Cerebellar-Cortical Connectivity and Prediction of Antipsychotic Treatment Response in First-Episode Psychosis

Abstract

Background

The cerebellum has traditionally been associated with motor functions, but recent evidence highlights its critical role in cognitive and emotional regulation, contributing to the neuropathology of schizophrenia. Our previous data-driven research demonstrated that cerebellar-cortical functional connectivity can predict antipsychotic treatment outcomes in first-episode psychosis (FEP). The present study aimed to investigate specific cerebellar functional systems involved in treatment prediction.

Study Design

This study included 127 patients with FEP who underwent 12 weeks of antipsychotic monotherapy (either risperidone or aripiprazole). Baseline resting-state functional MRI data were collected from two 3T scanners, and functional connectivity between 10 predefined cerebellar functional systems and the whole brain was analyzed. Psychotic symptom changes were measured using the Brief Psychiatric Rating Scale-Anchored version (BPRS-A). Connectivity patterns were examined in relation to treatment outcomes.

Study Results

Higher baseline connectivity between the cerebellar auditory system and cortical regions, including the visual cortex, dorsolateral prefrontal cortex, and the hippocampus, predicted worse treatment outcome. In contrast, stronger connectivity between cerebellar cognitive systems (default mode and frontoparietal networks) and the anterior cingulate cortex (ACC) and medial prefrontal cortex was associated with better treatment outcome. These findings were consistently present in data acquired from both scanners and both drugs.

Conclusions

Our results identify specific cerebellar-cortical circuitries as prognostic biomarkers for predicting psychosis treatment outcomes, and suggest that cerebellar auditory and cognitive systems may be potential targets for future interventions aimed at improving treatment efficacy in FEP.

Introduction

The integration of cognitive processes, emotion, and movement within the cerebellum plays a pivotal role in the pathophysiology of schizophrenia, as elucidated by Andreasen’s “cognitive dysmetria” hypothesis. This theory suggests that cerebellar dysfunction leads to a significant discoordination within these domains, a notion further supported by the observed alterations in cerebellar-cerebral loops among schizophrenia patients.^1,2 Such disruptions not only compromise the individual’s mental and emotional equilibrium, but also underscore cerebellar connectivity as a potential biomarker for assessing responsiveness to antipsychotic treatments. Emerging studies, including those from our own work,^3–5 demonstrate the significant predictive value of baseline cerebellar-cortical connectivity as a robust biomarker for psychosis onset among individuals at clinical high risk and for psychosis treatment outcome among first-episode psychosis (FEP) patients.

The heterogeneity in treatment responses to antipsychotic medication among individuals with FEP highlights the critical need for early identification of prognostic biomarkers.⁶ Given the demonstrated benefits of early intervention for patients’ long-term health and functionality, unveiling early predictors becomes indispensable. Antipsychotic medications, principally acting through the modulation of dopamine receptor activity, induce alterations in brain circuitry. Therefore, neurobiological measures of such circuitry functions may serve as viable predictors of therapeutic outcomes.⁷ In our previous data-driven, brain connectome-wide analysis, we revealed baseline connectivity between the cerebellum and the cerebral cortex as the most valuable predictors across the entire brain.^4,5 This is in line with the recent findings that the cerebellum is a key regulator of the brain’s dopaminergic system and sends direct monosynaptic output to the midbrain VTA neurons to modulate their activities.⁸

These findings raise a further question: Which cerebellar systems would play a critical role in predicting treatment response in FEP? The answer to this question is important as it will help refine and develop system-specific predictors that may eventually benefit clinical translation. Notably, the cerebellum is a highly heterogeneous brain structure in terms of its function. Recent studies have identified that the cerebellum can be divided into different functional subsystems that highly resemble those discovered in the cerebrum, including attention, auditory, language, visual, memory, sensory integration, cognitive control, and default mode.^9–11 The majority of these systems have been shown to be implicated in schizophrenia. For instance, the function of the auditory and visual networks is strongly related to hallucinations, a hallmark of psychosis and a core symptom target of antipsychotic medications,^12,13 bringing the possibility that their connectivity may predict treatment response. The default mode and cognitive control networks, known for their roles in modulation of internally focused and externally oriented thoughts and memory, may exert their controls over sensory systems and in turn regulate the brain’s capacity to respond to antipsychotic treatment.^14–16 However, while only a curated number of cerebellar nodes were studied in the prior work, it remains unknown which cerebellar systems would show the best predictive value.

In this study, we sought to establish a clear link between functional connectivity of specific cerebellar subsystems and their impacts on treatment effectiveness. To this end, we collected data from 2 first-episode patient samples with different imaging scanners. All patients received antipsychotic monotherapies with either risperidone or aripiprazole for 12 weeks. Resting-state fMRI scans were performed at baseline to quantify functional connectivity between each cerebellar system and the whole brain, whose relationship with the treatment outcome was further examined. This approach provides an opportunity to expand the findings from our previous study⁵ and give a more fine-grained delineation of specific cerebellar systems involved in the prediction. We expected to see the connectivity of sensory systems (especially auditory and visual) and cognitive systems (especially frontoparietal and default mode) to be the most predictive of treatment response in patients.

Materials and Methods

Subjects

This study involved 2 samples of FEP patients with a similar design, both recruited from the Zucker Hillside Hospital. One sample received imaging scans with a 3T GE Signa scanner (hereafter the “GE sample,” 85 subjects) as part of their participation in a double-blind, randomized trial comparing aripiprazole and risperidone in FEP (ClinicalTrials.gov ID: NCT00320671),¹⁷ and the other received scans with a 3T Siemens Prisma scanner (hereafter the “Siemens sample,” 91 subjects) during a study that included naturalistic treatment with aripiprazole or risperidone (ClinicalTrials.gov ID: NCT02822092). All participants were diagnosed with a psychosis spectrum disorder based on the Structured Clinical Interview for DSM-IV, and were at the early stage of their illness, defined as having been exposed to antipsychotic medications for less than 4 weeks in their lifetime. Exclusion criteria included serious neurological or endocrine disorders, a medical condition requiring medication with psychotropic effects, significant risk of suicidal or homicidal behavior, drug abuse or dependence, cognitive or language limitations, inability to sign an informed consent, and contraindications to antipsychotic monotherapy or MRI scans. All participants gave written informed consent in accordance with the ethical standards of the Northwell Health Institutional Review Board. A flowchart summarizing the overall process is shown in Figure 1.

At the start of the study, all participants underwent a clinical assessment using the Brief Psychiatric Rating Scale–Anchored version (BPRS-A) and a resting-state fMRI scan. They subsequently received 12-week treatment with either risperidone or aripiprazole. Follow-up visits occurred weekly for the first 4 weeks and then every 2 weeks for the remaining eight weeks, with the BPRS-A readministered at each visit to monitor symptom progress. Since antipsychotic medications typically take up to 4 weeks to reach their full therapeutic effects, in the following analysis we only included participants with at least 4-week follow-up, which is consistent with prior publication.¹⁸ After subject filtering and image quality control, the final dataset comprised a total of 66 individuals from the GE sample (mean age 21.88 ± 5.5 years, 49 males) and 61 individuals from the Siemens sample (mean age 23.89 ± 5.66 years, 33 males). Note that the excluded patients had similar demographic (age, sex) and clinical profiles (baseline psychotic and total symptom scores) compared with the included sample (GE sample: P > .31; Siemens sample: P > .34). The included sample details are shown in Table 1.

Definition of Treatment Response

Treatment response was defined using the same approach as our previous work.⁵ Specifically, we calculated a “psychosis score” by summing 5 items from the BPRS-A, namely, conceptual disorganization (item 4), grandiosity (item 8), suspiciousness (item 11), hallucinatory behavior (item 12), and unusual thought content (item 15). Using these scores, we generated individual slopes representing change in psychosis symptoms over time. The slopes were calculated using a linear mixed model, where time point was treated as a fixed-effect variable and subject as random-effect variable. This approach allowed us to quantify the rate of symptom change, providing a quantitative measure of treatment response that could be correlated with neuroimaging data.

MRI Data Acquisition

The BOLD images of the GE sample were collected using a single-band EPI sequence: repetition time (TR) of 2000 ms, echo time (TE) of 30 ms, field of view (FOV) of 240 mm × 240 mm, slice thickness of 3 mm, 40 contiguous slices, and voxel size of 3.75 mm × 3.75 mm × 3 mm. The total time for acquiring resting state images was 5 minutes. Additionally, T1 structural images were captured using an inversion-recovery prepared 3D fast spoiled gradient (IR-FSPGR) sequence, with TR of 7.5 ms, TE of 3 ms, inversion time (TI) of 650 ms, FOV of 240 mm × 240 mm, slice thickness of 1 mm, and 216 contiguous slices.

The Siemens sample employed a multi-band accelerated echo-planar imaging technique detailed in the Human Connectome Project. Every participant underwent a T1-weighted scan (with TR of 2400 ms, TE of 2.22 ms, voxel size of 0.8 mm³, and duration of 6 minutes and 38 seconds) alongside 2 resting-state fMRI runs (with opposite phase encoding directions, one with anterior-posterior, the other posterior-anterior). Each resting-state fMRI run lasted 7 minutes and 17 seconds (i.e., 594 volumes). The sequence included 72 consecutive axial/oblique slices aligned with the anterior commissure-posterior commissure (AC-PC) line (720 ms TR, 33 ms TE, 208 mm field of view (FOV), and 2 mm × 2 mm × 2 mm voxel size, and eightfold multi-band acceleration).

Image Preprocessing

The preprocessing of the GE sample adhered to a conventional protocol facilitated by the SPM12 software suite (available at https://www.fil.ion.ucl.ac.uk/spm/software/spm12/). This pipeline encompassed initial slice timing correction to adjust for temporal discrepancies between slices, subsequent realignment to compensate for any head motion artifacts, the co-registration of structural and functional imaging data, and the normalization of these images to the Montreal Neurological Institute (MNI) standard space. In addition, spatial smoothing was performed using a Gaussian kernel with a full width at half maximum (FWHM) of 4 mm to enhance the signal-to-noise ratio and compensate for intersubject anatomical variability.

The Siemens sample was preprocessed by utilizing the Human Connectome Project (HCP) pipeline as delineated by Glasser et al.¹⁹ This process comprised 5 principal stages: PreFreeSurfer, FreeSurfer, PostFreeSurfer, fMRI Volume, and fMRI Surface. Briefly, the initial steps involved the correction of images for distortions caused by gradient nonlinearity, head motion, and phase-encoding. Subsequently, these images were registered to individual T1-weighted images and normalized to the Montreal Neurological Institute (MNI) standard space to facilitate consistent anatomical framework for analysis, and spatially smoothed using 4 mm FWHM.

All patients’ functional data were inspected for excessive head motions by checking their framewise displacement (FD) values. Patients with a mean FD value higher than 0.5 were considered to have excessive head motion and were excluded from the study. Consequently, 3 patients from the GE sample and 2 patients from the Siemens sample were excluded due to head movement issues.

Seed-Based Connectivity Analysis

Seed-based connectivity was conducted using FSL (ver. 6.0.6, FMRIB’s Software Library). Here, seeds were defined as the ten cerebellar functional systems introduced by Ji et al., 2019, including the attentional functions, auditory functions, cingulo-opercular tasks, default mode network, frontoparietal network, language processing, multimodal integration system, sensorimotor functions, and the primary and secondary visual processing systems²⁰ (Figure 3). For each participant, time-series data from these cerebellar seeds were extracted using the fslmeant tool. To mitigate potential confounds arising from artifacts, time-series data from white matter (WM) and cerebrospinal fluid (CSF) were also extracted and used as control regions.

The analytical framework employed was a general linear model (GLM), where the time-series from each cerebellar seed were included as the independent variable, and time-series of each voxel across the brain were included as the dependent variable. This GLM included 27 explanatory variables (EVs): one representing the cerebellar seed region’s time-series, 24 parameter regressors for participant movement, and 2 for signal fluctuations from WM and CSF, respectively. The 24 movement regressors used in the analysis included the following: 6 head-motion parameters (3 translation and 3 rotation parameters) derived from the realignment step, along with their first derivatives. Additionally, the square of these 12 parameters was also included. These regressors were incorporated to mitigate the effects of motion and physiological noise, ensuring the correlations observed were due to neural connectivity rather than artifacts.

Following the first-level analysis, we collected all contrast of parameter estimate (COPE) images quantifying whole-brain connectivity with each cerebellar seed from both the GE and Siemens samples. To control for scanner-related effects, the NeuroCombat²¹ tool was used to harmonize the images across all subjects from the 2 samples. This was conducted for each brain voxel in the COPE images derived from each cerebellar seed. The harmonized images were then entered into the group-level statistical analysis, where individual symptom slope scores were included as the variable of interest, controlling for age, sex, and FD. Significance was tested by 5000 permutations, where subjects were randomly shuffled during each permutation. We applied a significance threshold of Family-Wise Error Rate (FWE) corrected P < .05 using the Threshold-Free Cluster Enhancement (TFCE) approach.

RESULTS

The linear mixed model showed a significant decrease in psychosis scores following antipsychotic treatment across patients. The group-level slope was −0.67 (P < .0001), indicating that, on average, psychosis scores decreased by 0.67 points per week (Figure 2). This finding suggests a consistent reduction in psychotic symptoms over the course of the treatment. Consistent with prior studies on antipsychotic medications,^22–24 we observed no significant reduction in negative symptoms. Consequently, our analysis focused on positive symptoms, where significant treatment-related changes were observed.

Our analysis revealed that connectivity of 3 cerebellar seeds (default mode, frontoparietal, and auditory) significantly predicted treatment response in patients. Specifically, higher connectivity between the cerebellar auditory seed and 3 cortical areas, namely, the visual cortex, dorsolateral prefrontal cortex (DLPFC), and hippocampus was related to larger (i.e., less negative) slope coefficients, suggesting that their connectivity strength negatively predicted treatment outcome in patients. In contrast, the opposite effects were observed for the cerebellar default mode network (DMN) and frontoparietal network (FPN) seeds. Specifically, for both seeds, higher connectivity with the anterior cingulate gyrus (ACC) and medial prefrontal cortex was associated with smaller (ie, more negative) slope coefficients, and therefore better treatment outcome in patients. Detailed results of the identified brain regions are present in Figure 3 and Table 2.

To further confirm that the results above were not driven by data from a single scanner, we conducted a Supplementary Analysis. Here, for the observed significant regions, we extracted their connectivity values and correlated these values with individual slopes, separately for the GE and Siemens scanners. We found significant associations between cerebellar connectivity and individual slopes in both scanners and for all 3 seeds (Auditory Seed: GE data r = 0.403, P < .001; Siemens data r = 0.376, P = .003. Default Mode Seed: GE data r = -0.393, P = .001; Siemens data r = −0.490, P < .001. Frontoparietal Seed: GE data r = −0.309, P = .012; Siemens data r = −0.451, P < .001), suggesting that the observed findings are consistently present in both scanners (Figure 3).

We additionally examined whether results derived from different cerebellar systems were intercorrelated. We found that results of the auditory seed were moderately associated with the results of other seeds (r = −0.247, P = .005 for default mode and r = −0.282, P = .001 for frontoparietal). In contrast, results of the default mode and frontoparietal seeds exhibited a strong correlation (r = 0.774, P < .001), suggesting a potentially shared mechanism underlying these 2 systems.

No significant differences in treatment response were observed between risperidone and aripiprazole. Analysis of cerebellar seed connectivity by medication type (risperidone vs. aripiprazole) revealed no significant differences in connectivity patterns between the 2 groups. Specifically, connectivity between auditory seed and BPRS-A psychosis symptoms yielded correlation coefficients of r = 0.348, P < .001 for risperidone and r = 0.4, P = .019 for aripiprazole. Similarly, the default mode seed showed correlations of r = −0.319, P = .002 for risperidone and r = −0.516, P = .002) for aripiprazole, while the frontoparietal seed demonstrated correlations of r = −0.321, P = .002 for risperidone and r = −0.444, P = .008 for aripiprazole.

Discussion

In this study, we identified that resting-state functional connectivity of 3 cerebellar systems (auditory, DMN, FPN) was predictive of antipsychotic treatment outcomes in first-episode psychosis. Specifically, higher connectivity of the cerebellar auditory system with the visual cortex, dorsolateral prefrontal cortex, and hippocampus predicted worse treatment response; while higher connectivity of the cerebellar DMN and FPN systems with the ACC and medial frontal cortex predicted better treatment response. These findings suggest nuanced cerebellar-cortical circuitries as prognostic biomarkers for psychosis outcome.

The results of our study align with a growing body of research that implicates cerebellar-cortical connectivity as a significant factor in the pathophysiology of schizophrenia and its treatment. Previous studies have highlighted the cerebellum’s role beyond motor control, emphasizing its involvement in cognitive and emotional regulation, which are often disrupted in schizophrenia.^1,2 Moreover, in our previous paper, we demonstrated that cerebellar-cortical connectivity may serve as a potential treatment response biomarker in a connectome-wide machine-learning analysis.⁵ The present work extended our previous finding by pinpointing specific cerebellar systems in the prediction of antipsychotic treatment, thereby highlighting refined cerebellar connectivity measures as potential prognostic biomarkers in patients with first-episode psychosis.

The findings from our study highlight that higher connectivity between the cerebellar auditory system and various cortical regions, including the visual cortex, DLPFC, and hippocampus is significantly associated with poorer treatment outcomes in FE psychosis patients. The connection between the auditory cerebellar subsystem and the DLPFC may reflect disruptions in sensory prediction processes, a core feature of schizophrenia pathophysiology. According to sensory prediction error models,²⁵ the cerebellum plays a critical role in integrating sensory inputs and providing error signals to higher-order regions like the DLPFC. These processes are essential for maintaining accurate top-down control over sensory information and preventing the generation of hallucinations. Hyperconnectivity between the cerebellar sensory system and the prefrontal cortex may indicate impaired processing in prediction errors and in turn abnormal top-down control over the auditory cortex, leading to an amplification of internally generated auditory signals. This aligns with prior evidence showing that aberrant prefrontal-auditory connectivity contributes to hallucinations through impaired top-down modulation.^26,27 This suggests a dysregulation in sensory-cognition integration, where excessive connectivity could overwhelm or distort sensory perceptions leading to hallucinatory experiences. In addition, the prefrontal cortex is heavily involved in the dopaminergic system, which plays a significant role in the pathophysiology of schizophrenia, particularly in the manifestation of positive symptoms such as delusions and disorganized thinking. The abnormal connectivity patterns observed might reflect a dysregulation of dopamine pathways, which can disrupt cognitive processes and decision-making, worsening the patient’s clinical symptoms.^28,29 The hippocampus, part of the limbic system, displays abnormal connectivity linked to its roles in memory and emotional regulation, which are pivotal in schizophrenia. Stronger hippocampal connectivity is particularly associated with exacerbating hallucinations and delusions, reflecting its critical involvement in the disorder’s cognitive and emotional disturbances.^30,31 Related to this during the hallucinatory behavior hippocampus shows stronger connectivity with the prefrontal cortex and visual cortices.³² These interpretations are consistent with existing literature, which reports significant dopaminergic, pathological, and cognitive roles of these cerebro-cortical areas in schizophrenia’s neural network disruptions.^3,33,34

Together, stronger connectivity between these cortical regions with the cerebellar auditory system may reflect an interference in the brain’s ability to efficiently filter and process information, leading to worse clinical outcomes. Moreover, this directionality may also suggest an excessive cerebellar influence on the brain’s dopamine system that contributes to dopaminergic dysregulation, further impairing cognitive functions. Consequently, our results underscore the need for further investigation into the cerebellum’s contribution to cortical dysfunctions and its broader role in modulating dopaminergic sensitivity, which may influence the clinical trajectories of schizophrenia patients.^2,5

Our findings also indicate that cerebellar cognitive systems, specifically the DMN and FPN, play a key role in treatment prediction. Here, stronger connectivity between these systems and the ACC and medial frontal cortex predicted better treatment responses. Both the anterior cingulate cortex and the medial frontal cortex (MFC) are critical for emotional regulation, decision-making, and conflict monitoring. These regions have been implicated in the cognitive control processes that are often disrupted in schizophrenia, and their higher connectivity may reflect a more adaptive response to antipsychotic treatment.³⁵ The ACC is known for its role in error detection and motivational responses, which are essential for adjusting behaviors in complex environments.³⁶ Meanwhile, the MFC contributes to the regulation of self-referential mental activity and emotional responses.³⁷ Their stronger connectivity with the cerebellum may suggest enhanced cognitive flexibility and emotional stability in patients responding positively to treatment. Moreover, it may imply that the cerebellum’s modulation of these frontal regions helps to stabilize the cognitive and emotional processes that are otherwise dysregulated in psychosis.^4,38 Furthermore, studies highlight that the cerebellum directly projects to the prefrontal cortex, including the ACC and MFC, suggesting a structural basis for the observed functional connectivity and its impact on cognitive and emotional regulation in psychiatric conditions.^39,40 These projections are thought to facilitate predictive coding frameworks within the brain, enhancing the prediction and integration of behavioral outcomes, which is crucial for effective treatment responses.^5,41,42

Clinically, our findings underscore the potential of specific cerebellar systems as neural targets for future interventions in psychosis. Crus I and Crus II, regions of the posterior lateral cerebellar hemispheres, are involved in higher-order cognitive functions. They are an integral part of the cerebellum's DMN and FPN systems, which regulate cognitive processes. Crus I is linked to the DMN and is associated with self-referential thoughts and social cognition, often disrupted in schizophrenia and related to symptoms like delusions.⁴³ Crus II connects to both the DMN and FPN, playing a role in cognitive control and decision-making, with its altered activity affecting cognitive flexibility in schizophrenia.^44,45 The strong connectivity of Crus I and II with these networks suggests they are crucial in modulating cognitive and emotional functions such as working memory⁴⁶ and language⁴⁷ processing. Structural and functional changes in Crus I and II have been implicated in the progression of schizophrenia, with alterations contributing to cognitive deficits and impaired network connectivity across different clinical stages.^44,48,49 Similarly, lobules V and VI, associated with the auditory cerebellar sensory system, are crucial for processing auditory information and integrating sensory input.⁵⁰ These lobules play a role in modulating attention and sensory perception, processes that are often disrupted in psychosis. Their connectivity with auditory networks suggests they are important for maintaining auditory processing fidelity, which can influence the treatment response in auditory-related symptoms of schizophrenia. Neuromodulation techniques, such as transcranial magnetic stimulation (TMS), may be employed to modulate cerebellar connectivity of these specific areas to improve treatment outcomes.⁵¹ This approach brings possibilities for personalized schizophrenia treatment, where individual response to antipsychotics may be boosted by targeting system-specific cerebellar-cortical networks.⁵²

Our study has several limitations that should be acknowledged. First, data were collected using different scanners and protocols. Although we harmonized the data using NeuroCombat and further confirmed a similar effect between the 2 samples, the scanner and protocol-related confounding effects may not be completely removed. Future studies with larger, more homogeneous samples collected from the same center and scanner are preferred to further validate these findings. Second, due to the sample size limit, we did not apply a multiple comparison correction across all seed regions, which should be addressed in future research to ensure robustness when analyzing multiple seed regions simultaneously. However, in a Supplementary Analysis, we reduced the number of multiple comparisons by grouping cerebellar seeds into two systems (sensory and cognitive), and observed similar findings to single seeds, suggesting that the reported findings are unlikely to be false positives. Third, for ethical reasons, no placebo group was included in this study, making it difficult to determine whether the detected predictors could relate to a placebo effect. Moreover, the use of only 2 types of antipsychotic medications may not capture potential differences in treatment response patterns between different antipsychotic medications. However, all approved antipsychotic medications block dopamine D2 receptors, making it likely that findings could be generalizable across different antipsychotic medications, with the exception of clozapine, which has higher response rates and a more unique profile in terms of receptor occupancy. Future studies with larger, more diverse populations and a broader range of medications are needed to validate and expand on our findings.

In conclusion, this study underscores the importance of cerebellar-cortical connectivity in prediction of response to antipsychotic treatment in first-episode psychosis, and points to the cerebellar auditory, DMN, and FPN systems as key contributors to the prediction. These findings may potentially advance the approach to managing first-episode psychosis, leading to more effective and personalized therapeutic interventions in the future.

Funding

This work was funded by the NIH grant R01MH138682, the Alkermes Pathways Research Award and the Feinstein Institutes Barbara Zucker Emerging Scientist Award to Dr Cao, the NIH grants P50MH080173 and R01MH108654 to Dr Malhotra, and NIH grant R01MH060004 to Dr Robinson.

Author contributions

Hengyi Cao and Anil K. Malhotra share senior authorship.

Conflict of interest

Dr Cao receives funding from Alkermes. Dr Gallego has served as a speaker for Tecnoquimicas. Dr Birnbaum has served as a consultant for Northshore Therapeutics and HearMe. Dr Robinson has served as a consultant for Acadia, Advocates for Human Potential, Amalyx, APA, C4 Innovations, Costello Medical Consulting, Health Analytics, Innovative Science Solutions, Janssen, Lundbeck, Neurocrine, Neuronix, Otsuka, Teva, and US WorldMeds and has received grant support from Otsuka. Dr Malhotra has served as a consultant for Acadia Pharma, Clarivate, Genomind, Health Advances, InformedDNA, Iqvia, and Janssen Pharma. The other authors report no financial relationships with commercial interests.

References

Author notes