EQ-5D™-derived utility values for different levels of migraine severity from a UK sample of migraineurs

Migraine sufferers were recruited through advertisements posted on migraine support group websites in the UK, and by using support groups’ databases. Researchers trained in study procedures screened 166 interested participants via the telephone using a structured clinician-validated screening form, which included study eligibility criteria and, to confirm a diagnosis of migraine (with or without aura), International Headache Society Criteria [14]. Participants were enrolled in the study if they were over 18 years of age, had at least one migraine in the past seven days, and had a history of physician-diagnosed migraine headaches for at least six months. Migraine sufferers with visual or cognitive impairments, as assessed by the researcher and those experiencing residual migraine symptoms from the most recent migraine were excluded. Tension-related component of the headache was not evaluated.

Descriptive statistics were calculated for sociodemographic and MIDAS data. All EQ-5D data were summarised and grouped according to the migraine state (mild pain, moderate pain, severe pain, and current health [without migraine]), with basic statistics given for the continuous data (number, mean, and SD). Utility values were calculated according to standard procedures, and using the York preference tariff [6].

Pearson's correlation coefficient was calculated to assess whether there was a relationship between migraine-related disability and utility using the participant's MIDAS score and his/her utility for the current health (without migraine) state.

Utility values were compared between severity levels for participants who had non-missing values for both data points in the comparison of interest. Paired t-tests were used to compare mean utility values for mild, moderate, severe, and inter-ictal periods against the inter-ictal values to evaluate whether there were significant differences in mean patient utility by migraine severity. In addition, if the EQ-5D data were not normally distributed, non-parametric analyses (Wilcoxon signed test) were planned.

It was estimated that a sample size of 100 would be sufficient for estimating significant differences in utility scores between four migraine states at a significance level of p < 0.05 and with 90% power to detect significant differences.

Participants were required to report on at least one of the four health states being evaluated (current health [without migraine], mild migraine pain, moderate migraine pain, or severe migraine pain) to be eligible for analysis. In the event of missing responses within a questionnaire, guidelines from instrument developers were used to handle missing data where available [15, 17].

Participant characteristics are shown in Table 1. A total of 106 migraine sufferers were recruited. The mean age of our sample was 47.4 years, with 76% female. The mean length of time since initial migraine diagnosis was 19.2 years. Over half of our sample (52.9%) was prescribed medication to treat migraine. Eighty-four participants (79.2%) reported at least one comorbidity; the number of comorbidities per participant ranged from 1-7. Of the comorbidities listed in the sociodemographic form, the most frequently experienced were depression (33%), followed by anxiety (31%). Congestive heart failure was reported the least frequently (0.9%). Over a third of the sample (34.9%) reported “other” comorbidities that were not otherwise listed.

Almost half of the participants (51 out of 106) could be categorised as suffering the highest level (Grade IV) of migraine-related disability, based on their MIDAS scores in the past three months. The number of participants that could be categorised as Grades I, II, and III was 9, 14, and 32, respectively. The mean (SD) number of lost days (due to either missed work or impaired productivity) in the last three months due to headache for Grades I, II, III, and IV of disability was 2.78 (2.44), 7.79 (1.53), 15.31 (2.72), and 48.00 (29.55), respectively.

Significant negative correlations between current health (without migraine) utility and MIDAS scores were found (r = -0.29, p < 0.003); as such, higher MIDAS scores (more disability due to migraine) were associated with lower current health (without migraine) utility values. Correlations were not very high, but were statistically significant at the 1% level.

The mean utility of current health (without migraine) in this UK study (0.87) was similar to the EQ-5D utilities elicited in a study of US migraineurs (sampled either outside of or during an attack) estimated at 0.82 by Luo et al. (2009) [13]. Statistically significant differences (p < 0.0001) were also found between 1.0 (no problems on any EQ-5D dimension) and each level of migraine pain severity, and also between current health (without migraine) and every level of migraine pain severity (p < 0.0001)—indicating that migraine significantly and negatively affects participants’ utility.

We also found significant negative correlations between EQ-5D utility for current health (without migraine) and MIDAS scores. This suggests that patients’ experience of migraine pain-related disability may significantly and negatively affect their utility, with greater migraine-related disability associated with lower health utilities.

We found that differences in utility for moderate and severe migraine pain and current health without migraine (0.34 and 1.05, respectively) were greater than those found for a decrease in health utility following hip fracture when compared to pre-fracture, as measured using the EQ-5D (0.12-0.20) [18, 19], and utility decrements of 0.13 and 0.14 for rheumatoid arthriti s and hand osteoarthritis, respectively, as measured using the SF-6D [20]. Xu et al. (2010) [12] reported similar findings.

This study further demonstrated that the levels of severity of migraine examined here are tangible: patients were able to value different levels of migraine pain severity of a recent migraine attack.

While this study aimed to examine the utility values only for migraine pain, participants may have considered other aspects of their migraine headaches, such as nausea and sensitivity to light and sound. However, previous studies have generally found a positive association between the severity of migraine pain and the accompanying symptoms of migraine [21, 22], and the migraine-specific triptan class of medications has been shown to relieve these symptoms along with reducing the severity of migraine pain [23]. Thus, the extent and directionality of any bias in the practical application of these utility values to therapeutic interventions for migraine is unclear, to the degree that treatment benefits also encompass these symptoms.

The results of our study suggest that having a migraine results in a significantly reduced utility (health state preference), with more severe migraine pain resulting in lower utility. Other studies have yielded similar results. Xu et al. (2010) [12] also found that EQ-5D-derived utility worsened with increasing migraine pain severity. However, unlike Xu et al. (2010) [12] who found the disutility of mild and moderate migraine to be relatively similar in magnitude (0.14 vs. 0.18), our study indicates a statistically significant difference of nominally greater magnitude between the disutility for mild (0.20) and moderate (0.34) migraine. Our analysis also found that, while eliminating a migraine yields the greatest improvement in health utility, the difference between moderate and mild severity in our analysis was significant, suggesting that reducing migraine severity to a mild would still be of value to migraine sufferers. These different results may be due to differences in study design.

Disutility values relative to current health without migraine estimated from this study for mild, moderate, and severe levels of migraine pain were higher (0.21, 0.34, and 1.07, respectively) than those estimated by Xu et al. (2010) [12]: 0.14, 0.18, and 0.49, respectively. This may also be a result of the methods used to obtain utilities in the respective studies. Patients in Xu et al.’s (2010) [12] study were evaluated while experiencing moderate/severe migraine pain at the baseline of the clinical trial, within 24 hours of first treating a migraine attack, and 24 hours later when the participants were pain free. By contrast, patients in our study were evaluated within seven days of a migraine attack, while they experienced no residual symptoms. Also, Xu et al. (2010) [12] applied the scoring algorithm for the US population [24], whereas the present study used the UK population scoring algorithm. The same health state tends to be valued more severely in the UK. For instance, the worst health state was scored as -0.11 using the US algorithm, which was higher than the UK valuation for severe migraine pain reported in this study (-0.20). These observed differences may also be a result of the context within which utilities were elicited (clinical trial vs. naturalistic, observational study) and differences in sample characteristics; for example, a higher percentage of females and participants of white or Asian origin, and a lower percentage of participants of black origin and no participants of Hispanic origin in our sample, as compared with Xu et al. 2010 [12].

The reported utility value for severe migraine that is "worse than dead" (which is in the range of the EQ-5D utility scale) is based on time-tradeoff utilities for members of the UK general public who evaluated profiles for the underlying EQ-5D health states corresponding to those selected by the migraine patients in our study when they considered the severe level of pain of their recent migraine. In ascribing a utility value of worse than dead, the members of the UK general public provided valuations for states regarded as worse than dead by making a choice between dying immediately or spending a length of time (10-x) in the target state, followed by x years in the 11111 state (no health problems on any domain) and finding the point of indifference. The utility score was given by the formula -x/(10-x) [6]. Thus, these states were regarded by members of the UK general public as being less desirable than a loss of life for that time span.

Another analysis eliciting migraine-related utilities from a US adult patient sample with 4-week recall (potentially covering time with and without migraine) using the HUI3 found migraine severity (self-assessed on a scale of 1 to 10) was not a significant predictor of utility [11]. However, authors note that this lack of a relationship between migraine-related pain and utility may have been due to the low variation in pain seen in the dataset.

Some limitations should be noted with respect to our analysis. Deciding on the optimal recall period in which a participant can and should reflect on their experiences is complex; and there is a large amount of literature which demonstrates that memory bias (such as confounding or over/under estimating experience) is inherent in the task of recalling complex information retrospectively [25, 26]. This may be particularly true for the recall of pain [27]. Some design features of the current study (e.g., subjects were not primed to report on dates and hours of their recent migraine) may have resulted in such memory bias, and therefore error.

We reported data for a convenience sample of migraineurs who were willing to participate in the study. Data on geographical distribution was not collected. It was not a random sample and may not be representative of the UK migraine population as a whole. The mean age and gender distribution of our sample appears similar to that reported from a national sample of English migraineurs aged 16-65; however, the monthly attack frequency within our study population was higher (five vs. two attacks per month), along with the proportion of non-Caucasians [28]. We were also not able to obtain EQ-5D-derived utilities during a migraine attack. We collected this data retrospectively within a week of participants’ most recent migraine, and it is possible that the number of migraines experienced over the retrospective period would have influenced the recall of migraine pain severity. For example, recall may be different in participants who suffer many migraine attacks in a given four-week period, compared with those who experience relatively few. Bias may exist in participants’ recall of migraine-related disability and the effect of their migraine pain on their HRQL (as more recent experiences are likely to be recalled differently than those happening further in the past). It may also be important to assess intra-patient mean differences rather than group mean differences, where the latter may not enable inference of the population as whole.

Utility weights for EQ-5D index health state profiles have been previously estimated via the TTO method based on a 10-year timeframe [6], while individual migraines are of shorter duration (4-72 hours). However, as noted by Xu and colleagues [12], when modelled over a patient’s lifetime, the cumulative duration of episodes of migraine is likely to be within the range of length of many other conditions which are readily valued using the EQ-5D (e.g., treatment for various cancers, osteoporotic fracture).

Finally, it is also important to note that this study does not directly examine the utility associated with improved migraine management through treatment, and if used in economic analyses of migraine therapies, the assumption would be that any utility benefits observed are generalisable to those obtained with a treatment that reduces the number or severity of migraine hours experienced.