Disrupted Functional Connectivity of the Amygdala Predicts the Efficacy of Non-steroidal Anti-inflammatory Drugs in Migraineurs Without Aura

Abstract

Machine learning (ML) has been largely applied for predicting migraine classification. However, the prediction of efficacy of non-steroidal anti-inflammatory drugs (NSAIDs) in migraine is still in the early stages. This study aims to evaluate whether the combination of machine learning and amygdala-related functional features could help predict the efficacy of NSAIDs in patients with migraine without aura (MwoA). A total of 70 MwoA patients were enrolled for the study, including patients with an effective response to NSAIDs (M-eNSAIDs, n = 35) and MwoA patients with ineffective response to NSAIDs (M-ieNSAIDs, n = 35). Furthermore, 33 healthy controls (HCs) were matched for age, sex, and education level. The study participants were subjected to resting-state functional magnetic resonance imaging (fMRI) scanning. Disrupted functional connectivity (FC) patterns from amygdala-based FC analysis and clinical characteristics were considered features that could promote classification through multivariable logistic regression (MLR) and support vector machine (SVM) for predicting the efficacy of NSAIDs. Further, receiver operating characteristic (ROC) curves were drawn to evaluate the predictive ability of the models. The M-eNSAIDs group exhibited enhanced FC with ipsilateral calcarine sulcus (CAL), superior parietal gyrus (SPG), paracentral lobule (PCL), and contralateral superior frontal gyrus (SFG) in the left amygdala. However, the M-eNSAIDs group showed decreased FC with ipsilateral caudate nucleus (CAU), compared to the M-ieNSAIDs group. Moreover, the M-eNSAIDs group showed higher FC with left pre-central gyrus (PreCG) and post-central gyrus (PoCG) compared to HCs. In contrast, the M-ieNSAIDs group showed lower FC with the left anterior cingulate cortex (ACC) and right SFG. Furthermore, the MwoA patients showed increased FC with the left middle frontal gyrus (MFG) in the right amygdala compared to HCs. The disrupted left amygdala-related FC patterns exhibited significant correlations with migraine characteristics in the M-ieNSAIDs group. The MLR and SVM models discriminated clinical efficacy of NSAIDs with an area under the curve (AUC) of 0.891 and 0.896, sensitivity of 0.971 and 0.833, and specificity of 0.629 and 0.875, respectively. These findings suggest that the efficacy of NSAIDs in migraine could be predicted using ML algorithm. Furthermore, this study highlights the role of amygdala-related neural function in revealing underlying migraine-related neuroimaging mechanisms.

Imaging the brain and vascular reactions to headache treatments: a systematic review

“…, 2022 [ 16 ] 70 MO 33 HC Type: NSAID Duration: 2 attacks treated within 3 months of scan Response definition: 50% or greater reduction in pain intensity from pre-treatment Modality: RS fMRI Time-point: Pretreatment Baseline: Responders vs non-responders: ↑ FC between visual and auditory network in responders ↓ FC between sensorimotor and visual network in responders Type of NSAID not specified FC compared at p < 0.05, uncorrected for multiple comparison Overlap of regions belonging to distinct networks Wei et al . , 2022 [ 17 ] 73 MO 33 HC Type: NSAID Duration: 2 attacks treated within 3 months of scan Response definition: 50% or greater reduction in pain intensity between the pre-treatment level and 2 h after taking medication Modality: RS fMRI and T1w MRI Time-point: Pretreatment Baseline: Responders vs non-responders: ↑ FC between left amygdala and right superior frontal gyrus, left calcarine sulcus, left superior parietal gyrus and paracentral lobule in responders ↓ FC between left amygdala and ipsilateral caudate nucleus in responders No whole-brain GM volume differences Type of NSAID not specified Ahmed et al . , 2022 [ 18 ] 500 Mx (150 MA, 235 CM) Type: Ibuprofen 200–400 mg Duration: At least 2 months Response definition: Pain freedom within 2 h in ≥ 4 of 5 attacks while insufficient responders achieved pain freedom in ≤ 3 of 5 attacks Modality: T2w MRI Time-point: Pretreatment Baseline: Responders vs non-responders: Acute treatment responders less frequently had WMHs, had fewer WMHs, and smaller WMHs Only T2-w images were used to assess WMHs.…”

“…In addition, a support vector machine model based on pre-treatment FC reported a 93% accuracy in predicting responders. In another study [ 17 ] based on the same cohort and focused on the FC of the left amygdala, responders had increased FC with the left calcarine, superior frontal, and parietal areas, as well as a decreased FC with the ipsilateral caudate nucleus, compared to non-responders. Pre-treatment RS FC of the amygdala with the caudate, visual and frontoparietal areas predicted patients’ response with an accuracy of 89%.…”

Potential predictors for the efficacy of non-steroidal anti-inflammatory drugs in patients with migraine

“…It is shown that cortico-limbic interactions mediate maladaptive responses relevant to psychopathology ( Kovner et al, 2019 ) and pain progression with peripheral inflammation ( Ma et al, 2019 ). Furthermore, abnormal functional patterns of the limbic system were significantly associated with pain-related characteristics and psychiatric disorders ( Huang et al, 2021 , Wei et al, 2020 ), and had the potential for predicting NSAIDs efficacy in migraine treatment ( Wei et al, 2022 ). Therefore, our findings suggest that the effectiveness of known treatments that managing both migraine and psychiatric conditions should be considered in the clinical strategy.…”

Identifying the neural marker of chronic sciatica using multimodal neuroimaging and machine learning analyses

“…In recent years, SVM techniques combined with neuroimaging metrics have been applied to differentiate pain patients from HCs and to predict the outcome of certain interventions ( Bagarinao et al, 2014 ; Zeng et al, 2019 ; Huang et al, 2020 ; Tu et al, 2020b ; Gui et al, 2021 ; Wei et al, 2022 ). The patients with neuropathic pain and the HC were classified by the mean ALFF values of the frontal gyrus and the precuneus using the linear SVM classifier, and the classification accuracy was 86.36% between the PHN patients and HC ( Huang et al, 2020 ).…”