Introduction

Secondary adrenal insufficiency (AI) is characterized by suppression of the hypothalamic–pituitary–adrenal (HPA) axis, the most common cause of which is long-term glucocorticoid (GC) therapy.1,2 GC-induced AI has been known since the early 1950s3 but it still remains a major diagnostic and therapeutic challenge. Recent studies suggested that the prevalence of AI in patients treated systematically with GCs varies from 14% to 100%.3-10 To date, no guidelines have been proposed for GCs withdrawal schedules to reduce the risk of AI.4 Moreover, the adrenal function is often not evaluated after a gradual withdrawal, which may result in the risk of overlooking AI. Due to the fact that nearly 1% of people in the general population are treated with GCs on a long-term basis,4 many patients could potentially have undiagnosed AI, which may lead to a life-threatening adrenal crisis.

Clinicians also face diagnostic difficulties due to no definite recommendation on how to properly interpret the results of cortisol levels and stimulation tests after GC withdrawal. There are 4 available tests used to evaluate HPA axis function: the insulin tolerance test (ITT), the overnight metyrapone test, corticotropin-releasing hormone stimulation test, and the short Synacthen test (SST) (standard and low dose, 250 µg and 1 µg of Synacthen, respectively). The SST is considered to be a reliable and safe method11 and is now commonly used for the diagnosis of AI.1,4,12,13 However, the cutoff values for secondary AI are not clearly defined. The most recently used criteria are cortisol concentrations of 18 µg/dl (500 nmol/l) at minute 30 of the SST and peak cortisol concentrations of 18 µg/dl or 20 µg/dl (550 nmol/l),14 but it is believed that peak cortisol of less than 22 µg/dl (600 nmol/l) does not exclude AI.1

The aims of our study were to assess 3 aspects of AI in patients treated long term with oral GCs who remained on a low maintenance dose of GCs (≤5 mg per day of prednisone or equivalent) for at least 4 weeks. A dose of 5 mg prednisone is the equivalent of a daily adrenal GC output15 and has not been associated with a high risk of AI so far.16

Primarily, we evaluated the prevalence of AI before GC withdrawal and its persistence right after withdrawal and during a minimum of 1-year follow-up. Next, we aimed to identify the risk factors for the development of AI. Finally, we sought to assess the diagnostic accuracy of morning cortisol and dehydroepiandrosterone sulfate (DHEA-S) concentrations measured before GC withdrawal, without the need to perform stimulation tests.

Patients and methods

Study group

The study is part of a larger prospective study evaluating the relapse rates of the underlying disease that was an indication for the use of GCs, after GC withdrawal (autoimmune diseases or renalorliver transplantation—baseline characteristics, see Table 1). Forty participants were recruited from the Department of Immunology, Transplantology and Internal Diseases of Medical University of Warsaw. All patients were on long-term treatment with oral GCs and were qualified to withdraw the GC therapy. Initially, the treatment included higher doses (maximum 500 mg of prednisone in pulses). The patients were enrolled in the study after reducing the dose to a maximum 5 mg of prednisone daily (or equivalent), taken for a period of 4 weeks to 120 months (median 12 months). Before withdrawal, the entire study group was admitted for the first visit between January 2015 and June 2017 to the Department of Internal Diseases and Endocrinology of Medical University of Warsaw. Patients with diagnosed adrenal dysfunction had then follow-up visits completed in July 2018. The study protocol was approved by the bioethical committee of the Warsaw University of Medicine (no. KB/241/2013).

Work­up

We assessed the adrenal function at 3 time points: initially before GC withdrawal in the entire study group and in patients with adrenal dysfunction at follow-up visits; after the attempt of withdrawal (median [range] time, 9 [6–12] weeks); and at least at the last follow-up visit (median [range] time, 1.86 [1.25–3] years]).

Data on the following parameters were collected from all patients: sex, age, body weight, height, underlying and concomitant diseases, medications, duration and dose of GCs. The patients did not take any drugs affecting measurement of total cortisol level. We assessed symptoms and signs of Cushing syndrome (including hypertension, muscle weakness and atrophy, easy bruising, mental disorders, backache or fractures as symptoms of osteoporosis, polydipsia or polyuria as symptoms of diabetes, impotence, oligomenorrhea or amenorrhea, rounding of the face, facial plethora, truncal obesity, adipose tissue accumulation around the supra-clavicular and cervical regions, striae, thinning of the skin) and symptoms and signs of AI (including fatigue, loss of appetite, muscle weakness, weight loss >3 kg, nausea, vomiting, muscle pain, abdominal pain, hypotension).17-19 Patients who had at least 2 of 4 signs most specific for Cushing syndrome (plethora, striae, bruises, muscle atrophy)17 were regarded as having Cushingoid features.

Laboratory tests during each visit included: morning serum cortisol, DHEA-S, sodium, potassium, glucose, creatinine, lipids, and alanine transaminase.

Diagnostic algorithm

The diagnostic algorithm is shown in Figure 1. We evaluated morning cortisol levels in all patients before GC withdrawal and then performed the standard dose (250 μg) SST if the cortisol levels were between 3 to 16 μg/dl (83 nmol/l–440 nmol/l). We used the SST as it is considered to be safe and reliable,11 has been validated against the ITT,9,10 and is now commonly used for the diagnosis of AI.1,4,12,13 The described algorithm was repeated at each visit.

Figure 1. Diagnostic algorithm used to evaluate adrenal function in patients before and after glucocorticoid withdrawal
Abbreviations: AI, adrenal insufficiency; AS, adrenal sufficiency; SST, standard dose short Synacthen test

The patients were divided into 3 groups based on the following criteria1,14,20-22: 1) AI group (recognized AI) with morning cortisol levels of less than 3.00 µg/dl or cortisol levels of less than 18.0 μg/dl at minute 30 and 60 of the SST; 2) AS group (adrenal sufficiency, AI excluded) with morning cortisol levels of more than 16.0 μg/dl or cortisol levels of 22.0 μg/dl or more at minute 30 of the SST; 3) intermediate group (partial AI with decreased adrenal cortical reserve) with cortisol levels of 18.0 μg/dl or more at minute 30 or 60 of the SST and cortisol levels of less than 22.0 μg/dl at minute 30 of the SST.

At the first visit a total of 31 patients (out of 40) had the SST (Figure 1). Two patients were diagnosed with AI based on the morning cortisol levels of less than 3 μg/dl and in 7 patients AI was excluded based on morning cortisol of more than 16.0 μg/dl. The patients with recognized AI were recommended GC replacement therapy and patients with partial AI were recommended only stress doses of GCs.

Laboratory evaluation

Baseline cortisol values were measured in a blood sample taken between 7:00 and 9:00 in the morning. The SST was performed in a similar setting and at a similar time (8:00–10:00 morning) after an overnight fast and cessation of GC administration for at least 24 hours. Cortisol was measured in blood samples collected prior to intramuscular administration of 250 μg tetracosactide (1–24-adrenocorticotropic hormone [ACTH], Synacthen, Novartis, Naples, Italy and SIGMA-TAU, Roma, Italy) and then 30 and 60 minutes after administration.

Cortisol concentrations were measured by the electrochemiluminescence assay using Elecsys Cortisol II Cobas diagnostic kits (Roche Diagnostics, Mannheim, Germany). The immunoassay for the in vitro quantitative determination of cortisol in human serum or plasma has a detection range of 0.054 to 63.4 μg/dl (1.5–1750 nmol/l). Reference values for morning cortisol are: 4.82 to 19.5 μg/dl (133–537 nmol/l). DHEA-S concentrations were measured by the electrochemiluminescence assay using Elecsys DHEA-S Cobas diagnostic kits (Roche Diagnostics). The assay for the in vitro quantitative determination of DHEA-S in human serum or plasma has a detection range of 0.010 to 100 ng/ml (0.003–27.0 μmol/l). Reference values are age and sex dependent.

Statistical analysis

The results are expressed as medians with ranges for continuous variables, and as numbers and percentages for categorical variables. Comparisons between groups regarding continuous variables were made using the Mann–Whitney test. Categorical variables were compared using the χ2 test and the Fisher exact test. For comparisons between 2 groups, the t test for independent samples was used. Significance was assumed when a null hypothesis could be rejected at a P value of less than 0.05. We decided to compare only 2 out of 3 study groups (AI vs AS). We did not include the intermediate group in comparisons in order to present only 2 clearly defined groups with different clinical characteristics and background in terms of hormone levels.

To assess morning cortisol and DHEA-S ratios as predictors of AI diagnosis, receiver operator characteristic (ROC) curves analyses were performed with true positive results (sensitivity) plotted against false positive results (1-specificity). The area under the curve (AUC) analysis was used to express the overall diagnostic accuracy of the assessed criterion. The Youden index (J) was used to determine the optimal empirical cutoff value of basal cortisol (value for which [sensitivity + specificity – 1] is maximized). We then examined alternative cutoffs, considering a minimum specificity (ruling in) or sensitivity (ruling out) requirement of 95%. DHEA-S ratios were converted by dividing DHEA-S value by the lower limit of the reference range for all participants, to make a clear comparison between the results of the recent study.23

Patients who dropped out of the study (n = 3) were excluded from the analysis. Statistical analyses were performed using STATISTICA software version 13.3 (TIBCO Software, Palo Alto, California, United States) and R version 3.5.3 (The R Foundation for Statistical Computing, Wien, Austria).

Results

Overall, 40 patients (27 women and 13 men) were evaluated. The baseline characteristics are shown in Table 1.

Table 1. Clinical characteristics of study groups before glucocorticoid withdrawal (first visit)
VariableTotal (n = 40)AI (n = 10)Intermediate (n = 7)AS (n = 23)P value (AI vs AS)

Age, y

42 (21–75)

54 (25–70)

30 (21–61)

38 (21–75)

0.12a

Sex

Male

13 (32.5)

2 (20.0)

2 (28.6)

9 (39.1)

0.26b

Female

27 (67.5)

8 (80.0)

5 (71.4)

14 (60.9)

BMI, kg/m2

25.0 (17.3–41.8)

27.1 (18.9–35.5)

22.5 (18.6–29.7)

24.9 (17.3–41.8)

0.22a

GC indication

Nephropathyc

16 (40)

2 (20)

1 (14.3)

13 (56.5)

0.054b

Autoimmune diseasesd

16 (40)

5 (50)

4 (57.1)

7 (30.4)

After LTx

6 (15)

3 (30)

2 (28.6)

1 (4.35)

After KTx

2 (5)

00

2 (8.70)

GCs

Prednisone

23 (57.5)

4 (40.0)

3 (42.9)

16 (69.6)

0.06b

Methylprednisolone

15 (37.5)

7 (70.0)

1 (14.3)

7 (30.4)

Deflazacort

2 (5.0)

0

2 (28.6)

0

Duration of GC therapy, mo

75 (8–420)

74 (8–420)

72 (8–360)

78 (18–288)

0.81a

Duration of low-dose treatment, mo

12 (1–120)

10 (1–36)

6 (2–72)

12 (2–120)

0.19a

Other immunosuppressive drugse

18 (45)

5 (50)

4 (57.1)

9 (39.1)

0.42b

Concomitant diseases

Diabetes

7 (17.5)

4 (40)

0

3 (13)

0.10b

Hypertension

29 (72.5)

10 (100)

3 (42.9)

17 (73.9)

0.09b

Dyslipidemia

22 (55)

6 (60)

4 (57.1)

12 (52.2)

0.49b

Osteoporosis

5 (12.5)

1 (10)

1 (14.3)

3 (13)

0.65b

Data are presented as median (range min–max) or number (percentage).

a The Mann–Whitney test for comparisons between groups

b The Fisher exact test for comparisons between groups

c IgA nephropathy (n = 14), minimal change disease (n = 1), focal segmental glomerulosclerosis (n = 1)

d Systemic lupus erythematosus (n = 14), autoimmune hepatitis (n = 1), IgA vasculitis (n = 1)

e ≥1 of drugs included: cyclosporine (n = 8), tacrolimus (n = 5), mycophenolate mofetil (n = 6), azathioprine (n = 5)

Abbreviations: BMI, body mass index; GC, glucocorticosteroid; intermediate, partial adrenal insufficiency group; IgA, immunoglobulin A; KTx, kidney transplantation; LTx, liver transplantation; others, see Figure 1

The prevalence and persistence of adrenal insufficiency

The study flowchart is shown in Figure 2. Before GC withdrawal, AI was diagnosed in 10 patients (25%) and partial AI (intermediate group) in 7 patients (17.5%). After the first visit, 3 patients were excluded (1 dropped out and 2 due to relapses of underlying disease). After GC withdrawal, in 4 patients AI and in 5 patients partial AI persisted (28.6% and 35.7% of the remaining 14 patients, respectively). At late follow-up visit adrenal function returned to normal only in 1 out of 7 patients (2 patients were lost to follow-up).

Figure 2. Study flow­chart. Data are presented as number (percentage).
Abbreviations: see Figure 1 and Table 1

Risk factors of adrenal insufficiency

Cushingoid feature (regarded as 2 or more of the 4 most specific Cushing syndrome signs, such as facial plethora, striae, easy bruising, muscular atrophies)17 occurred more often in the AI compared with the AS group (60% vs 13%; P = 0.01). Among individual signs of Cushing syndrome muscular atrophy occurred more often in the AI group (Table 2). Moreover, 8 out of 11 participants (72.7%) with Cushingoid features had impaired adrenal function (6 AI and 2 partial AI).

Table 2. Prevalence of signs that best discriminate Cushing syndrome17 in all groups at first visit
ParameterAI (n = 10)Intermediate (n = 7)AS (n = 23)P value AI vs AS

Cushingoid featurea

6 (60)

2 (28.6)

3 (13)

0.01

Plethora

2 (20)

2 (28.6)

4 (17.4)

0.61

Striae

3 (30)

0

3 (13)

0.25

Easy bruising

7 (70)

6 (85.7)

10 (43.5)

0.15

Muscular atrophies

5 (50)

1 (14.3)

1 (4.35)

0.005

Data are presented as number (percentage). The Fisher exact test was used to compare groups.

a ≥2/4 described signs

Abbreviations: see Figure 1

No significant differences were found between the groups in anthropometric parameters (age, sex, body mass index), type of GC, duration of GC therapy, duration of treatment with a low dose, indications for GC therapy, and concomitant diseases (Table 1).

No differences were found between the groups regarding the prevalence of AI symptoms, which were present in 3 out of 10 patients in the AI group, 3 out of 7 patients in the intermediate group and in 4 out of 23 patients in the AS group (AI vs AS group, P = 0.42; AI vs intermediate group, P = 0.59; and AS vs intermediate group, P = 0.16). Morning cortisol values did not differ significantly in patients presenting with these symptoms compared with the remaining group (P = 0.18).

No significant differences between the groups in terms of other basic biochemical parameters (sodium, potassium, glucose, creatinine, lipids, alanine transaminase, C reactive protein, urinalysis) were found (data not shown).

The diagnostic accuracy of morning cortisol and dehydroepiandrosterone sulfate levels before glucocorticoid withdrawal

Laboratory findings of the study group before GC withdrawal (n = 40) are shown in Table 3. There were significant differences in baseline morning cortisol values in AI, intermediate, and AS groups (median cortisol levels, 4.68 μg/dl [129 nmol/l], 11.2 μg/dl [309 nmol/l], and 15.3 μg/dl [422 nmol/l], respectively; P <0.001 for comparison of AI and AS). During the SST, cortisol values at each time point and Δ cortisol were also higher in the AS group compared with the AI group (P <0.001). DHEA-S levels were significantly lower in the AI compared with the AS group (P = 0.002). Moreover, 12 out of 19 participants (63.2%) with DHEA-S levels below reference values had impaired adrenal function.

Table 3. Baseline laboratory findings
ParameterAI (n = 10)Intermediate (n = 7)AS (n = 23)P value (AI vs AS)

Morning cortisol, μg/dl

4.68

(0.68–8.13)

11.2

(7.83–14.9)

15.3

(10.8–20.9)

0.001a

Cortisol at 30’ of SST, μg/dld

12.2

(8.05–14.9)

19.1

(17.7–21.0)

25.2

(22.0–33.4)

0.001a

Cortisol at 60’ of SST, μg/dld

13.8

(9.31–17.6)

23.4

(20.3–26.2)

28.9

(24.7–41.2)

0.001a

∆Serum cortisol, μg/dlc,d

5.99

(3.71–9.70)

10.8

(8.1–24.4)

15.5

(7.64–27.8)

0.001a

DHEA-S, μg/dl

12.6

(1.4–166.1)

49.4

(9.2–261.4)

106.5

(14.1–244.1)

0.002a

DHEA-S below reference values

8 (80)

4 (57.1)

7 (30.4)

0.009b

Data are presented as median (range min–max) or number (percentage) of patients.

a The Mann–Whitney test

b The χ2 test

c Peak serum cortisol levels in the short Synacthen test subtracted by baseline serum cortisol

d Measurements not available for all patients. The short Synacthen test was performed in patients with a suspicion of AI, with morning cortisol values between 3.00 to 16.0 μg/dl (n = 31).

SI conversion factors: to convert cortisol to nmol/l, multiply by 27.59; DHEA-S to nmol/l, by 2.714.

Abbreviations: DHEA-S, dehydroepiandrosterone sulfate; others, see Figure 1 and Table 1

The ROC analyses were based on morning cortisol and DHEA-S measurements before GC withdrawal (n = 40) as predictors of AI diagnosis. To find the optimal cutoff values to rule out and to rule in AI, we compared the AS versus AI and intermediate group and the AI versus AS and intermediate group. The AUC of morning cortisol for these 2 comparisons was 0.93 and 0.997, respectively (ROC curves, Figure 3A and 3B). Table 4 shows sensitivities, specificities, positive predictive values (PPV), negative predictive values (NPV), positive likelihood ratios, and negative likelihood ratios of morning cortisol at different cutoffs, including those determined by the Youden index. A good cutoff to rule out AI was a baseline cortisol level of 14.91 μg/dl (411 mmol/l) with sensitivity of 100% and specificity of 52% (PPV 61%, NPV 100%) (Figure 3A). A good cutoff to rule in AI was a baseline cortisol level of 6.51 μg/dl (179.6 nmol/l) with specificity of 100% and sensitivity of 90% (PPV 100%, NPV 97%) (Figure 3B).

Figure 3. Receiver operating characteristic curves of morning cortisol: A – adrenal sufficiency (AS) group versus adrenal insufficiency (AI) and intermediate group; B – AI group versus AS and intermediate group
Table 4. Diagnostic yields of morning serum cortisol levels at various cutoff points
Morning cortisol, μg/dlSensitivity, %Specificity, %PPV, %NPV, %JLR+LR–

AS group vs AI and intermediate groups

11.69a

88

91

88

91

0.80

10.15

0.13

14.80

94

52

59

92

0.46

1.97

0.11

14.91b

100

52

61

100

0.52

2.09

0

15.27

100

48

59

100

0.48

1.92

0

AI group vs AS and intermediate groups

6.51c

90

100

100

97

0.90

NA

0.10

7.83

90

97

90

97

0.87

27.00

0.10

8.13a

100

97

91

100

0.97

30.00

0

9.40

100

93

83

100

0.93

15.00

0

a Optimal cutoff point according to Youden index (J)

b Optimal cutoff point to rule out AI

c Optimal cutoff point to rule in AI

SI conversion factors: to convert to nmol/l, multiply by 27.59.

Abbreviations: J, Youden index; LR+, positive likelihood ratio; LR–, negative likelihood ratio; NA, not applicable; NPV, negative predictive value; PPV, positive predictive value; others, see Figure 1

DHEA-S concentrations as the only criterion for the diagnosis of AI proved to be a worse parameter than morning cortisol levels. The AUC of ROC curves of DHEA-S ratios corresponding to Figure 3A and Figure 3B curves of baseline cortisol was 0.78 in both cases.

Discussion

The prevalence of GC-induced AI varies from 14% to 100%.3-10 Data about the prevalence in patients taking low doses of GCs are scarce. Schlaghecke et al,6 in their study on 60 patients taking less than 5 mg of prednisone per day, found a normal response to corticotropin-releasing hormone stimulation test only in 37% of patients. On the other hand, Hicklin et al15 found abnormal results of the SST only in 10% from 39 patients taking 5 mg of prednisone. Among our 40 patients, insufficient adrenal response (AI or partial AI) occurred in 42.5% patients before GC withdrawal, which is in agreement with the study by Borresen et al24 that demonstrated an insufficient response to the SST in 39% of 33 patients receiving 5 mg prednisolone daily. Such a high prevalence is alarming, especially considering that currently in patients taking these doses of GCs, it is suggested to test HPA axis function only when abrupt discontinuation is being considered or a patient is exposed to acute stress.16 To reduce the risk of overlooking AI, we advise evaluating adrenal function before GC therapy withdrawal in all cases, regardless of the dose. It could be helpful to identify patients who need extended withdrawal time and/or being at risk of persisted AI after GC withdrawal.

Our study suggests that impaired adrenal function can be detected in 22.5% of patients after GC withdrawal and in 15.0% of patients at late follow-up visit. Our data are in agreement with previous reports that GC-induced AI could last from few days to more than a year7,10,11 and with a study by Jamilloux et al7 who showed insufficient response to the SST in 10% of the study group 2 years after GC withdrawal.

The best known risk factors for AI development are: daily doses of more than 20 mg of prednisone (or equivalent) taken for more than 3 weeks, evening doses of 5 mg or more of prednisone taken for more than a few weeks, and a Cushingoid appearance.16 We found no significant differences between the study groups in terms of duration of GC therapy, but our results may be caused by a small sample size. On the other hand, our data are in agreement with evidence suggesting that Cushingoid features increase the risk of AI.

Despite known factors affecting the HPA axis function, it is still hard to predict which of the patients taking GCs will develop an iatrogenic Cushing syndrome and/or AI.25-27 It may be caused by individual variations in GC metabolism and genetic differences in regulation of GC receptor activity.27

As many as 90% of patients with GC-induced AI may be asymptomatic.28 Moreover, AI and adrenal crisis may also have atypical course.29,30 In our study, only 6 of 17 patients from the AI and intermediate groups had symptoms of AI. This shows that AI would be missed in 64.7% if adrenal function was evaluated in symptomatic patients only. In our study group, 1 patient with AI with no symptoms at first visit developed an adrenal crisis during follow-up in the course of gastrointestinal infection. For this reason, it is extremely important to evaluate adrenal function in all patients taking long-term GCs, even those who are asymptomatic.

For HPA axis evaluation, we used the SST which is now commonly used for the diagnosis of AI.1,4,12,13 Our decision about the cutoff values of the SST was based on the available literature and supported by assessment of clinical status. However, we assessed the value of morning cortisol and DHEA-S in facilitating AI diagnosis. Morning cortisol measurements are safe, accessible in all outpatient clinics, and may significantly reduce the cost of diagnostic workup if there is no need to perform stimulation tests. A strong positive correlation between morning cortisol and stimulated cortisol levels during ITT was described previously.31 Although several cortisol cutoff values have been used, most contemporary studies showed that the concentration of less than 3.6 μg/dl (100 nmol/l) strongly suggests AI.9,20,31-33 The cutoff value for AI exclusion, however, varies depending on the study, from 5.07 μg/dl (140 nmol/l) to 18.34 μg/dl (506 nmol/l).1,9,19-21,31,34,35 Woods et al,9 in their study in 404 patients taking GCs, found the best cutoff value to rule out AI at baseline cortisol was 14.86 μg/dl (410 nmol/l). Sbardella et al,34 in an analysis of a group of 2050 SSTs performed with the Roche Modular System assay, found this value at 18.34 μg/dl (506 nmol/l). The authors drew attention to the fact that this limit may be lower in patients receiving GCs, which is in agreement with our results, demonstrating that a morning cortisol level of 14.91 μg/dl (411 nmol/l) gave 100% NPV for ruling out AI. But in ruling in AI, we found with 100% specificity that the value of cortisol concentration may be higher, not 3.6 μg/dl, but 6.51 μg/dl. If we could adopt these 100% cutoffs in our study, 9 out of 31 SSTs (29%) would not have to be performed because the results of the SST could have been predicted based on the morning cortisol alone.

Previous studies have suggested that a normal DHEA-S level in patients with pituitary masses is a strong indicator of adrenal sufficiency and normal ACTH secretion.36,37 It is considered that GCs cause suppression of DHEA-S secretion, which could last even months after GC secretion returns to normal.36 Studies including patients treated with GCs demonstrated that DHEA-S still exhibits good prediction of HPA function, but a number of these patients was small (10 out of 46 patients)23 or unknown.38 To the best of our knowledge, our study is the first to evaluate only patients taking GCs. Despite significantly lower concentrations in the AI group compared with the AS group, we found the lower diagnostic value of DHEA-S measurement to diagnose AI compared with the study by Charoensri et al23, who showed the AUC of ROC curves at level 0.92, compared with 0.78 in our study. Our results indicate that DHEA-S may have lower diagnostic value after GC therapy, but there is a need for larger studies to validate this result.

We acknowledge several important limitations of our study: 1) small sample size; 2) heterogeneous group of patients in terms of indications for GC therapy, the duration and type of used GCs; 3) lack of SST and ACTH measurement in the entire study group; 4) no follow-up in the entire study group. Moreover, due to long-term treatment in most patients (up to 35 years) it was not possible to evaluate precisely the cumulative dose of taken GC. Finally, as a potential hypothesis-generating study, we did not apply corrections for multiple testing.39 Therefore, the results of our study need to be replicated in a larger, prospective trial.

In conclusion, the prevalence of secondary AI in patients who received long-term low-dose GC is high. It may persist even years after gradual withdrawal of GC. Due to the risk of adrenal crisis, it is truly important to evaluate HPA axis function in all patients before withdrawal. We suggest the morning cortisol measurement for initial assessment of HPA axis function as it is a safe simple and easily accessible test. Values of 14.91 μg/dl or higher with 100% sensitivity suggest adrenal sufficiency and values of 6.51 μg/dl or less with 100% specificity suggest AI. It should be interpreted together with a clinical presentation (features of iatrogenic Cushing syndrome, signs and symptoms of AI). Further studies are needed to assess the value of DHEA-S in the diagnosis of AI.