 diabetes. These are similar to previous reviews of healthy patients which report a median CV of 0.013 (95% CI 0.012–0.025) for IFCC (mmol/mol) units and 0.013 (0.12–0.21) for NGSP(%) [10]; 0.016 (95%CI 0.013–0.024) for IFCC(mmol/mol) and NGSP (%) 0.011 (0.002–0.019) for % [11]. This is consistent with previous research suggesting that patients with diabetes have a higher CVI of HbA1c than healthy patients [10, 22].
We demonstrated a correlation between mean population HbA1c and within-subject variation measured by CVi. This suggests subjects with prediabetes have a higher HbA1c variability than the healthy and poorly-controlled patients with diabetes have a higher variability than well-controlled patients with diabetes. This may be due to different levels of insulin resistance, more than variation in caloric intake. It also means clinical outcomes which correlate with mean HbA1c will also correlate with CVi. Researchers investigating variability as a predictor of outcomes should consider using a measure of variability which is independent of the mean.
The variability of HbA1c has implications for clinical decision making. Misclassification of type 2 diabetes may be considered a low risk since the HbA1c within-subject variability is lower, although with borderline results, a single reading may still be above the cutoff for a diagnosis of type 2 diabetes when the true mean is below the cutoff point. General practitioners’ understanding of this issue derived from clinical experience may explain why they often do not commence treatment with oral anti-diabetic drugs until they document a HbA1c reading well above the cutoff recommended by National Institute for Healthcare and excellence (NICE) guidelines [23]. Of even more importance are the implications for monitoring and treatment decisions, since the variability is higher in these patients. Metformin monotherapy lowers HbA1c by 1.1% NGSP units (12mmol/mol IFCC units on average) [24] which is within the usual range of within-subject variation. This is consistent with the finding that 15–18% of patients that commence treatment with oral anti-diabetic drugs demonstrate an increase in measured HbA1c –this can be explained by within-subject variation.
Another concern is that higher HbA1c variability is associated with an increased all-cause mortality due to a variety of mechanisms such as endothelial dysfunction, increased oxidative stress and increased release of cytokines [25], demonstrating the importance of reducing HbA1c variability by improving control of diabetes. Moreover, a strict glycemic control may play a cardioprotective effect [26].
Two obvious outliers were found in this review, one which showed a CVi of 0.59 [16]. This appeared to be an error and the CVi was recalculated for this review from data available in the study to give a result of 0.086. The other study showed a CVi of 0.0029 [27], although there is an error in the abstract where a different CVi result is given, which appears to be the result for males only. The authors acknowledge their result is lower than previously published estimates of HbA1c CVI and suggest this may be due to strict control of pre-analytic factors and Chinese ethnicity of the subjects. This latter seems unlikely since four other papers with Chinese subjects reported a CVI ranging from 0.075 to 0.17 [16, 28–30]. Nevertheless, it has been shown that ethnicity influences HbA1c levels [31]. This does illustrate the importance of taking ethnicity into account when studying variability, and only 34/111 studies in this review explicitly reported the ethnicity of their subjects, although it is not always clear whether nationality or ethnicity is being referred to. Future research in HbA1c variability should include the ethnicity of subjects.
There are a number of weaknesses with some of the studies in this review. Because variability was not the primary outcome of the studies some relevant studies may not have been identified. Some included studies had low numbers of subjects and/or low numbers of repeat measurements. It was not always clear whether studies were reporting total variability (i.e. CVA + CVi) or within-subject variability (CVi). S3 Fig in Appendix 7 of S1 File illustrates the CVs of studies split by whether they reported CVI, CVT, or whether this was uncertain. Subjectively there seemed to be minimal difference between the groups.
HbA1c may not be an ideal measure of glycaemic control because it does not capture hypoglycaemic events. In the future it may be replaced by flash glucose monitoring for some patients, but at present it is the main parameter used to assess glycaemic control and is likely to remain so for most patients.
Twenty-three studies did not report the method with which they calculated variability. This is particularly important with ICCs since there are at least ten different ways of calculating ICC which can all give different results [32].
Most studies reported results in units of % (NGSP method) despite mmol/mol (IFCC method) being recommended. In some studies, it was unclear whether results were expressed using NGSP or IFCC methods, which is important since these methods can give different results.
A bespoke risk of bias tool was developed and included in the original protocol recorded in Prospero, but it was ultimately decided preferable to use a validated risk of bias tool, and the COSMIN tool was selected. This has some limitations, particularly when it comes to the questions regarding blinding. Generally, in studies into biological variability the investigators are not blinded to previous measurements. These questions were therefore omitted when a study’s risk of bias score was assigned. However, this is unlikely to have affected findings as HbA1c is measured objectively.
Risk of bias due to missing publications was deemed to be low, since in most cases the outcome measure of this review was not the primary outcome of the individual studies, and therefore likelihood of publication is unlikely to be related to variability outcomes. Some papers with repeated measures do not report variability data, but this is unlikely to be related to variability outcomes.
As some publications included multiple populations, with information being given only on the subgroups and not the total, arbitrary decisions had to be made as to which populations to include in the primary analysis. However, the rules to decide which group to include were applied systematically, using the following hierarchy: 1. Healthiest population; 2. Most stable population; 3. First listed population. Some studies reported demographic data for the whole population but not for the subgroup for which variability data was reported. In these cases, the demographic data for the whole study population was used and was assumed to be close to the variability subgroup. Although a larger number of studies has been included than in previous reports, the quality of the studies and the other limitations of the review mean that an accurate real-world estimate of within-subject variability of HbA1c remains lacking. Further studies with larger numbers of subjects with a variety of health conditions is important to understand the extent to which HbA1c varies within a subject over time in a clinical setting. This would help inform clinicians when making diagnostic and treatment decisions, and the authors of clinical guidelines when making recommendations on diagnosis and treatment.
Conclusion
This comprehensive systematic review of estimates of HbA1c variability includes data from 111 papers. We provide separate estimates of HbA1c variability in healthy and diabetic populations. We also observe a positive correlation between mean HbA1c and within-subject variability.
Supporting information
S1 File.
https://doi.org/10.1371/journal.pone.0289085.s001
(DOCX)
S1 Checklist.
https://doi.org/10.1371/journal.pone.0289085.s002
(DOCX)
S1 Data.
https://doi.org/10.1371/journal.pone.0289085.s003
(XLSX)
Acknowledgments
Other information
CRD42020202387 Registered 17/09/2020
The review protocol is available at https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42020202387
The protocol describes a risk of bias tool which was adapted from the Quality Assessment of Diagnostic Accuracy Studies (QUADAS tool). The study used the COSMIN tool for risk of bias assessment as being more appropriate and validated.
Citation: Gough A, Sitch A, Ferris E, Marshall T (2023) Within-subject variation of HbA1c: A systematic review and meta-analysis. PLoS ONE 18(8): e0289085. https://doi.org/10.1371/journal.pone.0289085
About the Authors:
Alex Gough
Roles: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing – original draft, Writing – review & editing
E-mail: [email protected]
Affiliation: Institute of Applied Health Research, College of Medical and Dental Sciences, University of Birmingham, Birmingham, United Kingdom
ORICD: https://orcid.org/0000-0001-6954-7408
Alice Sitch
Roles: Conceptualization, Formal analysis, Methodology, Supervision, Writing – review & editing
Affiliation: Institute of Applied Health Research, College of Medical and Dental Sciences, University of Birmingham, Birmingham, United Kingdom
Erica Ferris
Roles: Data curation, Project administration, Writing – review & editing
Affiliation: Institute of Applied Health Research, College of Medical and Dental Sciences, University of Birmingham, Birmingham, United Kingdom
Tom Marshall
Roles: Conceptualization, Formal analysis, Methodology, Supervision, Writing – review & editing
Affiliation: Institute of Applied Health Research, College of Medical and Dental Sciences, University of Birmingham, Birmingham, United Kingdom
ORICD: https://orcid.org/0000-0001-9277-5214
1. Barr RG, N David M, M James B, S Daniel E. Tests of Glycemia for the Diagnosis of Type 2 Diabetes Mellitus. Ann Intern Med. 2002;137(4):263. pmid:12186517
2. American Diabetes Association. 2. Classification and Diagnosis of Diabetes: Standards of Medical Care in Diabetes—2021. Diabetes Care. 2020;44(Supplement_1):S15–S33.
3. Anonymous. Type 2 diabetes in adults: management, NICE guideline [NG28] 2022 [https://www.nice.org.uk/guidance/ng28/chapter/Recommendations.
4. Lenters-Westra E, English E. Evaluating new HbA1c methods for adoption by the IFCC and NGSP reference networks using international quality targets. Clin Chem Lab Med. 2017;55(9):1426–34. pmid:28432844
5. Dhatt GS, Agarwal MM, Bishawi B. HbA1c: A comparison of NGSP with IFCC transformed values. Clinica Chimica Acta. 2005;358(1):81–6. pmid:16018879
6. Lenters-Westra E, Roraas T, Schindhelm RK, Slingerland RJ, Sandberg S. Biological variation of Hemoglobin A1c: Consequences for diagnosing diabetes mellitus. Nederlands Tijdschrift voor Klinische Chemie en Laboratoriumgeneeskunde. 2015;40 (2):104.
7. Little RR, Rohlfing CL. The long and winding road to optimal HbA1c measurement. Clin Chim Acta. 2013;418:63–71. pmid:23318564
8. Raphael A, Friger M, Biderman A. Seasonal variations in HbA1c among type 2 diabetes patients on a semi-arid climate between the years 2005–2015. Prim Care Diabetes. 2021;15(3):502–6. pmid:33309124
9. Fraser Callum G. Biological Variation: From Principles to Practice. USA: AACC Press; 2001.
10. Gonzalez-Lao E, Corte Z, Simon M, Ricos C, Coskun A, Braga F, et al. Systematic review of the biological variation data for diabetes related analytes. Clinica Chimica Acta. 2019;488:61–7. pmid:30389455
11. Anonymous. European Federation of Clinical Chemistry and Laboratory Medicine (EFLM) Biological Variation Database 2022 [https://biologicalvariation.eu/about.
12. Braga F, Dolci A, Mosca A, Panteghini M, Sacco L, editors. Variabilità biologica delŒ emoglobina glicata: un analisi sistematica della letteratura2009.
13. Byron C. Wallace KS, Brodley Carla E., Lau Joseph and Trikalinos Thomas A., editor Deploying an interactive machine learning system in an evidence-based practice center: abstrackr. ACM International Health Informatics Symposium (IHI); 2012.
14. Mokkink LB, Boers M, van der Vleuten CPM, Bouter LM, Alonso J, Patrick DL, et al. COSMIN Risk of Bias tool to assess the quality of studies on reliability or measurement error of outcome measurement instruments: a Delphi study. BMC Med Res Methodol. 2020;20(1):293. pmid:33267819
15. Field AP. Is the meta-analysis of correlation coefficients accurate when population correlations vary? Psychol Methods. 2005;10(4):444–67. pmid:16392999
16. Wan EYF. Association of variability in hemoglobin A1c with cardiovascular diseases and mortality in Chinese patients with type 2 diabetes mellitus—A retrospective population-based cohort study. Journal of Diabetes & its Complications. 2016;30(7):1240–8. pmid:27318537
17. Segar MW, Patel KV, Vaduganathan M, Caughey MC, Butler J, Fonarow GC, et al. Association of Long-term Change and Variability in Glycemia With Risk of Incident Heart Failure Among Patients With Type 2 Diabetes: A Secondary Analysis of the ACCORD Trial. Diabetes care. 2020;15. pmid:32540922
18. Sugawara A, Kawai K, Motohashi S, Saito K, Kodama S, Yachi Y, et al. HbA1c variability and the development of microalbuminuria in type 2 diabetes: Tsukuba Kawai Diabetes Registry 2. Diabetologia. 2012;55(8):2128–31. pmid:22580991
19. Mellergard E, Johnsson P, Eek F. Sociodemographic factors associated with HbA1c variability in type 2 diabetes: a prospective exploratory cohort study. BMC Endocrine Disorders. 2020;20(1). pmid:32641021
20. Raman S, Dai H, DeLurgio SA, Williams DD, Lind M, Patton SR, et al. High hemoglobin A1c variability is associated with early risk of microalbuminuria in children with T1D. Pediatr Diabetes. 2016;17(6):398–406. pmid:26377593
21. Yu ZB, Zhu Y, Li D, Wu MY, Tang ML, Wang JB, et al. Association between visit-to-visit variability of HbA1c and cognitive decline: a pooled analysis of two prospective population-based cohorts. Diabetologia. 2020;63(1):85–94. pmid:31485707
22. Carlsen S, Petersen PH, Skeie S, Skadberg O, Sandberg S. Within-subject biological variation of glucose and HbA(1c) in healthy persons and in type 1 diabetes patients. Clinical Chemistry & Laboratory Medicine. 2011;49(9):1501–7.
23. Maguire A, Mitchell BD, Ruzafa JC. Antihyperglycaemic treatment patterns, observed glycaemic control and determinants of treatment change among patients with type 2 diabetes in the United Kingdom primary care: a retrospective cohort study. BMC Endocr Disord. 2014;14:73. pmid:25163796
24. Hirst JA, Farmer AJ, Ali R, Roberts NW, Stevens RJ. Quantifying the effect of metformin treatment and dose on glycemic control. Diabetes Care. 2012;35(2):446–54. pmid:22275444
25. Ma WY, Li HY, Pei D, Hsia TL, Lu KC, Tsai LY, et al. Variability in hemoglobin A1c predicts all-cause mortality in patients with type 2 diabetes. Journal of Diabetes & its Complications. 2012;26(4):296–300. pmid:22626873
26. Caturano A, Galiero R, Pafundi PC, Cesaro A, Vetrano E, Palmiero G, et al. Does a strict glycemic control during acute coronary syndrome play a cardioprotective effect? Pathophysiology and clinical evidence. Diabetes Res Clin Pract. 2021;178:108959. pmid:34280467
27. Liang L, He H, Zeng Y, Zhang M, Wang X, Li X, et al. Evaluation of biological variation of glycated hemoglobin and glycated albumin in healthy Chinese subjects. Journal of Clinical Laboratory Analysis. 2019;33(3):e22715. pmid:30461058
28. Chiu WC, Lai YR, Cheng BC, Huang CC, Chen JF, Lu CH. HbA1C Variability Is Strongly Associated with Development of Macroalbuminuria in Normal or Microalbuminuria in Patients with Type 2 Diabetes Mellitus: A Six-Year Follow-Up Study. BioMed Research International. 2020;2020 (no pagination). pmid:32047814
29. Lui DTW, Lee CH, Chan YH, Chow WS, Fong CHY, Siu DCW, et al. HbA1c variability, in addition to mean HbA1c, predicts incident hip fractures in Chinese people with type 2 diabetes. Osteoporosis International. 2020. pmid:32385660
30. Mo Y, Zhou J, Ma X, Zhu W, Zhang L, Li J, et al. Haemoglobin A1c variability as an independent correlate of atherosclerosis and cardiovascular disease in Chinese type 2 diabetes. Diabetes & Vascular Disease Research. 2018;15(5):402–8. pmid:29848065
31. Cavagnolli G, Pimentel AL, Freitas PAC, Gross JL, Camargo JL. Effect of ethnicity on HbA1c levels in individuals without diabetes: Systematic review and meta-analysis. PLoS One. 2017;12(2):e0171315–e. pmid:28192447
32. Koo TK, Li MY. A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. J Chiropr Med. 2016;15(2):155–63. pmid:27330520
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
© 2023 Gough et al. This is an open access article distributed under the terms of the Creative Commons Attribution License: http://creativecommons.org/licenses/by/4.0/ (the “License”), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Abstract
Background
Glycosylated haemoglobin (HbA1c) measurement is used to diagnose and to guide treatment of diabetes mellitus. Within-subject variability in measured HbA1c affects its clinical utility and interpretation, but no comprehensive systematic review has described within-subject variability.
Methods
A systematic review and meta-analysis was performed of within-subject variability of HbA1c. Multiple databases were searched from inception to November 2022 for follow-up studies of any design in adults or children, with repeated measures of HbA1c or glycosylated haemoglobin. Title and abstract screening was performed in duplicate, full text screening and data extraction by one reviewer and verified by a second. Risk of bias of included papers was assessed using a modified consensus-based standards for the selection of health measurement Instruments (COSMIN) tool. Intraclass correlation coefficient (ICC) results were pooled with a meta-analysis and coefficient of variation (CV) results were described by median and range.
Results
Of 2675 studies identified, 111 met the inclusion criteria. Twenty-five studies reported variability data in healthy patients, 19 in patients with type 1 diabetes and 59 in patients with type 2 diabetes. Median within-subject coefficient of variation (CV) was 0.070 (IQR 0.034 to .09). For healthy subjects the median CV for HbA1c % was 0.017 (IQR 0.013 to 0.022), for patients with type 1 diabetes 0.084 (IQR 0.067 to 0.89) and for type 2 diabetes 0.083 (IQR 0.06 to 0.10). CV increased with mean population HbA1c.
Limitations
Assessment of variability was not the main aim of many of the included studies and some relevant papers may have been missed. Many included papers had few participants or few repeated measurements.
Conclusions
Within-subject variability of HbA1c is higher for patients with than without diabetes and increases with mean population HbA1c. This may confound observed relationships between HbA1c variability and health outcomes. Because of its importance in clinical decision-making there is a need for better estimates and understanding of factors associated with of HbA1c variability.
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
Details
; Sitch, Alice; Ferris, Erica; Marshall, Tom