Substance and claimed effects

HbA1c (glycated hemoglobin) is the standard laboratory measure of chronic glycemia: the proportion of circulating hemoglobin that has irreversibly glycated through non-enzymatic reaction with ambient glucose, integrated over the ~120-day average lifespan of the red blood cell Sacks et al. 2011. It is the diagnostic threshold for diabetes (≥6.5%) and prediabetes (5.7–6.4%) per the American Diabetes Association ADA 2024, and the dominant treatment target. This entry covers the conditions and physiologies under which the HbA1c value diverges from a patient's actual mean glucose — iron-deficiency and hemolytic anemias, hemoglobinopathies (HbS, HbC, HbE, HbF), recent transfusion, erythropoiesis-stimulating agents, advanced chronic kidney disease, late pregnancy, splenectomy, and inter-individual variation in red-cell lifespan and intracellular glycation kinetics — and the alternative measures (fructosamine, glycated albumin, CGM-derived Glucose Management Indicator and Time in Range) that should replace it when it can’t be trusted. The consequence space is narrow but high-stakes: missed diabetes diagnosis, overtreated patients driven into hypoglycemia, and untreated patients walking around with mean glucose well above their lab number suggests.

Evidence by addressing question

Mechanism

HbA1c measures the fraction of adult hemoglobin (HbA) carrying a glucose adduct on the N-terminal valine of the β-chain. The reaction is non-enzymatic, irreversible, and proportional to time-averaged glucose exposure inside the red cell. The reported percentage is therefore the product of three factors: (1) mean intracellular glucose concentration, (2) red-cell lifespan (longer-lived cells accumulate more adduct), and (3) the intrinsic rate of glycation, which varies across individuals by ~25% even at identical glucose Cohen et al. 2008, Khera et al. 2015. Any perturbation of factor (2) or (3) decouples HbA1c from mean glucose. Shortened RBC lifespan (hemolysis, blood loss, ESA-driven erythropoiesis, late pregnancy hemodilution and turnover) lowers HbA1c at any given glucose; lengthened lifespan (splenectomy, aplastic anemia, iron deficiency — older cells persist) raises it.

Hemoglobin variants attack the system at a different layer. The assay itself — HPLC, immunoassay, capillary electrophoresis, boronate affinity — can mistake variant peaks for the HbA1c peak (or fail to detect them), producing falsely high or falsely low results that have nothing to do with the patient’s glucose Bry et al. 2001, NGSP 2023. The direction and magnitude depend on which variant and which assay; the NGSP maintains a current table by manufacturer because the picture changes as platforms update.

Fructosamine and glycated albumin measure glycation of serum proteins (~2–3 week half-life), bypassing the red cell entirely; CGM measures glucose directly in interstitial fluid every 1–15 minutes and computes GMI (Glucose Management Indicator) from a regression equation against mean glucose Bergenstal et al. 2018, plus Time in Range (% of readings in 70–180 mg/dL).

Evidence

Iron-deficiency anemia shifts HbA1c upward by a clinically meaningful margin in non-diabetics. Coban et al. found mean HbA1c ~0.9 percentage points higher in iron-deficient women than matched controls, with normalisation after iron repletion Coban et al. 2004. The NHANES analysis by Kim et al. showed iron deficiency in women without diabetes was associated with a ~0.16 percentage point shift in HbA1c at the population level, and increased the odds of crossing the diagnostic 5.7% prediabetes threshold Kim et al. 2010 — small absolute effect, but large enough to flip diagnostic categories at the margin.

Hemolytic anemias and any shortened-RBC state — hereditary spherocytosis, autoimmune hemolysis, G6PD deficiency under oxidant stress, paroxysmal nocturnal hemoglobinuria, recent significant blood loss, erythropoiesis-stimulating agents in CKD — lower HbA1c by reducing average cell age. Effect sizes vary widely; the NGSP and ADA both flag these as situations where HbA1c should not be used and an alternative (fructosamine, glycated albumin, or CGM) substituted NGSP 2023, ADA 2024.

Hemoglobinopathies. Sickle cell trait (HbAS) is carried by ~8–10% of African Americans. Lacy et al. analysed 4,620 African Americans across cohort studies and found HbA1c was ~0.3 percentage points lower in HbAS carriers than non-carriers at the same fasting glucose, and ~0.4 lower at the same 2-hour glucose, with a corresponding under-diagnosis of prediabetes and diabetes Lacy et al. 2017. HbC, HbE, HbD trait produce variant-and-assay-dependent shifts; HbF elevations (thalassemias, hereditary persistence of fetal hemoglobin) and HbSS (sickle disease) make HbA1c uninterpretable on most platforms Bry et al. 2001, NGSP 2023.

Recent transfusion. Transfused units carry donor hemoglobin with donor glycation history. Spencer et al. studied diabetic patients receiving red-cell transfusions and showed HbA1c dropped by an average of ~1.3 percentage points within days post-transfusion — an immediate, transient artefact unrelated to the patient’s actual glycemia Spencer et al. 2011. Effect lingers for ~2–3 months as transfused cells turn over.

Pregnancy. Nielsen et al. measured HbA1c longitudinally in 651 pregnancies (diabetic and non-diabetic) and showed HbA1c falls progressively from baseline through the second and third trimesters, by ~0.5 percentage points in non-diabetics, driven by increased red-cell turnover, hemodilution, and shortened cell lifespan Nielsen et al. 2004. ADA guidelines explicitly recommend that gestational diabetes diagnosis and pregnancy glycemic monitoring not rely on HbA1c after the first trimester — use OGTT for diagnosis, fingerstick or CGM for management ADA 2024.

Chronic kidney disease. Multiple factors collide: shortened RBC survival from uremia, anemia of CKD, ESA therapy, iron supplementation cycles, and carbamylated hemoglobin from urea cross-reacting on some assays. Lo et al. analysed paired CGM and HbA1c in 90 patients with T2D and stages 3–5 CKD and showed HbA1c consistently underestimated mean glucose, with the bias growing as eGFR fell — HbA1c of 7.0% in CKD-5D corresponded to a true mean glucose roughly 30 mg/dL higher than the same HbA1c in CKD-3 Lo et al. 2014. Inaba et al. showed glycated albumin tracked mean glucose far better than HbA1c in hemodialysis patients, especially when ESA dosing changed Inaba et al. 2007.

Inter-individual variation at the same mean glucose. The ADAG study (a key piece of evidence behind the HbA1c-to-eAG conversion) showed that even after controlling for confounders, individuals with the same mean glucose could differ in HbA1c by ~1.0–1.5 percentage points Nathan et al. 2008. Hempe et al. formalised this as the hemoglobin glycation index (HGI) — the residual after regressing HbA1c on mean glucose — and showed in the ACCORD trial that high-HGI patients had worse outcomes from intensive treatment, suggesting the discordance is biologically meaningful and not just measurement noise Hempe et al. 2015. Beck et al. quantified this directly using CGM-derived mean glucose paired with lab HbA1c in 387 T1D patients: at any single mean glucose, the HbA1c range spanned roughly 1.5 percentage points — meaning two patients with identical day-to-day glucose can land on opposite sides of a treatment threshold Beck and Hirsch 2017, Beck et al. 2019.

Race-stratified differences without underlying glycemia differences. Herman et al. in the Diabetes Prevention Program cohort found Black participants had HbA1c ~0.4 percentage points higher than white participants at the same fasting and 2-hour glucose Herman et al. 2007. Whether this reflects assay artefact, genuine glycation-rate differences, RBC-lifespan differences, or unmeasured confounding remains contested — but the practical implication is that race-blind HbA1c screening overdiagnoses Black patients at the margin and underdiagnoses Asian patients (whose HbA1c tends to read lower at the same glucose).

Protocol

The clinical workflow when HbA1c is suspect:

• Flag the condition. Any of: iron deficiency (low ferritin, low transferrin saturation), hemolysis indicators (high LDH, low haptoglobin, elevated reticulocytes), known hemoglobinopathy or family origin in high-prevalence regions, transfusion in the prior 3 months, second or third trimester pregnancy, eGFR <30 mL/min/1.73 m², or splenectomy.
• Confirm with a glucose-based test. Fasting plasma glucose ≥126 mg/dL or 2-hour OGTT ≥200 mg/dL remains diagnostic regardless of HbA1c reliability ADA 2024.
• Switch to an alternative for monitoring. Fructosamine (or glycated albumin where available) integrates glucose over ~2–3 weeks via serum protein glycation, independent of RBC physiology Wright and Hirsch 2012. CGM gives direct glucose with Time in Range and GMI — the cleanest substitute and now widely available, including for type 2 diabetes Battelino et al. 2019.
• Interpret GMI vs lab HbA1c discordance. When CGM-derived GMI and lab HbA1c disagree by >0.5 percentage points, that gap is itself the signal — one of the factors above is in play, and treatment decisions should anchor on GMI/TIR Bergenstal et al. 2018.

Contraindications

Using HbA1c as the primary glycemic measure is inappropriate — per ADA and NGSP — in: any hemolytic anemia, sickle cell disease, beta-thalassemia major, end-stage kidney disease (especially on hemodialysis with ESA therapy), the second and third trimester of pregnancy, recent (within 2–3 months) red-cell transfusion, recent significant blood loss, and on assays that don’t handle the patient’s specific hemoglobin variant ADA 2024, NGSP 2023, Sacks et al. 2011. The assay-specific caveat matters: a patient with HbE trait can get a valid result on one platform and a wildly wrong result on another — the ordering clinician usually doesn’t know which platform the lab runs.

Misconceptions

The dominant misconception is that HbA1c directly equals mean glucose. It is a proxy with two layers of variability (RBC lifespan and intrinsic glycation rate) on top of mean glucose. The Beck et al. analysis is the cleanest demonstration: at a CGM-measured mean glucose of 154 mg/dL, lab HbA1c spans 6.0%–8.0% across individuals Beck and Hirsch 2017. The corollary misconception is that two patients with the same HbA1c have the same glucose control — they don’t, because the variability cuts both ways.

A second misconception: HbA1c targets (e.g., <7%) are evidence-based individual targets. The ACCORD trial result — that intensive lowering was associated with excess mortality in high-HGI patients — suggests the same numeric target is not equally appropriate for everyone Hempe et al. 2015. Glycemic variability and time in hypoglycemia (captured by CGM but invisible to HbA1c) are independently associated with complications and quality of life Hirsch 2015.

Alternatives

Fructosamine reflects mean glucose over the prior 2–3 weeks via glycation of serum albumin and other proteins. Cheap, widely available on standard chemistry panels, but affected by anything that changes albumin turnover — nephrotic syndrome, hyperthyroidism, liver disease, severe protein malnutrition Wright and Hirsch 2012, Welsh, Kirkman, Sacks 2016. Not a diagnostic test for diabetes (no ADA threshold), but useful for tracking response to therapy when HbA1c is unreliable.

Glycated albumin is the more specific assay for albumin glycation, less affected by albumin concentration than fructosamine. Strong performance in dialysis patients Inaba et al. 2007. Availability is the constraint — not on most US lab menus.

CGM with Time in Range and GMI is the modern reference. The international consensus targets are TIR >70% (in 70–180 mg/dL) for most adults with diabetes, with explicit subtargets for time below range (<4% <70 mg/dL, <1% <54 mg/dL) Battelino et al. 2019. Beck et al. showed Time in Range correlates strongly with mean glucose (and therefore with development of retinopathy and microalbuminuria in DCCT data) Beck et al. 2019. GMI uses the regression equation GMI(%) = 3.31 + 0.02392 × mean glucose (mg/dL) derived from 528 patient-weeks of paired CGM and HbA1c Bergenstal et al. 2018.

Fasting plasma glucose and OGTT remain the diagnostic anchors when HbA1c is unreliable — they measure glucose directly, no proxy layer ADA 2024.

Failure modes

Three high-stakes clinical failure modes recur in the literature:

• Overtreatment. An older patient with CKD-4 has falsely elevated apparent control; their HbA1c reads 7.5% but mean glucose is in the 180s; titration to a lower HbA1c drives sulfonylurea- or insulin-mediated hypoglycemia. The Lo et al. data suggests this is common Lo et al. 2014.
• Undertreatment. A patient with iron-deficiency anemia and a falsely elevated HbA1c reads 6.8%; mean glucose is normal; they receive a diabetes diagnosis and metformin they don’t need Coban et al. 2004. Conversely, a sickle-trait carrier reads 6.0% while mean glucose is 170 mg/dL; diabetes is missed until complications surface Lacy et al. 2017.
• Missed glycemic variability. A T1D patient with TIR of 50% but matched hyper- and hypoglycemic excursions can have an HbA1c identical to a patient sitting steadily at the upper edge of range. The first patient has substantially higher symptom burden, hypoglycemia risk, and likely complication trajectory; HbA1c sees neither Hirsch 2015, Beck and Hirsch 2017.

Audience

Three populations carry concentrated risk of HbA1c error:

• People with sub-Saharan African, Mediterranean, South / Southeast Asian, or Middle Eastern ancestry — carrier frequencies for hemoglobinopathies are elevated. African Americans carry HbS trait at 8–10%; HbC trait at ~3%; HbE trait is common across Southeast Asia; β-thalassemia trait is common around the Mediterranean and South Asia. Most carriers are clinically silent, only revealed by an electrophoresis or a discordant HbA1c Lacy et al. 2017, NGSP 2023.
• Women of reproductive age — iron deficiency is the most common cause of anemia globally and disproportionately affects menstruating women. HbA1c can be biased upward by ~0.3–1.0 percentage points Coban et al. 2004, Kim et al. 2010.
• Patients with advanced CKD — the bias is dominant and clinically dangerous; CGM or glycated albumin should drive therapy decisions Lo et al. 2014, Inaba et al. 2007.

Practicalities

Fructosamine costs roughly the same as HbA1c on a standard chemistry panel and is broadly available. Glycated albumin is more specific but not on most US lab menus; commonly used in Japan and parts of Asia. Hemoglobin variant analysis (electrophoresis or HPLC) costs $50–200 and is one-time. CGM systems — Dexcom G7, Abbott FreeStyle Libre 3 — are now FDA-cleared for over-the-counter sale in the US (Stelo, Lingo, Libre Rio launched 2024), but most insurance coverage is still tied to a diabetes diagnosis with insulin therapy or documented hypoglycemia. Out-of-pocket cost ~$50–100 per 14-day sensor.

History

HbA1c emerged as a clinical assay in the 1970s after Rahbar identified the elevated glycohemoglobin fraction in diabetics in 1969. The DCCT trial (1993) and UKPDS (1998) established it as the dominant treatment target by tying it to microvascular complication rates. The ADA added it as a diagnostic test in 2010 alongside fasting glucose and OGTT. The NGSP, established in 1996, standardised assay performance against the DCCT reference method — this is what makes lab-to-lab HbA1c comparable. The CGM-derived metrics (Time in Range, GMI) were formalised by international consensus only in 2019, reflecting how recent the alternatives are despite the underlying biology being understood for decades.

Out of scope

This entry covers when HbA1c misleads. The broader question of what mean glucose target is right for whom belongs in a separate entry on glycemic targets; the question of which CGM device to use and how to interpret a daily glucose pattern belongs in a separate entry on CGM as a wellness tool; the question of screening cadence for diabetes belongs in a separate entry on diabetes screening.

Credibility range

Optimist case

The optimist position is the establishment one: HbA1c is the most-validated glycemic biomarker in medicine, standardised globally by NGSP, anchored to hard outcomes (retinopathy, nephropathy, cardiovascular events) in DCCT, UKPDS, ACCORD, and dozens of cohort studies. The interference conditions are well-catalogued by NGSP and ADA. Clinical workflow already accounts for them: any decent endocrinologist orders fructosamine or CGM when CKD or anemia is in play. For the >90% of patients without these conditions, HbA1c is cheap, reproducible across labs, and tracks complications. Switching wholesale to CGM-derived metrics introduces new problems: device variability, regression-equation drift (GMI has its own ~0.5pp error vs lab HbA1c), and outcome data tied to HbA1c doesn’t cleanly transfer to TIR thresholds.

Skeptic case

The skeptic position: the field has known about HbA1c’s limitations for decades and continues to use it as if it were ground truth. The Beck et al. analysis — ~1.5 percentage point spread at any given mean glucose Beck and Hirsch 2017 — means a substantial fraction of treatment decisions are being made on a number whose signal-to-noise ratio is worse than commonly assumed. ACCORD’s mortality signal in high-HGI patients Hempe et al. 2015 is direct evidence that intensive HbA1c lowering harms a subset who were never going to benefit from it. Race-stratified differences Herman et al. 2007 create systematic over- and under-diagnosis. The continued reliance on HbA1c as the primary screening test in non-CKD, non-pregnant adults is at least partly inertia — CGM-derived metrics are a better measure of what we actually care about (mean glucose, time in dangerous ranges, glycemic variability) and the data is now available cheaply.

Author’s call

HbA1c is a useful tool for most adults most of the time, and the alternatives are not free. But the conditions that bias it are common enough — iron deficiency in menstruating women, sickle trait in African Americans, CKD in older adults, hemoglobinopathies broadly across non-European-ancestry populations — that “ask, then test” is the right default. For people in any of those groups, or whose CGM-derived GMI and lab HbA1c diverge by >0.5 percentage points, or who have unexplained discordance between fasting glucose and HbA1c, the lab number should be treated as one data point, not the answer. The evidence supporting the limitations is strong (evidence: 4 — well-replicated, guideline-recognised), the controversy is moderate (controversy: 2 — specialists agree, primary care often defaults to HbA1c alone), and the consequence for the right patient (avoiding overtreatment-driven hypoglycemia, catching missed diabetes) is meaningful.

Stakeholder and incentive map

• NGSP, ADA, AACC. Standards bodies pushing for both HbA1c standardisation and explicit acknowledgement of its limits. Aligned with the patient interest; no commercial conflict.
• CGM manufacturers (Dexcom, Abbott). Direct commercial interest in expanding from T1D / insulin-using T2D to broader markets. Driving the Time in Range and GMI push. Generally produce solid science (Beck, Battelino) but the framing emphasises CGM’s strengths.
• Labs and primary care. HbA1c is cheap, fast, and a single number; alternatives require either more expensive assays or a CGM device. Inertia favours HbA1c.
• Hematology and nephrology specialists. Most attuned to interference conditions because they see them constantly; least likely to over-rely on HbA1c.
• Patients with hemoglobinopathies. Often unaware their HbA1c is unreliable. No advocacy infrastructure equivalent to the diabetes patient communities.

Population variability

The effect is most pronounced — and the failure modes most dangerous — in:

• African ancestry: 8–10% HbS trait, ~3% HbC, with assay-specific direction Lacy et al. 2017. Race-coded HbA1c differences at matched glucose may reflect a mix of these.
• Mediterranean / South Asian / Southeast Asian ancestry: β-thalassemia trait, HbE trait, HbD — carrier rates regionally 5–30%.
• Menstruating women: ~10–15% have iron deficiency at any time; the upward bias is small (0.3–1.0pp) but can flip prediabetes labels.
• People with CKD (eGFR <60): bias grows with stage; large in dialysis Lo et al. 2014.
• Pregnant women: after first trimester, HbA1c systematically underreads Nielsen et al. 2004.
• Anyone with intrinsically slow- or fast-glycating RBCs: Hempe’s HGI work suggests a substantial fraction of the general population sits 1–2 percentage points off the mean-glucose regression line for reasons not yet fully explained Hempe et al. 2015.

Knowledge gaps

• What drives intrinsic glycation-rate variability? Genome-wide association studies have identified loci near HK1, G6PC2, and others, but the picture is incomplete. Why a healthy non-diabetic with average glucose reads 5.9% while their sibling at the same glucose reads 5.2% is not fully explained.
• Are CGM-derived endpoints prognostic for the same outcomes HbA1c is? TIR is associated with retinopathy in retrospective DCCT analyses, but prospective trials with TIR as primary endpoint are only now reading out. Until then, the case for switching from HbA1c rests partly on inference.
• Glycated albumin in non-CKD populations. Solid evidence in dialysis; thin elsewhere, especially in primary-care screening contexts.
• Race-stratified targets. If race differences in HbA1c at matched glucose are real and biological, current screening thresholds are systematically miscalibrated for non-white populations — but adjusting thresholds by race is itself contested.

Scoping. The brief named six interfering conditions (anemia, hemoglobinopathies, transfusion, pregnancy, CKD, altered RBC turnover) plus the alternatives (fructosamine, CGM-derived metrics). All are covered in both the dossier and the article. The article additionally surfaces inter-individual glycation-rate variability (Cohen 2008, Beck 2017, Hempe 2015) and race-stratified differences (Herman 2007, Lacy 2017) because they share the same mechanism family and the same practical implication — the lab number diverges from mean glucose for biological reasons readers can do something about. No silent narrowing.

Hard scoping calls. Glycated albumin was kept as a brief mention in alternatives rather than its own treatment because availability in the US (the dominant reader market) is limited; in a future translation it deserves more weight for Japan / parts of Asia where it's standard. 1,5-anhydroglucitol was excluded entirely — niche, mostly used in research, and the dossier reasonably ends at three alternatives without it.

Rating difficulties. longevity and health_short_term were each scored at 2 rather than higher. The argument for higher: getting the right diagnosis affects decades of management. The argument for 2: the effect is conditional — for the majority of readers who aren't in an affected group, the entry is informational only. The 2 reflects the holistic effect across the whole reader base, with the per-affected-group impact carried in the justification. controversy at 2 captures the gap between specialist consensus (the limitations are real) and primary-care practice (HbA1c often still used alone). Could argue for 3 if the focus is on race-stratified thresholds, which remain genuinely contested.

Contraindications field. Deliberately left empty. The closed vocabulary doesn't have a token for "this is an informational entry, not an intervention" and none of the existing tokens apply to reading about a test's limits. The article-body contraindications section captures when the test is contraindicated as a clinical instrument — different concept from the structural-meta contraindications field.

Future-link candidates. Adjacent entries this should cross-link to once they exist: glycemic targets (personalised A1c / TIR thresholds), CGM-as-wellness-tool (using a patch without diabetes), diabetes screening cadence, prediabetes management. The out-of-scope section foreshadows all four.

Separate-entry candidates. "Time in Range as a diabetes outcome" is large enough on its own — Battelino et al. 2019 plus Beck et al. 2019/2020 plus the regulatory shift toward TIR as a trial endpoint warrant a standalone entry. Same for "race-stratified diagnostic thresholds in medicine," which surfaces here through Herman 2007 and Lacy 2017 but generalises beyond glycemia.

Audience field. Deliberately left absent (= applies to everyone). The conditions cluster in specific populations (women, African / Mediterranean / Asian ancestry, CKD patients), but the relevance of the entry — knowing whether your test is biased — applies to every adult who has ever had blood sugar measured. The article's audience section does the population-specific work; the meta field would over-narrow reach.