A new approach for the determination of reference intervals from hospital-based data

Reference limits are some of the most widely used tools in the medical decision process. Their determination is long, difficult, and expensive, mainly because of the need to select sufficient numbers of reference individuals according to well-defined criteria. Data from hospitalized patients are, in contrast, numerous and easily available. Even if all the information required for a direct reference interval computation is usually not available, these data contain information that can be exploited to derive at least rough reference intervals.

In this article, we propose a method for the indirect estimation of reference intervals. It relies on a statistical method which has become a gold-standard in other sciences to separate components of mixtures. It relies on some distributional assumptions that can be checked graphically. For the determination of reference intervals, this new method is intended to separate the healthy and diseased distributions of the measured analyte. We assessed its performance by using simulated data drawn from known distributions and two previously published datasets (from human and veterinary clinical chemistry).

The comparison of results obtained by the new method with the theoretical data of the simulation and determination of the reference interval for the datasets was good, thus supporting the application of this method for a rough estimation of reference intervals when the recommended procedure cannot be used.