A Bayesian nonparametric meta-analysis model for estimating the reference interval

Area of Science:

• Biostatistics
• Clinical Laboratory Science
• Medical Diagnostics

Background:

• Reference intervals define healthy population norms vital for clinical laboratory testing and disease differentiation.
• Single-study reference intervals lack broad applicability; meta-analysis offers generalizability but relies on restrictive assumptions.
• Existing meta-analysis methods for reference intervals often assume normally distributed study means and equal variances, which may not reflect real-world data.

Approach:

• Proposes a Bayesian nonparametric model to overcome limitations of traditional meta-analysis for reference interval estimation.
• Employs more flexible assumptions regarding the distribution of study-specific means and within-study variances.
• Extends random effects meta-analysis to enhance the accuracy and applicability of reference intervals.

Key Points:

• The proposed Bayesian nonparametric model offers greater flexibility than conventional meta-analysis techniques.
• Demonstrates improved performance in simulation studies where standard assumptions are violated.
• Validates the approach using two real-world datasets, showcasing its practical utility.

Conclusions:

• The novel Bayesian nonparametric approach provides a more robust method for estimating generalizable reference intervals.
• This method is particularly valuable when study-level data deviate from traditional meta-analysis assumptions.
• Enhances the reliability of laboratory testing by providing more accurate and widely applicable reference ranges.