Weighing Up Our Clinical Confidence in Reference Limits

In this issue of Clinical Chemistry, Hickman and colleagues (1) have presented a systematic analysis of the impact of various statistical approaches to removal of outliers in reference interval determination. They used a reference interval dataset from the Aussie Normals study (2). These data were also used by the Australasian Association of Clinical Biochemists as part of the evidence employed to derive ‘harmonized’ national reference intervals (3, 4). The conclusion in the new article is that the differing statistical approaches for removal of outliers give significantly differing reference limits, with a quarter of studied analytes shifting by over 20% and some by over 100%. The global guideline regarding reference interval determination, the Clinical Laboratory Standards Institute (CLSI) A28-A3 (5), mentions the two potential approaches to statistical outlier exclusion (Dixon and Tukey methods) used in the report by Hickman et al. (1). Outlier exclusion is a necessary step because, while the reference population is expected to be homogenous, some of the reference values may arise from a different population of test results. Unless outliers are known to be aberrant observations (due to mistakes in analysis or a lapse in pre-analytical controls), the emphasis should be on retaining outlier values. If there is heterogeneity in the reference population, however, this may lead to a need for outlier exclusion. Among the criteria listed by the Clinical Laboratory Standards Institute (CLSI) for possible exclusion from the apparently healthy population are illness, medications, and obesity. 1856 individuals in the Aussie Normals study were included; exclusion criteria included pregnancy, diabetes, asthma requiring medication with oral steroids, any history of malignancy, as well as other conditions involving significant systemic disease known to bias biochemical concentrations (2). Individuals over the age of 50 taking statins and antihypertensive medications were not excluded, nor were individuals classified as overweight or obese. In the new outlier study of Hickman and colleagues (1), the analytes showing differences included alanine aminotransferase (ALT), creatine kinase (CK), lipase, immunoglobulin (Ig)A, IgM, homocysteine (in men), ferritin (in men), thyroid stimulating hormone (TSH), and NT-pro-brain natriuretic protein (NT-proBNP). Interestingly, body mass index (BMI) is a potential determinant of ALT (6), CK (7), lipase (8), homocysteine (9), ferritin (10), TSH (11), and NT-proBNP (12). Despite our hard definitions of obesity (e.g., BMI >=30 kg/m2), body weight represents a continuous variable, and rather than having a dichotomous distribution of non-obese vs. obese, this continuum of BMI can potentially cause skewing of all laboratory data affected by body weight. Hickman et al. clearly recognize that mild obesity is associated with higher ALT activity, as this was a finding in the Aussie Normals study where statistically higher ALT (and ferritin) were found in overweight adults. The inclusion of overweight and obesity in the Aussie Normals study may be one determinant of heterogeneity that could have impacted the reported outlier challenges, and an issue deserving of further study. As stated in their recent article, “there is often a gradual transition from health to disease, it is very difficult to ensure that all persons included are truly healthy.” The CLSI guideline accepts that reference intervals are intended to be ‘health associated’ but that health is a relative condition lacking universal definition. The health of any reference population can be defined by the investigator, but the clinical history requirements and investigations to exclude disease should always be described and documented so others can evaluate the health status of the reference sample. The reference population is assumed to be healthy or normal, as implied in the ‘Aussie Normals’ study name. Nevertheless, obesity (or overweight) was not an exclusion criterion. The inclusion of overweight and obesity is a very common difficulty with many (most?) recent reference interval studies, including those from Turkey (13) and Saudi Arabia (14), where, just as in Australia and the USA, most of the apparently healthy population is overweight or obese. This is a complicated issue but one that lies at the heart of the purpose and use of reference intervals. The reference interval simply provides a likelihood that the results for any patient belong to the same homogenous distribution as the apparently healthy reference population. If we do accept obesity as consistent with apparent health, we have by definition also tried to exclude the subclinical manifestations of obesity as potential reasons to lie outside the reference limits. This may be clinically appropriate; after all, why do we need laboratory tests to confirm what a set of scales can often do more directly? While the obesity inclusion approach may apparently solve our clinical dilemma with obesity, we may still need to cope with the outlier impacts of population heterogeneity (at least until everyone is overweight or obese). The issues of subclinical diseases causing heterogeneity and outlier exclusion are intimately related. The clinical question is paramount when interpreting laboratory results. Hickman et al. have pointed out that clinicians vary in their use expectation of liver transaminase reference limits. They also recognize that the substantial overlap in results between apparently healthy persons and those with some degree of liver pathology, means no quoted reference interval will satisfy all parties. This stand-off is probably why reference intervals, despite their difficulties, have remained popular (i.e., because they avoid the plethora of disease definitions by concentrating on the apparently simpler problem of health). Hickman et al. suggest that the revision to the CLSI standard should pay more attention to the preferred method for statistical treatment of outliers, which would be helpful. The existing CLSI standard reported on the emerging trend (in 2008) of replacing reference intervals with decision limits, such as those used for cholesterol and glycated hemoglobin. The importance of this distinction has recently been highlighted by the expert committee of the International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) (15). It is just as important that the next CLSI revision expands on the strengths and weaknesses of the general situations that ‘healthy’ reference intervals may be considered useful, as well as expanding on the clinical situations where disease-based clinical decision limits may be more appropriate. Author Contributions: The author confirms contribution to the intellectual content of this paper and has met the following 4 requirements: (a) significant contributions to the conception and design, acquisition of data, or analysis and interpretation of data; (b) drafting or revising the article for intellectual content; (c) final approval of the published article; and (d) agreement to be accountable for all aspects of the article thus ensuring that questions related to the accuracy or integrity of any part of the article are appropriately investigated and resolved. Authors' Disclosures or Potential Conflicts of Interest: No authors declared any potential conflicts of interest.