# Acceptability Limits for Inter-instrument Variability in Viral Serological Tests on Two Enhanced Chemiluminescence Immunoassay Instruments: Evidence From a Hospital Laboratory-Based Study

## TL;DR

This study determines acceptable limits for differences in viral test results between two lab instruments, suggesting a 5% variability threshold for reliable diagnostics.

## Contribution

Proposes a 5% variability threshold for inter-instrument agreement in eCLIA viral serology tests using Bland-Altman and bootstrapping analyses.

## Key findings

- Strong linear correlations for HCV and HBsAg (r = 0.85), but moderate for HIV (r = 0.64).
- Bland-Altman analysis showed ~95% agreement across all three viral markers despite varying correlation strengths.
- Bootstrapping confirmed ≤5% error across datasets, supporting a 5% variability threshold as acceptable.

## Abstract

Background

Reliable viral serological testing is essential for accurate diagnosis and public health management of HIV, hepatitis B, and hepatitis C. Chemiluminescent immunoassay platforms, particularly enhanced chemiluminescence (eCLIA) systems, are widely used; however, inter-instrument variability remains a challenge in laboratories operating multiple analyzers. Defining acceptable limits of variation is crucial for ensuring diagnostic accuracy, laboratory quality assurance, and clinical decision-making.

Objectives

To evaluate the correlation and agreement between two eCLIA platforms (Vitros ECi and Vitros 3600) in detecting HIV, HBsAg, and HCV, and to establish data-driven acceptability limits for inter-instrument variability.

Methods

An analytical cross‑sectional study was conducted in a diagnostic laboratory of a medical college in Bengaluru, India (January 2019-December 2024). Each day, one patient serum sample was selected at random and tested simultaneously on both eCLIA analyzers. Hemolyzed, lipemic, or insufficient samples were excluded. This random sampling approach, with one sample per day, was considered appropriate for inter‑instrument comparison, as most accreditation bodies recommend testing 1-2 samples only every three months. Results were expressed as S/CO values. Descriptive statistics, Pearson correlation, and Bland-Altman analyses were performed. Jarque-Bera testing assessed normality. Bootstrapping (1000 iterations) was conducted to evaluate the stability of out-of-limit findings and propose acceptable error thresholds.

Results

Of 830 samples processed, the final dataset (after excluding incomplete or zero values) showed high variability across parameters (CV > 50%; HCV > 100%). Pearson correlation demonstrated strong linear relationships for HCV (r = 0.85, R² = 0.72) and HBsAg (r = 0.85, R² = 0.73), but only moderate correlation for HIV (r = 0.64, R² = 0.41). Jarque-Bera statistics (p<0.0001) indicated non-normal distribution, supporting the use of Bland-Altman analysis. Bland-Altman plots revealed approximately 95% agreement for all three markers, with similar proportions of out-of-limit values despite differing correlation strengths. Bootstrapping analysis showed ≤5% error across simulated datasets, suggesting a 5% inter-instrument variability as a reasonable acceptability limit.

Conclusion

Correlation alone is insufficient to assess agreement between CLIA instruments, particularly for non-normally distributed diagnostic data. Bland-Altman analysis combined with bootstrapping provides a more robust approach for evaluating inter-instrument variability. Based on empirical error estimates, an acceptability threshold of ≤5% variability is proposed for laboratories using dual eCLIA platforms for viral serology. These findings support evidence-based quality assurance and instrument interchangeability in clinical diagnostics.