Accuracy of repeated measurements of late-night salivary cortisol to screen for early-stage recurrence of Cushing's disease following pituitary surgery

ObjectiveThe performance of late-night salivary cortisol (LNSC) to accurately screen for postoperative recurrence of Cushing's disease (CD) at an early stage is unknown. The aim of this study was to compare the accuracy of multiple sampling strategies to suggest the optimal number of LNSC samples needed for diagnosing post-surgical recurrences of CD at an early stage. DesignRetrospective analysis in a single centre. Patients and measurementsThirty-six patients in surgical remission of CD had successive measurements of LNSC, defined as sequences', using a locally modified RIA assay as part of long-term follow-up (692106months). Patients underwent an extensive biochemical evaluation within 3months before or after a sequence of saliva sampling and were classified as being in remission or in early-stage recurrence. The accuracy of three diagnostic strategies combining two, three or four LNSC results from a sequence was estimated using areas under the ROC curves (AUC), sensitivity, specificity and predictive values. ResultsForty-four sequences of LNSC measurements were available. Fifty-two percent of sequences were performed during early-stage recurrence. The intrasequence variability of LNSC was higher during recurrence than during remission (medians of SDs: 21 vs 05nm; P<00001). AUCs from ROC curves ranged from 093 to 096 depending on the strategy. For 90% sensitivities, the best specificities (929% and 909%) were achieved by strategies taking into account three or four measurements summarized either by their mean or their maximum value. ConclusionsIncrease in LNSC concentration is an early abnormality during post-surgical recurrence of CD. However, due to a major within-patient variability of LNSC from 1day to another, a screening strategy using three or four samples collected on successive days may be recommended to detect early-stage recurrence of CD with a high accuracy.