To the best of our knowledge, this is the first CBMA of CTh studies in migraine. Using the SDM-PSI meta-analytical approach, our CBMA that included 17 datasets comprising 872 patients and 949 controls detected no statistically significant consistency of CTh alterations in patients with migraine relative to healthy controls. This lack of specific CTh alterations indicates that CTh analysis is not a reliable and reproducible metric as a potential biomarker of migraine. Although little is known about the exact reasons for the absence of consistency of CTh alterations in migraine, we will discuss the possible sources and factors from the variability of sample size and heterogeneous patient selection criteria to imaging collection and methodological differences across independent studies.
There is an increasing concern regarding the reliability and reproducibility in neuroimaging research [41]. A small sample size with low statistical power undermines the reliability of neuroscience [42]. A power calculation is applauded to estimate the appropriate sample size before the study initiation. A well-powered cross-sectional CTh study required approximately 50 subjects per group to detect a 0.25-mm CTh difference [43]. Sample size estimates were heterogeneous over the cortical surface [43]. Of the 17 datasets included in the CBMA, the sample sizes range from 11 to 166 (mean 51.3) in the patient groups and from 11 to 309 (mean 59.3) in the HC groups, of which the majority (n = 13) enrolled participants with small sample size less than 50 subjects per group. Only three studies included in the CBMA conducted prior statistical power calculations with different sample sizes required [25–27]. The findings from these studies with small sample sizes were probably false that affected the generalizability of the obtained results. Although it is challenging in practice, data sharing or multi-center collaboration to increase the sample size (and therefore power) is highly needed [41, 42].
Heterogeneous patient selection criteria make it difficult to define consistent migraine characteristic alterations of CTh. Migraine is a heterogeneous neurological disease. Some datasets only enrolled episodic migraineurs [12, 15, 16, 22, 24–26, 28], while some other datasets included both episodic and chronic migraineurs [13, 14] or only chronic migraineurs [11, 44]. Majority of the datasets (n = 14) in the CBMA included patients with mixed gender [10, 11, 13–17, 22, 25–28, 44], while two datasets only included female migraine patients [23, 24] and one only male migraine patients [12]. Some datasets only included patients with aura [16, 23, 25, 28] or without aura [13, 22, 24–26], while some other studies included both patients with migraine without aura and with aura [10, 11, 15, 17]. Individual CTh studies showed that the observed regional pattern of CTh abnormalities in patients with migraine was influenced by age [10, 11], gender [12], gender [12], disease duration [11, 13–15], attack frequency [13–15, 44], pain intensity [14], the presence of aura [15–17], and photosensitivity [18]. In addition, two datasets were cross-sectional population-based studies [23, 27] and the rest are clinic-based studies that the former minimized the selection biases compared to the later. Migraine is a recurrent headache disorder characterized by a cycle of attacks including pain-attack ictal and pain-free interictal phases. Different patterns of morphometric GM changes detected via VBM and dynamic variations in the anatomical microstructure of the thalamus detected via diffusion tensor imaging between ictal and interictal phases were observed in migraine, which suggests that abnormal structural plasticity may be an important mechanism of migraine pathology [45, 46]. While no CTh studies have been conducted to explore headache phase-related cortical plasticity in migraine. An extensive literature has shown that a wide range of psychiatric disorders, especially anxiety and depression, can accompany migraine [4, 47–49]. Previous studies revealed cortical abnormalities in depression [35, 50, 51] and anxiety disorders [52–55]. However, these psychiatric problems are often under-diagnosed and have not been thoroughly assessed in CTh studies in migraine. Only a few studies in the CBMA included patients at the medication-free state [10, 16, 22]. Medication status and type are other potential confounders that may influence CTh findings in migraine; however, no CTh studies have attempted to evaluate such effects.
Differences in imaging collection and methodology of CTh analyses may also have contributed to the absence of consistency from CTh studies in migraine. Previous reports showed that results of CTh analyses can be influenced by scanner platform [56, 57], field strength [58–60], pulse sequence [58, 61, 62], the number of coil channels [61], scanner relocation [63], and imaging sites [56, 64]. As shown in Table 2, differences in scanner manufacturer and platform (Siemens, Philips, and GE), field strength (3.0 T and 1.5 T), head coil (8-, 12-, 32-, and 64-channel), MR sequence (MPRAGE, FFE, TFE, FLASH, and FSPGR), TR/TE, and voxel size (from 1.33 × 1.0 × 1.0 to 0.89 × 0.89 × 0.8 mm3) across studies were noted. Besides, variations in computing workstation types [65], operating systems [65, 66], processing pipelines and software packages [65, 67, 68], the extent of smoothing [69], and statistical strategies [10, 16, 17, 22] may produce inconsistent results. These differences make direct comparisons between the different studies difficult. A multi-center study from four academic headache centers used different MRI systems and vendors, MRI sequences, TR/TE, and voxel size [15], which were controlled in the subsequent analyses. However, most individual studies did not explicitly state the computing workstation types and operating systems used in the CTh analyses. It has been shown that variability at various levels of processing pipeline influences cortical thickness measurement [70]. The CTh studies in migraine included in the CBMA used divergent processing pipelines and software packages (different versions of FreeSurfer and CAT12), smoothing kernels, and statistical strategies were used. Specially, four studies revealed that the use of a more liberal uncorrected threshold produced more positive results [10, 16, 17, 22], which may be false positive. Moreover, there is increasing awareness that image quality can systematically bias the results [71–73]. Quality control of imaging data in the processing pipelines should be applied in all CTh studies to achieve reliable results [71, 72]. However, only three of the studies included in the CBMA explicitly conduct a visual inspection and manual correction of topological errors for quality control [10, 14, 15].
As discussed above, many potential confounders may contribute to the inconsistencies of CTh alterations in migraine, which merit attention in future studies. Of the 17 CTh datasets included in the CBMA, 9 reported null finding in patients with migraine relative to healthy controls using corrected thresholds for multiple comparisons [10–12, 16, 25–28]. Is migraine indeed not associated with CTh alterations? Are significant CTh alterations observed in the studies secondary, or specified to migraine subgroups, or just a reflection of structural plasticity of the migraine cycle? To answer these questions and to obtain reliable results, we need to design longitudinal population-based studies at different migraine phases that recruit homogeneous patients with appropriate sample size using standardized imaging collection protocols with high field strength, multi-echo sequence, and a high number of coil channels and latest well-validated processing and analysis pipelines controlling for the age, gender, comorbidities, and medication. Besides, longitudinal multimodal neuroimaging studies would contribute to elucidate whether CTh alterations are secondary to chronic functional abnormalities.
Several limitations to our CBMA must be considered. First, given the clinical heterogeneity of migraine and the lack of sufficient original studies, we were unable to conduct separate subgroup CBMA to identify the effects of potential moderators, such as migraine with aura vs. migraine without aura, male migraine vs. female migraine, and episodic migraine vs. chronic migraine. More CTh studies in migraine with homogeneous subtypes are needed to characterize the CTh patterns. Second, the present meta-analysis is coordinate-based rather than image-based or mixed coordinate- and image-based, which may lead to biased results. Future studies with imaging data sharing would be helpful to obtain more accurate results.