Establishing Thyroid Reference Intervals Through Hierarchical Cluster Analysis: A Comparative Evaluation of Limit...

Area of Science:

• Clinical Biochemistry
• Endocrinology
• Medical Diagnostics

Background:

• Manufacturer reference intervals for thyroid function tests (TSH, fT4, fT3) often lack age or sex stratification.
• This limits their diagnostic accuracy in diverse adult populations.
• Establishing indirect, stratified reference intervals is crucial for improved clinical decision-making.

Purpose of the Study:

• To establish indirect, age-stratified reference intervals for TSH, fT4, and fT3 in a large adult cohort.
• To compare unsupervised clustering and conventional partitioning methods for subgroup identification.
• To evaluate the diagnostic performance of derived intervals against an independent external validation cohort.

Main Methods:

• Analyzed data from 37,255 adults, establishing a reference population of 5870.
• Used hierarchical clustering and Random Forest analysis to identify age-based subgroups.
• Calculated reference intervals using multiple algorithms and population frameworks, validated using NHANES data.

Main Results:

• Age was the dominant variable influencing TSH, fT4, and fT3 distributions.
• TSH-only classification showed positive discrimination; fT4-only and combined TSH+fT4 showed poor performance, suggesting inter-platform harmonization issues.
• Age-stratified intervals showed potential advantages, but external validation indicated inconsistent reproducibility dependent on clinical thresholds.

Conclusions:

• Age stratification and algorithm choice require further clinical validation for broad adoption.
• Unsupervised clustering presents a practical, objective alternative to manual subgrouping for laboratories.
• Inter-platform harmonization issues for fT4 need addressing to improve diagnostic accuracy.