Validated Criteria for the Interpretation of a Single Measurement of Serum Cortisol in the Investigation of Suspected Adrenal Insufficiency

Scott D. Mackenzie; Robert M. Gifford; Luke D. Boyle; Mike S. Crane; Mark W. J. Strachan; Fraser W. Gibb


Clin Endocrinol. 2019;91(5):608-615. 

In This Article


Using what is to our knowledge the largest data set of its kind for a single assay, we have shown that using a cut-off of <275 nmol/L for samples drawn in the morning, or <250 nmol/L in the afternoon, basal serum cortisol is >95% sensitive for the detection of a subnormal-stimulated cortisol. This sensitivity is maintained when applying the criteria to a second outpatient cohort, and to a third cohort comprising inpatients, although for inpatients the test only meets the prespecified sensitivity threshold in morning samples. Although specificity and positive predictive value are low, the use of the proposed thresholds would result in a significant reduction in the number of patients requiring dynamic testing with the SST.

Previous efforts to define similar thresholds for basal serum cortisol have produced values which vary widely.[6–18] Assay variability plays a significant role in limiting the ability to define unified criteria. With this in mind, our thresholds cannot be applied to assays other than the Abbot Architect assay used in this study. A recent study by Sbardella et al compared three assays, including 449 outpatients for whom serum cortisol was measured using the Abbott Architect assay, and found a serum cortisol of <295 nmol/L achieved 95% sensitivity for the detection of AI.[17] In our study, comprising a total of over 3000 patients, this sensitivity is exceeded using the same assay and 430 nmol/L cut-off for stimulated cortisol, and receiver operating characteristics are similar.

We have targeted 95% sensitivity in order to enable comparison with recently published literature.[6,7,17] Our data demonstrate NPV is very high using this threshold, although because of the dependence of NPV on pretest probability, caution is required in those with a particularly high index of clinical suspicion. Although setting a higher sensitivity would come at the expense of a greater proportion of patients requiring confirmatory testing, we found cut-offs of <308 and <312 nmol/L would achieve 99% and 100% sensitivity, respectively, for morning samples. In the afternoon, a cut-off of <312 nmol/L is required to meet both 99% and 100% thresholds. Even using the lower 95% threshold however, the few individuals in the group of 'false negatives' are likely to have a degree of adrenal reserve, and in our study, the majority in this group were subsequently found to have intact adrenal function upon repeat testing.

We found a basal serum cortisol of below 50 nmol/L to be strongly predictive of a subnormal-stimulated cortisol, with only two 'false positives' across all cohorts using this approach. In one of these, a basal serum cortisol alone may have been the more appropriate test, given the suspicion of a recent pituitary insult. The second patient demonstrated a pattern previously reported in association with opioid analgesia,[21,22] and we would recommend dynamic testing as a first-line approach for this patient group.

Diurnal variability is a concern when interpreting single measurements of serum cortisol.[19] Due to this, previously published studies in outpatients have assessed morning cortisol alone. Recently, Brown et al produced time adjusted criteria,[7] although samples were drawn from 0700–1200 only. We have shown the test is equally sensitive when serum cortisol is sampled in the afternoon and overall performance, assessed by ROC analysis, is similar. The requirement for a slightly lower cut-off, and the reduced specificity of the test in the afternoon reflect the diurnal variation in serum cortisol. This variation is slightly smaller than reported in other studies,[23,24] perhaps reflecting a degree of stress in patients attending outpatient clinics compared with studies in healthy volunteers. Despite the reduction in specificity with afternoon testing, use of the criteria would still avoid the requirement for dynamic testing in a significant proportion of patients when tested in the afternoon.

For inpatients, comparatively little attention has been given to the use of a single measurement of serum cortisol as an initial step in the approach to suspected AI, although an approach which avoids the need for a carefully timed dynamic test in busy inpatient units is clearly advantageous. Our study addresses this question using predefined criteria and in a much larger cohort than previously published studies.[16,18] We have found morning basal serum cortisol is highly sensitive and carries the potential for a large reduction in the need for SSTs. Sensitivity was inferior in the afternoon, and while it is possible that performance of the test here would be improved by defining a higher cut-off, this would require study of a second inpatient cohort to validate separate criteria.


The main limitation of our study is the retrospective nature of the data collection and possible institutional referral bias for outpatients. For inpatients, interpretation of serum cortisol is complicated by difficulty defining an 'appropriate' response to acute illness in a heterogeneous group, and the findings cannot be generalised to critically ill patients. Equally, alterations in cortisol binding in acute illness further complicate interpretation of total serum cortisol measurements[25] and we have not been able to measure free cortisol. Finally, the data need to be interpreted in light of the limitations of the SST compared with the gold standard insulin tolerance test (ITT). Although the SST is generally well validated,[1,2] it provides a less physiological stimulus than the ITT, and should be interpreted with particular caution where results are borderline. Importantly, in those with a strong preclinical probability, and particularly in those with pituitary disease,[26] a normal SST does not exclude AI.