Individual Reference Intervals for Personalized Interpretation of Clinical and Metabolomics Measurements

Abstract

The Population Reference Interval (PRI) refers to the range of outcomes that are expected in a healthy population for a clinical or a diagnostic measurement. This interval is widely used in daily clinical practice and is essential for assisting clinical decision making in diagnosis and treatment. In this study, we demonstrate that each individual indeed has a range for a given variable depending on personal biological traits. This Individual Reference Interval (IRI) can be calculated and be utilized in clinical practice, in combination with the PRI for improved decision making where multiple data points are present per variable. As calculating IRI requires several data points from the same individual to determine a personal range, here we introduce novel methodologies to obtain the correct estimates of IRI. We use Linear Quantile Mixed Models (LQMM) and Penalized Joint Quantile Models (PJQM) to estimate the IRI s upper and lower bounds. The estimates are obtained by considering both the within and between subjects variations. We perform a simulation study designed to benchmark both methods performance under different assumptions, resulted in PJQM giving a better empirical coverage than LQMM. Finally, both methods were evaluated on real-life data consisting of eleven clinical and metabolomics parameters from the VITO IAM Frontier study. The PJQM method also outperforms LQMM on its predictive accuracy in the real-life data setting. In conclusion, we introduce the concept of IRI and demonstrate two methodologies for calculating it to complement PRIs in clinical decision making.