Reference interval estimation: Methodological comparison using extensive simulations and empirical data.

Area of Science:

• Clinical chemistry and laboratory medicine
• Statistical analysis in healthcare

Background:

• Accurate reference intervals are crucial for clinical diagnosis.
• Commonly used methods for estimating reference intervals include parametric, non-parametric, and robust approaches.
• The optimal method depends on data distribution characteristics.

Purpose of the Study:

• To statistically compare and evaluate common reference interval estimation methods.
• To determine the best method based on data distribution characteristics.
• To provide guidance for laboratories on selecting appropriate estimation techniques.

Main Methods:

• Comparison of parametric, non-parametric, and robust methods.
• Use of simulated Gaussian and non-Gaussian data sets.
• Evaluation based on bias and precision measures.
• Illustration with real-world data.

Main Results:

• Parametric approach yielded least biased and most precise estimates for Gaussian data.
• No single method was optimal for all non-Gaussian scenarios; non-parametric performed best in most cases.
• Method performance hierarchy was influenced by sample size and data skewness.
• Variability inflated differences between interval estimates.

Conclusions:

• Transforming data to a Gaussian distribution and using the parametric approach is recommended for optimal reference intervals.
• When Gaussian transformation is not feasible, laboratories should consider sample size and skewness in method selection.
• Potential consequences of false positives or negatives should also inform the choice of method.