Random cortisol levels, obtained based on our previous protocol to determine when it is safe to stop steroids, are likely not associated with central AI in neonates exposed to prolonged steroids, as there was no difference between cortisol levels obtained in neonates who developed signs of AI and those who did not. In addition, clinical AI was rarer than expected, despite many of these patients having an ICD-10 code for AI in the medical record. Only 8.5% of cortisol levels obtained were associated with the patient having clinical signs of AI requiring intervention or for which steroids were restarted by the clinical team due to signs of AI. Only 5.6% of ACTH stimulation tests performed resulted in a failed test (peak cortisol <18 µg/dl at 30 or 60 min). The true incidence of secondary AI among all neonates exposed to steroids may be even lower than these numbers suggest, because patients with longer and repeated steroid courses had more cortisol levels obtained, so the results are skewed toward those patients with the highest steroid exposure. However, our data are also skewed because the protocol mandated restarting steroids if the cortisol level was <3 ug/dl, so possibly the most at-risk group was not given the chance to show signs of AI. Although the initial indication for starting steroids was not collected, given our typical unit practices it can be assumed that most were started and continued on steroids for prevention or treatment of bronchopulmonary dysplasia or chronic lung disease.

There have been several prior studies evaluating cortisol levels in preterm neonates, and most have also shown that cortisol levels are not predictive of adrenal insufficiency, though only two small studies looked at cortisol levels obtained after steroid exposure, as our study does. In one study of 25 premature infants, patients who received dexamethasone treatment (5 patients, for a mean of 16 days) had cortisol levels (measured at 1, 2, 4, and 8 postnatal weeks) that were not significantly different from those babies who did not receive dexamethasone, though ACTH levels were lower in infants who received dexamethasone [7]. This suggests that ACTH levels may indicate some degree of hypothalamic-pituitary axis suppression, but cortisol levels did not appear to be affected by steroid treatment. In another study of 23 very low birth weight infants who received a 3-week tapering course of dexamethasone, both ACTH and cortisol levels (obtained during a corticotropin-releasing hormone stimulation test) were significantly lower at the 3-week mark compared with prior to the steroid course. Four weeks after treatment, ACTH levels had recovered, while stimulated cortisol levels remained lower than baseline [8].

Two other studies examined cortisol levels in preterm infants, unrelated to steroid exposure. One study measured cortisol levels every 2 weeks in all neonates born <29 weeks’ gestation, and found that there was no difference in the cortisol levels obtained in neonates who later went on to develop AI versus those obtained in neonates who did not [2]. Another study looked at 11 preterm infants (with matched controls) with clinical signs of late-onset AI and measured cortisol and cortisol precursor levels before giving hydrocortisone, and found that cortisol levels did not differ between the cases and controls, but the concentration of steroid precursors was higher in the group with adrenal insufficiency, indicating that there was a limited ability to synthesize sufficient cortisol for the degree of clinical stress in those infants [3].

There are some limitations to our study. First, our initial sample size calculation determined that we would need 80 total cortisol levels (8 in babies with AI and 72 in babies without AI, assuming 10% of the cohort would have AI) to not miss a difference of 5 µg/dl in cortisol levels between AI and non-AI babies. We were only able to identify 71 random cortisol levels that qualified despite looking over a 5-year period. The steroid weaning protocol was not in place before that time, and is no longer being used, limiting the time period available. In addition, there was a lower incidence of clinical AI (8.5%) than we initially predicted (10% was used for the sample size estimate). The standard deviation of cortisol levels (7.5 µg/dl) was also higher than predicted (4.7 µg/dl, calculated as a weighted average from limited data available in the literature) [9, 10]. With this new information, the power to not miss a difference of 5 µg/dl in cortisol levels with an α of 0.05 was only 33%. However, based on our actual data, we had 80% power to not miss a difference of 9 µg/dl in cortisol levels between patients with and without AI.

Another limitation to this study is that for many neonates with cortisol levels <3 µg/dl, it could not be assessed whether they would go on to develop clinical signs of AI, because steroids were immediately resumed, per protocol. This could bias the results because those patients with the lowest cortisol levels presumably would be the ones most likely to develop AI, yet those patients were not followed prospectively. Thus, our estimate of clinical AI is reflective of infants with moderate (>3 mcg/dl) random cortisol levels.

Despite these limitations, our study had enough power to not miss a difference of 9 µg/dl in cortisol levels between patients who developed AI and those who did not in the population of babies with levels >3 mcg/dl, suggesting that cortisol levels may not be associated with secondary AI in neonates after prolonged steroid exposure. In addition, AI and failing an ACTH-stimulation test were uncommon (8.5% and 5.6% respectively). Based on our protocol, babies with levels <3 mcg/dl automatically resumed maintenance hydrocortisone. These neonates were likely receiving unnecessary continued steroid exposure in response to low, yet non-diagnostic, cortisol levels, and received a median of an additional 25 days of steroids due to these low levels. Based on the findings presented here, we have revised our NICU protocol with the goal of reducing unnecessary steroid treatment, blood draws, and associated cost of cortisol level testing. We have based our new approach on that of the “Buffalo” protocol, published in a review article [1] which heavily relies on observation time in hospital post-steroid course as an important factor. In general, we believe this problem can be managed without cortisol levels.