The main findings in this blinded controlled study were that migraineurs reported more sleep related symptoms than controls while the corresponding differences in sleep diaries and PSG sleep quality variables were much smaller. Interestingly, migraineurs had lower PT and a strong tendency to more N3 sleep. We hypothesize that the latter findings are explained by an increased sleep pressure in the interictal phase.
Subjective and objective sleep quality in controls and interictal migraineurs
In accordance with other studies migraineurs presently report more subjective symptoms of anxiety [
19], insomnia [
20] and subjective tiredness [
21], than healthy subjects. However, in contrast to these clear differences in subjective symptoms the differences in objective sleep diary variables, daytime hypersomnia scales, and PSG sleep parameters between migraineurs and controls were smaller. A similar discrepancy has previously been found among chronic insomniacs [
22]. Insomnia is also a risk factor for migraine [
23]. Although patients with known and previously diagnosed sleep disorders were excluded from the present study we can hardly rule out that some migraineurs had an undetected insomnia. Indeed, the high PSQI suggest that some migraine patients may suffer from a sleep disorder. However, PSQI is a summary measure that incorporates insomnia, hypersomnia, snoring and other symptoms and can not be interpreted as a specific sign of insomnia alone. In this study we have focused on sleep differences between our groups, not individualized sleep diagnoses. Furthermore, we excluded those with a clear history of comorbid sleep disorders to characterize the migraine-related sleep dysfunction (including insomnia).
The more objective findings were somewhat inconclusive for insomnia as more migraineurs reported long sleep latencies and long awakenings while PSG sleep latency and total sleep times in both PSG and sleep diaries did not differ from controls.
Despite excessive daytime tiredness among migraineurs we found sleepiness (ESS) to be normal, in agreement with Seidel et al. [
24]. However, slightly elevated ESS abnormality rates (ESS cutoff ≥10) in migraine have been observed in a larger population study [
1], suggesting that migraine patients on the average achieve somewhat less sleep during nighttime than they actually need.
In the present study both controls and patients with interictal migraineurs had increasing amount of N1 sleep with increasing symptoms of anxiety. However, in the migraine group we also found increasing symptoms of insomnia and decreasing subjective sleep quality with increasing amount of fast arousals. We also found decreasing amount of SWS with increasing amount of anxiety in the interictal migraine group. Hence sleep quality seems to be more vulnerable (i.e. disturbed by anxiety or subjective insomnia) among migraineurs than healthy controls.
In PSG recordings we found almost significant (p=0.05), tendencies towards less fast arousals and more N3 sleep among migraineurs in the interictal phase than among controls. Because fast arousals occur more frequently in superficial than deep NREM sleep [
25] it is possible that less fast arousals partly is a passive reflection of more N3 sleep. More deep sleep is one factor that usually indicates better sleep quality [
26], but in migraineurs it was paradoxically accompanied by more awakenings. It is possible that awakenings increase sleep pressure and induce compensatory more deep sleep between awakenings. On the other hand, it seems illogic that nights with more awakenings is the explanation for more deep sleep because negative correlations were found between N3 sleep and both awakening index and fast arousals. Another explanation for the paradoxical findings could be subgroup differences. Our findings are only partly in line with findings of Karthik et al. [
27]. In that study thirty migraineurs without aura, recruited from a tertiary university hospital, were found to have longer latency to sleep onset, less NREM sleep, lower arousal-index, more awakenings, longer time in bed and more awake time compared to 32 controls. In PSG, we did not find longer latency to sleep onset or less NREM sleep among migraineurs, but we have included migraineurs from advertisement and not hospital patients. Furthermore, Karthik et al. [
27] did not say anything about blinding or the temporal relation to the attack.
De Gennaro et al. [
28] found reduced amount of fast arousals and increased SWS the night after sleep deprivation. Hence, our PSG findings could indicate sleep deprivation among the interictal migraineurs. However, according to the sleep diaries, sleep deprivation was probably not present in this study, unless one postulates too low power to detect small differences for sleep time in diaries or that migraineurs need more sleep than controls. Even though more SWS is found after sleep deprivation, Sforza et al. [
29] did not find more slow bursts while Nicholas et al. [
30] found increased numbers of K-complexes per minute in sleep stage 2 after sleep deprivation. A lower number of low frequency, high amplitude EEG bursts in NREM and a lower index of high frequency EEG arousals during REM sleep have previously been found among interictal migraineurs with attacks related to sleep compared to controls [
31]. These findings have been interpreted as hypofunction of the arousal system [
31], but we suggest that this may be related to a relative sleep deprivation and that migraineurs are relatively sleep deprived and need more sleep than healthy controls.
Differences in objective sleep among migraineurs in inter, pre and post ictal phases
A shorter latency to sleep onset in the preictal phase compared to interictal phase was the only significant phase difference in PSG variables we found among migraineurs. As far as we know, reduced sleep latency in the preictal phase has not been described before. Hence, preictal migraineurs seem to have greater sleep pressure that fits with premonitory symptoms reported in the preictal phase [
10] and daytime EEG changes in the preictal phase [
4]. We found no other significant differences in our PSG measures between migraineurs in different phases, probably due to too low power, because we captured only 9 patients in the preictal and 8 in the postictal phase.
Less arousals [
6] and more SWS (and also REM sleep) [
32] have been found during the night before an attack. We could not confirm these findings in our study. Göder et al. [
6] and Dexter [
32] included patients with morning- or sleep- related migraine attacks and compared intra individually nights before an attack with nights without ensuing attacks in order to improve the statistical power of their studies. In contrast we compared subgroups inter individually and we also included those with sleep recordings between 24 and 48 h before attack. Because only one of our nine preictal patients had attacks usually initiated by sleep, the study populations are not quite comparable, possibly explaining the discrepant results. As we did not find changes in sleep quality in the postictal group, our result do not suggest that the polysomnographic changes are caused by the attack or sleep disturbances that may accompany the attack in some patients.
Pain and sleep
We found decreased PT in the interictal migraine group 48 h from any attacks in accordance with Schwedt et al. [
33]. However, the latter studies did not exclude preictal migraine from their “interictal” groups. Previously, we did not find significant thermal PT differences between interictal migraine and controls, neither with 24 nor 72 h cutoffs [
5], possibly caused by lack of power. The same study showed decreased pain thresholds 24 h before the attack [
5] compared to the interictal period. However, the present study was not designed to re-address the latter question as longitudinal observations were not included. Allodynia during a migraine attack has also been described previously [
34].
As slow bursts are frequent before and during slow wave sleep, especially in the first sleep cycles [
35], these slow bursts are possibly related to sleep pressure or sleep need. Then, both increased N3 sleep and more slow arousals in a stable interictal phase could be assumed to be measures of increased sleep pressure. N3 sleep correlated negatively with PPT and slow burst correlated negatively with TPT in the present study, hence signs of increased sleep pressure seems to be associated with lower pain thresholds. The relationship between PSG signs of a tired (or “weary”) brain and PT does not seem unreasonable as increased pain sensitivity has been found among healthy volunteers after sleep deprivation [
8]. Lovati et al. [
36] also found in a questionnaire study that migraineurs with head allodynia during attack reported more symptoms of insomnia.
Apparently it seems inconsistent that HPT in the preictal migraine group was negatively correlated to superficial sleep, that is N1 (Figure
2), and not to N3 as in the interictal phase. The reason is unclear, but is should be noted that, compared to controls, interictal migraineurs both had increased SWS and reduced PT. As explained above, we apprehend increased SWS as a probable compensation for increased sleep pressure. Hence N3 could be a measure of both foregoing cerebral tiredness and the ability to compensate for it. As the needs accumulate, in our study indicated by preictally reduced latency to sleep onset in PSG, lack of compensation might induce, or contribute to a migraine attack. When maximum N3 sleep is achieved, any further need for repose could be reflected by the less restful N1 sleep.
Migraine and sleep
With respect to a hypothesized relative sleep deprivation it makes sense that migraineurs have increased symptoms of tiredness, have either normal EEGs or subtle findings that may reflect drowsiness [
37], and tend to have low PT. Even though small undetected differences can not be ruled out, sleep diaries revealed normal sleep times among migraineurs. But why should migraineurs need more rest than controls and how should lack of sleep initiate a migraine attack? More deep sleep may possibly be induced either by increased awake neuronal activity [
38] or an absolute sleep deprivation [
28]. Since the latter option apparently is not the case here (although a relative deprivation can be operative), increased neural activity might be the link between mental stress (both as state and trait factors (i.e. anxiety symptoms)) and the onset of a migraine attack [
39]. Partial correlation adjusted for age indicates that migraineurs have reduced SWS with increasing headache duration. If amount of SWS goes down to a “normal level” with increasing duration of the migraine, this might be a parallel phenomena to reduced migraine attack prevalence after menopause [
40].
Strengths and limitations
The strengths of this study are the blinded, controlled, and prospective design. The participants were mainly recruited by advertising in local newspapers, in contrast to a hospital-based migraine population which may include more severe and longstanding cases. A study design with repeated PSGs may be more powerful for the detection of phase-related differences. In addition, a so-called first-night effect should be considered [
41]. Even though a first-night effect has been shown to occur in inpatient PSGs, there are also findings indicating reverse first night effect for some subjects (i.e., increased sleep quality) [
42]. Even with two PSGs it is not possible to eliminate the first-night-like effects, because these might last more for than one night in some subjects [
41]. In addition, a single PSG design may also have some advantages because serial PSGs can be affected by individually variable order-effects [
43]. A slight underestimation of sleep problems, as compared to the general migraine population, is expected in our study because patients with any known and diagnosed previous sleep disorders were excluded, however, this exclusion is a strength regarding the major aims of the study. This study was exploratory and we did not adjust for multiple comparisons because we did not want to increase type II failures on the cost of reducing type I failures [
44,
45]. Another limitation is the rather low power for comparing interictal, pre- and postictal subgroups and the possibility of both type I and type II errors is acknowledged. Independent replication of our results is accordingly needed.