We included randomized and non-randomized controlled trials and non-controlled studies that were in accordance with the criteria below.

To minimize the risk of selection bias, at least 10 patients had to be included in the studies. In case of overlapping populations, the article with the largest sample and more complete reporting of data was included.

Three general search strategies were developed for the main electronic health databases: Embase (1980–August 20, 2020), PubMed/Medline (1966–August 20, 2020), and Cochrane Collaboration Controlled Trials Register (1982–August 20, 2020). A second search on all databases was conducted on January 16, 2021. The strategies for PubMed and Embase were reviewed by a medical librarian (MD) using the PRESS 2015 Evidence-Based Checklist tool (22). The search strategies included the following descriptors and synonyms: Cushing’s disease, cabergoline, pasireotide and ketoconazole, osilodrostat, levoketoconazole, metyrapone, and temozolomide. The complete search strategy for Pubmed/Medline is provided in the supplementary material ( Supplementary Tables S1A , S2A ). To search for gray literature, we checked for ongoing studies on ClinicalTrials.gov, references of articles selected for full reading, and annals of congress. There was no language restriction.

Two reviewers (JSCG and VSNN) independently reviewed the titles and abstracts. Potentially eligible studies were selected for full reading, to be assessed for adequacy to the PICO previously established. In case of disagreement, a consensus meeting was made.

Two reviewers (JSCG and ANSCN) used a standardized form to independently extract relevant data of the included studies and to assess the risk of bias of the included studies. In case of disagreement, a consensus meeting was made. To assess the risk of bias of the included studies, the critical appraisal tool from Joanna Briggs’s Institute was adapted to check the included studies with regard to the following aspects: (i) clear inclusion criteria, (ii) diagnostic criteria stated, (iii) description of valid biochemical assays to measure hypercortisolism, (iv) consecutive and complete inclusion of participants, (v) complete reporting of baseline information, (vi) complete reporting of outcomes, (vii) complete reporting of demographics of the site, and (viii) appropriate statistical analysis. For each aspect, we assigned yes, no, or unclear (23).

Homogeneous endpoints in at least two studies were plotted in meta-analyses using the Stata Statistical Software 16 (Stata Statistical Software: Release 16, College Station, TX, StataCorp LLC, USA). Proportional meta-analyses were performed for dichotomous data. We used the updated command metaprop_one and fit the logistic-normal random-effects model to the data (20). Continuous data were expressed as means and standard deviation (SD), and the pre- and post-intervention standardized mean difference (SMD) were calculated with respective 95% confidence interval (CI).

Inconsistencies between the results of the studies included were ascertained by a visual inspection of forest plots and by applying the Higgins statistic (I ²) and the chi-square test (χ ²). Moderate heterogeneity was ascertained if I ² >35%. For χ ², statistic heterogeneity was considered if p <0.10 (21). In order to explore the potential sources of heterogeneity, meta-regression was performed using logit transformed outcomes and logit transformed with study SD. The study sample size, study design (i.e., randomized, prospective), mean age of the study participants, and doses of the intervention were considered as potential explanatory variables. The Knapp–Hartung correction was used to calculate the significance of the meta-regression coefficients (24).

Prediction interval (PI) was calculated for the random-effect meta-analysis, if χ ² p <0.1 or I ² >35% and more than five studies. PI predicts the possible treatment effect in an individual study setting, whereas the random effect meta-analysis summarizes the average effect across the studies (25). Because the potential treatment effect when applied within an individual study setting may differ from the average effect, the PI provides interesting insights for clinical practice (25).

For the outcomes from controlled studies, the quality of evidence for estimating the effect of intervention was generated in accordance with the Grading of Recommendations Assessment, Development, and Evaluation Working Group (26).

The search strategies resulted in 2,102 and 1,712 articles after duplicates were removed using the Endnote software. We selected 55 articles for full reading, of which 27 (1,405 patients) were included. Although we set out to include patients with CD only, we included for full reading studies that sampled other etiologies of Cushing’s syndrome and tried contacting authors to retrieve the data for CD only. Among the excluded studies, nine studies had overlapping population with other already included studies (27–35), seven studies were cohorts of less than 10 patients (36–42), three did not match the study population (43–45), six did not comply with the outcomes (46–51), and three assessed outcomes before six months of follow-up (13, 52, 53). The selection process is summarized in Figure 1 .

The characteristics of the included studies are reported in the supplementary material ( Supplementary Table S3A ). We included eight studies on pasireotide (518 patients) (10, 11, 54–59), six studies on cabergoline (139 patients) (12, 60–64), 10 studies on ketoconazole (559 patients) (60, 65–73), five studies on metyrapone (160 patients) (66, 69, 74–76), and one study on osilodrostat (36 patients) (77).

There were four randomized controlled studies (10, 54, 60, 77) and 23 single-arm studies, from which 13 were prospective (11, 12, 55–58, 61, 62, 65, 67, 68, 73, 75) and 10 were retrospective (59, 63, 64, 66, 69–72, 74, 76). To confirm the pituitary origin of Cushing’s syndrome, the selected studies considered dynamic tests, in addition to the criteria pre-established in our review protocol. Four studies considered the 8-mg/day high-dose dexamethasone suppression test (12, 63, 71, 72), four studies considered the corticotropin-releasing-hormone test (10, 68, 73, 75), and 10 considered both (59–61, 64–67, 69, 70, 74).

Among the three randomized controlled studies included, only one compared two different medications. Data plotted in the meta-analysis refers to the 6-month follow-up of the study by Barbot et al., in which both cabergoline and ketoconazole were used as monotherapy. No statistical difference was observed in the CD control (relative risk: 0.53, 95% CI: 0.15 to 1.87, 14 participants, very low certainty of evidence). The quality of evidence was rated down due to imprecision (i.e., wide CI and no achievement of optimal information size) and high risk of selection bias (60). Two randomized studies compared different dosages of pasireotide (i.e., Colao et al. compared 600 vs. 900 mcg and Lacroix et al. compared 10 vs. 30 mg long-acting release) (10, 54).

The description of inclusion criteria was adequate in all the included studies. In 65% of the studies, it was unclear if the inclusion of patients was consecutive and complete. Reporting was unsatisfactory in 24% of the included studies regarding baseline information and in 48% regarding outcomes. Statistical analysis was considered inappropriate in 20% of the studies. Table 1 shows the risk of bias of the included studies.

The treatment effects of pasireotide, cabergoline, ketoconazole, and metyrapone on disease control, as defined by the individual included studies, were pooled in the proportional meta-analyses. Although UFC was most commonly used to measure disease control, Lila et al. used midnight salivary cortisol (MNSC) and low-dose dexamethasone suppression (LDSC) test to define disease control (62), and Daniel et al. used mostly 9 AM cortisol and mean cortisol from a cortisol day-curve (76). The morning serum cortisol was also used by three studies (66, 67, 74).

Figure 2 shows the proportional random meta-analysis on disease control. The pooled proportion of control was 35% (27–43%) for cabergoline and 41% (36–46%) for ketoconazole, with low heterogeneity. Although statistic heterogeneity was not confirmed for metyrapone (p = 0.12), the small sample of included studies yielded a large CI for`66% of the observed disease control (95%CI: 46–87%, four studies, 66 patients). A subgroup analysis considering studies that considered UFC as the only criteria of disease control was performed for cabergoline (36%, 95% CI: 28–45%, p = 0, five studies, 121 patients) and ketoconazole (41%, 95% CI: 36–45%, P = 0, 5 studies, 434 patients). With regard to pasireotide, 44% of patients had disease control (30–60%, χ ² = 21.3, p = 0, PI = 18–74%, eight studies, 522 participants; Supplementary Figure S1A ). To investigate the heterogeneity, meta-regression was conducted, and it showed that the number of included patients was the variable that explained 50% of heterogeneity. A meta-analysis with the two larger randomized studies showed that the proportion of disease control after pasireotide was 29% (25–35%, χ ² = 0, Figure 2 ).

Because different units of measure were applied to report UFC (i.e., nmol/24 h, µg/24 h, or number of times above the upper limit of normality), the SMD was used as a measure of effect size to plot pre- and post-intervention data in a meta-analysis. The meta-analysis of studies with pasireotide, cabergoline, and ketoconazole consistently showed a reduction on UFC, although with high heterogeneity. The pooled results for cabergoline, ketoconazole, and pasireotide are shown on Table 2 (forest plots in Supplementary Figure S2A ).

The high heterogeneity on the meta-analysis for cabergoline could be explained by one outlier study (12), in which a higher weekly dose (7 mg) was used. The visual analysis of the forest plot for ketoconazole likewise had one outlier study (65), which had a high risk of bias. A sensitivity analysis was performed to explore the heterogeneity for pasireotide. Considering randomized studies only, pasireotide reduced the UFC in -0.94 SD (CI: -1.14, -0.74, I ² = 7%) ( Supplementary Figures S2A–C ).

Pre- and post-intervention data on systolic and diastolic blood pressure, cholesterol, triglycerides, body mass index, and waist circumference were extracted when available. Among these secondary outcomes, improvement of blood pressure (BP) levels was the most commonly reported, although there was variability in reporting and measurement (i.e., proportion of controlled BP, reduction of the parameter itself). Hence, due to the lack of data and heterogeneity on the reporting of outcomes, a meta-analysis for pasireotide was performed considering the reduction of the parameter itself, disregarding the number of medications, which was poorly described in most studies. Table 3 shows a metanalysis on the improvement of BMI, waist circumference, and systolic and diastolic BP with pasireotide.

With regard to cabergoline, Lila et al. observed that four out of 18 patients showed a decrease of 20 mmHg on SBP and 10 mmHg on DBP after 5 months of treatment with cabergoline (62). In another series, a mean reduction on SBP from 141.5 to 118 mmHg after 12 months of follow-up was found (12). Moreover, ketoconazole was also reported to reduce the mean blood pressure from 148/105 mmHg at baseline to 115/85 mmHg after a mean follow-up of 23 months (72). A second study reported that 40% of patients had controlled hypertension after treatment with ketoconazole (71).

The proportion of different AE associated with pasireotide, cabergoline, ketoconazole, and metyrapone was also plotted in a proportional meta-analysis, as summarized in Table 4 . SAEs were reported exclusively for pasireotide, probably because two very low bias randomized studies on pasireotide were included, whereas for other medications, the included studies were mostly prospective or retrospective cohorts.

In the proportional meta-analyses of main AE associated with pasireotide (i.e., diabetes, hyperglycemia, cholecystitis, nausea, abdominal pain, and headache), a high heterogeneity was identified. The meta-regression showed that the type of study was the explanatory variable for this heterogeneity, and sensitivity analysis including only randomized trials was performed.

Cabergoline and metyrapone were mainly associated with vertigo and nausea, with low heterogeneity in the meta-analysis. Studies with ketoconazole reported mainly elevated transaminases, rash, and adrenal insufficiency. Diarrhea and/or abdominal pain were assessed as a composed outcome in the meta-analysis.

No study with temozolomide and levoketoconazole fulfilled our inclusion criteria. Osilodrostat was evaluated in one prospective, open-label, single-arm study with a placebo randomized withdrawal period (77). At 48 weeks, 91 (66.4%, 95% CI: 57.9, 74.3) enrolled patients had a complete response. Sixty-four out of 97 patients (66%) who were treated with osilodrostat throughout the 48 weeks and had a complete response and maintained a complete response for at least 6 months. The most common adverse events included nausea (42%), headache (34%), fatigue (28%), and adrenal insufficiency (28%). Moreover, symptomatic hypocortisolism was reported by 70 (51%) patients, and 58 (42%) patients reported adverse events related to adrenal hormone precursors.