Do Our Two Haematology Analysers Agree? CBC Method Comparison with Bland–Altman and Westgard QC

Many Indian hospital laboratories run two haematology analysers side by side — typically a newer 5-part differential analyser in the main lab and an older 3-part differential analyser used for emergency or night samples. Clinicians assume that a haemoglobin of 9.8 g/dL means the same thing whichever machine produced it. This project tests that assumption.

The two-member team runs 100 EDTA whole-blood samples, selected to cover low, normal and high values, on both analysers within four hours of collection. For haemoglobin, WBC count, RBC count, platelet count, haematocrit and MCV, agreement is assessed with Bland–Altman plots (mean bias and 95% limits of agreement), Passing–Bablok regression (constant and proportional bias) and correlation, and the observed bias is judged against allowable-error limits derived from biological variation. In parallel, both analysers' internal QC is monitored daily with three-level controls on Levey–Jennings charts and interpreted with Westgard multirule logic, so that any disagreement can be traced to calibration drift rather than guessed at.

The design follows the logic of CLSI EP09 (measurement procedure comparison using patient samples) at a scale suitable for a B.Sc MLT project. The deliverable is a short, practical recommendation to the lab: which parameters are interchangeable between the analysers, and which need a flag on the report or a recalibration.

Abstract

A laboratory-based method-comparison study will compare complete blood count (CBC) results from a 5-part and a 3-part differential haematology analyser using 100 paired EDTA samples. Agreement will be evaluated by Bland–Altman analysis, Passing–Bablok regression and Pearson correlation, and bias will be compared with desirable allowable total error based on biological variation. Daily internal QC on both analysers will be charted on Levey–Jennings plots and interpreted with Westgard rules for the duration of the study.

Introduction

Automated analysers use different principles — electrical impedance for RBC and platelets, cyanide-free SLS or cyanmethaemoglobin methods for haemoglobin, and flow cytometry with light scatter for 5-part differentials. Differences in principle, calibration and reagent lots produce systematic differences between instruments. When a patient's serial CBCs come from different machines, a small bias can look like a clinical change: a falling platelet count, a rising haemoglobin.

Literature gap

Published comparisons usually evaluate new models against a reference instrument in large labs. Small and mid-sized Indian hospitals rarely check agreement between their own two analysers after installation, and correlation coefficients are often misused as evidence of agreement. Correlation measures association, not agreement; Bland–Altman analysis is the appropriate method.

Proposed work

• Select 100 samples across the analytical range, run on both analysers within four hours.
• Compute bias, limits of agreement and regression per parameter.
• Compare bias with allowable error; classify each parameter as interchangeable or not.
• Monitor QC daily and document any Westgard rule violations and corrective actions.

Feasibility

• Technical: both analysers are in routine use; no extra instrument is needed.
• Economic: only the reagent cost of re-running 100 samples on the second analyser (roughly ₹3,000–5,000) and QC material already bought by the lab.
• Time: 100 samples can be collected in four to five weeks.
• Ethical: anonymised residual samples after routine reporting; IEC waiver of consent.

The study laboratory reports CBCs from two different haematology analysers depending on the time of day and workload. Patients admitted to wards often have serial CBCs run on both machines, and clinicians compare the values directly to monitor anaemia, infection or thrombocytopenia. No documented comparison of the two analysers has been done since installation, and the lab relies on each analyser's internal QC being "within range" without checking whether the two instruments agree with each other on patient samples.

If a systematic bias exists — for example, platelet counts consistently lower on the 3-part analyser — a patient may appear to deteriorate or improve simply because the sample went to a different machine. The laboratory needs a documented, statistically sound comparison of the two analysers on real patient samples, interpreted against accepted quality specifications, together with a record of QC performance, so that it can decide which parameters are interchangeable.

In scope

• Adult and paediatric EDTA samples received for routine CBC, run on both analysers within four hours.
• Six parameters: Hb, WBC, RBC, platelets, HCT, MCV.
• Daily internal QC for the study period.
• Statistical agreement analysis and a practical recommendation.

Out of scope

• Differential leucocyte count comparison (the 3-part analyser gives only three populations; comparison with manual differential would be a separate study).
• Comparison with a reference method such as manual haematocrit or haemocytometer counts for every sample.
• Linearity, carry-over and precision studies beyond daily QC.
• External quality assessment scheme data.

Study design: laboratory-based analytical method-comparison study (paired design), following the principles of CLSI EP09.

Setting: central clinical laboratory, haematology section, of a 300-bed multispecialty hospital (fictional: Kaveri Multispeciality Hospital), five weeks of sample collection within a 12-week project.

Sample size: CLSI EP09 recommends at least 40 patient samples covering the reportable range for a method comparison, and more for better precision of limits of agreement. We use 100 samples, which narrows the 95% confidence interval of each limit of agreement to about ±0.34 SD of the differences (Bland and Altman's approximation, √(3s²/n)).

Inclusion criteria: EDTA (K2/K3) whole-blood samples with adequate volume (≥ 1 mL), analysed on the first analyser within two hours of collection; deliberate selection so that about 20% of samples have low and 20% high values for Hb and platelets.

Exclusion criteria: clotted, haemolysed or lipaemic samples; samples with platelet clumps on smear; samples older than four hours; analyser flags for cold agglutinins or interference.

Procedure

1. Run QC (three levels) on both analysers each morning; plot on Levey–Jennings charts using the lab's own mean and SD from the previous 20 runs.
2. After routine reporting on analyser A, mix the sample by gentle inversion (8–10 times) and run it on analyser B within the four-hour window.
3. Record both results in a coded Excel sheet (study number only).
4. Repeat any sample with an analyser flag; exclude it if the flag persists.

Statistical analysis: Excel/SPSS for descriptive statistics; Bland–Altman plots (difference vs mean, absolute or percentage difference as appropriate); Passing–Bablok regression with 95% CI of slope and intercept (MedCalc or an Excel template — an industry-standard extra); paired t-test or Wilcoxon signed-rank test for mean differences; Pearson r reported only as a measure of range adequacy. Bias judged against desirable total allowable error from the EFLM biological-variation database.

Ethics: IEC approval with waiver of consent — residual anonymised samples, no additional blood drawn.