Impulse Control Disorders by Dopamine Partial Agonists: A Pharmacovigilance-Pharmacodynamic Assessment Through the FDA Adverse Event Reporting System

Abstract Background The dopaminergic partial agonism of the so-called third-generation antipsychotics (TGAs; aripiprazole, brexpiprazole, cariprazine) is hypothesized to cause impulse control disorders (ICDs). Relevant warnings by the Food and Drug Administration (FDA) were posted on aripiprazole (2016) and brexpiprazole (2018). Our study investigated the FDA Adverse Event Reporting System and the pharmacodynamic CHEMBL database to further characterize TGA-induced ICDs. Methods We downloaded and pre-processed the FDA Adverse Event Reporting System up to December 2020. We adapted Bradford Hill criteria to assess each TGA’s —and secondarily other antipsychotics’—causal role in inducing ICDs (pathological gambling, compulsive shopping, hyperphagia, hypersexuality), accounting for literature and disproportionality. ICD clinical features were analyzed, and their pathogenesis was investigated using receptor affinities. Results A total of 2708 reports of TGA-related ICDs were found, primarily recording aripiprazole (2545 reports, 94%) among the drugs, and gambling (2018 reports, 75%) among the events. Bradford-Hill criteria displayed evidence for a causal role of each TGA consistent across subpopulations and when correcting for biases. Significant disproportionalities also emerged for lurasidone with compulsive shopping, hyperphagia, and hypersexuality, and olanzapine and ziprasidone with hyperphagia. Time to onset varied between days and years, and positive dechallenge was observed in 20% of cases. Frequently, co-reported events were economic (50%), obsessive-compulsive (44%), and emotional conditions (34%). 5-Hydroxytryptamine receptor type 1a agonism emerged as an additional plausible pathogenetic mechanism. Conclusions We detected an association between TGAs and ICDs and identified a new signal for lurasidone. ICD characteristics are behavior specific and may heavily impact on life. The role of 5-Hydroxytryptamine receptor type 1a agonism should be further explored.


Third-Generation Antipsychotics (TGAs) Aim
The scope of this pharmacovigilance study is to extend current knowledge about TGA-induced ICDs, both as specific behaviors and as a unified diagnostic entity. Relying on FAERS spontaneous reports from real-world clinical practice and on CHEMBL pharmacodynamic measures, we aim to characterize ICDs' clinical features and explore their underlying pharmacological basis. Our focus will be primarily on aripiprazole, brexpiprazole, and cariprazine, including a comparison with other antipsychotics.

FDA Adverse Event Reporting System
The FAERS is a freely available spontaneous reporting system that collects worldwide reports of suspected adverse drug reactions. Its raw quarterly data (FDA, 2022) include demographic, therapeutic, and outcome details. Reactions and indications are coded using Medical Dictionary for Regulatory Activities (MedDRA) Preferred Terms.
All FAERS Quarterly Data (ASCII files from January 2004 up to December 2020) were downloaded and pre-processed for duplicate removal.
Although we acknowledge that spontaneous reports do not allow to fully apply diagnostic criteria [in fact the need for diagnostic scales-for both ICDs and the individual behaviors affected-is still largely unanswered (Evans et al., 2019)], we presumed that any suspected adverse drug reaction, to be reported, must have some impact on the quality of life.

Descriptive Analysis
To characterize TGA-related ICDs-both as a single heterogeneous diagnostic construct and as individual behaviors-we compared their demographics (sex, age, outcome, continent, reporter) with 2 Reference Groups (RGs): RG1, all other TGA reports (see Table 1); and RG2, all other ICD reports (see Table 2). Given that the population reporting TGA exposure was not equally split between the sexes and the tendency to report to the FDA is also skewed, even an equal tendency to report the adverse event would not result in equal distribution of cases between the sexes. Therefore, directly comparing the number of men with the number of women in the case group would not be informative. Women tend to report more often than men to the FAERS, and conditions for which a specific drug is prescribed may be distributed asymmetrically between men and women. To consider both reporting biases (e.g., higher reporting by women) and the sex distribution in the underlying indication for use, we chose as comparison the population of non-ICD TGA reports (RG1). In the same way, we compared TGA-related ICD reports with all the other reports of ICDs (RG2) to investigate whether ICDs related with TGAs share their features with other ICDs.
Differences in categorical variables were assessed using a chi-squared test of independence performed on a 2 × 2 contingency table with Yates' continuity correction (see supplementary Material-chi-squared statistics for the 2 × 2 tables comparing reports of men and women treated with TGAs with and without the ICD conditions). Significance was assumed when the P value, corrected for multiple comparisons using the Holm-Bonferroni method, was less than .05. To identify co-occurring conditions and the psychosocial impact of ICDs, co-reported psycho-social events were retrieved and discussed.

Bradford Hill Causality Assessment
Adapted Bradford Hill criteria , which account for already accrued evidence and disproportionality, were used to systematically gather evidence for TGA-induced ICDs (see Table 3 for a schematic presentation of these criteria and the strategies applied to assess them). Relying on the literature, we appraised biological plausibility and analogy to drugs already known for inducing ICDs. Through disproportionality analyses of the FAERS (reporting odds ratio [ROR]), we evaluated strength (extent of the ROR), consistency across sensitivity analyses (accounting for reporter, country, notoriety, and confounding by indication biases), coherence with the impulsivity substrate ("impulsive behavior," "impulse-control disorder"), and specificity of the adverse event (assessing the association of ICDs with other antipsychotics). The ROR was calculated, using the 2 x 2 contingency table, whenever at least 3 reports with both the event and the drug investigated were found. A disproportionate reporting (i.e., an augmented probability of reporting an event with a specific drug) was considered significant when the lower limit of its 95% confidence interval was higher than 1. The ROR was reported as median (CI 2.5% -CI 97.5% ) [n]. In particular, the analyses limited to reports submitted before the FDA warnings concerning ICD development with (1) aripiprazole (March 05, 2016); and (2) brexpiprazole (February 09, 2018), together with the analyses stratified by reporter occupation (i.e., medical doctor, consumer, lawyer), allowed us to partly adjust for the notoriety bias, likely amplified by the many related lawsuits. Furthermore, the possibility that TGAs are associated with ICDs not because they induce them but because of a common cause (i.e., their psychiatric indication for use) was considered through a subpopulation analysis limited to patients administered with antipsychotics. We also investigated temporal relationship (time to onset), biological gradient (association with daily dose, when available), reversibility of the adverse event (dechallenge and rechallenge), and the potential confounding role of drugs already acknowledged to cause ICDs (dopamine replacement therapy and previous treatment with other TGAs).
To explore the pharmacological basis of ICDs, we integrated pharmacosurveillance and pharmacodynamic data. We interpolated intraclass RORs (limited to patients administered with antipsychotics) with Homo Sapiens affinity data (pKi) from the ChEMBL database (Mendez et al., 2019).

Bradford Hill Criteria
Strong disproportionality was found for ICDs (strength) and for general impulsivity (coherence) with each TGA and was consistent across subpopulations (consistency). The RORs were higher for aripiprazole, followed by brexpiprazole and cariprazine (see Table 4). Even if RORs increased when calculated from lawyers' reports and after FDA warnings, the significance was preserved when restricting to reports submitted before the warnings and by physicians. The intraclass RORs The larger the association, the more likely that it is causal. Verified both on the entire database and considering only suspected drugs.

Consistency
Consistent findings are observed on different subpopulations taking the drug. The difference may be in reporter type (physician, patient, or lawyer) and country of occurrence (North America or elsewhere).

Coherence
Coherent findings are observed when investigating the association of the same drug with related events, in particular with an impulsivity substrate identified as "impulsive behavior" and "impulse-control disorder."

Exclusion of bias b
The association between drug and event persists when correcting for notoriety bias (before the regulatory warnings on aripiprazole [before March 5, 2016], between the warning for aripiprazole and that for brexpiprazole [March 5, 2016-February 2, 2018, after the warning for brexpiprazole [after February 2, 2018]), and for channeling bias (only patients administered with antipsychotics).

Specificity
Among patients administered with antipsychotics, the association is specific for the drug considered and is not common to other antipsychotics.

Temporality
The event has to occur after the drug is administered, and the time to onset (delay from the first administration of the drug to the date of occurrence of the event) is coherent with biological and clinical notions.

Descriptive
Biological Gradient There is a direct or inverse proportion between dose and occurrence of the event.

Descriptive Reversibility
If the drug is stopped the event stops (dechallenge), and if the drug is reintroduced the event occurs again (rechallenge).

Exclusion of confounders b
The event is not always explained by the co-administration of dopamine replacement therapy or other drugs known to induce ICDs (aripiprazole and brexpiprazole).

Descriptive
Abbreviations: ICD, impulse control disorder; ROR, reporting odds ratio. a Table showing the criteria adapted from Bradford Hill to assess causality in pharmacovigilance databases.
(disproportionality limited to patients using N05A drugs) showed a significant association between TGAs and ICDs, both as a unitary diagnostic entity and as individual behaviors (correction for bias). Other antipsychotics associated with the ICD unitary construct (specificity) were lurasidone (121 cases; associated with hyperphagia, hypersexuality, and compulsive shopping), olanzapine and ziprasidone (associated with hypherphagia), and iloperidone (associated with pathological gambling). Refer to supplementary Figure 3 for the distribution of antipsychoticrelated ICDs among different behaviors, and supplementary Figures 4 and 5 for antipsychotics intraclass disproportionality. Possible confounders were co-reported in 55 reports of aripiprazole (2.2%, mostly ropinirole and pramipexole). Thirtysix reports of brexpiprazole (20.2%) recorded an exposition to aripiprazole. Of the 32 reports of cariprazine, 9 also recorded aripiprazole or brexpiprazole, and 2 with both (exclusion of confounders). Dechallenge (ICD resolution after TGA discontinuation) occurred in more than 20% cases, reaching 40.6% for cariprazine (reversibility), whereas a positive rechallenge (ICD reappearance after TGA reintroduction) was recorded in only 1%-3% of the reports. The delay from the first administration of the TGA and the onset of the ICD ranged between days and years. The median time to onset was 89, 8, and 15 days for aripiprazole, brexpiprazole, and cariprazine, respectively (temporality), with Specificity c Even if on the whole FAERS disproportionality can be found for many antipsychotics, among patients administered with antipsychotics, only a few remain significant: the 3 TGAs and lurasidone, and with a more ambiguous signal olanzapine and ziprasidone (and iloperidone with a broader interval and the lower limit near to 1). third-generation antipsychotic. a Using the procedure described in Table 3, a causality assessment was performed for the 3 dopamine partial agonists. Reporting odds ratios were reported as median median daily doses of 10, 1, and 4.5 mg (other suspected reactions to cariprazine developed within an interquartile dose range of 1.5-3 mg; biological gradient). Literature evidence consisted of analogy to dopamine agonists; empirical, mostly observational, evidence; and biological plausibility related to the catecholaminergic activity. In conclusion, aripiprazole and brexpiprazole satisfied Bradford Hill criteria for causality assessment, apart from biological gradient (brexpiprazole had also little empirical evidence). Cariprazine, instead, satisfied all the criteria apart from empirical evidence. An extract of the 32 cariprazine cases is shown in the supplementary Table 2. The interpolation of disproportionalities and affinity towards different receptors (Figure 1) highlighted a common agonism for dopaminergic receptors (especially D2) shared by TGAs. The only molecular target shared also with lurasidone and other ICDassociated antipsychotics was 5-HT1A.

Further Evidence for TGA-Induced ICDs
To gather further evidence on the potential association between TGAs and ICDs, we adapted Bradford Hill criteria into a 12-point scheme for evidence assessment by combining literature and pharmacosurveillance disproportionality analyses.
The literature had already accrued strong observational evidence supporting the plausibility of a causal link between aripiprazole and ICDs (Moore et al., 2014;Mahapatra et al., 2016;Etminan et al., 2017;Lertxundi et al., 2018), whereas brexpiprazole and cariprazine were supported by only weak pharmacosurveillance evidence [7 cases each and significant RORs on the EudraVigilance database (Zazu et al., 2021)]. Nonetheless, all the TGAs partially share their pharmacodynamic profile with dopamine agonists, known to induce ICDs; partial agonism on D2/D3 receptors, along with a variegate catecholaminergic activity, could be responsible for influencing the neuroanatomical correlates of impulsivity.
Our study contributes with stronger, multifaceted, pharmacosurveillance evidence. TGAs were recorded in onequarter of the ICDs in the FAERS. A total 94% of the cases concerned patients exposed to aripiprazole. This drug has a long marketing life, a well-established notoriety, and is more commonly administered due to guideline promotion. Interestingly, the FDA warning concerning aripiprazole resulted in a strengthening of the ROR limited to aripiprazole. After the warning concerning brexpiprazole, a strengthening of the ROR was also observed for the other TGAs (known as ripple effect), maybe because it highlighted the plausibility of a drug classrelated effect. Time to onset for cariprazine and brexpiprazole was coherent with what is known of aripiprazole (days to several months) (Grall-Bronnec et al., 2016), and a high dechallenge rate was documented. The co-recording of multiple TGAs in 1.7% of the reports, plausibly referring to antipsychotic switching, complicated the assessment of newer TGAs. Doses were coherent with approved posology, and cariprazine cases reported higher doses than non-cases. Disproportionalities were significant for all TGAs, satisfying strength, consistency, coherence, and exclusion of bias criteria. The lower but significant intraclass ROR supported the existence of a disease-related susceptibility to ICDs not sufficient to explain the disproportionality of TGAs, lurasidone [a recent antipsychotic for which a case of hypersexuality has already been recorded (Reddy et al., 2018)], and olanzapine and ziprasidone (limited to hyperphagia). The high proportion of aripiprazole-related ICD cases in which the primary indication for use was a mood disorder is in line with the hypothesis that depression may be a synergistic risk factor for impulsive-compulsive behaviors (Cao et al., 2021).
Notably, TGAs and lurasidone are often considered safe alternatives to SGAs in patients fearing a bodyweight increase. In fact, both aripiprazole and lurasidone are classified in the antipsychotics group with the lowest risk for weight increase (Musil et al., 2015). Nonetheless, the occasional development of TGAinduced hyperphagia may be sufficient to determine weight gain, even in the lack of metabolic dysfunction. It may therefore be indicated to maintain careful weight monitoring in the early phases after the switch. Furthermore, a strict monitoring should be extended to all the drugs with a pharmacodynamic profile overlapping with that of TGAs. For example, lumateperone is a new multimodal agent characterized by dopaminergic presynaptic partial agonism and postsynaptic antagonism (Snyder et al., 2015). In part because of its recent approval (FDA, 2019), we found only 151 reports concerning lumateperone, with no record of ICDs.

Clinical Feature of TGA-Induced ICDs
A total of 42.3% of the TGA-related ICD reports recorded more than 1 ICD. This proportion is higher than that observed in dopamine-agonist-related ICDs in Parkinson's disease (approximately 25%) (Weintraub et al., 2010) and may be explained both by differences in the underlying population and by under-reporting of individual, less-impactful ICDs.
The influence of culture on the insurgence and clinical manifestation of ICDs is strong and already acknowledged (El Otmani et al., 2019). In agreement with past evidence, the high contribution of reports from North America, particularly in compulsive shopping and pathological gambling, could be related to a strong consumeristic culture (Parra-Díaz et al., 2020). FAERS data were also consistent with expected gender patterns (Weintraub et al., 2010;Santangelo et al., 2013), with men significantly more likely to experience hypersexuality and pathological gambling relative to other TGA reports. Instead, we failed to find a significantly higher proportion of binge eating and compulsive shopping in females. Gender patterns may be in part explained by gender norms, intended as "social norms defining acceptable and appropriate actions for women and men in a given group or society" (Cislaghi and Heise, 2020), and may therefore change across countries and time. Although ICDs are rarely related to Heat map comparing disproportionalities and receptor activity of multiple antipsychotics. Agonism and partial agonism are shown in warm tones, antagonism in cold tones, with intensity proportional to the affinity (pKi). The adjacent plot shows significant (in red) and non-significant (in grey) disproportionalities. fatal outcomes, disability and hospitalization are reported far more frequently than in other TGAs adverse events. Disability may reflect the strong impact that ICDs have on the life and functioning of patients and their caregivers and can be also observed in the pattern of co-reported events: "economic circumstances," "anxiety," "employment issues," "disability," "family and partner issues," "housing circumstances," and "criminal activity." The fact that binge eating results in less disability raises the question as to whether weaker stigma could reduce the disability related to other ICDs. The higher proportion of hospitalization may instead be related with the fact that 29% of the ICD cases (vs 5% non-cases; data not shown) recorded also self-injurious behaviors, with 734 reports recording "suicidal ideation," 670 "suicide attempt," and only 6 "suicide completed" (which explains the low proportion of deaths).
The strong contribution of reports by lawyers, instead, does not seem to be related only to the seriousness of ICDs, because ICDs induced by other drugs (e.g., dopamine agonists) have a low contribution by lawyers. Instead, this could be the result of both the over-signaling of lawyers, using the ICD diagnosis to defend their clients, and the sub-signaling of physicians, who consider ICDs as part of the illness and not a drug reaction. Hopefully the increase in the rate of reports by doctors and consumers regarding brexpiprazole and cariprazine, compared with aripiprazole, may imply an increasing awareness.
Finally, other conditions frequently co-reported support the use of an extended query for ICDs cases (eating disorders; ICDs; conduct disorders; obsessive-compulsive symptoms, such as hoarding, trichotillomania, dermatillomania; and suicidal and self-injurious behaviors). Therefore, we call for an effort to better define ICDs toward more standardized and sensitive studies.

Insights in ICD Pathogenesis
Of note, lurasidone, along with brexpiprazole, does not share with aripiprazole and cariprazine the partial agonism on D3 receptors, which is strongly advocated as the primary mechanism in the onset of dopaminergic agent-induced ICDs (Seeman, 2015). Instead, it shares a strong agonism on 5-HT1A (Huang et al., 2012), an inhibitory auto-receptor located in the dorsal raphe. 5-HT1A inhibits the serotonergic circuit, which is usually responsible for the inhibition of the dopaminergic neurons departing from the ventro-tegmental area and directed to the ventral striatum. Therefore, 5-HT1A may inhibit the pathway responsible for impulse control. This is coherent with the finding, in Parkinson's disease, that ICDs are associated with an increased release of dopamine in the ventral striatum (Martini et al., 2018), which opposes the prevailing hypothesis of a primary role related to the exogenous dopaminergic activity. Interestingly, different studies have already demonstrated 5-HT1A-induced impulsivity and addiction in mice (Humby et al., 2020;Chu et al., 2021) and investigated the relationship between impulsivity and 5-HT1A polymorphisms in humans (Benko et al., 2010;Stamatis et al., 2020). Furthermore, 5-HT1A agonists such as buspirone and vortioxetine may cause impulsive-compulsive behaviors in rats (Liu et al., 2004;Lu et al., 2013;Martis et al., 2021), and a 5-HT1A partial agonist (tandospirone) may result in reduced impulsivity in rats because of its 5-HT1A antagonistic activity (Ohmura et al., 2013). Therefore, based on our pharmacovigilance analyses and on the supporting literature, we raise the hypothesis of 5HT1A agonism as a plausibly pivotal mechanism of the disinhibition proper of TGA-related ICDs, and 5-HT1A antagonism as a promising approach for therapeutic intervention. Nonetheless, we acknowledge that a high pKi does not always result in high receptor occupancy and that positron emission tomography occupancy studies found undetectable occupancy for brexpiprazole on both D3 and 5-HT1A receptors (Girgis et al., 2020). Therefore, further studies investigating the pharmacodynamics of TGAs and other drugs involved in the development of ICDs are required.

Strengths and Limitations
The analysis of spontaneous adverse events reports, peculiar to pharmacosurveillance studies, does not allow formal risk assessment and requires considering multiple biases. In particular, researchers must contend with unverified reports, interreporter heterogeneity in the choice of terms, under-reporting, and the influence of mass media and regulatory warnings (i.e., notoriety bias). Nonetheless, it allows to collect great numbers of cases from a large heterogeneity of environments and to account for clinical errors, populations under-represented in clinical trials, comorbidity, and interactions. Through our in-depth analysis and adapted Bradford Hills Criteria, we integrated multiple data from the FAERS database to provide more evidence regarding TGA-induced ICDs. Although acknowledging the heterogeneous compilation of the reports, we investigated the potential role of the drugs in determining the event without neglecting the predisposition of the individual, related to concomitants and underlying disease. Note that other kinds of predisposition, such as genetic or environmental, could not be assessed through the limited information of the reports. As Hill himself clearly understood, the validation of these criteria supports evidence of a causal relationship, but no amount of evidence will ever be enough to confirm a suspicion, nor will a lack of evidence will ever be enough to dismiss it. Hence, further clinical studies are required.

Conclusion
This study gathered FAERS data and literature evidence to assess the link between TGAs and ICDs and raised several insights and hypotheses deserving further investigation. Our findings, even if not sufficient to establish a causality, are adjusted for major biases and can help to better define the path for regulatory actions towards a more conscious and safer use of TGAs. The fulfilled Bradford Hill criteria support the warning already issued and highlight, for the first time-to our knowledge-the plausible role of cariprazine in inducing ICDs. In particular, when switching to TGAs because of SGA-induced weight gain, early detection is important of any potential surge in time spent eating, which could itself be responsible for weight gain. Lurasidone also displays associations with individual and grouped ICDs.
A better ICDs syndrome characterization is pivotal to support monitoring of the patient by the caregivers. The features of ICD reports are consistent with what is already known from clinical studies, with sex differences in the ICD manifestation. The analysis of co-reported events supports the strong impact of ICDs on the social and psychological life of individuals and their families and points to the need for an extended Standardized MedDRA Query able to detect more ICD reports.
Finally, a promising target to understand and treat ICDs is also represented by 5-HT1A, which may be responsible for the disinhibition, resulting in the progression from an everyday gratifying behavior toward an ICD and, eventually, a compulsion. Further knowledge about ICD mechanisms may be extremely valuable to a more conscious switch between antipsychotics and to the search of pharmacological treatments for ICDs among new and repurposed drugs.