The promise of artificial intelligence and machine learning for migraine treatment outcome prediction: A narrative review

Abstract

Background

Migraine is a highly prevalent neurological disorder with many treatment options, both pharmacological and non-pharmacological. Artificial intelligence (AI) has great potential to optimize treatment selection strategies for individual patients. This review provides an overview of AI models and the techniques used to predict migraine treatment outcomes.

Methods

We conducted a literature search in PubMed and examined studies that reported employing AI models to predict migraine preventive and acute treatment outcomes. We also explored incorporating AI/machine learning to enhance personalized migraine treatment strategies, including forecasting migraine attacks. Additionally, we summarized future research directions, including incorporating multimodality data, using AI frameworks for the discovery of novel treatment targets, and advancing the field with innovative AI techniques such as digital twins, conversational AI and virtual AI agents.

Results

Studies have employed ML and deep learning on a combination of clinical features and imaging data to predict acute or preventive migraine treatment outcomes with reported success. Continued model optimization, validation, and prospective assessment of the clinical utility of deploying ML models in real-world settings are crucial.

Conclusions

While AI has demonstrated success in predicting migraine treatment responses, future research incorporating novel AI techniques and diverse data sources could pave the way to advance personalized migraine treatment.

Graphical abstract

This is a visual representation of the abstract.

Keywords

artificial intelligence machine learning precision migraine treatment response prediction

Introduction

Migraine is a highly prevalent neurological disorder, leading to a significant disease burden on patients, families and healthcare systems (1). This presents both a challenge and a compelling opportunity for researchers and healthcare professionals to advance more effective treatment solutions for patients with migraine.

Over the years, various migraine treatment options have emerged, including non-specific pharmacological treatments, migraine-specific medications, such as triptans and calcitonin gene-related peptide (CGRP) targeting medications and non-pharmacological interventions (2,3). For patients with frequent headaches, a preventive therapy to reduce the frequency and severity of headaches is usually indicated. However, not all patients respond to each treatment, and response rates depend on multiple factors, whether these are genetic, molecular, structural, personal lifestyle related or depending on migraine subtypes (4). The current trial-and-error approach for identifying an effective treatment can be a journey that lasts months, or years, for many patients, during which they continue to experience migraine attacks and, in some cases, adverse treatment reactions (5).

The ability to prescribe an effective, personalized migraine treatment early in the disease course would shorten the trial-and-error process, reduce disease burden and improve patients’ quality of life (6). Several studies have identified various clinical and demographic factors associated with a good response to CGRP monoclonal antibodies (mAbs) (7 –13). However, more studies incorporating advanced techniques are needed to determine the most suitable treatment for an individual patient.

In recent years, artificial intelligence (AI) and machine learning (ML) techniques have been employed widely in medicine, including in the fields of neurology (14,15) and headache (16,17). AI/ML tools can facilitate data extraction from medical records (18), obtain insights from large-scale, real-world datasets (19,20), process diverse information for accurate diagnosis (21,22), identify prognostic factors (23) for disease progression, predict treatment responses in acute and preventive treatment (24,25) and even aid in drug discovery (26). AI tools can also assist in clinical decision-making, prioritizing tasks and triaging patients based on disease severity and urgency (16,27 –30).

This review summarizes current studies employing AI and ML techniques to predict outcomes of preventive and acute migraine treatments. Furthermore, we explore future applications of AI and ML in migraine care, with a focus on personalized treatment. We discuss novel technologies, such as agentic AI systems, generative AI and digital twins, and also include articles from other medical fields to highlight potential applications of emerging AI techniques in headache medicine. Overall, this review provides a comprehensive overview of how current and future AI approaches can be employed to optimize migraine treatment strategies.

Methods

We searched PubMed for keywords including machine learning, artificial intelligence and migraine, headache disorders, treatment response prediction, treatment outcome, forecast attacks, wearables and precision treatment from its inception to 25 March 2025. Original articles published in English that investigated treatment response prediction, migraine attack forecasting and the use of AI/ML in migraine, with a clear description of the research methodology, including the types of AI techniques utilized, and the results, including performance and evaluation metrics of the ML algorithms, were included. Articles not published in English, not focused on migraine or lacking a clear description of the AI methodologies or reporting of model performance were excluded. Additional articles known to the the investigators or identified from the references of the included articles were also added to address the topics of each section. Furthermore, when discussing important future directions for which few original research studies have been published in headache medicine, such as multimodality treatment response prediction, AI for drug discovery and novel AI technologies (including agentic AI, digital twins and conversational diagnostic AI tools), we included examples from other medical fields to illustrate their potential applications in the headache field. Two of the investigators (KP and CC) then carefully reviewed all included articles and summarized the key content in this review.

Results

Published studies using AI/ML for predicting migraine treatment response

Most published studies on ML-based migraine treatment prediction have employed supervised ML methods, including support vector machines (SVM), random forests, gradient boosting and other deep learning neural networks. Many of these studies not only focused on the performance of the model, but also examined the significance of various predictive features.

Chiang et al. (24) utilized headache questionnaires and demographic information collected before starting preventive medications to develop ML models to predict treatment outcomes to seven different types of commonly used migraine preventive medications, including tricyclic antidepressants, beta-blockers, topiramate, gabapentin, verapamil, onabotulinumtoxinA and CGRP mAbs. Incorporating deep learning and supervised ML models and using 146 detailed clinical and headache descriptions from 4260 patients, the models effectively predicted responses at the first follow up visit, particularly for CGRP mAbs with an area under receiver operating curve (AUC) of 0.83 and accuracy of 80% (24). Factors important for predicting treatment outcomes included age, headache frequency, body mass index (BMI), migraine-associated symptoms and migraine attack triggers.

Gonzalez-Martinez et al. (25) applied a random forest model to predict treatment outcomes to CGRP mAbs based on headache and migraine frequency at baseline and three months post-treatment initiation to predict treatment response at six, nine and 12 months. The study included 712 participants, and their model yielded AUC that ranged from 0.87 to 0.98 and an F1 score in the range 0.70–0.97 (25). Romozzi et al. (26) utilized a similar framework and constructed ML models with demographic and early post-treatment data, including monthly headache days (MHD), use of acute medication, and Headache Impact Test-6, collected at one, three and 6 months to predict CGRP mAbs response at subsequent timepoints, up to 12 months post-treatment, with accuracy in the range 0.40–0.73 and AUC in the range 0.56–0.76. The most important feature of the model, using early post-treatment outcomes, was the reduction in MHD from the previous data point (31). However, combining pre-treatment data with post-treatment outcomes could potentially inflate the model performance, and the ability to predict treatment outcomes before starting a preventive treatment might be of more clinical utility. Martinelli et al. (32) investigated outcome prediction to onabotulinumtoxinA treatment using various classifier models based on clinical characteristics. However, most of their models relatively underperformed, with only a sub-analysis of those with high frequency episodic migraine group achieving accuracy of 0.86 based on specific four baseline features. Martinelli et al. (32) concluded that routine clinical features alone were not enough for prediction of treatment response to onabotulinumtoxinA in migraine, and that a multimodal approach should yield better results. Stubberud et al. (33) utilized a gaussian model incorporating clinical factors, such as migraine phenotype and comorbidities, to predict response to common migraine treatments in a cohort of 1446 patients. CGRP-targeting therapies (mAbs or gepants) were not included. Their model demonstrated an increased true response rate and a 35% reduction in time to treatment response for patients (a reduction of more than 3.5 months), when compared with timelines that were based on expert guidelines and recommendations (33). Ferroni et al. (34) developed a ML model to predict medication overuse headache (MOH) in patients with migraine using clinical and demographic data collected from questionnaires and biochemical serum analysis. The SVM-based model predicted probable or definite risk for MOH with an accuracy of 0.87 and AUC of 0.83 (34).

ML techniques have also been applied to imaging modalities for treatment prediction. Tu et al. (35) employed a SVM model on functional magnetic resonance imaging (fMRI data), and identified migraine-related fMRI signal changes that were associated with good response to acupuncture. Fu et al. (36) used fMRI data to characterize migraine-specific imaging features with an accuracy of 0.79 and AUC of 0.83. They further utilized these features to predict the effectiveness of transcutaneous vagus nerve stimulation (tVNS) (36). Similarly, Wei et al. (37) utilized fMRI data to investigate functional network connectivity disruption associated with migraine, and developed a model that predicts response to non-steroidal anti-inflammatory drugs (NSAIDs) with an AUC of 0.93. Yang et al. (38) used a SVM model on structural MRI data, including gray matter volume in the frontal, temporal, parietal, precuneus and cuneus gyri, to predict 50% reduction in monthly migraine days after acupuncture treatment for migraine. The model achieved an accuracy of 0.83 and AUC of 0.79 (38).

Finaly, Chen et al. (39) conducted a meta-analysis incorporating many of the above-mentioned studies and concluded that AI models demonstrated strong predictive performance, with a pooled AUC of 0.86 (95% confidence interval = 0.67–0.95), for migraine treatment response. However, there was significant heterogeneity of the design, variables, modalities and treatment outcomes of the included studies. Furthermore, some studies included pre-treatment data, while some incorporated early post-treatment data in the prediction, which as expected would lead to a higher model performance. We have summarized the published studies, including the input variables, model performance and the treatment outcomes being predicted, in Table 1.

Table 1.

Published studies using machine learning to predict treatment outcomes for migraine.

Article	Input variables	Treatments (medications) outcome predicted	Model performance in the held-out test sets	Model type
Chiang et al. (24)	Detailed migraine description and demographics data collected before starting treatments	CGRP mAbs. OnabotulinumtoxinA, Topiramate, Tricyclic antidepressants, Beta blockers, Gabapentin, Verapamil	Model performance based on medications (30% responder vs non-responder at the first follow-up visit) CGRP mAbs AUC 0.83, OnabutulinumtoxinA AUC 0.58, Topiramate AUC 0.61, Gabapentin AUC 0.63, Verapamil AUC 0.67, Tricyclic Antidepressants AUC 0.61, Beta-blockers AUC 0.66	Deep learning and supervised machine learning, model types include gradient boosting machine, stacked ensemble models
Gonzalez-Martinez et al. (25)	Headache frequency and Headache Impact Test-6 (HIT-6) at baseline and three months post-treatment initiation	CGRP mAbs	AUC was calculated for different response rates (30%, 50%, 75%), at six, nine and 12 months AUC ranged between 0.87 and 0.98, depending on the combination	Supervised machine learning – random forests
Romozzi et al. (31)	Demographic and early post treatment data, including headache days (MHD), use of acute medication and HIT-6	CGRP mAbs	AUC was calculated to predict response at three, six and 12 months AUC ranged between 0.56 and 0.76	Supervised machine learning – random forests
Martinelli et al. (32)	Clinical and anamnestic characteristics	OnabotulinumtoxinA	Model performance based on model type: Random forest AUC 0.17, SMV AUC 0.23, ANFIS AUC 0.2, neural network 0.15, fuzzy clustering AUC 0.87 Random forest for high-frequency episodic migraine subgroup – AUC 0.91	Supervised machine learning, model types include neural network, SMV, random forest, ANFIS, fuzzy clustering
Stubberud et al. (33)	Migraine phenotype and comorbidities	OnabotulinumtoxinA, flunarizine, candesartan, tricyclic antidepressants, valproate, topiramate, serotonin noradrenaline reuptake inhibitors, acupuncture, beta blockers, serotonergic agents	Accuracy was calculated for the prediction of each drug and ranged between 0.53 and 0.87 *AUC of the models was not reported	Gaussian process model
Fu et al. (36)	fMRI data – resting-state BOLD fMRI (GE-EPI)	Transcutaneous vagus nerve stimulation mpa#nbsp;(four weeks of treatment). Response was evaluated by the difference in migraine attack frequency four weeks post treatment compare to baseline.	AUC 0.83	Supervised machine learning – SVM
Wei et al. (37)	fMRI data – Resting-state BOLD fMRI (GE-EPI)	NSAIDs. Response was evaluated by the change in VAS score two hours after treatment.	AUC 0.93	Supervised machine learning – SVM
Yang et al. (38)	Structural MRI data	Acupuncture treatment for four weeks. Responder: at least 50% reduction in monthly migraine days	AUC 0.79	Supervised machine learning – SVM

Abbrevations: ANFIS = adaptive neuro-fuzzy inference system; AUC = area under the curve; BOLD = blood-oxygen-level-dependent; CGRP = calcitonin gene-related peptide; fMRI = functional MRI; GE-EPI = gradient-echo echo-planar imaging; mAbs = monoclonal antibodies; MRI = magnetic resonance imaging; NSAIDs = non-steroidal anti-inflammatory drugs; rs-fMRI = resting-state functional magnetic resonance imaging; SVM = support vector machine; VAS = visual analog scale.

In summary, recent studies have employed ML and deep learning on a combination of clinical data, questionnaires, and imaging (such as MRI and fMRI) variables, to predict acute or preventive migraine treatment responses. While the studies demonstrated promising predictability and feasibility, future directions include optimizing and validating the model performances and using prospective study design to assess clinical utility of these models.

Utilizing AI/ML to facilitate big data research and study real-world treatment outcomes

Natural language processing (NLP) and large language models (LLMs) are advancing rapidly, enabling efficient data extraction from medical records. Chiang et al. (18) demonstrated that a LLM-based NLP model can effectively extract headache frequency data from free-text clinical notes in the electronic health records, with the best performing model reaching an accuracy of 0.92. Guo et al. (19) used a free-text classification NLP model to identify users that self-report their migraine-related experience on social media channels, with promising results of F1 score 0.9 and 0.93 on two different social media platforms, providing with a tool that could identify and capture patient self-reported migraine experiences in real-world settings. Another example is a study that utilized migraine attack data recorded in an e-diary smartphone application from 278,006 users. Using a nested logistic regression model, Chiang et al. (20) conducted a simultaneous comparison of 25 different acute headache medications on patient reported treatment effectiveness, and demonstrated that triptans were reported as the most effective class of acute headache medications. Notably, gepants and ditan were not included in this analysis. These published studies demonstrate the potential of employing AI/ML on large-scale, real-world datasets to gather insights and study outcomes directly reported from patients that could advance the understanding of real-world migraine treatment experiences.

Using AI/ML to forecast migraine attacks and facilitate precision acute treatment

Wearable biosensors, as well as detailed symptoms recorded in headache diaries, have been used for forecasting migraine attacks. Several groups of researchers, including Stubberud et al. (40), Kapustynska et al. (41) and Siirtola et al. (42), have investigated using wearable data including heart rate, peripheral skin temperature, muscle tension, blood volume response, respiratory rate and sleep data to forecast migraine attacks. While demonstrating feasibility and yielding insights into the types of data that might be useful for migraine attack prediction, these models require optimization before they could be implemented in the clinical setting. Going forward, wearable could be used not only in the context of forecasting migraine attacks. but also in assessing treatment outcomes, such as detecting adverse reactions, providing additional tools other than the current self-reporting methods. Additionally, the widespread use of smartphones allows for real-time data collection, that could be incorporated to forecast migraine attack (43,44), improve migraine characterization (45) and monitor treatment response (20,46).

Electroencephalography (EEG) also has the potential to forecast migraine attacks. Although early studies showed conflicting results, several studies have demonstrated EEG changes that occur with different migraine phases. During the interictal phase, there was diffuse slowing of background rhythm, including reduced high frequency bands (alpha and beta) and increased low frequency bands (theta and delta), which normalized closer to the migraine attacks (47). Enhancement in occipital EEG entropy (48) was also observed. Cao et al. (49) employed a SVM classifier model on EEG data and found that EEG complexity in the prefrontal area could potentially be used to forecast migraine attacks, as increased complexity in the premonitory stage and normalized complexity between attacks in the prefrontal cortex has been observed. The model could classify between ictal and interictal phases with an accuracy of 0.76 (49). A study by Martins et al. (50) using a single-channel wearable EEG device placed on the frontal area demonstrated an increase in delta waves and decrease in beta waves in the 24 hours prior to a migraine attack. These studies demonstrate a proof-of-concept for using EEG as a tool to forecast migraine attacks. Currently, there are no published studies using EEG changes to predict treatment response for migraine, but some success has been reported using EEG to predict treatment outcomes for depression (51). The incorporation of EEG in treatment outcome prediction for migraine could be considered.

As patients do not always recognize early signs of an upcoming attack (44,52), real-time prediction of an attack could help with activity planning or early intervention with acute treatments, which are typically more effective when initiated early, with some medications showing efficacy when given during the prodromal phase (53). These predictive tools could also be employed for non-pharmacological treatments, such as neuromodulation devices (54 –57), muscle relaxation (58), biofeedback (59,60) or even emerging novel methods such as virtual reality, which has shown some benefit for migraine treatment (61 –63). These promising technologies enable us to envision a future where patients with migraine receive real-time alerts and personalized acute treatment recommendations based on a variety of physiological and environmental data inputs.

Toward multimodality treatment outcome prediction models

The inclusion of multiple data types in ML models might optimize migraine treatment prediction models. For example, models could include clinical symptoms, laboratory results, imaging findings, genetics and other omics data (64). Some of these data are typically collected during the course of usual clinical practice, whereas others would need to be collected for possible treatment response prediction, such as genetic, proteomic and radiomic data (65,66). Optimally, prediction models would rely on data that are easy and inexpensive to collect.

Prior to integrating a specific type of data in multimodality prediction models, it is important to demonstrate that the data type is useful for measuring or predicting migraine treatment responses. Studies have reported brain imaging features that are associated with or predictive of treatment responses. Wei et al. (67) demonstrated that altered connectivity in specific brain regions (e.g. anterior cingulate cortex, lingual gyrus and occipital cortex), measured by fMRI, could be predictive of treatment response to NSAIDs. Basedau et al. (68) observed reduced hypothalamic activation in fMRI in all 26 patients with migraine who received galcanezumab, with a stronger reduction observed in responder versus non-responders. Furthermore, specific pre-treatment activity of the spinal trigeminal nucleus was associated with response to galcanezumab (68). Lee et al. (69) used changes in structural connectivity in the thalamic nucleus, as measured using diffusion-tensor imaging to both distinguish between migraine patients and healthy controls, and to identify changes between sumatriptan responders versus non-responders. Wu et al. (70) used MRI regional volumetric measurements and identified that the left hippocampal volume to be significantly larger in sumatriptan responders versus non-responders. Furthermore, Ahmed et al. (71) have used white matter hyperintensities (WMH) on MRI to characterize migraine phenotype and demonstrated that a higher WMH burden was associated with not responding to topiramate preventive treatment and ibuprofen acute treatment. Vascular imaging can also be evaluated as predictive factors. Nowaczewska et al. (72,73) utilized transcranial Doppler and demonstrated an association between baseline cerebral blood flow alterations and treatment response to CGRP mAbs. Despite that most of these studies had a relatively small sample size, the results demonstrated the potential of using imaging parameters to predict treatment response for patients with migraine.

In addition to specific imaging features, computer vision techniques, including image classification and feature extraction, are increasingly used in research and clinical settings (74). These methods have been applied in structural MRIs for headache classification and imaging biomarker extraction (75) to detect acute neurological events (76) and classify brain regions of interest in neurodegenerative diseases (77,78). The application of computer vision to characterize patients with migraine may facilitate the integration of imaging modalities into research aimed at advancing precision migraine treatment.

Genetic factors associations with migraine treatment outcomes have also been studied. Cutrer et al. (79) used whole-exome sequencing from 225 migraine patients and identified single nucleotide polymorphisms (SNPs) associated with response to verapamil. Out of the 524 SNPs that were found to be associated with verapamil response, 39 of them were also validated in the confirmatory cohort. Chase et al. (80) demonstrated a positive association between polygenic risk score and treatment response to CGRP mAbs, and that RAMP1 and COMT polymorphisms were associated with non-responsiveness. Similarly, a high genetic risk score was associated with better treatment response to triptans as demonstrated by Terrazzino et al. (81), Cargnin et al. (82) and Kogelman et al. (83, 64), with specific genetic variants pointing towards a poor response. Epigenetics and DNA methylation patterns have also been associated with treatment response. In a study by Mehta et al. (84), specific DNA methylation patterns, including lower pre-treatment methylation in MARK3 and lower methylation in HDAC4 post-treatment, were associated with treatment response to discontinuation from the overused acute medication in patients with chronic migraine with medication overuse. These findings suggest that targeting chromatin remodeling and synaptic plasticity might be a helpful treatment target for patients with medication overuse.

There are examples that combining different data types, such as imaging, genetic, clinical and demographic information, could yield good results in prediction of treatment response. In a study by Tso et al. (85), employing ML on a combination of clinical and imaging data, aiming to predict verapamil response in patients with cluster headache, resulted in a model AUC of 0.69, demonstrating feasibility in the prediction. The study also identified specific brain regions associated with treatment efficacy (85). Additional examples can be found outside the headache field. In a study by Jong et al. (86), a ML model predicting treatment response in patients with epilepsy using a combination of clinical and genetic data achieved an AUC of 0.76. For patients with acute ischemic stroke, Jo et al. (87) developed a ML model that combined clinical and neuroimaging features to predict three-month functional outcomes and achieved an AUC of 0.79. There are also examples outside the field of neurology. In a study conducted by Sammut et al. (88), integrating clinical, digital pathology, genomic, and transcriptomic data in a ML model predicting breast cancer therapy treatment response achieved an AUC of 0.87.

Using AI/ML for drug development and discovery of new treatment target

AI-driven drug discovery could accelerate the identification of new treatment targets, which can be particularly helpful for patients who do not respond to existing therapies (89,90). Successful examples can be found outside of the headache field. Novel DDR1 kinase inhibitors were designed using novel frameworks that combine reinforcement learning with generative models to learn from existing chemical data and create a new molecular structure with the desired properties that could potentially inhibit DDR1 kinase, a critical treatment target in cancer and fibrosis (91). For CDK20 inhibitors, a structure-based drug discovery system was utilized, starting with a deep learning model predicting the structure of CDK20, then several models were employed to identify potential drug targets followed by an AI-powered compound generation model to generate the novel CDK20 inhibitor, which can be used for the treatment of hepatocellular cardinoma (92,93). AI-driven approaches have also been used to facilitate the discovery of new therapeutic indications for existing drugs. For example, baricitinib, an existing drug for rheumatoid arthritis, was discovered as a potential treatment for COVID-19 infection. The researchers used graph representation learning techniques, employing deep learning to a large medical knowledge graph of drugs, diseases, genes, molecules and proteins (94). Through this process, baricitinib was identified due to its dual action in reducing cytokine signaling and inflammation and inhibiting viral entry (89,95). It's efficacy to treat hospitalized patients with moderate to severe COVID-19 was later demonstrated in randomized controlled studies (96). Similar big data and systemic AI-based approaches could be employed in headache medicine.

There are several known potential targets for new migraine treatments, with pituitary adenylate cyclase–activating polypeptide being the most established with promising phase 2 clinical trial results published (97). In addition, there are other candidates of potential interest that play a role in the pathogenesis of migraine attacks and pain signaling. These include substance P, a neuropeptide expressed in trigeminal neurons (98); histamine and glyceryl trinitrate, both implicated in headache induction (99); vasoactive intestinal polypeptide (100) and amylin (101), which have been shown to induce headache in clinical trials; various intracellular targets such as nitric oxide and phosphodiesterase; hormones such as oxytocin and prolactin; and ion channels, including potassium, calcium and transient receptor potential channels (102,103). The use of AI, in combination with the methods described above, could expedite the identification of new therapeutic targets or establish the relevance of these potential targets reported.

Other emerging AI techniques to facilitate personalized migraine treatment

As AI continues to evolve, its role in medicine, including migraine treatment, will expand, bringing us closer to truly personalized care. Digital twins, a virtual replica of an actual patient that updates in real time to simulate and analyze its real-world counterpart (104), are gaining popularity in medicine. Researchers have utilized digital twins to model brain aging on MRI data for patients with multiple sclerosis (105). Furthermore, studies have developed virtual patients, digital twins, to model individualized parameters for personalized fentanyl therapy for cancer patients with chronic pain. Such approaches could optimize appropriate pain relief and drug dosing (106) and could additionally be used in drug target discovery (107,108).

Another AI field that is gaining more attention is agentic AI, which refers to AI systems with the capacity to operate autonomously and take initiative in complex decision and task execution, with minimal human intervention. These AI agents could proactively monitor and extract data, identify issues and propose solutions, and interact with various necessary tools (109). Most AI agents are LLMs or chains of LLMs. For example, a virtual AI agent could be assigned to function as a hospitalist, with the task of orchestrating and facilitating communications among a virtual team of AI agents assigned with different expertise, such as a neurologist, a cardiologist, a pharmacist, a social worker, a billing specialist and an administrator. The human clinicians would then interact with the lead hospitalist AI agent to facilitate treatment care decisions. Careful human supervision and cautious monitoring of performance, especially confabulation or hallucination from AI agents would be required. This new and promising tool could facilitate scientific discovery and improve clinical efficiency.

Finally, conversational diagnostic AI, a tool based on LLM that aims to replicate the diagnostic and empathetic capabilities of physicians, is recently on the rise. These AI conversational agents are generally achieving good results and high user satisfaction (110). A recent study by Zhang et al. (111) demonstrated that a dialog based chatbot using LLM for diagnosis of chronic diseases achieved high accuracy in diagnosing 24 common chronic diseases while also demonstrating high user satisfaction. In a study by Tu et al. (112), an LLM based AI system, developed through multiple agentic AI framework named AMIE was trained using simulated dialogues created using online datasets. It was trained to obtain relevant history and generate differential diagnosis. After its development, a randomized, double-blind crossover study comparing the performance of AMIE to primary care physicians was conducted and showed that AMIE achieved better diagnostic accuracy and quality of care (112). While several limitations of the study exist, these studies demonstrate the potential to employ carefully designed LLM systems as a solution to expand medical expertise to medically underserved regions. Similar systems could be used for early diagnosis and treatment of headache disorders, as well as for treatment response monitoring, facilitating high quality migraine treatment. A glossary of AI related terms is presented in Table 2.

Table 2.

Glossary of AI terms included in this review.

Term (30,113,114)	Description
Artificial intelligence (AI)	An area of computer science focused on building systems that can perform tasks requiring human intelligence
Machine learning (ML)	A subset of AI that uses statistical techniques to enable computers to learn patterns from data and make predictions or identify subgroups
Supervised learning	A type of machine learning where the model is trained on a labelled dataset Example algorithms: - Decision tree: splits data into subsets based on feature values, forming a tree-like structure, each branch represents a decision and each leaf an outcome - Random forest: Constructs multiple decision trees and aggregates their predictions - Gradient boosting: builds decision trees sequentially, with each new tree trained to correct the errors of the previous ones - Support vector machine (SVM): identifies the optimal hyperplane to separate data into classes
Unsupervised learning	A type of machine learning where the algorithm is given data without explicit labelling. This is often used to detect patterns or similar characteristics of groups within the data Example algorithms: - Clustering: a type of unsupervised learning that groups data points into clusters based on similarity, without the use of predefined labels Examples: Gaussian mixture model, K-means, dimensionality reduction
Deep learning	A subfield of machine learning involving neural networks with multiple layers, allowing them to learn complex patterns and representations from large datasets
Neural networks	Computational models inspired by the structure and function of the human brain, designed to learn from data and perform tasks
Natural language processing (NLP)	A branch of AI that enables machines to understand, interpret and generate human language
Large language models (LLMs)	Advanced NLP systems built on transformer architectures, trained on massive text datasets. They generate human-like text by predicting the most probable next word or sequence of words, enabling tasks such as conversation, summarization and translation
Agentic AI	AI systems designed to operate as autonomous agents that can make decisions, take actions and pursue goals, rather than only following direct human instructions. Agentic AI systems often interact with tools, environments or other agents
Digital twins	Virtual models of real-world objects that mirror their state and behaviour, enabling simulation, monitoring and analysis in real time

Discussion

Limitations and other considerations

While the use of AI- and ML-based models for migraine treatment outcome prediction is promising, several limitations exist. Studies using AI and ML techniques to predict treatment outcomes varied in the features included and the outcomes measured, making it challenging to compare results across studies. It should also be noted that using early post-treatment outcomes as input features may influence results and potentially inflate performance metrics due to “data leakage”, allowing the models to learn from early post-treatment data when making predictions. Another limitation of the current literature is that only a minority of studies have included external validation cohorts, raising questions about the generalizability of their findings. Validation of research results, whether through prospective or external validation, is an important next step. Furthermore, most studies discussed in this review represent preliminary research findings. Prospective, randomized controlled studies of real-world implementations are needed to evaluate the clinical utility of AI/ML models.

Beyond the limitations of published studies, there are also regulatory, ethical and data harmonization challenges inherent in the use of AI and ML in research and clinical practice. AI/ML algorithms rely heavily on the quality and diversity of training data. When the training data lack diversity, the models trained could be biased with limited generalizability in under-represented populations. ML models are also prone to overfitting, especially when trained on small or highly imbalanced datasets. Overfitting occurs when a model captures not only the underlying patterns in the data, but also noise and random fluctuations, resulting in excellent performance on the training dataset but poor generalizability to unseen cases. Training models on more diverse datasets, utilizing cross-validation and applying careful feature selection could help mitigate this challenge. Some AI and ML models, especially deep learning models, function as “black boxes”, making their decision-making processes difficult to interpret. This lack of transparency can undermine trust and accountability. Given the rapidly evolving nature of AI techniques, regulatory and reimbursement guidance for their use in medical practice also face challenges because current frameworks are mostly designed for static algorithms. This may hinder the development and implementation of adaptive AI models, meaning those that continue to learn from new data. Furthermore, the liability for medical malpractice related to the use of AI algorithms remains unclear, which may reduce physicians’ willingness to incorporate these tools into clinical practice.

Data harmonization is also an important factor to consider. As the diagnosis and practice of headache medicine rely mostly on symptom descriptions and patient-reported outcomes, the datasets and variables used are highly heterogeneous, increasing the challenges of cross-institutional model validation. Ensuring interoperability through standardized data formats and leveraging federated learning approaches may help mitigate some of these challenges by optimizing the generalizability and clinical utility of AI models. In addition to optimizing AI/ML performance, clinicians and researchers must also be educated on the structure, capabilities and limitations of the models to ensure their effective and responsible use in clinical and research settings.

By addressing current limitations and carefully developing and validating novel AI techniques for the use cases discussed in this review, AI tools hold the potential to expand access to headache specialty care, particularly when implemented by non-specialists such as primary care providers. Such tools could facilitate earlier diagnosis and more timely treatment recommendations. In addition, AI applications may enhance what headache specialists can offer by enabling personalized treatment plans that reduce the trial-and-error process of medication selection, forecasting migraine attacks to help patients better plan their daily activities, and streamlining processes across research, clinical care, education and administration.

Conclusions

The effectiveness of migraine treatment varies among patients. With the wide range of pharmacological and non-pharmacological treatments available today, it is crucial to identify the most suitable option for each individual. AI and ML hold great promise for transforming the current approach to selecting acute and preventive migraine treatments. Research studies have reported success and feasibility using ML to predict response to preventive and acute migraine treatment and gather real-world patient experiences on treatment outcomes. Going forward, incorporating multimodality dataset could optimize the predictability of the ML models. While this review predominantly focus on migraine, similar concepts and techniques could be applied to other primary and secondary disorders. Furthermore, novel generative AI systems could be employed to facilitate drug discovery and identify new therapeutic targets for migraine. Innovative techniques, such as digital twins and conversational and agentic AI systems also hold great promise to facilitate the early diagnosis and treatment of headache disorders and advance the headache field toward precision treatment. As AI tools holds great promise and become more integrated into healthcare, it is essential that clinicians, researchers and patients develop a clear understanding of their capabilities and limitations.

Public health relevance

Studies have employed ML and deep learning on a combination of clinical features and imaging data to predict acute or preventive migraine treatment outcome with reported success.

Incorporating multimodality data sources, optimizing ML models that can forecast migraine attacks, and employing AI to discover new treatment targets are future directions to advance toward personalized migraine treatment.

Novel AI techniques, such as digital twins, agentic AI, foundation models and conversational AI tools, could be incorporated to optimize the diagnosis and treatment of headache disorders.

Footnotes

Author contributions

Conceptualization and design: CC, KP. Data Acquisition, literature review and analysis: CC, KP. Manuscript drafting: CC, KP. Manuscript revision and commentary: CC, KP, FMC, TJS. Final approval: CC, KP, FMC, TJS.

Data availability statement

All information and data analyzed in this narrative review are fully referenced in the article and are publicly available through their respective journals.

Declaration of conflicting interests

KP and FMC reports no financial disclosures.

TJS reports, over the last 36 months, personal fees for consulting from: AbbVie, Allergan, Amgen, Axsome, Collegium, Eli Lilly, Linpharma, Lundbeck, Salvia, Satsuma, Scilex and Theranica. He has received royalties from UpToDate and holds/held stock options in Allevalux, Aural Analytics and Nociria. His employer has received research grants on his behalf from American Heart Association, Flinn Foundation, Henry Jackson Foundation, National Headache Foundation, National Institutes of Health, Patient Centered Outcomes Research Institute, Pfizer, Spark Neuro and United States Department of Defense.

CC has received personal fee for consulting from: Satsuma, eNeura, Pfizer, Amneal and AbbVie. She has received research grants from the American Headache Society, Pfizer and Lundbeck, with funds paid to her institution.

Funding

The authors received no financial support for the research, authorship and/or publication of this article.

ORCID iDs

Keshet Pardo

Todd J. Schwedt

Chia-Chun Chiang

References

Feigin

Nichols

Alam

, et al. Global, regional, and national burden of neurological disorders, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet Neurol 2019; 18: 459–480.

Puledda

Sacco

Diener

, et al. International Headache Society global practice recommendations for the acute pharmacological treatment of migraine. Cephalalgia 2024; 44: 3331024241252666.

Puledda

Sacco

Diener

H-C

, et al. International headache society global practice recommendations for preventive pharmacological treatment of migraine. Cephalalgia 2024; 44: 3331024241269735.

Raggi

Leonardi

Arruda

, et al. Hallmarks of primary headache: part 1 – migraine. J Headache Pain 2024; 25: 189.

Kawata

Shah

Poon

J-L

, et al. Understanding the migraine treatment landscape prior to the introduction of calcitonin gene-related peptide inhibitors: results from the assessment of TolerabiliTy and effectiveness in MigrAINe patients using preventive treatment (ATTAIN) study. Headache 2021; 61: 438–454.

Gallagher

Kunkel

. Migraine medication attributes important for patient compliance: concerns about Side effects may delay treatment. Headache 2003; 43: 36–43.

Baraldi

Castro

Cainazzo

, et al. Predictors of response to erenumab after 12 months of treatment. Brain Behav 2021; 11: e2260.

Silvestro

Tessitore

Scotto di Clemente

, et al. Refractory migraine profile in CGRP-monoclonal antibodies scenario. Acta Neurol Scand 2021; 144: 325–333.

Barbanti

Egeo

Aurilia

, et al. Fremanezumab in the prevention of high-frequency episodic and chronic migraine: a 12-week, multicenter, real-life, cohort study (the FRIEND study). J Headache Pain 2022; 23: 1–11.

10.

Bottiroli

De Icco

Vaghi

, et al. Psychological predictors of negative treatment outcome with Erenumab in chronic migraine: data from an open label long-term prospective study. J Headache Pain 2021; 22: 114.

11.

Ihara

Ohtani

Watanabe

, et al. Predicting response to CGRP-monoclonal antibodies in patients with migraine in Japan: a single-centre retrospective observational study. J Headache Pain 2023; 24: 1–10.

12.

Raffaelli

Fitzek

Overeem

, et al. Clinical evaluation of super-responders vs. non-responders to CGRP(-receptor) monoclonal antibodies: a real-world experience. J Headache Pain 2023; 24: 16.

13.

Caronna

Gallardo

Egeo

, et al. Redefining migraine prevention: early treatment with anti-CGRP monoclonal antibodies enhances response in the real world. J Neurol Neurosurg Psychiatry 2024; 95: 927–937.

14.

Voigtlaender

Pawelczyk

Geiger

, et al. Artificial intelligence in neurology: opportunities, challenges, and policy implications. J Neurol 2024; 271: 2258–2273.

15.

Beam

Drazen

Kohane

, et al. Artificial intelligence in medicine. N Engl J Med 2023; 388: 1220–1221.

16.

Ihara

Dumkrieger

Zhang

, et al. Application of artificial intelligence in the headache field. Curr Pain Headache Rep 2024; 28: 1049–1057.

17.

Petrušić

Chiang

C-C

Garcia-Azorin

, et al. Influence of next-generation artificial intelligence on headache research, diagnosis and treatment: the junior editorial board members’ vision – part 2. J Headache Pain 2025; 26: 2 .

18.

Chiang

C-C

Luo

Dumkrieger

, et al. A large language model–based generative natural language processing framework fine-tuned on clinical notes accurately extracts headache frequency from electronic health records. Headache J Head Face Pain 2024; 64: 400–409.

19.

Guo

Rajwal

Lakamana

, et al. Generalizable natural language processing framework for migraine reporting from social Media. AMIA Jt Summits Transl Sci Proceedings AMIA Jt Summits Transl Sci 2023; 2023: 261–270.

20.

Chiang

C-C

Fang

Horvath

, et al. Simultaneous comparisons of 25 acute migraine medications based on 10 million users’ self-reported records from a smartphone application. Neurology 2023; 101: e2560–e2570.

21.

Katsuki

Shimazu

Kikui

, et al. Developing an artificial intelligence-based headache diagnostic model and its utility for non-specialists’ diagnostic accuracy. Cephalalgia 2023; 43: 3331024231156925.

22.

Cowan

Rapoport

Blythe

, et al. Diagnostic accuracy of an artificial intelligence online engine in migraine: a multi-center study. Headache 2022; 62: 870–882.

23.

Hindiyeh

Riskin

Alexander

, et al. Development and validation of a novel model for characterizing migraine outcomes within real-world data. J Headache Pain 2022; 23: 124.

24.

Chiang

C-C

Schwedt

Dumkrieger

, et al. Advancing toward precision migraine treatment: predicting responses to preventive medications with machine learning models based on patient and migraine features. Headache 2024; 64: 1094–1108.

25.

Gonzalez-Martinez

Pagán

Sanz-García

, et al. Machine-learning-based approach for predicting response to anti-calcitonin gene-related peptide (CGRP) receptor or ligand antibody treatment in patients with migraine: a multicenter Spanish study. Eur J Neurol 2022; 29: 3102–3111.

26.

Hasselgren

Oprea

. Artificial intelligence for drug discovery: are we there yet? Annu Rev Pharmacol Toxicol 2024; 64: 527–550.

27.

Yang

Meng

Torben-Nielsen

, et al. A machine learning approach to support triaging of primary versus secondary headache patients using complete blood count. PLoS One 2023: 6;18(3):e0282237

28.

Dumkrieger

Chiang

C-C

Zhang

, et al. Artificial intelligence terminology, methodology, and critical appraisal: a primer for headache clinicians and researchers. Headache 2025; 65: 180–190.

29.

Petrušić

W-S

Labastida-Ramirez

, et al. Influence of next-generation artificial intelligence on headache research, diagnosis and treatment: the junior editorial board members’ vision – part 1. J Headache Pain 2024; 25: 151.

30.

Stubberud

Langseth

Nachev

, et al. Artificial intelligence and headache. Cephalalgia 2024; 44: 3331024241268290.

31.

Romozzi

Lokhandwala

Vollono

, et al. An evolving machine-learning-based algorithm to early predict response to anti-CGRP monoclonal antibodies in patients with migraine. Cephalalgia 2024; 44: 3331024241262751.

32.

Martinelli

Pocora

De Icco

, et al. Searching for the predictors of response to BoNT-A in migraine using machine learning approaches. Toxins (Basel) 2023; 15: 364.

33.

Stubberud

Gray

Tronvik

, et al. Machine prescription for chronic migraine. Brain Commun 2022; 4: fcac059.

34.

Ferroni

Zanzotto

Scarpato

, et al. Machine learning approach to predict medication overuse in migraine patients. Comput Struct Biotechnol J 2020; 18: 1487–1496.

35.

Zeng

Lan

, et al. An fMRI-based neural marker for migraine without aura. Neurology 2020; 94: e741–e751.

36.

Zhang

, et al. Predicting response to tVNS in patients with migraine using functional MRI: A voxels-based machine learning analysis. Front Neurosci 2022; 16: 937453.

37.

Wei

H-L

Yang

W-J

Zhou

G-P

, et al. Altered static functional network connectivity predicts the efficacy of non-steroidal anti-inflammatory drugs in migraineurs without aura. Front Mol Neurosci 2022; 15: 956797.

38.

Yang

X-J

Liu

Z-L

, et al. Baseline brain gray matter volume as a predictor of acupuncture outcome in treating migraine. Front Neurol 2020; 11: 111.

39.

Chen

Zhang

Cao

, et al. Prediction models for treatment response in migraine: a systematic review and meta-analysis. J Headache Pain 2025; 26: 32.

40.

Anker

Hegna

Eiliv

, et al. Forecasting migraine with machine learning based on mobile phone diary and wearable data. Cephalalgia 2023; 43: 3331024231169244.

41.

Kapustynska

Abromavičius

Serackis

, et al. Machine learning and wearable technology: monitoring changes in biomedical signal patterns during Pre-migraine nights. Healthcare 2024; 12: 1701.

42.

Siirtola

Koskimäki

Mönttinen

, et al. Using sleep time data from wearable sensors for early detection of migraine attacks. Sensors 2018; 18: 1374.

43.

Katsuki

Tatsumoto

Kimoto

, et al. Investigating the effects of weather on headache occurrence using a smartphone application and artificial intelligence: a retrospective observational cross-sectional study. Headache 2023; 63: 585–600.

44.

Houle

Turner

Golding

, et al. Forecasting individual headache attacks using perceived stress: development of a multivariable prediction model for persons with episodic migraine. Headache 2017; 57: 1041–1050.

45.

Vives-Mestres

Casanova

Buse

, et al. Patterns of perceived stress throughout the migraine cycle: a longitudinal cohort study using daily prospective diary data. Headache 2021; 61: 90–102.

46.

Minen

Jaran

Boyers

, et al. Understanding what people with migraine consider to be important features of migraine tracking: an analysis of the utilization of smartphone-based migraine tracking with a free-text feature. Headache 2020; 60: 1402–1414.

47.

Puledda

Viganò

Sebastianelli

, et al. Electrophysiological findings in migraine may reflect abnormal synaptic plasticity mechanisms: a narrative review. Cephalalgia 2023; 43: 3331024231195780.

48.

Cao

Lin

C-T

Lai

K-L

, et al. Extraction of SSVEPs-based inherent fuzzy entropy using a wearable headband EEG in migraine patients. IEEE Trans Fuzzy Syst 2020; 28: 14–27.

49.

Zehong

Kuan-Lin

Chin-Teng

, et al. Exploring resting-state EEG complexity before migraine attacks. Cephalalgia 2017; 38: 1296–1306.

50.

Martins Isabel

Marissa

Marco

, et al. Brain state monitoring for the future prediction of migraine attacks. Cephalalgia 2019; 40: 255–265.

51.

Cohen

Zantvoord

Wezenberg

, et al. Electroencephalography for predicting antidepressant treatment success: a systematic review and meta-analysis. J Affect Disord 2023; 321: 201–207.

52.

Pellegrino

ABW

Davis-Martin Rachel

Houle Timothy

, et al. Perceived triggers of primary headache disorders: a meta-analysis. Cephalalgia 2017; 38: 1188–1198.

53.

Dodick

Goadsby

Schwedt

, et al. Ubrogepant for the treatment of migraine attacks during the prodrome: a phase 3, multicentre, randomised, double-blind, placebo-controlled, crossover trial in the USA. Lancet 2023; 402: 2307–2316.

54.

Diener

H-C

Goadsby

Ashina

, et al. Non-invasive vagus nerve stimulation (nVNS) for the preventive treatment of episodic migraine: the multicentre, double-blind, randomised, sham-controlled PREMIUM trial. Cephalalgia 2019; 39: 1475–1487.

55.

Yarnitsky

Dodick

Grosberg

, et al. Remote electrical neuromodulation (REN) relieves acute migraine: a randomized, double-blind, placebo-controlled, multicenter trial. Headache 2019; 59: 1240–1252.

56.

Starling

Tepper

Marmura

, et al. A multicenter, prospective, single arm, open label, post-market, observational study to evaluate the use of sTMS in reduction of migraine headache (ESPOUSE study)(S52. 004). Neurology 2017; 88: S52–004.

57.

Riederer

Penning

Schoenen

. Transcutaneous supraorbital nerve stimulation (t-SNS) with the cefaly® device for migraine prevention: a review of the available data. Pain Ther 2015; 4: 135–147.

58.

Minen

Adhikari

Padikkala

, et al. Smartphone-Delivered progressive muscle relaxation for the treatment of migraine in primary care: a randomized controlled trial. Headache 2020; 60: 2232–2246.

59.

Nestoriuc

Martin

. Efficacy of biofeedback for migraine: a meta-analysis. Pain 2007; 128: 111–127.

60.

Treadwell

Tsou

Rouse

, et al. Behavioral interventions for migraine prevention: a systematic review and meta-analysis. Headache 2025; 65: 668–694.

61.

Cuneo

Yang

Zhou

, et al. The utility of a novel, combined biofeedback-virtual reality device as add-on treatment for chronic migraine: a randomized pilot study. Clin J Pain 2023; 39: 286–296.

62.

Connelly

Boorigie

McCabe

. Acceptability and tolerability of extended reality relaxation training with and without wearable neurofeedback in pediatric migraine. Children 2023; 10: 329.

63.

Bottiroli

Matamala-Gomez

Allena

, et al. The virtual “enfacement illusion” on pain perception in patients suffering from chronic migraine: a study protocol for a randomized controlled trial. J Clin Med 2022; 11: 6876.

64.

Kogelman

LJA

Esserlind

A-L

Francke Christensen

, et al. Migraine polygenic risk score associates with efficacy of migraine-specific drugs. Neurol Genet 2025; 5: e364.

65.

Ashina

Terwindt

Al-Karagholi

MA-M

, et al. Migraine: disease characterisation, biomarkers, and precision medicine. Lancet 2021; 397: 1496–1504.

66.

Keiko

Francesco

Ahmed

, et al. Are we closer to achieving precision medicine for migraine treatment? A narrative review. Cephalalgia 2024; 44: 3331024241281518.

67.

Wei

H-L

Yang

Zhou

G-P

, et al. Abnormal causal connectivity of anterior cingulate cortex-visual cortex circuit related to nonsteroidal anti-inflammatory drug efficacy in migraine. Eur J Neurosci 2024; 59: 446–456.

68.

Basedau

Sturm

L-M

Mehnert

, et al. Migraine monoclonal antibodies against CGRP change brain activity depending on ligand or receptor target – an fMRI study. Elife 2022; 11: e77146.

69.

Lee

Kim

Lee

H-J

, et al. Predicting sumatriptan responsiveness based on structural connectivity in patients newly diagnosed with migraine. J Clin Neurol 2023; 19: 573–580.

70.

J-W

Lai

P-Y

Chen

Y-L

, et al. The use of neuroimaging for predicting sumatriptan treatment response in patients with migraine. Front Neurol 2022; 13: 798695.

71.

Ahmed

Mohamed

AAM

Salem

, et al. Association of white matter hyperintensities with migraine phenotypes and response to treatment. Acta Neurol Belg 2023; 123: 1725–1733.

72.

Nowaczewska

Straburzyński

Waliszewska-Prosół

, et al. Cerebral blood flow and other predictors of responsiveness to erenumab and fremanezumab in migraine-A real-life study. Front Neurol 2022; 13: 895476.

73.

Nowaczewska

Straburzyński

Meder

, et al. Changes in cerebral blood flow after erenumab treatment in good and non-responders—a pilot study of migraine patients. J Clin Med 2021; 10: 2523.

74.

Friedrich

Relton

Wong

, et al. Computer vision in clinical neurology: a review. JAMA Neurol 2025; 82: 407–415.

75.

Rahman Siddiquee

Shah

Chong

, et al. Headache classification and automatic biomarker extraction from structural MRIs using deep learning. Brain Commun 2023; 5: fcac311.

76.

Titano

Badgeley

Schefflein

, et al. Automated deep-neural-network surveillance of cranial images for acute neurologic events. Nat Med 2018; 24: 1337–1341.

77.

Chung

Y-C

Kim

, et al. Classification and visualization of Alzheimer’s disease using volumetric convolutional neural network and transfer learning. Sci Rep 2019; 9: 18150.

78.

Ghahremani

Boorboor

Mirhosseini

, et al. Neuroconstruct: 3D reconstruction and visualization of neurites in optical microscopy brain images. IEEE Trans Vis Comput Graph 2022; 28: 4951–4965.

79.

Cutrer

Moyer

Atkinson

, et al. Genetic variants related to successful migraine prophylaxis with verapamil. Mol Genet Genomic Med 2021; 9: e1680.

80.

Chase

Semenov

Rubin

, et al. Characteristics associated with response to subcutaneously administered anti-CGRP monoclonal antibody medications in a real-world community cohort of persons living with migraine: a retrospective clinical and genetic study. Headache 2024; 64: 68–92.

81.

Terrazzino

Viana

Floriddia

, et al. The serotonin transporter gene polymorphism STin2 VNTR confers an increased risk of inconsistent response to triptans in migraine patients. Eur J Pharmacol 2010; 641: 82–87.

82.

Cargnin

Viana

Sances

, et al. Using a genetic risk score approach to predict headache response to triptans in migraine without aura. J Clin Pharmacol 2019; 59: 288–294.

83.

Kogelman

LJA

Falkenberg

Buil

, et al. Changes in the gene expression profile during spontaneous migraine attacks. Sci Rep 2021; 11: 8294.

84.

Mehta

de Boer

Sutherland

, et al. Alterations in DNA methylation associate with reduced migraine and headache days after medication withdrawal treatment in chronic migraine patients: a longitudinal study. Clin Epigenetics 2023; 15: 190.

85.

Tso

Brudfors

Danno

, et al. Machine phenotyping of cluster headache and its response to verapamil. Brain 2021; 144: 655–664.

86.

de Jong

Cutcutache

Page

, et al. Towards realizing the vision of precision medicine: aI based prediction of clinical drug response. Brain 2021; 144: 1738–1750.

87.

Kim

Gwon

, et al. Combining clinical and imaging data for predicting functional outcomes after acute ischemic stroke: an automated machine learning approach. Sci Rep 2023; 13: 16926.

88.

Sammut

S-J

Crispin-Ortuzar

Chin

S-F

, et al. Multi-omic machine learning predictor of breast cancer therapy response. Nature 2022; 601: 623–629.

89.

Patne

Dhulipala

Lawless

, et al. Drug discovery in the age of artificial intelligence: transformative target-based approaches. Int J Mol Sci 2024; 25: 12233.

90.

Pun

Ozerov

Zhavoronkov

. AI-powered therapeutic target discovery. Trends Pharmacol Sci 2023; 44: 561–572.

91.

Zhavoronkov

Ivanenkov

Aliper

, et al. Deep learning enables rapid identification of potent DDR1 kinase inhibitors. Nat Biotechnol 2019; 37: 1038–1040.

92.

Ambreen

Umar

Noor

, et al. Advanced AI and ML frameworks for transforming drug discovery and optimization: with innovative insights in polypharmacology, drug repurposing, combination therapy and nanomedicine. Eur J Med Chem 2025; 284: 117164.

93.

Ren

Ding

Zheng

, et al. Alphafold accelerates artificial intelligence powered drug discovery: efficient discovery of a novel CDK20 small molecule inhibitor. Chem Sci 2023; 14: 1443–1452.

94.

Zhou

Wang

Tang

, et al. Artificial intelligence in COVID-19 drug repurposing. Lancet Digit Heal 2020; 2: e667–e676.

95.

Richardson

Robinson

BWS

Smith

, et al. The AI-assisted identification and clinical efficacy of baricitinib in the treatment of COVID-19. Vaccines (Basel) 2022; 10: 951.

96.

Kalil

Patterson

Mehta

, et al. Baricitinib plus remdesivir for hospitalized adults with COVID-19. N Engl J Med 2021; 384: 795–807.

97.

Messoud

Ravinder

Melanie

, et al. A monoclonal antibody to PACAP for migraine prevention. N Engl J Med 2024; 391: 800–809.

98.

Pellesi

Edvinsson

. Revisiting substance P in migraine: a methodological approach inspired by anti-CGRP and anti-PACAP success. J Headache Pain 2025; 26: 22.

99.

Olesen

. Provocation of attacks to discover migraine signaling mechanisms and new drug targets: early history and future perspectives – a narrative review. J Headache Pain 2024; 25: 105.

100.

Pellesi

Al-Karagholi

MA-M

Chaudhry

, et al. Two-hour infusion of vasoactive intestinal polypeptide induces delayed headache and extracranial vasodilation in healthy volunteers. Cephalalgia 2020; 40: 1212–1223.

101.

Ghanizada

Al-Karagholi

Walker

, et al. Amylin analog pramlintide induces migraine-like attacks in patients. Ann Neurol 2021; 89: 1157–1171.

102.

Al-Hassany

Boucherie

Creeney

, et al. Future targets for migraine treatment beyond CGRP. J Headache Pain 2023; 24: 76.

103.

Chiang

C-C

Porreca

Robertson

, et al. Potential treatment targets for migraine: emerging options and future prospects. Lancet Neurol 2024; 23: 313–324.

104.

Fekonja

Schenk

Schröder

, et al. The digital twin in neuroscience: from theory to tailored therapy. Front Neurosci 2024; 18: 1454856.

105.

Cen

Gebregziabher

Moazami

, et al. Toward precision medicine using a “digital twin” approach: modeling the onset of disease-specific brain atrophy in individuals with multiple sclerosis. Sci Rep 2023; 13: 16279.

106.

Bahrami

Rossi

De Nys

, et al. An individualized digital twin of a patient for transdermal fentanyl therapy for chronic pain management. Drug Deliv Transl Res 2023; 13: 2272–2285.

107.

Ren

Pieper

Cheng

. Utilization of precision medicine digital twins for drug discovery in Alzheimer’s disease. Neurotherapeutics 2025; 22: e00553.

108.

Lee

Lilja

, et al. A dynamic single cell-based framework for digital twins to prioritize disease genes and drug targets. Genome Med 2022; 14: 48.

109.

Zou

Topol

. The rise of agentic AI teammates in medicine. Lancet 2025; 405: 457.

110.

Milne-Ives

de Cock

Lim

, et al. The effectiveness of artificial intelligence conversational agents in health care. Systematic review. J Med Internet Res 2020; 22: e20346.

111.

Zhang

Song

. A chatbot based question and answer system for the auxiliary diagnosis of chronic diseases based on large language model. Sci Rep 2024; 14: 17118.

112.

Schaekermann

Palepu

, et al. Towards conversational diagnostic artificial intelligence. Nature 2025; 642: 442–450.

113.

Haug

Drazen

. Artificial intelligence and machine learning in clinical medicine, 2023. N Engl J Med 2023; 388: 1201–1208.

114.

Liu

Chen

P-HC

Krause

, et al. How to read articles that use machine learning: users’ guides to the medical literature. JAMA 2019; 322: 1806–1816.