Guest Essay
HbA1c on the Sigma Scale: What Passing Proficiency Testing Did Not Tell Us
What happens when you can pass Proficiency Testing, but your analytical Sigma metric still says your assay is not meeting the goal? A Tale of Two Expectations, this time focusing on HbA1c.
HbA1c on the Sigma Scale: What Passing Proficiency Testing Did Not Tell Us
November 2026
Jagdish Chandarana
Founder and Managing Director, Vedant Diagnostic Center, Rajkot, India
Why we looked
HbA1c is one of the tests a laboratory most needs to trust. A result of 6.5% can establish a diagnosis of diabetes, and 7.0% is a common treatment target, so a few tenths of a percentage point carry clinical weight. Our laboratory had every routine reason for confidence: an NGSP-certified HPLC method, NABL accreditation under ISO 15189, and external quality assessment (EQAS) results with every z-score inside ±1. We wanted to know whether that confidence held up on the Sigma scale.
The setting
Vedant Diagnostic Center is a stand-alone ambulatory laboratory in Gujarat, India. HbA1c is measured on a Bio-Rad D-10 HPLC analyzer. We analysed eight consecutive months of internal QC (October 2025 to May 2026): 208 results at each of two control levels, normal (pooled mean 5.03%) and high (9.85%), all from a single control lot. Bias was estimated two ways: from three EQAS cycles against the peer-group mean, and from the deviation of our pooled QC mean from the assigned target. We calculated Sigma as (TEa − |bias|)/CV and the Quality Goal Index (QGI) as bias/(1.5 × CV).
What the numbers showed
Against the 6% TEa used by CAP and NGSP, with an EQAS-derived bias of 2.37%, Sigma was 1.19 at the normal level and 2.70 at the high level. Using the QC-target bias instead, the values were 1.46 and 4.22. Relaxing TEa to 8% raised the normal level only to 1.85–2.11. Whichever combination we chose, the normal control never reached 3 Sigma.
Meanwhile, the three EQAS deviations from the peer mean were −1.56%, +4.10% and −1.46%, all acceptable. The laboratory was passing its external assessment while performing below 2 Sigma at the lower end of the range.
This is not a contradiction. EQAS grading limits are wider than what the Sigma framework demands, and an EQAS sample is a snapshot, while eight months of QC data describe the process. Proficiency testing told us we were not grossly wrong. It did not tell us we were good enough.
Three choices that change the verdict
Anyone repeating this exercise will meet three decisions, and each one moves the result.
TEa. At 6%, 7% and 8%, the normal-level Sigma ranged from 1.19 to 1.85 (EQAS bias), and the high level from 2.70 to 4.19. We reported all three rather than choose the most flattering.
Bias source. Three EQAS cycles gave 2.37%; the QC-target comparison gave 1.58% and 0.33% at the two levels. Neither is a reference measurement procedure. The EQAS samples clustered at lower concentrations, and QC target values carry commutability uncertainty. We reported both.
CV. Our main estimate was the long-term CV pooled across all eight months: 3.04% at the normal level and 1.34% at the high level. The pooled within-month CV was lower, at 2.30% and 1.24%. Using it lifted the normal level only to 1.57 Sigma, and the conclusion did not change. Because all data came from one control lot, between-lot variation is not included in either estimate.
In our view the long-term CV is the honest one, because it is the imprecision patients actually experience across months. The gap between the two estimates turned out to be the most useful finding in the study.
Where the error lived
QGI at the normal level was 0.52 (EQAS bias) and 0.35 (QC-target bias), both well below 0.8. The problem was imprecision, not bias. The difference between long-term and within-month CV showed where that imprecision came from: the monthly control mean itself moved, between 4.90% and 5.20% at the normal level, while monthly CV ranged from 1.46% to 2.77%. Scatter within a month was only part of the story. Month-to-month instability inflated the long-term CV.
That is a different problem from a noisy analyzer, and it points to different actions: calibration, reagent cartridge lots, column condition and maintenance, all of which change between months.
What QC rules can and cannot do
The usual instinct is to tighten the control rules. On a process below 2 Sigma, no practical control procedure can reliably detect medically important errors without unacceptable false rejections. The Sigma metric tells you, before you redesign your QC, that the process itself needs fixing.
We therefore did not propose level-specific rules. Most middleware supports one control procedure per test, so we recommended a single procedure selected on the poorer-performing level, and deferred Sigma-based rule redesign until imprecision improves. The corrective plan we proposed targets the sources of between-month drift:
- calibration verification at defined intervals, not only at calibrator lot changes;
- formal comparison of each new reagent cartridge lot against the lot in use;
- preventive maintenance to schedule, with documented column performance monitoring;
- review of the monthly control mean as a drift indicator in its own right
Five lessons for laboratories
- Passing EQA is not a Sigma assessment. Calculate Sigma from your own QC data, even when proficiency results look clean.
- Report your choices. State the TEa, the bias source and the CV estimate, and show how the result moves when each one changes.
- Use long-term CV, and compare it with within-month CV. A large gap means the process drifts between months, and the fix lies in calibration, lots and maintenance rather than in the QC rules.
- Let QGI direct the effort. Ours pointed away from bias adjustment and towards precision.
- Track the monthly control mean. It is already in your QC data, costs nothing to review, and is an early signal of the drift that inflates long-term imprecision.
Closing thoughts
HbA1c is a demanding test for Sigma: the TEa is tight and the decision limits are close together. A low Sigma value is not a reason to stop calculating it. It is the reason to calculate it. Our EQAS results said the laboratory was acceptable. The Sigma metric showed where it was not, and the QGI and the long-term versus within-month comparison showed what to fix. For a medium-volume laboratory without a dedicated quality department, that sequence needs nothing beyond the QC data already being collected.
This essay draws on work carried out with Dr. Kinjal Butani, Laboratory Head at Vedant Diagnostic Center.
Reference
Chandarana J, Butani K. Sigma metric and Quality Goal Index assessment of hemoglobin A1c analytical performance in an ambulatory accredited laboratory. Lab Med. 2026;57(5):lmag052. doi:10.1093/labmed/lmag052
About the author
Jagdish Chandarana is Founder and Managing Director of Vedant Diagnostic Center, an NABL/ISO 15189-accredited laboratory in Rajkot, Gujarat, India, with about 28 years in diagnostics. His published work covers Sigma-metric and risk-based quality management across the pre-analytical, analytical and post-analytical phases. ORCID: 0009-0002-8796-6432.