High-Dose and Low-Dose Cosyntropin Stimulation Tests for Diagnosis of Adrenal Insufficiency

IN RESPONSE:

The paper by Gandhi and associates (1) was excluded from our analysis because of inadequate stratification of patients by gold standard testing (only 7 of 31 patients had an insulin tolerance or metyrapone test). The paper by Choi and coworkers (2) appeared on MEDLINE in April 2003, after our final draft was submitted, and includes 72 patients who underwent the 1-µg test and the insulin tolerance test. The paper by Rose and colleagues (3) was not detected by our MEDLINE search and includes 2 separate samples of pediatric patients: 38 who had 250-µg testing and metyrapone testing and 120 who at a later time had the 1-µg test and metyrapone testing. Thus, none of these 3 studies are useful for head-to-head testing of the 250-µg and 1-µg tests. When the data of Choi and coworkers (2) and Rose and colleagues (3) are included in our Tables 2 and 4 and subjected to a summary receiver-operating characteristic (ROC) analysis, there is no change in our conclusion that the operating characteristics of the 250-µg and 1-µg tests are virtually identical (no difference in area under the curve, the value of sensitivity, or sensitivity at a specificity of 95%; P > 0.2 for all comparisons).

In the paper by Abdu and colleagues (4), the data in their Table 1 for all 42 patients who had an insulin tolerance test as the gold standard revealed that the sensitivity was 100% for both cosyntropin stimulation tests at cutoff cortisol levels of 500 nmol/L and 600 nmol/L. In the paper by Mayenknecht and colleagues (5), an analysis of paired ROC data in our Appendix Figure firmly establishes that the performance characteristics of the 250-µg and 1-µg tests are indistinguishable.

Averaging sensitivities and specificities to perform a t-test is inappropriate for meta-analysis. For this method to be valid, specificities would have to be equalized, which is not possible with the available data. Summary ROC methods were designed to deal with these difficulties and facilitate comparison between studies because they assess performance characteristics independent of cut-score. Dr. Dickstein's Figure shows inequality of specificities, and the data are underpowered to conclude that specificities are not different (P = 0.092). Calendar time is less of an issue, since both the 250-µg and the 1-µg tests were compared with stable gold standard tests. To address the issue of direct comparisons, we analyzed the 7 studies that met our inclusion criteria and compared the 1-µg and the 250-µg tests head-to-head in the same sample of patients (Tables 2 and 4); we found no difference between the tests for any parameter of test performance (area under the curve, value of sensitivity, or sensitivity at a specificity of 95%; P > 0.2 for all comparisons).

Perhaps the 1-µg test has marginally superior performance characteristics in carefully selected samples, but our inclusive analysis using appropriate methods demonstrates no difference in test performance or sensitivity when specificity is equalized. The apparent increase in sensitivity associated with the 1-µg test derives from selection of a high cortisol cut-score, which shifts up along the ROC curve and improves sensitivity at the expense of lower specificity. For a disorder such as secondary adrenal insufficiency, where pretest probability is typically 10% to 30%, the utility of a diagnostic test with low specificity is limited because of the large number of false-positive results. As a numerical example, a cut-score that achieves 95% sensitivity for the 1-µg test in our Figure 1 matches a specificity of 60% and yields a positive likelihood ratio of only 2.4. At a 30% prevalence of adrenal insufficiency in 100 patients tested, there would be 28 false-positive test results and an equal number of true-positive test results. All 56 patients with positive results on the 1-µg test would then require additional sorting by gold standard tests to distinguish true- and false-positive results. Alternatively, normal patients without abnormal hypothalamic–pituitary–adrenal function would be inappropriately managed with lifelong corticosteroid replacement. On the other hand, a specificity of 95% matches a sensitivity of 61% for the 1-µg test; at a prevalence of 30% in 100 tested patients, there would be 3.5 false-positive results versus 18 true-positive results. At a clinically practical level of specificity (for example, 95%), neither the 250-µg test nor the 1-µg test achieve a sufficient level of sensitivity to obviate the need for integrated tests of hypothalamic–pituitary–adrenal function when the results of the cosyntropin test are negative and clinical suspicion for adrenal insufficiency based on pretest probability is substantial. In the event that a test with high sensitivity but low specificity finds a role in the screening algorithm for adrenal insufficiency, our results, which show equivalency of the 1-µg and 250-µg tests, indicate that the algorithm could be made more practical simply by raising the cortisol cut-score for the 250-µg test.

• Copyright ©2004 by the American College of Physicians