The challenge of the reference intervals determination: An example from two articles published in the present issue

This issue presents two different research articles on the establishment of reference intervals (RI) for serum progastrin releasing peptide (proGRP) measured with the same methodology on the Chinese population.^1,2

Yang et al. ¹ established the RI as the 95th percentile of the healthy population, finding a value <53.9 ng/L in adults aged 21–70 years and <75.7 ng/L for subjects aged >70 years, respectively, with no differences related to gender. Zhao et al. ² found values approximately 20% higher, also in this case without differences between the sexes.

Both studies show a positive association between proGRP concentration and age; differences were statistically significant between groups of 10-year intervals in subjects younger than 70 years in the study by Zhao et al. ²

The two studies followed methods that seem to be accurate, and are homogeneous regarding both the analytical platform used and the population studied. Nevertheless, there are some differences in the setting that can justify the discrepancies found; although, considering the respective confidence intervals of the RI and the intrinsic analytical variability of the method, these differences are probably not as relevant.

Zhao et al. ² evaluated a remarkably high number of subjects, considering that they are individually and clinically selected and not merely extracted from the laboratory database as in the methodologies of indirect RI determination. ³ Yang et al. ¹ examined a more limited number of cases, but used more stringent criteria, according to the classical “direct” approach, ⁴ making the selection also on the basis of laboratory evaluations and diagnostic imaging. As expected, the more focused selection of the subjects led to the identification of lower RI levels in the study by Yang et al. ¹ In contrast, the large number of cases evaluated by Zhao et al. ² allowed the identification of differences between different age intervals, which are statistically significant—though quantitatively modest—at around 5%–10% between groups.

Some considerations can emerge from the comparison of these two interesting studies. The number of cases studied by Zhao et al. ² are comparable to those of some studies on the determination of RI by indirect methods,^5,6 but the selection of included subjects is much more focused. Nevertheless, the identified RI remain higher than those from the more classical study of Yang et al., ¹ which is as expected in view of the less selective recruitment criteria. This could suggest that the mere use of wide subject series is not sufficient to mitigate the confounding effect in the examined cohort of the possible presence of diseases in a subclinical form, such as diabetes, hypertension, or cardiovascular or renal disorders, thus suggesting caution on the actual comparability between the “direct” and the “indirect” methodological approaches for RI determination. In fact, the indirect methodologies, although generally based on a number of cases—that can also be one order of magnitude higher ⁷ than those examined by Zhao et al. ² , use selection criteria even less rigorous from the clinical point of view, based mainly upon statistical considerations.

On the other hand, if the objective is to distinguish a specific pathology in the general population, perhaps the determination of RIs in selected groups of subjects could not be the most appropriate strategy, and the approach of Zhao et al. ² and the use of large, apparently healthy, populations seems more realistic, although possibly less correct from a formal point of view.

Moreover, particularly in the case of tumor biomarkers, it is known that the need to identify patients with cancer from those with benign diseases, which may present similar clinical findings, would still require a different approach based on Bayesian parameters, such as analytic sensitivity and specificity, or the likelihood ratio, and tools such as receiver operating characteristic curves.

In summary, the identification of correct decisional limits in clinical biochemistry, despite the many well-established statistical methodologies, remains an intricate challenge.