Association Between Electronic Diary–Rated Sleep, Mood, Energy, and Stress With Incident Headache in a Community-Based Sample

Abstract

Background and Objectives

The aim of this study was to examine the diurnal links between average and changes in average levels of prospectively rated mood, sleep, energy, and stress as predictors of incident headache in a community-based sample.

Methods

This observational study included structured clinical diagnostic assessment of both headache syndromes and mental disorders and electronic diaries that were administered 4 times per day for 2 weeks yielding a total of 4,974 assessments. The chief outcomes were incident morning (am) and later-day (pm) headaches. Generalized linear mixed-effects models were used to evaluate the average and lagged values of predictors including subjectively rated mood, anxiety, energy, stress, and sleep quality and objectively measured sleep duration and efficiency on incident am and pm headaches.

Results

The sample included 477 participants (61% female), aged 7 through 84 years. After adjusting for demographic and clinical covariates and emotional states, incident am headache was associated with lower average (ß = −0.206*; confidence intervals: −0.397 to −0.017) and a decrease in average sleep quality on the prior day (ß = −0.172*; confidence interval: −0.305, −0.039). Average stress and changes in subjective energy levels on the prior day were associated with incident headaches but with different valence for am (decrease) (ß = −0.145* confidence interval: −0.286, −0.005) and pm (increase) (ß = 0.157*; confidence interval: 0.032, 0.281) headache. Mood and anxiety disorders were not significantly associated with incident headache after controlling for history of a diagnosis of migraine.

Discussion

Both persistent and acute changes in arousal states manifest by subjective sleep quality and energy are salient precursors of incident headaches. Whereas poorer sleep quality and decreased energy on the prior day were associated with incident morning headache, an increase in energy and greater average stress were associated with headache onsets later in the day. Different patterns of predictors of morning and later-day incident headache highlight the role of circadian rhythms in the manifestations of headache. These findings may provide insight into the pathophysiologic processes underlying migraine and inform clinical intervention and prevention. Tracking these systems in real time with mobile technology provides a valuable ancillary tool to traditional clinical assessments.

Background

Migraine headaches are highly prevalent across the life span and are often associated with significant functional impairment,¹ yet migraine remains underdiagnosed and untreated worldwide. Even when diagnosed and treated, controlled studies of pharmacologic agents for acute and prophylactic treatment have generally not resulted in more than a 50% reduction in headaches.² More accurate prediction could, therefore, advance our ability to prevent migraine attacks and associated impairments.

One of the major challenges to effective treatment of migraine is pervasive comorbidity with a range of disorders, particularly cardiovascular disease³ and mood disorders.⁴ However, there is limited understanding of potential explanations regarding how mood disorders and symptoms involving sleep, mood, and energy may contribute to the dynamic manifestations of migraine. Although the association between sleep disturbances and migraine has been well-recognized, the mechanisms involved are complex. Disturbed sleep could be a manifestation of mood disorders or symptoms or a reflection of underlying dysregulation of homeostatic and autonomic systems that are more broadly related to migraine susceptibility.^5,6

Electronic diaries are a powerful approach to characterize the dynamics of prodromal manifestations and/or triggers of acute headache attacks.^7,8 Daily diaries that have been routinely used to collect real-time data on the precursors and sequelae of headaches in both clinical practice and empirical research have identified a wide range of triggers of headache attacks including fatigue,^9,10 stress,^10,11 altered sleep patterns and sleep deprivation,^12,13 dietary factors,¹⁴ and menstrual cycle fluctuations.¹⁵ Several studies have also examined circadian variation in migraine, with convergence of evidence that migraine attacks are more likely to occur in the morning.^11,12,16

Despite the compelling evidence from previous studies that have used prospective diaries to characterize the predictors, features, and postdrome of headaches,^9,10,13,17 there are a number of methodologic issues that limit the generalizability and clinical implications of this work. First, most of the samples have been conducted in specialty headache clinics or nonsystematic samples that may not generalize to migraine in primary care or the general population. Second, most prior studies were based exclusively on female participants and did not collect data across the full age spectrum. Third, non-migraine control groups that are critical for studying the specificity of factors associated with migraine have not been included in most prior diary research on headaches. Finally, many studies include only one assessment per day that is subject to retrospective reporting bias.

In this study, we evaluated the real-time associations between incident headaches and sleep, mood, and energy in people with a lifetime diagnosis of migraine and/or mood disorders using combined actigraphy and ecological momentary assessment 4 times per day for 2 weeks. We used generalized linear mixed-effects models that included both the average level and within-subject day-specific changes in sleep, emotional, and energy states and stress across two-week epochs. The objectives of this study were as follows:

1. To evaluate whether the average levels of anxiety, depression, energy, stress, and sleep quality reported with electronic diaries across 2 weeks are associated with incident headache;

2. To examine whether changes in the average levels of anxiety, depression, energy, stress, and sleep quality are associated with incident headache on the following day; and

3. To assess whether the predictors of incident headache differ for AM and PM onset and by demographic and clinical history of migraine or mood and anxiety disorders.

Methods

Sample

Participants in this cross-sectional study were enrolled in the National Institute of Mental Health (NIMH) Family Study of Affective Spectrum Disorders, a large community-based controlled family study with participants recruited from a screening of the greater Washington, DC metropolitan area and other local sources through the NIH Clinical Center general volunteer referral core (see Table 1, Results). More details of the family study methods are presented elsewhere.¹⁸

Table 1.

Demographic and Clinical History of Participants With and Without Lifetime Migraine Diagnosis (n = 477; 4,974^b)

Variables	Males (n = 186)			Females (n = 291)
	No migraine (n = 127)	Migraine (n = 59)	p Value	No migraine (n = 130)	Migraine (n = 161)	p Value
Age group, n (%)			0.017			0.6
<45	65 (51)	42 (71)		76 (58)	87 (54)
45–59	28 (22)	11 (19)		34 (26))	43 (27)
>= 60	34 (27)	6(10)		20 (15)	31 (19)
Race and ethnicity, n (%)
White	108 (85)	51 (86)	0.81	99 (76)	130 (81)	0.342
Non-White	19 (15)	8 (14)		31 (24)	31 (19)
American Indian or Alaska Native	0 (0)	0 (0)		1 (1)	2 (1)
Asian	3 (2)	2 (3.5)		0 (0)	0 (0)
Black or African American	9 (7)	3 (5)		19 (15)	11 (1)
Hispanic	0 (0)	0 (0)		1 (1)	3 (2)
More than one Race	4 (3)	1 (2)		3 (2)	3 (2)
Unknown/Other	3 (2)	2 (3.5)		7 (5)	12 (7)
Mood/anxiety, n (%)
Mood and/or anxiety disorders (lifetime)	73 (57)	42 (71)	0.044	86 (66)	147 (91)	<0.001
Current mood or anxiety disorders	55 (43)	37 (63)	0.010	69 (53)	125 (78)	<0.001
Migraine characteristics, n (%)
Current Migraine	0 (0)	45 (76)	<0.001	0 (0)	138 (86)	<0.001
Presence of any AM headache	31 (24)	26 (44)	0.005	51 (39)	103 (64)	<0.001
Presence of any PM headache	43 (34)	34 (58)	0.002	70 (54)	120 (75)	<0.001
Subjective emotional states
Average (happy to sad mood) (1–7)	2.61 (0.89)	2.95 (0.96)	0.11	2.60 (0.88)	2.79 (0.83)	0.03
Average (calm/anxious mood) (1–7)	2.31 (0.87)	2.58 (1.04)	0.14	2.27 (0.93)	2.53 (0.87)	0.01
Average energy (lethargic to energetic) (1–7)	3.98 (0.77)	3.69 (0.72)	0.044	3.83 (0.80)	3.48 (0.82)	0.002
Average stress level (relaxed to highly stressed (1–7)a	2.76 (0.98)	2.94 (1.02)	0.2	2.75 (1.07)	3.01 (0.94)	0.05
Sleep measures
Average sleep duration (actigraphy)	7.28 (1.17)	7.58 (1.25)	0.3	7.50 (1.42)	7.74 (0.98)	0.15
Average sleep midpoint (actigraphy)	3:50 am	3:51 am	0.8	3:49 am	3:42 am	0.90
Average sleep quality (EMA) (1–7)	4.39 (1.19)	4.16 (1.06)	0.2	4.46 (1.20)	3.96 (1.22)	0.002

^an = 389.

^bn of assessments.

Standard Protocol Approvals, Registrations, and Patient Consents

Ethics

This research was approved by the Combined Neuroscience IRB at the NIH (NIH; protocol no. 03-M-0211).

Informed Consent

All participants provided written informed consent for the protocol.

Procedures

Diagnoses

DSM-IV mood and anxiety disorders were assessed by comprehensive diagnostic assessment with the NIMH Family Study Diagnostic Interview for Affective Spectrum Disorders (DIAS).¹⁸ A total of 11.8%, 13.6%, and 38.6% of those with migraine and 10.5%, 8.2%, and 24.9% of controls had a lifetime history of bipolar I, bipolar II, and MDD, respectively. The lifetime rates of anxiety disorders were 21.8% for panic or Generalized Anxiety Disorder, 7.7% for social phobia or agoraphobia, and 13.8% for both panic/Generalized Anxiety Disorder and phobias. There was also substantial comorbidity between mood and anxiety disorders. We controlled for both mood and anxiety disorders in our analyses, and the only significant associations with incident headaches were found in the univariate models. ICHD-III headache subtypes were assessed with the NIMH Diagnostic Interview of Headache Syndromes (DIHS).¹⁹ Both the DIAS and DIHS were administered by experienced clinical interviewers and reviewed by clinical experts. Systematic evaluation of the reliability of the DIHS compared with clinical expert diagnostics yielded an area under the curve of 0.86, with a sensitivity of 91.3% and specificity of 81.0%.

Ecological Momentary Assessment (EMA)

EMA procedures were used from 2008 to 2019 to collect a comprehensive set of domains including the Mood Circumplex, Context, Life Events, Sleep, Physical Activity, Eating/Drinking, Pain, and Headache. Variables included in these analyses were collected from the mood circumplex that includes analog ratings of multiple domains underlying emotional states as defined by Larsen and Diener.²⁰ Participants were asked to rate their current emotional states 4x/day on a Likert scale from 1 to 7 for Anxiousness [(1) very calm to (7) very anxious]; Mood [(1) very happy to (7) very sad]; and Energy [(1) very tired to (7) very energetic]. Participants rated sleep quality for the previous night [(1) not at all rested to (7) fully rested] at the first assessment of the day. A subset of the sample that only was assessed in an earlier version of the mobile app, Perceived Stress [(1) relaxed, calm to (7) highly stressed] was rated once per day at the end of the day. The Headache Module of the NIMH EMA application has now been incorporated on the Open Source Mindlogger Platform.²¹

Throughout the study, participants were prompted to complete an EMA assessment 4 times per day over a two-week period. A fixed schedule was used for each participant with the timing of the interviews occurring within a sampling window ranging from 7:00 am to 10:00 pm, with a minimum spacing of one hour between any 2 assessments and an average delay of 4 hours between assessments. For example, scheduling could be at 7:15 am, 11:30 am, 3:45 pm, and 7:30 pm. For each participant, sampling schedules were adjusted to accommodate their typical sleep and wake schedules so as not to modify usual daily life activities. Responses to EMA assessments that were made 45 minutes after the signal or beyond were considered as missing data. The NIMH EMA Headache Module is now available on the Mindlogger Platform.²¹ Additional details of EMA procedures have been described previously.^22-24

Actigraphy-Estimated Sleep

Participants wore actigraphy monitors (Respironics and Actiwatch Score; Philips Respironics) on their nondominant wrist for 2 weeks to collect data on their minute-by-minute activity counts. From the actigraphy data, we derived sleep duration and sleep midpoint (sleep and wake-up times). Wake and sleep periods were defined by a default threshold of less than 40 counts per period, for more than 10 minutes.²⁵

Definitions of Incident Headache

The presence of a headache was collected at each of the 4 daily assessments across the 56 epochs. If the participant endorsed the presence of a headache, questions on the symptoms, severity, timing, and medication use were then collected. “Incident headache” was defined as the occurrence of a new-onset headache that was not present at the 2 previous assessments. Incident morning headache was rated positive if no headache was reported at the last assessment of the previous day. Incident headaches were then divided into morning (am) headache at the first assessment of the day and afternoon/evening (pm) headache for those that were reported at the 3 subsequent assessments.

Statistical Analyses

A series of generalized linear mixed models assessed the association between AM and PM headache incidence with history of migraine, history of mood or anxiety disorders, and EMA ratings of sleep quality at the first assessment, and self-perceived stress once/day, and the mood circumplex (sad mood, energy, and anxiety) ratings 4 times/day. All models were adjusted for age in years, sex, and weekend vs weekdays and associations with a history of diagnoses of migraine, mood, and anxiety disorders. Interaction terms were included in the models to examine whether the association between subjectively rated emotional states and sleep variables with incident headache differed between people with lifetime diagnoses of either migraine and/or mood/anxiety disorders. Separate models were estimated for each EMA variable of interest to avoid potential multicollinearity issues followed by multivariate models that included significant effects from the univariate models. Nonstatistically significant terms were removed from the final models. Objective measures of sleep midpoint and duration were not associated with incident AM or PM headache and were, therefore, not included in the models presented below. Results are reported from models which showed statistically significant within (changes from usual) and between (subject average) effects. As described in the eAppendix 1 (links.lww.com/WNL/D387), this decomposition approach that incorporates simultaneous effects of both within and between-day effects has only rarely been used in EMA research. Point estimates, p values, and 95% confidence intervals are reported for all the selected models. Confidence intervals were constructed using Wald's method. Statistical significance was assessed using a threshold of 0.05.

Missingness was handled at the data processing phase. We had excellent compliance with more than 80% of the sample completing the full two-week period of EMA assessments. Systematic analysis of missing data revealed that youth were more likely to miss the second assessment of the day during weekdays and adults more likely to miss weekend assessments. However, there were no differences in the proportion of missing data among cases than controls. We, therefore, assumed that the missing data were missing at random.

Data Availability

Deidentified participant data and R analytic programs will be shared on request for research purposes through request to the study PI.

Results

Demographic and Clinical Correlates and Outcomes

There were a total of 477 participants and 4,974 assessments. A summary of the sample characteristics, including age; diagnoses; and averages and standard deviations of subjective ratings of mood, anxiety, energy, sleep quality, and stress stratified by sex and a lifetime history of migraine is presented in Table 1. The sample consisted of 161 participants (61% female) older than 50 years and 316 participants (61% female) 50 years or younger. Approximately 84% of the sample considered themselves as White or Hispanic, 8% Black, 2% Asian, 2% American Indian, and 4.2% of multiple races and ethnicities. A greater proportion of participants with migraine were female, and male participants with a history of migraine tended to be younger than those without migraine. Both lifetime history and current (past year) diagnosis of mood and anxiety disorders were more common in those with a history of migraine. Most of the sample with lifetime migraine continued to have migraine attacks during the past year.

During the two-week assessment period both male and female participants with a history of migraine were more likely to experience at least 1 AM and 1 PM headache attack as compared with those without a history of a diagnosis of migraine. Both male and female participants with a history of migraine reported significantly lower levels of subjective energy. Female participants had significantly higher levels of both sad and anxious moods across the study duration. Average sleep duration and sleep midpoint did not differ between those with and without migraine, but female participants with migraine had poorer sleep quality than controls across the assessment period.

Prediction of Incident Headache

The results of the generalized linear mixed-effects models are presented in Tables 2 and 3. There were no significant interactions between lifetime diagnoses of migraine or mood/anxiety disorders (Aim 3). Although people with migraine tended to have higher levels of sadness, anxiety, and stress levels in univariate models, their influence on headache incidence was not statistically significant in the multivariate models that included energy and sleep quality (Aim 1). The Figure summarizes the results of both the univariate and multivariate models of the effects of the subjective emotional states, stress, and sleep quality on incident AM and PM headache.

Table 2.

Predictors of Incident AM Headaches

Models	Model 1a	Model 2a	Model 3a	Model 4a
Variables	Demographic and Clinical	Energy	Sleep quality	Energy and sleep quality
(Intercept)	−4.537***,a,b (−5.189 to −3.886)	−3.033*** (−4.084 to −1.983)	−3.436*** (−4.269 to −2.604)	−2.968*** (−4.016 to −1.920)
N's and demographics
Sex (Female vs Male)	0.358 (−0.026 to 0.741)	0.273 (−0.108 to 0.653)	0.332 (−0.048 to 0.711)	0.293 (−0.088 to 0.675)
Age	0.006 (−0.003 to 0.015)	0.009 (−0.000 to 0.018)	0.012* (0.002 to 0.021)	0.012* (0.002 to 0.21)
Diagnosis
Migraine Dx	0.955 *** (0.594 to 1.317)	0.833*** (0.471 to 1.195)	0.860 *** (0.499 to 1.220)	0.823*** (0.461 to 1.185)
Mood/Anxiety Disorder	0.439 (−0.003 to 0.881)	0.358 (−0.081 to 0.797)	0.346 (−0.093 to 0.785)	0.333 (−0.106 to 0.772)
Energy
Average		−0.392** (−0.622 to −0.163)		−0.209 (−0.492 to 0.074)
Scaled Dev prior day		−0.146 * (−0.287 to −0.006)		−0.145* (−0.286 to −0.005)
Subjective sleep quality
Average			−0.290*** (−0.443 to −0.136)	−0.206* (−0.397 to −0.017)
Scaled Dev from Average			−0.172* (−0.305 to −0.039)	−0.172* (−0.305 to −0.039)

Model parameters
AIC	2,319.768	2,308.341	2,303.465	2,301.341
BIC	2,365.352	2,366.949	2,362.072	2,372.973
R-square (fixed)	0.082	0.099	0.105	0.110
R-square (total)	0.340	0.340	0.345	0.346

^ap Value; 95% CI.

^b(p Value: ***p < 0.001; **p < 0.01; *p < 0.05).

Table 3.

Predictors of Incident PM Headache

Model	Model 1b	Model 2b	Model 3b	Model 4b	Model 5b
Variables	Demographic/Clinical	Energy	Sleep Quality	Energy and Sleep Quality	Energy and Stress
(Intercept)	−3.019***a,b (−3.503 to −2.534)	−2.000*** (−2.867 to −1.133)	−2.431*** (−3.110 to −1.752)	−1.974 *** (−2.842 to −1.107)	−2.409*** (−3.490 to −1.329)
Demographics
Sex (Female)	0.668 *** (0.343 to 0.993)	0.615 *** (0.291 to 0.94)	0.66 *** (0.337 to 0.983)	0.625 *** (0.300 to 0.950)	0.450* (0.104 to 0.796)
Age	−0.017 *** (−0.025 to −0.009)	−0.015 *** (−0.023 to −0.007)	−0.014 *** (−0.022 to −0.006)	−0.014 *** (−0.022 to −0.006)	−0.018*** (−0.026 to −0.009)
Disorders
Migraine Dx	0.658*** (0.355 to 0.961)	0.576*** (0.270 to 0.881)	0.602 *** (0.298 to 0.910)	0.570*** (0.264 to 0.876)	0.555*** (0.228 to 0.883)
Mood/Anxiety Disorder	0.325 (−0.041 to 0.691)	0.274 (−0.091 to 0.640)	0.290 (−0.076 to 0.655)	0.270 (−0.096 to 0.635)	0.260 (−0.156 to 0.675)
Subjective states
Average energy		−0.263** (−0.544 to −0.073)		−0.198 (−0.436 to 0.040)	−0.245* (−0.459 to −0.032)
Scaled deviation from Ave		0.156* (0.031 to 0.281)		0.157* (0.032 to 0.281)	0.109 (−0.036 to 0.255)
Sleep quality
Average			−0.153* (−0.282 to −0.025)	−0.073 (−0.233 to 0.088)
Scaled dev from Ave			−0.036 (−0.153 to 0.082)	−0.036 (−0.154 to 0.081)
Stressc
Average					0.243** (0.081,0.406)
Scaled dev from Ave					0.037 (−0.100 to 0.173)

Model parameters
AIC	2,814.310	2,804.753	2,812.461	2,807.597	2,120.906
BIC	2,859.894	2,863.361	2,871.068	2,879.228	2,188.901
R-square (fixed)	0.097	0.107	0.103	0.109	0.134
R-square (total)	0.291	0.294	0.293	0.295	0.273

^ap Value; 95% CI; ^b(p value: ***p < 0.001; **p < 0.01; *p < 0.05).

^cSubsample.

Figure. Predictors of Incident Headache.

AM Headache

Table 2 shows the results of the predictors of incident AM headache (models 1a–4a). In the univariate models, increased average levels of sad mood and anxiety and decreased levels of average energy and sleep quality were significantly associated with am headache (Table 2). Lower average energy (p < 0.001) and an acute decrease in average energy on the previous day were associated with an increased likelihood of incident am headache (p < 0.001, Table 2: model 2). Lower average sleep quality (p < 0.001) and poorer than average sleep quality on the previous night (p < 0.01) were associated with an increased likelihood of am headache. When energy and sleep quality were included in the same model, the effect of average energy was no longer significant, but the effect of lower energy level on the previous day persisted (models 2a vs 4a, respectively). These results suggest that lower sleep quality on average and poorer than average subjectively rated sleep quality on the prior night were associated with a greater likelihood of incident am headaches. These associations persisted after adjusting for individuals' energy, although the estimated association between average sleep quality and likelihood of incident am headache was attenuated ($\widehat{\beta }=$−0.290 vs −0.207 in models 3a and 4a, respectively) (Aims 2, 3).

PM Headache

Table 3 (models 1b–4b) summarizes the associations between energy and sleep quality with incident pm headache compared with am headache. All the EMA ratings including higher sad and anxious moods, greater stress, lower energy, and poorer sleep quality were associated with pm headache. In the multivariate models, higher than average energy level on the prior day was associated with an increase in incident pm headache (p = 0.014, model 2b). In addition, lower average sleep quality, but not a decline in average sleep quality, on the prior night was significantly associated with an increased likelihood of pm headache (model 3b). However, this effect was no longer significant after adjusting for the energy levels (model 4b). Therefore, the only significant predictor of pm headache was higher than average energy level on the prior day. In the subsample of participants who were administered the dimensional stress question, there was a significant association between average self-perceived stress and PM headache (p < 0.01). However acute prior day changes in average stress levels were not associated with incident pm headache (Aim 2, 3).

Discussion

This study advances our understanding of the short-term dynamics of migraine through prospective assessment of the influence of within and between-day fluctuations of energy, mood, stress, and sleep on incident headache. The most salient finding is the prognostic role of prior day changes in energy and sleep quality beyond usual levels on incident morning headache attacks. Differences in the demographic and clinical correlates of migraine between morning and later-day incident headache highlight the role of circadian rhythms in the manifestations of headache attacks.²⁶ Despite the well-established links between mood disorders and migraine,^27,28 we show that neither lifetime depressive disorder nor higher average or acute changes in depressive or anxiety symptoms were associated with incident headache attacks. The unique application of electronic diaries in a controlled community-based sample of people with migraine also enhanced our ability to examine specificity of predictors among those with a history of migraine.

The significance of both average and change in perceived sleep quality as a predictor of incident am headache highlights the central role of sleep in the pathogenesis of migraine as documented by numerous prior controlled studies of associations between headaches and disturbed sleep patterns and disorders.^29-32 However, the lack of association between actigraphically derived sleep measures and incident migraine, as shown previously,¹³ indicates that objective sleep characteristics may not be the primary source of decreased energy and poor perceived sleep quality. There are several possible explanations for the association between incident morning headache with poorer perceived sleep quality including (1) sleep quality could reflect a disruption in neural processes involved in impending headache that may lead to disrupted sleep; (2) psychological processes that interfere with sleep quality may be a direct trigger of the migraine attack; (3) sleep difficulties may be a residual influence of earlier episodes or their longer term effects^5,6; or (4) sleep quality may be a correlate of another primary headache predictor such as acute stress, diet, or exposure to sensory stimuli that may disturb sleep. Ironically, because sleep is also one of the most effective interventions for migraine attacks,³³ these potential explanations warrant further investigation.

The associations between sleep and migraine should also be interpreted in the context of the strong associations between subjective energy and incident headache. The influence of an acute reduction in energy, but not average energy level, suggests that the direct influence of sleep quality on headache may be a consequence of internal changes reflected in subjective energy on the day preceding headache. The role of changes in prior day energy, but different valence for AM and PM headache, would support the central role of arousal-sleep systems in the pathogenesis of headache.^34,35 In fact, tiredness and fatigue have been the most potent reported predictors of acute headache attacks in several prior diary studies of people with migraine.^10,12,36,37

The differences in predictors of morning and later-day headache suggest that timing of onset of headache may be an important source of heterogeneity in their etiology that could reflect physiologic systems involved in differential manifestation of subtypes of headache. The greater incidence of morning headaches among those with migraine compared with controls confirms prior retrospective and prospective studies of morning onset as a characteristic of migraine in clinical samples of people with headache.^12,16,26,29 The association with sleep quality further supports the occurrence of changes in neural systems that emerge during sleep. Although the morning onset of migraine parallels that of several other episodic disorders such as myocardial infarction, stroke, and asthma attacks, the potential mechanisms have not been elucidated. By contrast, the association between higher levels of stress and increased energy with pm headache suggests that later-day onset of headache could be synonymous with “tension-type” headache exacerbated by higher levels of arousal and reactivity to daily events. The short-term dynamics of the underlying physiologic systems involved in migraine and other acute health events such as cortisol, catecholamines, and platelet aggregability and related systems have not been systematically interrogated in the diurnal variation in migraine.³⁸ Future studies that measure dynamics of these biologic correlates as a function of changes in energy and sleep may advance our understanding of the pathophysiology of migraine.

This work contributes to our understanding of both the stable and proximal predictors of headache. First, this study provides systematic evaluation of incident headache in people with comprehensive diagnostic interviews of both headache syndromes and mood and anxiety disorders. This enabled us to further test whether predictors of incident migraine differ in those with comorbid depression and anxiety compared with migraine alone. Second, the large heterogeneous community-based sample that spans a broad age range provided information on incident headache both in people with a history of migraine and controls that have not been included in prior prospective diary studies of migraine. Although we did not specifically compare those with migraine and tension-type headache, prior studies have shown that some of the same factors that were associated with incident headache extend to prediction of tension-type headache, thereby suggesting a lack of specificity with respect to migraine.³⁹ Third, this is one of the only prospective diary studies that included a large sample of male participants. However, although female participants had greater rates of migraine than male participants, we did not find sex differences in predictors of acute migraine attacks. Fourth, in contrast to prior studies that were almost exclusively based on one assessment per day that is subject to substantial retrospective bias,⁴⁰ this study systematically included multiple assessments within each day with a validated set of EMA modules that has been administered in several other large population-based and clinical samples. Fifth, our study included both objective and subjective measures of sleep characteristics that have only been included in one prior electronic diary study of a clinical sample of migraine.^6,13 Sixth, few prior studies have included systematic tracking of the stability, fluctuations, and inter-relationships of several emotional states, including mood, energy, and anxiety, which are well-established correlates of sleep patterns.¹⁶ Finally, we used a novel statistical method that simultaneously adjusted for fixed covariates such as sex, age, and diagnoses; average and acute changes in emotional states; and both objective and subjective sleep measures. Very few of the prior prospective diary studies jointly considered both between and within-subject effects of predictors of incident headache.^11,13

There are also several methodologic limitations of this work that should be considered in the interpretation of these findings. First, this study reflected a relatively brief period of observation that only provides a cross-sectional snapshot of migraine that occurs over longer periods. Previous studies of clinical samples of migraine tended to track precursors over longer periods ranging from one to three months. Although such longer periods of assessment are far more feasible in people undergoing treatment of migraine, the frequency of missing data that ranged from 25% to 70% in prior studies is an important trade-off that may also seriously jeopardize the generalizability and validity of the findings. Second, our study did not reflect the economic and ethnic/racial diversity of the broad geographic region. Efforts to expand our sample are underway. Third, although we controlled for ongoing medication use in the sample, we could not examine the associations between specific medications and incident headache. Few prior studies of migraine predictors in specialty headache clinics considered whether ongoing treatments may have masked the influence of naturalistic precursors of headache. Fourth, we could not study the full range of headache predictors because of insufficient power for simultaneous multidomain models. Fifth, the dimensional measure of stress was only collected in a subsample, so the results may not apply to the full sample. Future analyses of these data and other studies in our consortium will test the generalizability of these findings and their bidirectional influences, longer cross-lagged epochs, as well as physiologic, dietary, social contextual, and stressful events that were measured 4 times per day.

These intriguing findings suggest that both the longer term and short-term evaluation of the dynamic inter-relationships between the rhythms of energy, sleep, and headaches should be a priority for future research. There is a need for systematic studies to determine whether clinical interventions that target sleep and low energy or their physiologic correlates may reduce the incidence, progression, and severity of migraine attacks.⁴¹ Greater harmonization of ecologic research on migraine that uses common measures, systematic samples, and analytic methods could address the widespread variability in findings regarding proximal predictors of migraine. The growing sophistication of ambulatory tools to track clinical and environmental factors associated with headaches in daily life will transform our ability to derive individual profiles that characterize circadian dynamics of headaches that can inform both the clinical management and prevention of headache.

Glossary

DIAS

Diagnostic Interview for Affective Spectrum

DIHS

Diagnostic Interview of Headache Syndromes

EMA

ecological momentary assessment

Appendix. Authors

Name	Location	Contribution
Tarannum M. Lateef, MD, MPH	Children’s National Health System, Pediatric Specialists of Virginia, and George Washington University of Medicine; Intramural Research Program, Section on Developmental Genetic Epidemiology, National Institute of Mental Health	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data
Debangan Dey, PhD	Intramural Research Program, Section on Developmental Genetic Epidemiology, National Institute of Mental Health	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Andrew Leroux, PhD	Department of Biostatistics and Informatics, University of Colorado School of Public Health	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Lihong Cui, MSc	Intramural Research Program, Section on Developmental Genetic Epidemiology, National Institute of Mental Health	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Mike Xiao, BS	Child Mind Institute, New York	Major role in the acquisition of data; study concept or design; analysis or interpretation of data
Vadim Zipunnikov, PhD	Department of Biostatistics, Johns Hopkins Bloomberg School of Public Health	Drafting/revision of the manuscript for content, including statistical writing for content; analysis or interpretation of data
Kathleen R. Merikangas, PhD	Intramural Research Program, Section on Developmental Genetic Epidemiology, National Institute of Mental Health, Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data

Study Funding

This work was supported by the Intramural Research Program of the National Institute of Mental Health. This work was supported by the National Institute of Mental Health of the National Institutes of Health (grant numbers 1ZIAMH002804 and 1ZIAMH002954). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Dr. Dey's contribution to this work was supported by the Joint Training Program of the Johns Hopkins Department of Biostatistics and the NIMH Intramural Research Program.

Disclosure

The authors report no relevant disclosures. Go to Neurology.org/N for full disclosures.