Subjects and Methods
Subjects enrolled under protocol 97-CH-0076 designed for genetic investigation and treatment of pituitary tumours were screened for eligibility in the study. Patients were included based on the following criteria: children (<18 years at presentation) with diagnosis of CD as defined by the Endocrine Society Guidelines and previously described [abnormal measurements in at least two of the following criteria: elevated 24-hour urinary free cortisol (UFC), elevated midnight serum cortisol [>121 nmol/L (4.4 mcg/dl) in children] and failure to suppress cortisol to 1 mg (or weight-based equivalent dose) overnight dexamethasone suppression test (post–dexamethasone cortisol >50 nmol/L (1.8 mcg/dL)],[11,12] successful surgical treatment [defined as postoperative cortisol nadir level <138 nmol/L (5 mcg/dL)] and minimum postoperative follow-up of 3 months. All patients underwent transsphenoidal surgery (TSS) for resection of the pituitary adenoma, either at our institute or at another hospital. Only patients after first TSS were included in the study, to avoid dependent events in the survival analysis or effect of multiple surgical interventions on the outcome. Patients who received intracranial/pituitary radiation or medical therapy after surgery for inhibition of cortisol production, such as steroidogenesis inhibitors, were likewise excluded. Some of the patients included in the current manuscript (n = 44) were previously described by Lodish et al After surgery, all patients were prescribed oral hydrocortisone at 8–12 mg/m2/day and weaning of the dose was performed according to previously described recommendations and guided by their biochemical results and symptoms.
Informed consent was obtained from parents and assent from patients if developmentally appropriate. All study procedures were approved by the Eunice Kennedy Shriver National Institute of Child Health & Human Development (NICHD) Institutional Review Board (IRB).
Clinical and Biochemical Data
Clinical data and biochemical results were extracted from electronic medical records. Duration of disease was calculated as the time between the age at weight gain in excess of previous percentile, as noted on patient's growth chart, and the age at first biochemical diagnosis. Height, weight and body mass index (BMI) z-scores were calculated based on age- and sex-matched references from the 2000 CDC growth charts. Serum cortisol and plasma ACTH levels were calculated as the average value of the corresponding levels performed at 23:30 hours and 00:00 hour (reported as midnight values) and 07:30 hours and 08:00 hours (reported as morning values). Twenty-four-hour urinary free cortisol (UFC) was calculated as the average of the first 2–3 samples reported in the electronic medical records. Given the possible differences in the assays and reported reference range for UFC, we calculated the increase in UFC based on the upper limit of normal (ULN) according to the following formula: UFCxULN = UFC/Upper limit of the reference range. Serum cortisol was measured with solid-phase, competitive chemiluminescent enzyme immunoassay on Siemens IMMULITE 2500 analyser (Malvern, PA, USA). Plasma ACTH was measured with chemiluminescence immunoassay on Siemens IMMULITE 200 XPi analyser (Malvern, PA, USA). UFC was measured with chemiluminescent enzyme immunoassay until 2011 and with high-performance liquid chromatography/tandem mass spectrometry since 2011(LC-MS/MS).
Primary and Secondary End-points
The primary outcome was time from first TSS to recovery, as defined by a serum cortisol level of >500 nmol/L (18 mcg/dL). All patients had a morning cortisol measurement (8:00 AM). If the serum morning cortisol was >500 nmol/L (18 mcg/dL), no further evaluation was performed, and it was considered that the patient had recovered. If morning cortisol was <500 nmol/L (18 mcg/dL), a standard-dose ACTH stimulation test (performed with IV administration of 250 mcg of cosyntropin) was performed and peak cortisol >500 nmol/L (18 mcg/dL) was consistent with recovery. The secondary outcome was time to recurrence as defined by the period between TSS and first biochemical confirmation of recurrence.
Categorical data are presented as counts and proportions and compared between subjects with vs. without recurrence using chi-square test or Fisher's exact test, as appropriate. Continuous data were checked for normality based on histogram distribution and the Shapiro-Wilk statistical test. Continuous data with normal distribution are described as mean (standard deviation) and were compared between groups using Student's t test. Continuous data without normal distribution are presented as median (interquartile range) and were compared between groups using Wilcoxon rank-sum test. Median time to recovery and 95% confidence intervals (CI) were calculated using the Kaplan-Meier method. To investigate potential collinearity of the covariates of interest, the variance inflation factor (VIF) was calculated and covariates with VIF > 4 were considered as potentially collinear with others; the covariates morning cortisol vs. midnight cortisol, and year of surgery vs. surgeon met this criterion. After excluding morning cortisol and year of surgery from the model, all remaining VIFs were < 4. The proportional hazards (PH) assumption of the Cox PH test was tested by calculating the Schoenfeld residual using the cox.zph function. Covariates with P-values < .05 were further investigated through direct inspection of the zph plot. No evidence of nonproportionality was found. For continuous covariates, the linearity assumption was assessed by performing a lack-of-fit test comparing a linear fit with a nonlinear fit based on a restricted cubic spline with 3 knots, and no evidence of a nonlinear relationship was found. For the 11 model covariates considered in the regression analyses, 19.2% subjects were missing a value for at least one of the 11 covariates. However, only 7 of the 11 covariates were missing values and the per cent of missing values ranged between 2.3% and 7.7%, well below the recommended maximum of 20% to 25%. Multiple imputation was implemented on the model covariates. Firstly, the missing data were filled in five times to generate five complete data sets as per the full conditional specification method. Secondly, the five complete data sets were analysed using standard statistical analyses, here Cox proportional hazards regression. Lastly, the results from the five complete data sets were combined using Rubin's rules to produce the final inferential results to account for the additional variability from the imputation. Results were considered significant if P-value was < .05. Statistical analyses were performed in R. Plots were created in R using the ggplot2 package.
Clin Endocrinol. 2021;94(1):40-47. © 2021 Blackwell Publishing