An interesting editorial on the topic of reference ranges can be found in the January edition of the Annals of Clinical Biochemistry (46) 1-2, 2009, which also includes a couple of interesting papers on the approaches using "healthy" populations and real patient data. The editorial writer comments on these various approaches to defining reference ranges:
1. A classic based approach is to draw samples from a well defined cohort(s) of healthy people and typically take the middle 95% of values (mean +/-2 SD). This is required if there are variations between people of different age, gender, race etc. There are limitations to this approach: Who is healthy? Should there be a separate reference range for symptomatic people? How do we define ethnicity in populations that are composed of people of mixed ethnicity? For how long are samples stable prior to analysis?
2. The use of patient data to establish reference ranges. This may seem at first sight to be an odd way to establish a reference range. Almost by definition, patients have some health problems. On the other hand, many patients do in fact have concentrations of many analytes that are within the ranges found in healthy people. If one excludes the "outlier" values, it is possible to derive a reference range that at least approximates that seen in healthy individuals. The other objection to the use of patient samples is that there may be pre-analytical variation that is not seen in carefully collected samples used solely for the purpose of establishing a reference range. The answer to this objection is that "real" patient samples are usually drawn and transported under less than ideal conditions. If there is variation seen because of pre-analytic factors, then shouldn't it be reflected in the reference range?
A big advantage of using real patient data is that large numbers of data points are available and, importantly, from groups such as children, from whom it is generally extremely difficult to get samples to do a reference range study.
So we have a "purist" approach and a "pragmatist" approach to reference range establishment. (Yes, Laboratory Medicine has its own set of disputes!) Moreover, those are not the only ways to think about lab values that require a clinical response. Some people advocate Decision Limits. In a nutshell, these are values that trigger a clinical decision (e.g., a diagnosis or a therapeutic intervention). We talked about some of these in the context of cholesterol and glucose, and how those values that have clinical significance are not evident from a simple statistical analysis of the distribution of values in health or disease. These are what I refer to as a "reference range by committee".
When it comes to reference ranges, all of these different approaches need to be considered; there is no one approach that is clearly right or wrong in all circumstances.