Introduction: Thyroid dysfunction is associated with several diseases and mortality. Due to the within-individual variation of TSH and free T(4) (FT(4)) levels, the association between thyroid function and disease or mortality rates might be underestimated in studies based on one single measurement of TSH and/or FT(4). This kind of bias is called regression dilution bias. The aim of this study was to examine the within-individual variability of TSH and FT(4) measurements several years apart in different study cohorts and to determine the magnitude of the underestimation of the association between thyroid function and disease rates in population-based studies using only one baseline measurement.
Methods: We used a pair of measurements of serum TSH and FT(4) levels of subjects of the Nijmegen Biomedical Study (NBS) and the Rotterdam Study (RS) to calculate the regression dilution ratio (RDR) with a nonparametric method, the MacMahon's method. Risk estimates could be corrected for regression dilution bias by dividing them by the RDR.
Results: The RDR of serum TSH in the NBS and RS were 0.74 and 0.78, respectively. The RDR of serum TSH were similar for women and men, 0.75 and 0.74 in the NBS and 0.80 and 0.77 in the RS. The RDR of serum FT(4) in subjects in the NBS was 0.77.
Conclusion: The relationship between thyroid function and disease rates is underestimated by studies using only one measurement of TSH and FT(4). The true association will be about 33% (1/0.75) higher for studies with a follow-up time of 2-4 yr.