A statistical basis for harmonization of thyroid stimulating hormone immunoassays using a robust factor analysis...

Area of Science:

• Clinical Chemistry
• Biostatistics
• Immunoassay Technology

Background:

• Achieving between-method equivalence for thyroid stimulating hormone (TSH) measurements typically relies on SI-traceable reference procedures.
• Mid-term achievement of SI-traceability for TSH measurement is unlikely, necessitating alternative harmonization strategies.

Purpose of the Study:

• To investigate a statistical alternative for TSH measurement harmonization using a factor analysis (FA) model.
• To evaluate the efficacy of a robust alternating regressions (RAR) method for harmonizing TSH immunoassay data.

Main Methods:

• Applied a factor analysis (FA) model to TSH results from 94 samples across 14 immunoassays.
• Utilized a robust alternating regressions (RAR) method to address non-homogeneous data, outliers, and missing values in the FA model.
• Standardized the dataset prior to FA model application due to deviations from elliptical symmetry assumptions.

Main Results:

• The RAR method successfully estimated key FA model quantities despite data complexities.
• Mathematical recalibration using the FA model eliminated method-specific differences in TSH measurements.
• Improved between-method coefficient of variation (CV) for TSH from 11% to 6%.

Conclusions:

• Robust alternating regressions (RAR) is suitable for harmonizing multimethod TSH comparison datasets.
• The FA model, when applied with RAR, effectively addresses challenges posed by outliers and missing values.
• This statistical approach offers a viable alternative for achieving TSH measurement harmonization.