The Role of Attachment Style, Adverse Childhood Experiences and Dissociation in Migraine

ABSTRACT Migraine and chronic migraine are caused by a combination of modifiable and non-modifiable genetic, social, behavioral and environmental risk factors. Further research of possible modifiable risk factors for this headache disorder is merited, given its role as one of the leading causes of years lived with disability per year. The first aim of this online cross-sectional study was to investigate the psychosocial risk factors that predicted chronic migraine and severe migraine-related disability in 507 Irish and UK participants, focusing specifically on childhood maltreatment, attachment and tendency to dissociate, or experience depressed mood and/or anxiety. Additionally, this study aimed to examine variables that mediated the relationships between these psychosocial risk factors and migraine chronicity or severe migraine-related disability. Adjusted binary logistic regression revealed that shutdown dissociation (Odds Ratio [OR] 4.57, 95% Confidence Interval [CI] 2.66–7.85) and severe physical abuse (OR 4.30, 95% CI 1.44–12.83 had significant odds of predicting migraine chronicity, while depression (OR 3.28, 95% CI 1.86–5.77) significantly predicted severe migraine-related disability. Mediation analyses indicated that shutdown dissociation mediated the relationship between seven predictor variables and both chronicity and severe disability including possible predisposing factors emotional abuse, physical neglect, avoidant attachment and anxious attachment. These findings suggest that early life stressors (such as childhood trauma and avoidant attachment style), shutdown dissociation and depression may impact on migraine trajectory. To investigate whether these psychosocial factors are risk factors for migraine chronicity or disability, prospective research should be conducted in this area to account for fluctuations in migraine chronicity over time.


The burden of migraine
Despite 30 years of global research documenting the burden of headache disorders, they have remained one of the leading cause of years lived with disability worldwide (Feigin et al., 2019;James et al., 2018). Migraine and headache disorders are highlighted as an area of focus for reducing global disability (Saylor & Steiner, 2018;World Health Organization [WHO], 2011) and have been undiagnosed, underestimated and undertreated across healthcare systems (Buse et al., 2009;Lipton et al., 1992;Wang et al., 2000). Chronic migraine, affecting roughly 2-3% of the global population (Buse et al., 2012;Natoli et al., 2010), has been associated with changes in brain structure (Planchuelo-Gomez et al., 2020), increased economic burden (Blumenfeld et al., 2011;Munakata et al., 2009), psychiatric comorbidity (Chen et al., 2012), medication overuse , headache-related disability (Adams et al., 2015) and risk of further health problems including respiratory (Buse et al., 2010;Chen et al., 2012) and cardiovascular disorders Schurks et al., 2009).

Risk factors for chronic migraine
Progression from episodic to chronic migraine originates from a combination of genetic, social, behavioral and environmental risk factors (Breslau & Rasmussen, 2001;Lipton & Bigal, 2005;Probyn et al., 2017). Research on risk factors for chronic migraine progression demonstrated strongest evidence for headache frequency, depression and medication overuse, but called for further investigation into modifiable risks; including psychosocial factors Xu et al., 2020). Depression is the only significant psychosocial risk factor identified thus far, with anxiety and stress mentioned as having an insufficient number of prospective or case-control studies Xu et al., 2020).

Predisposing psychosocial risk factors in migraine
Despite these findings, exposure to repeated stress has been implicated in contributing to allostatic load and causing lasting structural and functional changes in the brain (Borsook et al., 2012). Psychosocial constructs measuring early stress such as childhood trauma and insecure attachment styles could offer additional insight into mechanisms that explain the association between stress and migraine (Ehrlich & Cassidy, 2019). Retrospective research into childhood maltreatment has demonstrated significant odds ratios for migraine, identifying emotional abuse as having the greatest impact on migraine above all other forms of trauma (Tietjen, 2016;Tietjen et al., 2010aTietjen et al., , 2015Tietjen et al., , 2017 even after adjusting for anxiety and depression (Tietjen et al., 2015(Tietjen et al., , 2017. However, such research has not compared emotional, physical and sexual adverse experiences using one validated measure, in episodic and chronic migraine. The attachment system has been described as a hardwired biological system arising from the threat or fear system, which can foster adequate careseeking and caregiving behavior (McCluskey & O'Toole, 2019). Insecure attachment has been identified as more prevalent in migraine groups across five studies (Esposito et al., 2013;Savi et al., 2005;Tarantino et al., 2017aTarantino et al., , 2018Williams et al., 2017). Anxious attachment and not avoidant attachment is more associated with increased headache intensity and frequency in child and adult samples (Berry & Drummond, 2014;Tarantino et al., 2017a), however, there is a need for additional research comparing both dimensions using larger samples.

Perpetuating psychosocial risk factors in migraine
Investigating the interplay of historical predisposing risk factors in migraine, such as ACE and insecure attachment, with current, perpetuating factors in migraine such as dissociation and psychiatric comorbidity could help better explain the onset and chronification of this disorder. For instance, a history of childhood trauma (Tietjen et al., 2007(Tietjen et al., , 2010a and insecure attachment (Mula et al., 2016;Rossi et al., 2005;Tarantino et al., 2017a;Williams et al., 2017) has been associated with comorbid anxiety and depression in migraineurs, highlighting a possible explanatory role in impacting migraine. However, these findings merit clarification both with mediation analyses and adjusted regression models using migraine characteristics such as chronicity and disability as outcomes. Similarly, dissociation has been implicated in explaining the relationship between ACE or trauma and migraine, including psychoform dissociative symptoms (e.g., amnesia, depersonalization, derealization and absorption) and somatoform dissociative symptoms (e.g., anesthesia, analgesia, pain symptoms). However, research has generally been limited in analyses and requires further investigation using adjusted regression or mediation analyses to (Arik et al., 2008;Kucukgoncu et al., 2014;Özsoy & Taşci, 2021;Saçmacı et al., 2020).
Theories integrating biopsychosocial factors to explain the impact of recent or historical stress on intense emotions or defense mechanisms such as dissociation merit consideration, particularly with physiological (Steppacher et al., 2016;Szabó et al., 2019) and psychological (Galli et al., 2017;R. A. Nicholson et al., 2007) differences documented in how migraineurs manage emotion. There remains a lack of consensus in defining dissociation, given the purported range of related phenomena and etiological causes (Van der Hart, 2021). However, research has offered insight into the biopsychosocial determinants of dissociation resulting from recent or historical trauma (Lanius et al., 2012;Nijenhuis et al., 2002;Schauer & Elbert, 2010;Van Dijke et al., 2015).
One such construct known as the defense cascade model provides insight into six behavioral reactions to danger and sympathetic and parasympathetic responses of the autonomic nervous system (ANS) including shutdown dissociation which is considered a progression on this model that enhances survival (Schalinski et al., 2015;Schauer & Elbert, 2010). Shutdown dissociation is considered a neurobiological response to stress associated with a shutting down of motor sensory and speech systems and parasympathetic dominance of the ANS (Schauer & Elbert, 2010). Given the impact of dissociation and the ANS in migraine (Miglis, 2018), measuring shutdown dissociation may offer additional insight into the psychosocial risk factors of this headache disorder.

Study aims
Research has argued for continued work into understanding and reducing the burden of chronic migraine (May & Schulte, 2016;Steiner et al., 2018). Limited research has investigated the impact of psychosocial risk factors across episodic and chronic migraine groups, opting instead for control or mixed headache groups. This study had two aims, the first of which was to investigate whether psychosocial risk factors such as childhood maltreatment, attachment and having a tendency to dissociate, or experience depressed mood and/or anxiety predicted chronic migraine or severe migraine-related disability. The second aim was to examine what variables mediated the relationships between psychosocial risk factors and migraine chronicity or severe migraine-related disability.

Participants
Participants were recruited to take part in a cross-sectional online study involving nine self-report questionnaires. Participants were excluded if they did not attend a health professional for migraine, reported having a different headache disorder and not migraine, or confounding health problems, such as space-occupying tumors. Participants were included if they were age 18 or older, living in Ireland or the UK, experienced migraine and had attended a health professional regarding migraine. Participants were classified as having episodic or chronic migraine and little, mild, moderate or severe disability respectively.

Demographic questionnaire
A 14-item demographic questionnaire was developed and included age, sex, occupation and working hours per week, and health-related information such as headache diagnosis, and health professional attended.

Chronic migraine
The Identify Chronic Migraine (ID-CM; Lipton et al., 2016) is a 12-item screening tool for chronic migraine, measuring headache frequency, headache symptoms, medication use and the impact of headache on activities and making plans. Six items relating to symptoms and making plans use fourpoint Likert scales and six items are open-ended questions for migraine frequency, medication and activity. This measure has a sensitivity of 80.6%, a specificity of 88.6%, a negative predictive value of 75.0% and a positive predictive value of 91.5%.

Migraine-related disability
The Migraine-Related Disability Assessment Questionnaire (MIDAS; Stewart et al., 2001) is a five-item questionnaire that measures missed days off work or school, household and family or social activities and reduced productivity in work or school and house activities in the last 3 months (Stewart et al., 2001). It classifies participants into four grades, Grade 1: Little or no disability (0-5), Grade 2: Mild disability (6-10), Grade 3: Moderate disability (11-20) and Grade 4: Severe disability (21+). The MIDAS has demonstrated acceptable internal consistency Stewart, Lipton, Whyte et al., 1999) and validity in comparison to a diary-based assessment tool (Stewart et al., 2000).

Childhood trauma questionnaire-short form
The Childhood Trauma Questionnaire-Short Form (CTQ-SF; Bernstein et al., 2003) is a 28-item retrospective measure of five areas of maltreatment: physical, sexual and emotional abuse and emotional and physical neglect. It has demonstrated good internal consistency and criterion-related validity (Bernstein et al., 2003). Responses for each item are given on a 5-point Likert scale. Levels of maltreatment were quantified as "None or Minimal," "Low to Moderate," "Moderate to Severe" and "Severe to Extreme" using cut off scores from Bernstein et al. (2003). Cronbach's alpha scores for emotional abuse, physical abuse, sexual abuse, emotional neglect and physical neglect were of .890, .861, .954, .935 and .715 respectively.

Attachment
The Experiences in Close Relationships -Relationship Structures Questionnaire (ECR-RS; Fraley et al., 2011) incorporates nine items that measure current attachment anxiety and avoidance in close relationships in general on a seven-point Likert scale. Using continuous scales that measure avoidant and anxious attachment dimensions gives a more precise representation of adult attachment than the four categories of attachment (e.g., secure, or preoccupied attachment; Brennan et al., 1998). This is particularly the case given that true attachment typology is uncertain and instead categories are considered regions in a two-dimensional space (Fraley & Waller, 1998). One month test-retest reliability of the individual scales has been found to be approximately 0.65 for romantic relationships and 0.80 for parental relationships (Fraley et al., 2011). Higher average scores suggest increased levels of anxious and avoidant attachment with suggested cut offs based off a normative sample of over 17,000 online participants. A Cronbach's alpha of .872 was found for the ECR-RS in this sample.

Dissociation
The Dissociative Experiences Scale-II (DES-II; Carlson & Putnam, 1993) is a 28-item screen for psychoform dissociative symptoms namely amnesia, depersonalization or derealization and absorption. Participants rate each item according to what percentage of the time these symptoms happen to them using an 11-point Likert scale from 0 to 100. The DES-II has been used in a range of mental health populations (Lyssenko et al., 2018). A cut off score of 30 or more is considered in the severe dissociative range and indicative of a need for follow up with structured clinical assessment. Good internal reliability and construct validity have been demonstrated for this measure (Carlson & Putnam, 1993;Zingrone & Alvarado, 2001). A Cronbach's alpha of .930 was found for this measure in the present study.
The Shutdown Dissociation Scale (Shut-D; Schalinski et al., 2015) is a 13item measure of dissociative experiences in the last six months. Responses were recorded on a four-point Likert scale and summed scores range from 0 to 39. Originally this questionnaire was used as an interview to be applicable in resource-poor settings, but has been used as a self-report questionnaire in this study for large scale data collection. Shut-D has been shown to have good psychometric characteristics, demonstrating good internal consistency, excellent retest reliability and high convergent validity and satisfactory predictive validity (Schalinski et al., 2015). Notably, its high convergent validity was documented with significant associations with all subscales of the Dissociative Experiences Scale (DES; Bernstein & Putnam, 1986). A Cronbach's alpha of .837 was found for the SHUT-D in this study.

Depression
The Patient Health Questionnaire-8 (PHQ-8; Kroenke & Spitzer, 2002) measures eight of nine criteria for depressive disorders (American Psychiatric Association [APA], 2013; Kroenke et al., 2009). It omits the 9 th criteria relating to suicidal or self-injurious thoughts to account for the use of online anonymous questionnaires in this survey with no opportunity for follow-up. Participants rate symptoms experienced over the last 2 weeks using a fourpoint Likert scale. Items are summed and range from 0 to 24. The PHQ-8 has minimally reduced sensitivity but similar specificity and cut offs for depression severity as the Patient Health Questionnaire-9 (PHQ-9; Kroenke & Spitzer, 2002;Wu et al., 2019). A Cronbach's alpha of .894 was found for the PHQ-8 in this study.

Anxiety
The General Anxiety Disorder-7 (GAD-7; Spitzer et al., 2006) measures seven anxiety symptoms experienced over the last 2 weeks. Participants rate symptoms experienced over the last 2 weeks using a four-point Likert scale with responses summed and ranging from 0 to 21. Cut off scores of five, 10 and 15 have been considered for mild, moderate and severe symptoms, with 10 being a reasonable score indicative of GAD. The GAD-7 is reported to have excellent internal consistency and good test-retest reliability (Spitzer et al., 2006). A Cronbach's alpha of .918 was found for the GAD-7 in this sample. Procedure Ethical approval was granted by the National University of Ireland, Galway Research Ethics Committee. Participants were invited to take part in an online study via press release, social media and organization mailing lists such as The Migraine Association of Ireland. The online survey was accessed in LimeSurvey (Schmitz, 2012) and took approximately 30 minutes to complete. Participants gave informed consent prior to beginning this survey. Data were inputted into SPSS Statistics version 26 (IBM Corp, 2019) for analysis.

Data analysis
A priori analyses were conducted on G*Power for regression models depending on best fit (Faul et al., 2009). For instance, for a two-tailed binomial logistic regression, the minimum required sample size of 325 participants was determined using the input parameters: effect size OR = 2; α = 0.05; 1-β = 0.95. Data were first explored using descriptive statistics reviewing demographic and psychological variables such as gender and CTQ respectively. Bivariate correlations, t-tests, one-way analyses of variance (ANOVA) were then conducted, followed by adjusted binary logistic regression to assess relationships between predictor variables and the outcome variables: migraine chronicity and migraine-related disability. Alternative outcome variables were each included as predictors in adjusted regression models, in accordance with past research Scher et al., 2017). Mediation analyses using PROCESS version 3.4 (Hayes, 2019) were conducted to assess indirect effects between outcome and predictor variables.

Preliminary analysis
A total 1461 individuals took part in the online survey, of which 507 were eligible for analysis. The other 954 participants did not begin (424), or complete (493) the survey, or met exclusion criteria (37). Demographic and outcome variables of participants, are presented in Table 1. Participants meeting criteria for episodic migraine and chronic migraine were 339 and 168 respectively; while 64, 64, 75 and 299 participants were categorized as having little, mild, moderate and severe disability]. Five participants provided invalid responses for all questions in the MIDAS regarding the number of days they were impacted by their migraine and as such were excluded from analysis of migraine-related disability.

Binary logistic regression
Two hierarchical logistic regression analyses were performed to determine which variables independently predicted headache chronicity and disability, using episodic headache and little/no-moderate disability as reference categories. Collinearity diagnostics were conducted using the mean VIF of the logit for each model; there was no evidence of multicollinearity (mean VIF of 1.52 for both outcomes respectively). The logit linearity assumption was determined by multiplying continuous predictors by their respective log transformations, with significant interactions indicating a violation of linearity (Stoltzfus, 2011). There was evidence of log linearity violations for all continuous variables other than participant age. Therefore, other than age, all continuous variables with established cut off thresholds (CTQ, PHQ-8, GAD-7, ECR-RS) were categorized accordingly. However, continuous variables with no standardized cut offs (DESII, SHUT D) were dichotomized based on the empirical optimal cut off point, whereby a threshold with the maximum  discriminatory capability based on the area under the receiver operating characteristic (ROC) curve was used (Habibzadeh et al., 2016;Unal, 2017). Although categorizing continuous predictors necessitates larger samples, the number of events in both regressions (203 and 168 for the disability and chronicity outcomes respectively) should be sufficient to avoid over-fitting in accordance with the events-per-variable criteria (Vittinghoff & McCulloch, 2007). Significant predictor variables were determined after adjusting for age, gender, anxiety, depression and headache features (ID-CM, MIDAS), as per Tables 9 and 13.

Migraine chronicity logistic regression
Unadjusted and adjusted logistic regression models for migraine chronicity are presented in Tables 7-9 and for migraine disability in Tables 11-13. Each model significantly contributed to the variance based on log likelihood as shown in Tables 6 and 10. Physical abuse and shutdown independently predicted chronic migraine, after adjusting for disability and psychological and demographic variables such as depression, anxiety, gender and age. As per Table 9, disability was the strongest predictor of chronicity (OR 8.67, 95% CI 4.74-15.87), followed by shutdown dissociation (OR 4.57, 95% CI 2.66-7.85)

Mediation analyses
Parallel multiple mediation analyses using PROCESS (Hayes, 2019) were conducted to investigate the relationship between psychosocial predictors and migraine chronicity and disability. Dichotomous variables such as migraine disability and gender, could not be included as mediators, however, available continuous variables such as MIDAS total scores were incorporated into mediation analyses. Exploratory models were conducted using a maximum of seven mediators, however, given the number of variables available, only mediators with significant indirect effects, or approaching significance, were included. Including multiple mediators in a model allows for the analysis of specific indirect effects of a predictor variable (X) on an outcome variable (Y) while holding constant other mediators, however this comes at a cost of decreased power and increased sampling variance (Hayes, 2017). Therefore, these analyses focused on indirect effects of individual mediators rather than total effects for two reasons. Firstly, as outcome variables were dichotomous standardized effect sizes could not be calculated, therefore indirect effects were examined using bootstrapping (BCa) with 5000 bootstrapped samples. Secondly, measuring total effect in mediation is not considered a requirement given its purpose is to test for indirect effects (Agler & De Boeck, 2017) and that competing direct and indirect effects can supress their effects and result in near-zero total effects (MacKinnon et al., 2000). Bootstrapped samples were considered significant (p < .05) if 95% confidence intervals (CIs) did not cross zero. A path model conceptualizing a parallel multiple mediation analyses with three mediators is presented in Figure 1.

Mediation analyses for predictors of migraine chronicity
Mediation models for significant indirect effects of psychosocial variables on migraine chronicity are presented in Tables 14-17 respectively. Shutdown dissociation mediated the relationship between seven predictor variables and ID-CM, namely emotional abuse, b = .20, Bootstrap 95% Bias Corrected Confidence Interval (BC CI) .01-04, physical neglect, b = .25 95% Bootstrap

Key findings
The aims of this study were to investigate the predictive nature of psychosocial factors in chronic migraine and severe migraine-related disability and examine indirect effects between these variables. There are several significant findings from this research, the first of which indicated psychosocial difficulties were generally more prevalent in chronic than episodic migraine and in severe than in less severe migraine-related disability. Adjusted binary logistic regression revealed that physical abuse, shutdown dissociation and migraine-related disability had significant odds of predicting migraine chronicity, while chronicity and depression significantly predicted severe disability. Not surprisingly,  disability was the highest predictor of chronicity and vice versa given their association and that MIDAS was one of the measures used in developing the ID-CM Manack et al., 2011;Scher et al., 2019). Similarly, while migraine-related disability is considered a secondary outcome, both outcome variables in this study were associated with a range of direct and indirect effects, highlighting the role of this construct in offering additional insight into the impact of migraine alongside chronicity (Andrasik et al., 2005;D'Amico et al., 2013;. Interestingly, physical abuse was the only childhood trauma to predict either outcomes after controlling for confounding variables. Previous research to date has suggested that emotional maltreatment was more prevalent in migraine compared with physical or sexual abuse (Tietjen et al., 2010b(Tietjen et al., , 2015. Only one study to date had compared childhood maltreatment between episodic and chronic migraine, documenting emotional abuse as being the highest predictor for chronic migraine and for those that had recently progressed from episodic to chronic, while physical abuse was the   (Tietjen et al., 2010b). In contrast, this study highlights physical abuse as being a markedly better predictor of chronicity than both emotional neglect and abuse, with notable differences in odds ratios. Depression was the only predictor of severe disability after migraine chronicity and gender, a finding which is not surprising given its high prevalence in migraine (Minen et al., 2016). Previous research using the Hospital Anxiety and Depression scale demonstrated that migraineurs with both depression and anxiety, but not anxiety or depression alone, have demonstrated increased disability (Lanteri-Minet et al., 2005;Peterlin et al., 2009). However, this study found that increased depression as measured with the PHQ-8 predicted severe disability, consistent with the findings of Seo and Park (2015). Depression, can be argued as being a possible by-product of a natural trajectory. However, at the very least it highlights a complex relationship with migraine, especially given the role depression plays in mediating nine individual relationships between predictor variables and disability including possible predisposing risk factors such as emotional neglect, emotional abuse, physical neglect, sexual abuse, avoidant attachment and anxious attachment. In a similar vein, previous research has documented that depressive symptoms and insecure attachment were the most significant predictors of disability in episodic and chronic migraine (Rossi et al., 2005). Heretofore, psychosocial difficulties in migraine were most commonly assessed under the auspices of psychiatric comorbidity. While assessing mental health in migraine is important, it is not enough. Research has argued that modern classification systems fail to capture other psychosocial variables prevalent in migraine such as DSM-5 (Cosci et al., 2020).
Shutdown dissociation and not psychoform dissociation predicted chronic migraine which can be considered a noteworthy finding, requiring further investigation. This is especially the case as it mediated seven predictor variables and chronicity and seven predictor variables and severe disability. Importantly, this indicates that these predisposing factors are related to increased chronicity if migraineurs engage in shutdown dissociation, which is a novel finding in the migraine field. Higher rates of childhood neglect and somatoform and psychoform dissociation have been documented in chronic headache samples (Arik et al., 2008;Yücel et al., 2002), but this research has grouped together multiple forms of neglect. Another study demonstrated significantly higher levels of emotional abuse and somatoform dissociation in migraine than in healthy controls, but did not investigate this finding further than between group analysis (Kucukgoncu et al., 2014).
The SHUT-D is argued to be distinct from other measures of dissociation such as DES-II in that it was constructed based on a theoretical model that includes the biological and the neural system and not based on clinical heuristic observations. This model therefore provides an opportunity for additional integrated biopsychosocial research into dissociation and migraine.
For instance, left-side migraineurs are identified as having predominant parasympathetic activation in response to pain compared to right-side migraineurs (Avnon et al., 2004). Given the role of the parasympathetic system in dissociation, future research could examine shutdown dissociation, the autonomic nervous system and migraine location.
Incorporating biopsychosocial constructs, such as shutdown dissociation, into migraine treatment offers new strategies in reducing the risk of chronification and disability in migraine. For instance, measuring ANS functioning through the vagus nerve, an important connection between body and brain could offer additional insight in assessment of stress, trauma and migraine pain (Porges & Dana, 2018;Yuan & Silberstein, 2016). Similarly, growing evidence concerning neurofeedback in PTSD samples has demonstrated changes in amygdala connectivity and emotion regulation regions of the brain (Nicholson et al., 2020b(Nicholson et al., , 2020a. In a therapy setting this could include psychoeducation of possible physiological responses in shutdown, comparing sensory similarities between the trauma context and a therapeutic setting, promoting activation such as applied muscle tension and not relaxation to maintain arousal and prevent immobility, and importantly not terminating exposure to the trauma before integration with contextual cues such as time and location of danger (Hembree & Cahill, 2007;Schauer & Elbert, 2010;Schauer et al., 2011).

Limitations and considerations for future research
This research employed an online cross-sectional design and offers only a brief snapshot of a migraineur's experience. Recall bias using the retrospective CTQ, is therefore a probable source of bias arising from this design. Underor over-reporting are also possible limitations associated with the CTQ owing to the sensitivity of this area. Furthermore, given this cross-sectional design these findings cannot demonstrate causality for the direct or indirect role of psychosocial factors on chronic migraine or severe disability. This is especially important given that the purported limitations in screening tools for chronic migraine (Potter et al., 2019) and the diagnostic boundary of chronic migraine being created by consensus rather than empirical investigation with fluctuations documented in participant criteria for chronic migraine Caronna et al., 2020;Serrano et al., 2017). Not having access to migraine diagnoses could have impacted on analysis and interpretation of results. One important example of this is the possibility of symptoms relating to migraine with aura confounding the high rates of shutdown dissociation, predicting chronic migraine.
Undercoverage and self-selection bias are other important limitations to consider with the use of an online survey, given that they only capture data of those with internet access or adequate computer literacy and may not meet the principles of probability sampling (Bethlehem, 2010;Couper, 2000;Wright, 2005). This bias may partially explain the proportionally higher rates of chronic and severely disabled migraine participants in this study given that these participants have a more active presence online for the management of migraine.
The limitations of this study point to the need for further research into psychosocial variables using prospective designs from multiple recruitment pathways including online and clinic or GP based pathways. Studying biopsychosocial risk factors for chronic migraine has been proposed to offer targets for further understanding of and intervention for this condition . Doing so could help stem the onset of chronic migraine progression or improve remission to episodic migraine (May & Schulte, 2016). However, research into psychological intervention is constrained by a lack of highquality research and insufficient efficacy in improving the primary outcome of migraine frequency (Sharpe et al., 2019), a finding reminiscent of early pharmacological interventions in headache (Tfelt-Hansen et al., 2000). An important caveat to consider with psychosocial factors in this condition is that migraineurs may also not adhere to psychological interventions (Gewirtz & Minen, 2019;Minen et al., 2020), nor consider them a high priority compared to fast-acting medication (R. A. Nicholson et al., 2007). Continued research is ongoing (Klan et al., 2019;Mansueto et al., 2018), however, furthering this area requires systematic evaluation of interventions targeting modifiable risk factors in episodic migraine over sufficient time to discern whether they can adequately reduce progression to chronic migraine status .

Disclosure statement
No potential conflict of interest was reported by the author(s).

Funding
The author(s) reported there is no funding associated with the work featured in this article.