Multimodal Brain Function in Migraine Patients With Patent Foramen Ovale

This is an investigator-initiated, single-center, prospective, longitudinal, self-controlled cohort study evaluating multimodal brain function changes in adults with migraine with aura and echocardiographically confirmed patent foramen ovale (PFO) who undergo clinically indicated percutaneous PFO closure as part of routine care. The study is designed to characterize within-person trajectories of resting-state neurophysiology, brain network organization, and neurocognitive performance from pre-closure baseline through post-closure follow-up, using standardized acquisition and analytic pipelines for high-density electroencephalography (hdEEG), resting-state functional MRI (rs-fMRI), and cognitive testing. Overall design and clinical workflow integration Participants are identified through routine clinical pathways (cardiology/structural heart disease and headache/neurology services). Study procedures are embedded within the standard pre-procedural assessment and post-procedural follow-up schedule whenever feasible to reduce participant burden and improve data completeness. The PFO closure procedure itself, device selection, peri-procedural management, and post-procedure antithrombotic therapy follow local clinical practice and are not assigned by the study. Participants complete a pre-closure baseline assessment and then undergo repeated assessments during follow-up. Study visits are time-locked to clinically meaningful recovery phases after PFO closure and are intended to capture early neurophysiologic changes, intermediate adaptation, and longer-term stabilization of brain network measures. Multimodal assessments and acquisition procedures Resting-state fMRI acquisition and quality control Resting-state fMRI is acquired on a 3.0T clinical MRI system using standardized sequences suitable for network-level analyses. Participants are instructed to remain still, stay awake, and follow a uniform resting condition (e.g., eyes closed or eyes open with fixation) across all visits to maximize longitudinal comparability. Imaging quality control emphasizes minimization of head motion and physiological artifacts. Data are screened for completeness, scanner protocol deviations, and motion outliers; scans failing prespecified QC thresholds (e.g., excessive motion) are flagged for exclusion from network analyses, while retaining other usable outcomes when appropriate. hdEEG acquisition and quality control High-density EEG is recorded under standardized resting-state conditions. The acquisition protocol is harmonized across visits, including participant preparation, impedance checks, recording duration, and environment control (e.g., low-noise room, standardized instructions). Recording quality control includes monitoring for excessive muscle activity, eye movement contamination, channel artifacts, and drowsiness. Raw data are archived to enable reproducible preprocessing and re-analysis if needed. Neurocognitive and symptom-related assessments Neurocognitive performance is assessed using the MATRICS Consensus Cognitive Battery (MCCB) administered by trained staff following standardized procedures and scoring rules. Additional clinical assessments collected as part of the study support interpretation of multimodal brain measures over time (e.g., headache diary-derived metrics and validated patient-reported measures). These measures are collected with consistent timing relative to imaging/EEG when feasible to reduce measurement variability. Neuroimaging and EEG preprocessing and network construction rs-fMRI preprocessing Resting-state fMRI preprocessing follows established pipelines for connectome-based analyses, including: removal of initial volumes (if applicable), slice timing correction (if used), motion correction, coregistration to individual anatomical space (when available), normalization to standard space, nuisance regression (e.g., motion parameters; optionally physiological or WM/CSF signals depending on the finalized pipeline), temporal filtering, and spatial smoothing consistent with the selected atlas and connectivity approach. Motion is quantified using standard metrics (e.g., framewise displacement). Scrubbing/censoring strategies and sensitivity analyses (e.g., with/without global signal regression) may be applied to evaluate robustness. hdEEG preprocessing EEG preprocessing includes band-pass filtering, bad channel detection/interpolation, artifact attenuation (e.g., independent component analysis or equivalent methods to address ocular and muscle components), re-referencing strategy consistent with high-density recordings, and segmentation into resting epochs. Frequency-domain features (e.g., power spectral density in canonical bands) and connectivity measures are derived using standardized parameters across visits. Parcellation (brain "atlas") and network generation For rs-fMRI, the brain is parcellated into regions of interest (ROIs) using a predefined atlas (e.g., a functionally defined multi-network parcellation such as a Schaefer-based atlas, finalized prior to database lock). Regional time series are extracted per ROI and pairwise connectivity matrices are constructed using correlation-based metrics (with Fisher z-transformation where appropriate). For EEG, connectivity is computed using phase-based measures that reduce volume-conduction bias (e.g., phase-lag index-type metrics), with a focus on prespecified frequency bands relevant to resting-state network physiology. Network representations may include whole-brain connectivity matrices and derived within-network/between-network summaries aligned to canonical functional systems (e.g., default mode, salience, executive control networks), finalized a priori. Statistical considerations and analysis approach (high level) Primary analytic framework The primary analyses are within-person comparisons between pre-closure baseline and follow-up timepoints for multimodal brain function and cognition endpoints. Longitudinal modeling will be implemented using approaches appropriate for repeated measures, such as linear mixed-effects models with participant-specific random effects, allowing inclusion of all available timepoints and accommodating incomplete follow-up under a missing-at-random assumption. Time is modeled as categorical (visit-based) or continuous (time since closure) depending on the finalized specification. Multiple primary endpoints and multiplicity control Because the study includes two primary domains (neuroimaging/EEG-derived brain function and MCCB cognition), multiplicity control is applied to preserve the overall type I error rate. A Bonferroni-adjusted significance threshold is used for confirmatory testing across the two primary endpoints (i.e., α split across two primary tests). Secondary and exploratory analyses are interpreted with appropriate caution and are primarily hypothesis-generating. Network-level inference and feature selection (exploratory components) For connectome-wide rs-fMRI analyses, network-based inference methods (e.g., network-based statistics with permutation testing) may be used to identify subnetworks showing coherent changes across edges while controlling family-wise error at the component level. For EEG connectivity, analogous cluster-based or network-based permutation approaches may be used depending on the finalized feature set. To relate multimodal network features to cognitive performance and clinical measures, multivariable modeling strategies may be applied. If the feature space is high-dimensional relative to sample size (e.g., edge-wise connectome features), regularization and feature-selection methods (e.g., LASSO or elastic net) can be used in an exploratory framework with cross-validation to reduce overfitting. Selected features can then be summarized (e.g., network strength indices) and examined for longitudinal association with MCCB change or symptom trajectories. These models are explicitly exploratory unless prespecified and adequately powered for prediction. Covariates and confounding control Models will consider key covariates that can influence brain network measures and cognition (e.g., age, sex, education, head motion metrics for fMRI, medication changes, and other clinically relevant factors). Sensitivity analyses may evaluate the impact of motion thresholds, preprocessing choices, and alternative atlas resolutions. Missing data and data quality Missingness may arise from loss to follow-up or unusable imaging/EEG data. The analysis plan prioritizes (1) preventing missing data through scheduling integration and QC feedback loops, and (2) transparent handling of missingness. Mixed-effects models enable inclusion of participants with partial follow-up, while imaging/EEG-specific QC exclusions are documented with reasons. Where applicable, sensitivity analyses will compare complete-case and all-available-data approaches. Safety oversight and adverse event documentation Although the study is non-interventional with respect to treatment assignment, it prospectively captures adverse events temporally associated with PFO closure and routine post-procedure management. Adverse events and serious adverse events are recorded throughout follow-up using standardized definitions, grading, attribution (e.g., procedure-related, device-related, medication-related, or unrelated), and reporting pathways consistent with institutional requirements. This safety documentation is observational and does not alter clinical management. Data management, confidentiality, and reproducibility Study data are recorded in a structured database with prespecified variable definitions and audit trails. Imaging and EEG data are stored in de-identified formats with controlled access. Data linkage between clinical records and research identifiers is maintained in a secured location accessible only to authorized personnel. Preprocessing pipelines, parameter settings, and analysis scripts are version-controlled to support reproducibility and transparent reporting. Dissemination Results will be disseminated through peer-reviewed publications and academic presentations. Reporting will follow relevant observational study guidelines and will distinguish prespecified confirmatory analyses from exploratory modeling.