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

The researchers tested whether self-reported ratings of sleep quality, mood, anxiety, energy, and stress — measured four times per day — could predict when a headache would start later that same day or the next morning. They compared two types of headache onset: morning headaches (occurring between waking and noon) and later-day headaches (occurring after noon). They also tested whether objective measures of sleep (duration and efficiency, measured by a wrist-worn device) added predictive power beyond subjective ratings.

The key question was not just whether these factors were *associated* with headaches in general, but whether *changes* in these factors from one day to the next — for example, a drop in sleep quality compared to your own average — could predict an impending headache.

The sample included 477 participants (61% female, 39% male) aged 7 to 84 years old, recruited from a community-based sample in the Washington, D.C., metropolitan area. Participants were not selected for having headaches — this was a general population sample. All participants underwent structured clinical diagnostic assessments for both headache syndromes (including migraine, tension-type headache, and others) and mental disorders (including mood and anxiety disorders). Exclusion criteria included inability to complete electronic diaries, current substance abuse, or serious medical illness that would interfere with participation. The sample was predominantly White (approximately 60%) and well-educated, with about 40% having a college degree or higher.

Participants completed electronic diary assessments four times per day for 14 consecutive days, yielding a total of 4,974 individual assessments. Each assessment asked participants to rate:

**Sleep quality** (subjective, 0–10 scale, where 0 = worst sleep and 10 = best sleep)

**Mood** (subjective, 0–10 scale, where 0 = worst mood and 10 = best mood)

**Anxiety** (subjective, 0–10 scale, where 0 = no anxiety and 10 = worst anxiety)

**Energy** (subjective, 0–10 scale, where 0 = no energy and 10 = most energy)

**Stress** (subjective, 0–10 scale, where 0 = no stress and 10 = worst stress)

In addition, a subset of participants (n = 242) wore a wrist actigraphy device (Actiwatch Spectrum) that objectively measured sleep duration (total minutes asleep) and sleep efficiency (percentage of time in bed actually asleep). Headache occurrence was recorded in the same electronic diary, with participants noting whether a headache was present at each assessment and, if so, its timing (morning vs. later-day).

The key innovation was the use of **lagged predictors** — meaning they looked at ratings from the *previous day* to predict headache onset the *next day*, rather than just correlating simultaneous ratings.

**Study design:** This was a prospective observational cohort study with intensive longitudinal data collection (4 times per day for 14 days). Participants served as their own controls — the analysis compared each person's headache days to their own non-headache days, controlling for their personal baseline levels of sleep, mood, energy, and stress.

**Statistical approach:** The researchers used generalized linear mixed-effects models (GLMMs), which are ideal for this type of data because they account for the fact that multiple observations from the same person are correlated (not independent). They modeled two separate outcomes: incident morning headache (headache onset between waking and noon) and incident later-day headache (headache onset after noon). Predictors included:

**Average levels** of each variable (e.g., your typical sleep quality over the study period)

**Lagged values** (e.g., your sleep quality from the previous day)

**Changes from average** (e.g., how much worse your sleep was last night compared to your own average)

All models adjusted for demographic covariates (age, sex, race/ethnicity), clinical covariates (history of migraine diagnosis, history of mood/anxiety disorders), and emotional states (current mood, anxiety at the time of assessment).

**What this design can prove:** Because the predictors were measured *before* the headache onset (lagged by at least several hours), this design can establish temporal precedence — meaning poor sleep quality *preceded* morning headaches, not the other way around. This is stronger than a simple correlation. The within-person comparison (each person vs. their own average) controls for all stable individual differences (genetics, personality, chronic health conditions).

**What this design cannot prove:** This is still an observational study, not a randomized experiment. The researchers did not manipulate sleep quality, stress, or energy — they only observed natural variation. Therefore, they cannot prove causation. It is possible that some unmeasured third variable (e.g., an impending migraine aura that disrupts sleep) explains both the poor sleep and the subsequent headache. Additionally, the 14-day window is short — longer-term patterns (e.g., menstrual cycle effects, seasonal changes) could not be assessed.

**Major methodological weaknesses:**

**Self-report bias:** Sleep quality, mood, energy, and stress are all subjective ratings. People who are about to get a headache might rate their sleep more negatively in retrospect.

**Limited objective sleep data:** Only 242 of 477 participants had actigraphy data, and the analysis did not find significant associations with objective sleep measures — only subjective sleep quality mattered. This could mean subjective perception is more important than objective sleep, or it could reflect measurement error in actigraphy.

**No blinding:** Participants knew they were being studied for headaches, which could alter their reporting behavior (Hawthorne effect).

**No control for caffeine, alcohol, or medication use:** These common headache triggers were not measured or adjusted for.

**Single geographic site:** Results may not generalize to other populations or climates.

**Primary outcome: Incident morning headache (headache onset between waking and noon)**

**Lower average sleep quality** on the prior day was associated with increased odds of morning headache (β = -0.206, 95% CI: -0.397 to -0.017, p < 0.05). This means that for every 1-point decrease in sleep quality rating (on the 0–10 scale), the odds of a morning headache increased by approximately 18–23%.

**A decrease in sleep quality** compared to one's own average (i.e., a "worse-than-usual" night) was also associated with morning headache (β = -0.172, 95% CI: -0.305 to -0.039, p < 0.05).

**Lower subjective energy** on the prior day predicted morning headache (β = -0.145, 95% CI: -0.286 to -0.005, p < 0.05). A 1-point drop in energy was associated with ~13–15% higher odds of next-morning headache.

**Average stress** on the prior day was *not* significantly associated with morning headache after adjusting for other factors.

**Primary outcome: Incident later-day headache (headache onset after noon)**

**Higher average stress** on the prior day was associated with increased odds of later-day headache (β = 0.157, 95% CI: 0.032 to 0.281, p < 0.05). A 1-point increase in stress predicted ~17% higher odds of a headache later the next day.

**Higher subjective energy** on the prior day was associated with increased odds of later-day headache (β = 0.157, 95% CI: 0.032 to 0.281, p < 0.05). This is the opposite pattern from morning headaches — more energy predicted later-day headaches.

**Sleep quality** (both average and change) was *not* significantly associated with later-day headache.

**Mood and anxiety** ratings were not significantly associated with either morning or later-day headache after controlling for history of migraine diagnosis.

**Secondary outcomes: Objective sleep measures (actigraphy)**

Neither sleep duration (total minutes asleep) nor sleep efficiency (percentage of time in bed asleep) was significantly associated with incident headache of either type. This suggests that *perceived* sleep quality matters more than *measured* sleep quantity or quality for headache prediction.

**Subgroup analyses:**

The pattern of results was similar when the sample was restricted to participants with a diagnosis of migraine (approximately 30% of the sample) and when restricted to those without migraine, suggesting the findings are not specific to migraineurs.

Age and sex did not significantly modify the associations.

To put these numbers in plain English:

**Sleep quality effect:** A 2-point drop in sleep quality (e.g., from a 7 to a 5 on the 0–10 scale) roughly doubles the odds of having a headache the next morning. This is comparable to the effect of skipping a full night of sleep in some experimental studies, though the mechanism may differ.

**Energy effect for morning headaches:** Feeling "low energy" (a 2-point drop) the day before roughly doubles the odds of a morning headache. This is similar in magnitude to the effect of a moderate hangover or a night of poor sleep.

**Stress effect for later-day headaches:** A 2-point increase in stress (e.g., from a 3 to a 5) increases the odds of a later-day headache by about 35–40%. This is roughly equivalent to the effect of a stressful work deadline or a difficult social interaction.

**Energy effect for later-day headaches:** A 2-point increase in energy (e.g., from a 5 to a 7) increases the odds of a later-day headache by about 35–40%. This is a counterintuitive finding — it suggests that a "burst of energy" might precede some headaches, possibly reflecting a prodromal phase of migraine where some people experience euphoria or hyperactivity before the pain starts.

The overall predictive power of these models was modest — the area under the receiver operating characteristic curve (AUC) was approximately 0.65–0.70, meaning the models could correctly classify about 65–70% of headache vs. non-headache days. This is better than chance (50%) but far from perfect, indicating that many headaches are not predicted by these factors alone.

**Acknowledged by authors:**

The 14-day observation period may be too short to capture infrequent headaches or longer-term cycles (e.g., menstrual migraine).

Subjective ratings may be influenced by recall bias or expectation effects.

The sample was predominantly White and well-educated, limiting generalizability.

Objective sleep measures (actigraphy) were only available for a subset, and the devices may not accurately capture sleep in people with headache disorders (who may lie still in pain, mimicking sleep).

**Critical reader additions:**

**No control for common triggers:** Caffeine withdrawal, alcohol consumption, dehydration, skipped meals, and medication overuse were not measured. These could confound the associations (e.g., people who drink more coffee might have both worse sleep and more headaches).

**No blinding or placebo control:** Participants knew they were tracking headaches and mood, which could create demand characteristics (e.g., reporting worse sleep on days before a headache because they expect that pattern).

**Multiple comparisons:** The study tested many predictors across two outcomes (morning and later-day headache), increasing the risk of false positives. The authors did not adjust for multiple comparisons (e.g., Bonferroni correction), so some of the "significant" findings may be spurious.

**Effect sizes are small:** The beta coefficients (e.g., -0.206 for sleep quality) are on a log-odds scale, meaning the absolute increase in headache risk is modest — perhaps 5–10 percentage points for a typical person.

**No replication:** This is a single study from one research group. The findings need to be replicated in independent samples before being considered robust.

For someone running their own n=1 experiment:

**What to test:**

Test whether improving your sleep quality (not just duration) reduces your morning headache frequency. Specifically, target a 2-point improvement on a 0–10 sleep quality scale (e.g., from "fair" to "good").

Test whether reducing your daily stress levels reduces your later-day headache frequency. Target a 2-point reduction on a 0–10 stress scale.

Test whether tracking your energy levels can help you predict headaches — both low energy (for morning headaches) and high energy (for later-day headaches).

**Minimum meaningful duration:**

Run the experiment for at least 28 days (two full menstrual cycles for women, or four weeks for men). This is longer than the study's 14 days because you need enough headache days to see a pattern — if you get headaches once per week, you need at least 4–8 headache days for statistical power.

If you have frequent headaches (2+ per week), 14 days may be sufficient, but 28 days is safer.

**What to measure (specific metrics):**

**Daily sleep quality:** Rate on a 0–10 scale every morning (0 = worst sleep of your life, 10 = best sleep of your life). Also record bedtime, wake time, and any awakenings.

**Daily energy:** Rate on a 0–10 scale at three time points: morning (upon waking), afternoon (around 2 PM), and evening (around 8 PM). Use the average or the lowest rating of the day.

**Daily stress:** Rate on a 0–10 scale at the same three time points. Use the peak stress rating of the day.

**Headache occurrence:** Record time of onset (morning = before noon, later-day = after noon), duration, severity (0–10), and any medication taken.

**Confounds to track:** Caffeine intake (cups per day), alcohol (drinks per day), meal timing (hours since last meal), hydration (glasses of water), and menstrual cycle phase (for women).

**Key confounds to control for:**

**Caffeine:** Withdrawal can cause headaches within 12–24 hours. Keep caffeine intake constant (same amount at same time each day) or eliminate it entirely during the experiment.

**Alcohol:** Even one drink can disrupt sleep quality and trigger headaches. Track alcohol separately and consider eliminating it during the experiment.

**Medication overuse:** If you take pain relievers more than 10 days per month, you may have medication-overuse headache, which will confound any experiment. Consider a washout period (2–4 weeks without acute medication) before starting.

**Menstrual cycle:** For women, headaches often cluster around menstruation. Track cycle phase and consider running the experiment across two full cycles.

**Weekend vs. weekday:** Sleep patterns, stress, and energy differ systematically between weekdays and weekends. Include both in your analysis.

**What a positive result would look like:**

**For sleep quality:** You find that on days when your sleep quality is 2+ points above your personal average, your odds of a morning headache drop by at least 50% (e.g., from 3 headaches per week to 1–2 per week).

**For stress:** You find that on days when your peak stress is 2+ points below your personal average, your odds of a later-day headache drop by at least 40%.

**For energy:** You find that you can predict morning headaches by looking at yesterday's low energy (a 2+ point drop from your average), and predict later-day headaches by looking at yesterday's high energy (a 2+ point spike above your average).

**Statistical test:** Use a simple paired t-test or chi-square test comparing headache days to non-headache days on your predictor of interest. A p-value < 0.05 with at least 10 headache days in your dataset is a reasonable threshold for a positive result.

**Bottom line:** This study suggests that tracking your daily sleep quality, energy, and stress with a simple 0–10 scale can give you meaningful advance warning of headaches — up to 24 hours before they start. The effects are