Transference Validation of Clinically Utilized Immunoassay Reference Intervals: Evidence From Thyroid...

Thyroid test standardization is hindered by interplatform variability, especially for immunoassays lacking metrologic traceability. Direct method-derived reference intervals (RIs) are gold standard but resource intensive. RI transference allows cost-effective multiplatform RI establishment but necessitates balanced consideration of statistical relevance, precision, and clinical feasibility.

To validate an RI transference framework for harmonizing RIs across platforms, using thyroid function testing as a model for assays with inconsistent reference standards.

Dual-platform framework (Roche Cobas e601 versus Beckman DxI 800) using euthyroid community midrange specimens (n = 2423) and pathology-enriched clinical leftovers (n = 300). Weighted Deming regression (constant coefficient of variation [CV]: thyroid-stimulating hormone [TSH], free thyroxine [FT4], total triiodothyronine [TT3]) and standard Deming regression (constant standard deviation: free triiodothyronine [FT3], total thyroxine [TT4]) were applied. Precision was evaluated via multiday temporal stability and concentration-dependent Monte Carlo sample size optimization. Spearman ρ, CV(b), and absolute value of bias ratio (|BR|) were benchmarked against direct RIs.

To explore multiday sampling, weighted Deming enabled accurate TSH transfer [CV(b) = 2.07%-2.89%] and acceptable precision for FT4/TT3/TT4 [CV(b) = 5.97%-10.82%]. Randomized multiday sampling reduced extreme |BR| outliers by 51% to 89%. To compare concentration-driven data sources, community-derived specimens showed superior alignment with direct RIs (FT3: |BR| = 0.581 versus 1.161 in residuals), despite weaker correlations (ρ = 0.33 versus 0.92) at n = 40. When using community-derived specimens, FT3 exhibited persistent challenges [ρ = 0.33-0.36; CV(b) = 3.07%-23.44%] attributable to methodologic heterogeneity. To determine the adaptive sample size, optimization suggested that 45 prestratified residuals with bootstrap resampling (100 × 40 pairs) improved efficiency but introduced pathologic bias (FT3 |BR| = 1.161).

This study established a comprehensive thyroid testing RI transference framework integrating multiday sampling, adaptive sample size, and concentration-driven data selection, identifying optimal solutions to enable cross-platform RI establishment for immunoassays. Resource-limited laboratories can further leverage multicenter direct RIs via transference to balance cost and clinical validity.