Adrenal Insufficiency with Anticancer Tyrosine Kinase Inhibitors Targeting Vascular Endothelial Growth Factor Receptor: Analysis of the FDA Adverse Event Reporting System

Simple Summary This real-world post-marketing research described clinical features of adrenal insufficiency with anticancer drugs targeting vascular endothelial growth factor receptor reported to the Food and Drug Administration pharmacovigilance database. A robust signal emerged for multi-targeted tyrosine kinase inhibitors, especially in combination with immunotherapy, namely checkpoint inhibitors in renal cancer. These signals should promote both prospective research and a multidisciplinary proactive monitoring by healthcare professionals. These findings strengthen the role of timely pharmacovigilance to detect and characterize post-marketing adverse events of special interest, thus supporting patient care. Abstract Background: We described clinical features of adrenal insufficiency (AI) reported with tyrosine kinase inhibitors (TKIs) targeting vascular endothelial growth factor receptor (VEGFR) in the Food and Drug Administration Adverse Event Reporting System (FAERS). Methods: Reports of AI recorded in FAERS (January 2004–March 2022) were identified through the high-level term “adrenal cortical hypofunctions”. Demographic and clinical features were inspected, and disproportionality signals were detected through the Reporting Odds Ratio (ROR) and Information Component (IC) with relevant 95% confidence/credibility interval (CI), using different comparators and adjusting the ROR for co-reported corticosteroids and immune checkpoint inhibitors (ICIs). Results: Out of 147,153 reports with VEGFR-TKIs, 314 cases of AI were retained, mostly of which were serious (97.1%; hospitalization recorded in 44.9%). In a combination regimen with ICIs (43% of cases), VEGFR-TKIs were discontinued in 52.2% of the cases (26% as monotherapy). The median time to onset was 72 days (IQR = 14–201; calculated for 189 cases). A robust disproportionality signal emerged, also in comparison with other anticancer drugs (ROR = 2.71, 95%CI = 2.42–3.04; IC = 0.25, 95%CI = 0.07–0.39). Cabozantinib, sunitinib and axitinib generated robust disproportionality even after ROR adjustment. Conclusions: We call pharmacologists, internists, oncologists and endocrinologists to raise awareness of serious AI with VEGFR-TKIs, and to develop dedicated guidelines, especially for combination regimens with immunotherapy.


Introduction
Angiogenesis inhibitors targeting the vascular endothelial growth factor and relevant receptor (VEGF/VEGFR) represent an established anti-cancer strategy in the management of various solid tumors, either alone or in combination with chemotherapy/immunotherapy. Three main pharmacological classes can be identified, including tyrosine kinase inhibitors (TKIs), an essential therapeutic option especially in renal cell carcinoma and thyroid neoplasms, monoclonal antibodies (mAbs) such as bevacizumab in metastatic colorectal carcinoma, and VEGF soluble decoy receptor capturing free VEGF such as the fusion protein aflibercept [1]. In contrast to mAbs, VEGFR-TKIs inhibit additional tyrosine kinases, including, among others, platelet-derived growth factor receptor (PDGFR), fibroblast growth factor receptors (FGFR 1-4), platelet-derived growth factor receptor-alpha (PDGFRα), and mast/stem cell growth factor receptor (KIT), thus making them multi-targeted TKIs [2].
Endocrine toxicity with VEGFR-TKIs is an emerging area of interest: understanding the mechanisms affecting endocrine vasculatures is crucial to minimizing adverse events (AEs) associated with their use [3]. In particular, fatigue, diarrhea, and weight loss represent most frequent AEs. Notably, chronic severe fatigue in oncology may be a cancer-related complication but also possibly subtend underlying adrenal dysfunctions, thus leading to drug discontinuation. However, the effects of VEGFR inhibitors on adrenal function are incompletely characterized [4,5]. To the best of our knowledge, only a few case reports/series have described adrenal insufficiency (AI) during bevacizumab [6,7], sunitinib [8,9], pazopanib [10], lenvatinib [11], and vandetanib therapy [12,13]. Moreover, the Food and Drug Administration (FDA) recently warned in the label (January 2021) that, when cabozantinib is used in combination with nivolumab, primary or secondary AI may occur (https://www.accessdata.fda.gov/drugsatfda_docs/label/2021/208692s010lbl.pdf accessed on 21 September 2022). However, case reports can only provide a partial epidemiological perspective.
The analysis of spontaneous reporting databases, such as the FDA Adverse Event Reporting System (FAERS), allows a broader perspective by collecting unpublished reports of AEs submitted worldwide occurring in real-world unselected subjects with comorbidities, poly-pharmacotherapy and complex anticancer combination regimens, also in the long-term; this ensures rapid detection of even rare AEs and emerging clinical entities such as AI, especially for drugs authorized through conditional accelerated approval [14]. Moreover, the analysis of pharmacodynamic data may be used to explore the underlying pathways involved in drug-induced diseases with TKIs [15][16][17].
This pharmacovigilance study aims to characterize the main features of AI with VEGFR-TKIs in FAERS, including the underlying pharmacological basis through receptor affinity data.

Data Sources and Study Design
FAERS is one of the largest publicly available repositories of unsolicited reports, gathering more than 20 million reports worldwide, including the US, Europe and Asia. By virtue of this comprehensive catchment area and the opportunity to access raw data for customized analyses, it was largely used to identify and characterize AEs with anticancer drugs, including VEGFR-TKIs [18,19].
The protocol was pre-registered on the Open Science Framework (https://osf.io/ bmgc9/ created on 5 July 2022) and analyses were performed through the open-source R software (version 4.0.2; 22 June 2022).

Definition of Cases and Drugs of Interest
In the FAERS database, AEs are classified according to the Medical Dictionary for Regulatory Activities (MedDRA) terminology in terms of signs/symptoms, called preferred terms (PTs). AEs of interest were identified using PTs under the high-level term "adrenal cortical hypofunctions", in line with a recent study [20].

Descriptive Analysis
Reports of AI were first described in terms of patient demographics (sex, age, country, type of reporter) and clinical features: concomitant drugs, comorbidities, fatality proportion (i.e., death reported as the outcome), seriousness (focusing on events resulting in hospitalization), latency (i.e., time to onset), discontinuation and positive dechallenge (clinical improvement after the offending agent is suspended).
Differences in categorical variables were assessed using a chi-squared test of independence performed on a 2 × 2 contingency table with Yates' continuity correction. Differences in continuous variables were assessed using a Kruskal-Wallis test. The Holm Bonferroni method was used to correct the p-values for multiple comparisons. Significance was assumed when the p value, corrected for multiple comparisons using the Holm-Bonferroni method, was less than 0.05.

Disproportionality Analysis
We performed the so-called case/non-case approach, a validated concept in pharmacovigilance, to assess whether suspected adrenal AEs (cases) are differentially reported with VEGF/VEGFR inhibitors as compared to other AEs (non-cases) [22]. If an imbalance in the number of cases vs. non-cases emerges, a disproportionality signal can be claimed, thus informing physicians to proactive monitoring [23]. We calculated two different disproportionality measures to reduce the likelihood of false positives: (a) the frequentist reporting odds ratio (ROR), deemed statistically significant by the lower limit of the 95% of the Confidence Interval (CI) >1, with at least five cases, using the Bonferroni correction to reduce the likelihood of detecting spurious associations related to multiple testing [20]. (b) the Bayesian Information Component (IC), deemed significant by the 95% credibility interval, IC0 25 > 0, which is more accurate in case of low number of cases [24].
Both approaches were first performed using all other drugs in the FAERS database as a comparator, a common exploratory disproportionality analysis. To assess the robustness of disproportionality signals and account for potential confounders of the drug-event association, we planned a priori a sensitivity disproportionality analysis using anticancer drugs as a comparator (to reduce confounding by indication and provide a clinical perspective), and adjusting the ROR for possible drug-related confounders, namely concomitant corticosteroids and immunotherapy [20].

Exploring the Underlying Pharmacological Basis
We retrieved affinity data from the Chembl Database (https://www.ebi.ac.uk/chembl/ accessed on 5 July 2022), a consolidated source of pharmacodynamic data [25] (Supplementary Table S1). We interpolated the strength of disproportionality (IC values vs. anticancer drugs) with Homo Sapiens pKd and pIC 50 values (= negative decadic logarithm of Kd and IC 50 [M]). When multiple values were available, the geometrical mean was calculated.

Global Assessment of the Evidence
Causality assessment was carried out on the entire body of evidence by using adapted Bradford Hill Criteria used in epidemiology, which have been applied to pharmacovigilance data [25][26][27]: biological plausibility, strength, consistency, specificity, coherence, and analogy. The biological gradient was not analyzed due to missing data.

Global Assessment and Pharmacological Basis
Globally, Bradford Hill criteria were fulfilled, as indicated by the strength of disproportionality and its consistency throughout the analyses, temporal relationship, and biological plausibility, thus supporting a likely causal association for TKIs (Table 3). The interpolation of ROR and affinity towards different receptors highlighted a common antagonism for VEGFR-1/2/3, and PDGFR-alpha/beta, especially shared by TKIs with significant disproportionality (Figure 3).

Discussion
Targeting VEGF/VEGFR has emerged as an effective therapeutic strategy in different cancers, especially in combination with ICIs. In fact, TKIs exert immunomodulatory properties that may shift the tumor microenvironment from immunosuppressive to immune-permissive [28]. However, different life-threatening off-target toxicities have been described, including thyroid dysfunctions, cardiovascular events, bleeding, proteinuria/renal impairment, hepatotoxicity, gastrointestinal perforations, metabolic and wound healing complications with an impact on quality of life and monitoring strategies [29].
This study is the largest collection of AI cases attributed to VEGF/VEGFR inhibitors from a worldwide pharmacovigilance database so far, and raised awareness on the reporting of serious events. Apart from recently approved drugs such as tivozanib, there is a considerable amount of data for other agents with a consolidated time on the market.
Our study raised debate on whether AI should be considered a class effect of VEGFR-TKIs. Notably, we observed reporting differences between mAbs and TKIs: the former did not consistently generate disproportionality signals, and only bevacizumab was reported in a significant number of cases; the latter were associated with a consistent and robust increased reporting, even when accounting for potential confounders, thus suggested the existence of a plausible cause-effect relationship for VEGFR-TKIs, as corroborated by the global evaluation through Bradford Hill criteria.
Taken together, these data call for constant epidemiological surveillance and urgent clarification by means of large population-based studies. Although actual incidence and prevalence cannot be derived from spontaneous reporting, the fact that VEGFR-TKIs were reported in 314 cases over the past few years (3.25% of total adrenal hypofunction reports in FAERS) suggests that the estimated AI risk is not so rare, and we advocate for an update of the relevant summaries of product characteristics where AI is not mentioned. Of note, only the FDA warned in the product information that cabozantinib might increase the risk of AI when combined with nivolumab.
Different demographic and clinical features emerged from this study, including large male preponderance, seriousness (in nearly half of the cases, was hospitalization recorded), severity (in more than 1/3 of the cases discontinuation was needed), and frailty (higher number of concomitant drugs as compared to non-AI reports). The selective reporting of serious events suggested that our data only represented the "tip of the iceberg", since nonserious cases are unlikely to be diagnosed and submitted to pharmacovigilance databases, thus calling for dedicated prospective clinical pathways to capture subclinical AI and clarify the actual epidemiological burden. Moreover, the exponential increase in the number of AI reports (as compared to other events), especially in 2021, is noteworthy and may be ascribable to various reasons, including the increasing uptake in clinical practice and expanding therapeutic indications, especially in a combination regimen with immunotherapy.
The available clinical evidence on AI by VEGFR-TKIs is scant, jeopardized and limited to cases reports and small case series [4,5], including a recent pharmacovigilance study on axitinib [30]. In a small prospective study, the analysis of basal and stimulated adrenal function in 12 patients receiving lenvatinib and vandetanib for advanced radioiodine refractory differentiated or medullary thyroid cancer, respectively, revealed a gradual ACTH increase with normal cortisol levels in 10 patients complaining of fatigue. Notably, patients diagnosed with primary AI were treated with cortisone acetate replacement therapy, and fatigue improved [13]. As expected, we found a large proportion of AI reports with coreported asthenic conditions, diarrhea and vascular events, which should raise suspicion by clinicians to early diagnose an underlying adrenal hypofunction.
The issue of combination regimen (VEGFR-TKIs plus ICIs) deserves further discussion, especially in metastatic renal cell carcinoma, where this combination is the preferred firstline approach in patients with favorable, intermediate and poor risk disease [31]. VEGFR-TKIs and ICIs share different overlapping toxicities, possibly including AI. However, pivotal trials were not designed to test the actual contribution of individual TKIs, thus making it challenging to unravel the inherent risk of AI. For instance, cabozantinib plus nivolumab was compared with sunitinib in a CheckMate 9ER trial [32]; grade ≥ 3 AI occurred in 1.9% of subjects (vs. zero event), with a median time to onset of 37.3 weeks (highly variable; IQR = 4.1-76.7; n = 320) [33]. Minor imbalances were noted in a KEYNOTE-426 trial for pembrolizumab plus axitinib (0.7% vs. 0.2% for sunitinib) [34], and a CLEAR trial for lenvatinib plus pembrolizumab (1.1% vs. 0% for sunitinib) [35]. Our real-world data found a median latency of three months for the combination regimen, with remarkable high rates of discontinuation and dechallenge, thus suggesting the importance of implementing dedicated guidelines and screening protocols (ACTH test) through a close collaboration among pharmacologists, internists, oncologists and endocrinologists [4,36,37].
A recent pharmacovigilance study on FAERS (up to September 2020) specifically addressed whether the various combination regimens of VEGFR-TKI with ICIs increased the reporting of AI, as compared to monotherapies, using various approaches and models. No increased reporting was observed across combinations and analyses, except for bevacizumab when combined with atezolizumab, thus failing to demonstrate the existence of a pharmacodynamic interaction, namely additive or synergistic effect [36]. Disproportionality signals were found for cabozantinib, axitinib and lenvatinib, in line with our results, thus confirming the hypothesis that VEGFR-TKIs per se likely causes AI. Unraveling the contribution of VEGFR-TKIs vs. ICIs represents the next clinical and research challenge: actual understanding of pathophysiology would allow proper management (i.e., the interruption of a single drug or the combination).
From a mechanistic viewpoint, one important insight can be derived from the appraisal of pharmacodynamics (receptor affinities) with pharmacovigilance data: we put forward the inner role of VEGFR-1/2/3, highly expressed in the fenestrated capillaries of endocrine organs. Notably, axitinib, cabozantinib and lenvatinib emerged with the strongest association and also possess the highest potency towards VEGFR [38]. We encourage regular surveillance by clinicians and pharmacovigilance experts on these drugs, including the third-generation agent tivozanib (higher selectivity towards VEGFR and favorable pharmacokinetics), recently approved for advanced renal cell carcinoma and under additional regulatory monitoring.
We acknowledge inherent limitations of pharmacovigilance data, which do not allow to infer causality (ROR and IC are not risk measures); reports may be incomplete, with missing information (e.g., laboratory and other clinical elements to validate the diagnosis). Moreover, incidence cannot be estimated due to under-reporting and lack of population exposure. The lack of disproportionality should not be interpreted as a safety endorsement, since disproportionality measures are interdependent (a commonly reported AE may mask rare AEs) and the choice of comparator may have a substantial impact. Moreover, the various drugs have different marketing approvals and evidence-based treatment recommendations have changed over recent years. All of these aspects may explain why pazopanib and sorafenib were not associated with significant disproportionality notwithstanding the non-negligible number of reported AI cases. Finally, the contribution of additional drugs with underlying adrenal liability cannot be excluded, including antimicrobials, hormone therapy and drugs for Cushing or adrenocortical cancer diagnosis/treatment.
Nonetheless, several strengths can be identified. We used a large-scale publicly accessible pharmacovigilance database, supporting the generalizability of the results, and contributed to the cumulative knowledge about the safety of VEGFR-TKIs in an unselected real-world population, an evolving clinical issue. There is no reason a priori to support the existence of confounding by indication or notoriety bias (i.e., increased reporting following media attention or regulatory measures), although channeling bias (i.e., selective prescription towards more severe patients) cannot be ruled out. Moreover, we employed several sensitivity analyses to assess the robustness of results, accounting for potential co-reporting biases and using anticancer drugs as comparator, thus providing a clinical perspective. The fact that the signal for VEGFR-TKIs emerged even without accounting for the potential masking effect by previously known and largely reported AEs further strengthened the impact of AI in the overall risk-benefit evaluation. Our global pharmacovigilance assess-ment, encompassing disproportionality, approaches and Bradford Hill criteria evaluation consistently support the notion of a plausible association.

Conclusions
Although causality cannot be firmly inferred, this study found a plausible association between VEGFR-TKIs and AI; the selective reporting of serious events should raise awareness by clinicians and suggested the existence of subclinical AI cases. We call for a tight multidisciplinary collaboration between pharmacologists, internists, oncologists and endocrinologists to: (a) proactively monitor the occurrence of AI with VEGFR-TKIs; and (b) implement dedicated guidelines, especially for a combination regimen with immunotherapy.
Considering the evolving use and recent marketing approvals of VEGFR-TKIs, pharmacovigilance plays a key role in promoting targeted clinical surveillance and safer prescribing, especially for rare AEs of special interest such as AI.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/cancers14194610/s1, Table S1: Affinity data (pKd/pIC50) from Chembl database; Table S2: Disproportionality analysis. Reporting odd ratio (ROR) and information component (IC) with relevant 95% confidence/credibility interval (CI); Table S3: Disproportionality analysis (ROR with 95%CI) adjusted for concomitant glucocorticoid and immunotherapy. Institutional Review Board Statement: Ethical review and approval were waived for this study because it is based on anonymous data that can be downloaded from a publicly available source.
Informed Consent Statement: Patient consent was waived because the study is based on anonymous data that can be downloaded from a publicly available source.

Data Availability Statement:
The datasets analyzed during the current study are available in the following resource available in the public domain: https://fis.fda.gov/extensions/FPD-QDE-FAERS/ FPD-QDE-FAERS.html (accessed on 26 July 2022).