RESUMO
CONTEXT: Thyroid-stimulating hormone (TSH) levels within populations do not follow Gaussian distribution, and normal limits are derived after mathematical normalization. The clinical relevance of these limits is unknown. The objective of this study was to compare upper and lower TSH limits by four data normalization methods with non-normalized data and assess their clinical relevance. DESIGN, PATIENTS AND MEASUREMENTS: Results of blood samples taken by community physicians and stored in a computerized database were analysed after removing samples from patients with evidence of thyroid illness. TSH values were normalized by the Hoffmann and Tukey methods and each method with natural log transformation. Non-normalized data for TSH in the uppermost and lowermost percentile were also calculated. Clinical relevance was determined by alterations in thyroid hormone levels at, below and above the limits for each method. RESULTS: The maximal reduction from non-normalized data for the upper normal limit (UNL) was by the Hoffman method 43% = 3·1 mIU/l). The maximal increase for the lower normal limit (LNL) was also by the Hoffman method (708% = 0·81 mIU/l). There was very limited difference in average FT3 and FT4 between patients with TSH within, below or above the normal range for all methods. CONCLUSIONS: Different normalization methods alter the normal limits greatly. However, in individuals without thyroid illness, thyroid hormone values are stable over a wide range of TSH levels including beyond the UNL for all methods. Indeed, there may be no true universal upper TSH cut-off level and clinical decision-making cannot rely on these calculated limits.