Abstract
Background: The burden of migraine is determined by impairment during attacks due to pain or non-painful symptoms such as cognitive symptoms.
Objective: Development of a questionnaire to measure self-reported subjective cognitive symptoms during migraine attacks.
Methods: Item generation was accomplished through structured patient interviews analysed by a panel of experts. A set of 43 candidate items was applied to consecutive migraine patients. Test construction with factor analysis retained nine items. Internal consistency was assessed with Cronbach’s alpha and Spearman’s rho, and convergent and construct validity by correlation to spontaneous cognitive complaints, and the 43-item and the Cognitive Failures Questionnaires.
Results: The nine-item Mig-SCog covers two domains, executive functions and language. Cronbach’s alpha was 0.82. It correlates with spontaneous cognitive complaints (p < 0.001), the 43-item (rho = 0.69) and the Cognitive Failures Questionnaires (rho = 0.61). Test–retest reliability (Cohen’s kappa) was 0.55.
Conclusions: Mig-SCog is a valid, reliable, consistent working instrument of fast self-administration that quantifies subjective cognitive symptoms during migraine attacks.
Introduction
Migraine is a highly prevalent disorder (1–3) and causes disability affecting millions of patients daily. Its overwhelming impact in world health was recognized by the World Health Organization in its 2001 World Health Report (4), as migraine was listed in the top 20 causes of Years of Life Lived with Disability (YLDs) worldwide, in both genders and all ages, being responsible for 1.4% of total YLDs (4–6).
The disability imposed by migraine affects mostly young and active individuals, leading to a significant public health and economic impact (7). Direct costs include health services and medication (7,8), yet indirect costs represent 70% of the economic burden and result from reduced productivity at work/school or to absenteeism (7,9–11). Family and leisure time is also affected, with impact on both the patient and their personal relations (7). Adding to a documented decrease in quality of life both during and between attacks, there are also unaccounted indirect costs, due either to inability to participate or to phobic avoidance of leisure and social activities (7,12).
A patient’s degree of disability during attacks is determined by the frequency, duration and intensity of pain (7,12–16), but also from associated symptoms such as nausea and vomiting (9,14,16,17). In addition, many patients report disabling cognitive symptoms (14,16,18,19) and patient testing during attacks has revealed impairment in several cognitive domains such as processing speed, sustained attention/concentration, working memory, visual-spatial processing, alertness/fatigue (20–22), immediate and sustained attention and verbal learning (23).
Some authors have also documented interictal mild executive dysfunction in a subgroup of migraine patients, which was interpreted as a possible cumulative effect of repeated attacks (24). Further literature revealed conflicting results, showing no differences between migraineurs and controls in interictal cognitive function (25–27).
Patients often report that effective medication can relieve their pain and/or nausea but cognitive symptoms tend to persist (28), often through to the following day. Persisting symptoms are described by 80% of migraineurs and include mental tiredness, asthenia, somnolence, depressed mood and concentration difficulties (29,30). Acute migraine treatment with sumatriptan has been able to revert both pain and cognitive impairment in small uncontrolled trials (20,22).
The cause of cognitive symptoms and impairment during attacks remains elusive, yet patients often complain that this type of symptom can be as disabling as migraine pain itself. No measurement of this kind of subjective complaints exists (12–14).
Our aim was to develop a specific instrument to quantify subjective cognitive symptoms during migraine attacks. Such an instrument could contribute to the assessment of attack-related disability and to monitor the effect of acute medication.
Subjects and methods
Patients were recruited from Headache and Neurology Outpatient Clinics of two general hospitals in Lisbon, Portugal. Consecutive patients, either first or follow-up visits, who fulfilled the inclusion criteria, were invited to participate. The study protocol was approved by each hospital’s Institutional Review Board.
Inclusion criteria were: (a) age over 16 years; (b) at least second grade education (able to read and write); (c) history of episodic migraine with or without aura, as defined by ICDH-II (31); (d) migraine that had been present for at least one year with a minimum of two attacks in the 3 months preceding inclusion; (e) written informed consent. Exclusion criteria were chronic migraine, chronic daily headache with or without medication overuse, other headache diagnosis besides migraine, history of past or current alcohol or drug dependence or abuse, and severe or uncontrolled medical or psychiatric disorder.
Generation of scale items
Structured interviews with consecutive migraine patients (n = 37) from the outpatient headache clinic were conducted in order to identify cognitive symptoms during migraine attacks. From this patient-centred data, an expert panel of three neurologists with experience in headache and cognition selected relevant items and generated new items based both on cognitive complaints commonly described by patients during migraine attacks and on a relevant medical literature review. The panel also evaluated wording for complexity and ambiguity and supported item relevance and validity (32). This process resulted in a draft questionnaire that included an initial open-ended question: “Do you feel your mental capacities are different during your headache attacks? Please describe your main difficulties.” This was followed by 43 multiple choice questions, asking about several domains of cognitive function during the attacks. Applicability and understandability of questions was evaluated by a focus group of 10 migraine patients.
Patients had to self-rate each item-symptom in a three-option scale: occurring often (scoring 2 points), rarely (scoring 1 point) or not at all (scoring 0) during the attacks; it was also possible to answer “don’t know, don’t want to answer”, to access item comprehensiveness and adequacy. Some questions with reversed or clearly unrelated responses were included in order to access the no/yes-saying bias (32).
The draft 44-item self-administered questionnaire was applied to 93 consecutive migraine patients interictally, immediately after their routine clinical appointment. Demographic and clinical data were collected and analysed (age, gender, literacy, migraine diagnosis and characterization, disease duration, prophylactic medication use (yes/no)). One of the authors checked the forms for completeness.
For item reduction, items that performed poorly because of a high level (>10%) of “don’t know, don’t want to answer” responses were eliminated from the start. Factor analysis with varimax rotation of the remaining items was used to identify likely domains of cognitive function. Items with an eigenvalue of 0.400 or higher were retained unless they had an eigenvalue difference below 0.300 between any two factors. The result was a simplified nine-item multiple choice self-administered questionnaire – the Mig-SCog – with a total score varying from 0 to 18, the highest scores representing more expressive cognitive symptomatology during attacks.
Statistical analyses
Construct validity of the Mig-SCog was assessed by analysing the spontaneous symptoms evoked by the first open question in number and content, and by using this as an empirical measure to infer the meaning of the total questionnaire score. The average number of spontaneous cognitive symptoms reported during the attacks was correlated with demographic and clinical variables (age, gender, literacy, disease duration, type of migraine, average intensity, frequency and duration of attacks, current use of prophylactic medication) and with total score of the 43-item multiple choice questionnaire. Qualitative analysis was performed by an expert panel of neurologists with experience in cognitive testing, who categorized the symptoms reported in non-cognitive and cognitive domains in both the spontaneous cognitive complaints and the questionnaires.
Internal consistency was assessed with Cronbach’s alpha and the lower bound of the one-sided 95% confidence interval. Inter-item, item-test and item-rest correlations were tested with Spearman’s rank correlation (ρ) and Cronbach’s alpha. The item was included in the total score for item-test correlations, but the item was excluded in the item-rest correlations and Cronbach’s alpha. The same methods were used to analyse dimensions within the final instrument: inter-dimension, dimension-total and dimension-to-own correlations and Cronbach’s alpha.
In order to provide evidence of external validity, convergent validity was assessed by correlating the reduced Mig-SCog scores to scores of the cognitive failures questionnaire (33) applied to 31 patients selected at random. This is a self-rated questionnaire that measures frequency of memory and cognitive failure behaviours in daily life, spanning the most frequent cognitive symptoms and domains (33). To assess test-retest reliability, the simplified nine-item self-administered questionnaire was applied within a 3-month interval to a random sample of 33 patients. The agreement for each item was tested by Cohen’s kappa and the correlation between total scores was tested with Spearman’s correlation coefficient. The simplified nine-item Mig-SCog was applied to a subsample of patients (n = 33) who also rated themselves interictally, that is, referring to when they are not having an attack. Average scores were compared using the paired t-test.
Associations between patient variables and Mig-SCog scores were investigated with the Pearson’s correlation coefficient (continuous variables) or t-tests (binary variables). Using Ronald Fisher’s classic z-transformation to normalize the distribution of Pearson’s correlation coefficient (34), the sample size of this study has at least 80% power, with a 5% type-I error, to identify an association between Mig-SCog scores and patient variables having a true correlation coefficient of 0.30 or more.
Statistical analysis was done with SPSS v16.5 (Statistical Product and Service Solutions, Chicago, IL, USA) and STATA 11 (Stata Corporation, College Station, TX, USA).
Results
The preliminary study group consisted of 37 patients (28 females), of whom 12 had migraine with aura. The group had an average age of 36.4 years, mean disease duration of 16 years and an average MIDAS score of 16.4 on the previous 3 months. The majority of patients had 1–4 attacks monthly (67.6%), the attacks lasted less than 24 hours in 51.4% and were of moderate to severe intensity in 64.9%.On average, each patient described 4 frequent and 3.5 infrequent cognitive symptoms during migraine attacks. These data were used to select the initial candidate items for the self-administered questionnaire.
The main study group consisted of 93 patients (86 females), 18 having migraine with aura. Average age was 39.2 ± 11.6 years (range 18–83 years), average years in school were 11.7 ± 5 years (range 2–22 years) and mean disease duration was 18.4 ± 11.2 years (range 1–57 years). The majority of patients (53.8%) had 1–4 attacks monthly, most attacks lasting 4 to 24 hours (53.8%) and usually of moderate to severe intensity (98.9%). Sixty-seven patients (72%) were currently doing migraine prophylactics.
Answers to the first open question generated on average 3.3 ± 1.6 cognitive symptoms by patient, ranging from 0 to 9. The number of cognitive symptoms reported was not shown to be associated with any of the patient variables studied.
Qualitative analysis of cognitive symptoms allowed its grouping in 21 items. These items were then analysed and classified into cognitive and non-cognitive symptoms. We observed that 37% of spontaneous complaints were not purely cognitive yet all were related to known attack-related symptoms that are recognized as able to interfere with global function during attacks. Non-cognitive symptoms included humour/anxiety changes (feelings of impatience, irritability, intolerance, sadness, despair, panic, lack of self-control, n = 47), specific symptoms related to avoidance behaviours during the attacks (visual, noise, movement and physical effort intolerance, n = 44) and eviction itself (need for isolation and to stay still, eviction of social contact, n = 10) and global feelings of tiredness, exhaustion, dizziness, lack of balance or even changes in appearance (n = 13).
The most common reported symptoms were specific of cognitive domains (63%) and included attention (difficulty in thinking, decreased attention or concentration, mental confusion, trouble in studying, difficulty in performing mental calculation (n = 84), planning (difficulties in routine chores such as cooking, domestic chores, working, driving, in doing two tasks at a time, in resolving or organizing the day, n = 29), lack of initiative (feeling impaired, ‘anaesthetized’, diminished or blocked, ‘dumb’, decreased initiative, unable to take action, react or to decide, having trouble doing everything, n = 25), language (difficulty in speaking, talking, understanding when being spoken to, writing, forgetting the names of people and objects, n = 24), processing velocity (slowness of thinking, slowness when moving, needs more effort to do basic things, everything takes more time, more time to learn new information, n = 20) and memory (lack of memory, empty mind, forgetfulness, forgetting to take painkillers, n = 12).
None of the patient variables was related either to the number of spontaneous cognitive symptoms or to the score of the 43-item questionnaire (p not significant).
Factor analysis with varimax rotation retained four factors explaining 70.6% of the observed variance and allowed item reduction to a none-item questionnaire (Tables 1 and 2). The reduced nine-item Mig-SCog had a completion time of around 1 minute. The average score was 8.63 ± 4.04, ranging from 0 to 18 (Figure 1). The score of the reduced nine-item the Mig-SCog was not influenced by gender (p = 0.16), presence of aura symptoms (p = 0.54), current migraine prophylaxis (p = 0.43), age (p = 0.63), disease duration (p = 0.78), attack frequency (p = 0.10), duration (p = 0.44) and intensity (p = 0.98), but was associated with the total number of spontaneous cognitive complaints (p < 0.0001) and to lower literacy (p = 0.009).
Total score of the reduced nine-item Mig-SCog questionnaire. The Mig-SCog (English translation) Four factors of the Mig-SCog
Cronbach’s alpha of the reduced Mig-SCog was 0.82 (lower bound of the one-sided 95% CI ≥ 0.77). The median inter-item correlation coefficient was 0.34 (range 0.05–0.56), the median item-test ρ was 0.68 (range 0.46–0.73) and the median item-rest ρ was 0.56 (range 0.33–0.62). The median item-rest Cronbach’s alpha was 0.80 (range 0.79–0.82).
The median inter-dimension ρ was 0.49 (range 0.36–0.60). Dimension-total Cronbach’s alpha was 0.79 and correlation coefficients ranged from 0.71 to 0.84. The dimension-to-own alpha ranged from 0.70 to 0.79, and the correlation coefficients ranged from 0.50 to 0.70.
Construct validity of the reduced Mig-SCog scores was confirmed by the high correlations to the 43-item draft questionnaire scoring (Spearman’s ρ 0.69, p < 0.001) and to the total score of the cognitive failures questionnaire (33) (Spearman’s ρ 0.61, p < 0.01).
Test-retest reliability of the nine-item Mig-SCog revealed a highly significant (p < 0.0001) Cohen’s kappa for each item, ranging from 0.58 to 0.76 (respectively 0.58 for items 1, 2 and 5, 0.62 for item 9, 0.66 for item 8, 0.71 for item 7, 0.74 for items 4 and 6 and 0.76 for item 3). Cohen’s kappa for the nine-item Mig-SCog was 0.55 (p = 0.001) and Cronbach’s alpha ranged from 0.84 to 0.88, from first to second application of the questionnaire. Correlation of total scores in both applications was high (Spearman’s ρ 0.91, p < 0.01).
A significant difference was shown between ictal and interictal scores in the subsample of 33 patients (8.0 ± 4.3 vs 1.5 ± 2.7, respectively, p < 0.0001).
Discussion
Cognitive symptoms occur during migraine attacks and are often not reversed by effective pain treatment (28) or contemplated as a valid endpoint in acute treatment migraine trials (35). Cognitive symptoms during attacks contribute substantially to reduced ability to carry out activities (19) and therefore to migraine burden, yet no instrument exists to identify and quantify this type of symptom.
The use of detailed neuropsychological testing during migraine attacks is not practical and currently available generalist testing instruments (36) are long, unspecific and impractical for everyday clinical practice. Available subjective scales for cognitive symptoms are very often related to progressive neurological disorders and are developed to predict cognitive decline (37–41).
We aimed to develop an instrument that would be patient-centred and disease-related. It also should be self-administered, fast to apply, easily understandable, requiring a minimal literacy, and cross-cultural. Although subjective, it should allow quantification and be versatile, being used either in relation to the usual headache pattern or to a specific migraine attack.
As no such instrument existed previously, the main methodological difficulty of this study was item generation and selection. Item generation was clinical-based, relying on patients’ self-report of cognitive symptoms during migraine attacks, identified by open-ended questions in a structured interview. Items generated were complemented by a panel of headache experts after relevant literature review (42). An extensive item list was then produced, to ensure that infrequent yet possibly relevant symptoms were contemplated. After analysis for language adequacy and item comprehension by a focus group, the extensive questionnaire was applied, yet still including an initial open-ended question, to ensure no relevant cognitive domain had been missed and to access construct validity. This effort to be as comprehensive as possible had the objective of not missing potentially relevant items to migraine. No effort was made to include cognitive symptoms that were infrequent in migraine or to have the same representativity of each domain by defining the number of items allowed in each domain. It was expected that only a few cognitive domains would prevail, if the questionnaire was to be universally accepted by migraine patients. As it is a patient-based questionnaire, the domains identified were expected to most probably represent the main practical cognitive difficulties present in everyday life chores and not necessarily the cognitive functions that were demonstrated to be impaired in cognitive testing during attacks (20–23).
Item reduction was first accessed by the ability of items to generate responses: Items with non-response rates of 10% or higher were eliminated. Exploratory factor analysis was used to identify conflicting or confounding items that could be attributed to several cognitive domains – these were systematically eliminated. The purpose was to obtain a clear and short questionnaire that would perform adequately on internal consistency, concurrent validity and test–retest reliability. As the process went along, qualitative analysis of the spontaneous cognitive symptoms reinforced its construct validity.
The final nine-item Mig-SCog questionnaire is simple, reliable and internally consistent and it has good temporal stability. Its performance is in line with an existing cognitive functions questionnaire (33), a good measure of everyday cognitive difficulties for young adulthood that has good correlation with laboratory evaluation (43). Mig-SCog reflects only two cognitive domains – Executive functions (attention/processing speed/orientation/planning) and Language (naming and language), that are the most frequent spontaneous complaints of patients in everyday life and some of the range of executive defects identified in objective testing of migraine patients during attacks (20–23). This is not the same as stating that these are the only symptoms expected to occur during migraine attacks, but that these are the most probable to be consistently reported by patients and, therefore, to be representative of their most troublesome cognitive symptoms. The Mig-SCog rating is significantly higher when patients refer to attacks compared to the interictal period, suggesting that cognitive symptoms during attacks are beyond everyday cognitive difficulties. As an example, low processing speed and difficulties in planning may explain a common clinical observation that patients within a migraine attack are often unable to get around to take their acute medication, a fact that has important implications in efficacy of pain relief and attack impact (44,45).
Our results show that the Mig-SCog questionnaire is a new working instrument that is versatile and may be applicable, in the future, both in clinical practice and in research settings, without significantly increasing patient evaluation time. Its scoring was only influenced by literacy, despite literacy having no influence on the number of spontaneous cognitive symptoms reported. This suggests that less literate individuals tend to overrate the frequency of cognitive symptoms during attacks, probably because of a baseline lower cognitive performance (46,47).
The Mig-SCog score was independent of any other clinical variable, leading to speculation that the expression of cognitive symptoms is attack-related and not disease-related, much like the clinical expression of all other symptoms of a migraine attack.
It is recognized by the authors that more work needs to be done on validation of this instrument to other languages, such as English, and also to study its performance in cohorts of patients in different clinical settings and in other headache types. Limitations of this study are acknowledged, namely the possible effect of the recall bias, as the date of the last attack was not sought out. However, because neither the frequency nor the intensity of attacks was related to higher scoring on the Mig-SCog we think this bias is unlikely to be present. Another limitation is the absence of data regarding the type of migraine preventives used in this sample, as well as other relevant psychoactive drugs that could influence the occurrence of cognitive symptoms. The use of preventives was not related to any specific complaint, scale item or total score of the Mig-SCog.
Sometimes the obvious needs to be stated: (i) Migraine acute therapy is not only about pain control. The ideal migraine drug must also contemplate non-pain-related symptoms that contribute significantly to disability, so instruments that identify and quantify these symptoms are essential to improve migraine treatment strategies; (2) Migraine-related cognitive symptoms during attacks are real and disabling. Researchers need an instrument to evaluate the contribution of cognitive symptoms to impairment during attacks and physicians need a fast quantifiable report of cognitive symptoms by their patients, to redefine treatment strategies. Mig-SCog also offers patients an easy and quantifiable way of measuring ill-characterized, difficult to express disabling symptoms to their attending physicians.
The Mig-SCog could be the first practical contribution to allow the valorization of cognitive symptoms of migraine patients during attacks. Hopefully, the importance and impact of these symptoms will be recognized in future guidelines of migraine trials, either as a valid endpoint of acute treatment efficacy (35) or as part of the assessment of a migraine’s impact.
Footnotes
Acknowledgements
The preliminary study mentioned in the Methods section was awarded the 3rd Tecnifar – Portuguese Headache Society award in 2001.
Funding
This study was funded by the Tecnifar – Portuguese Headache Society Headache Grant.