We Analyzed 12,847 Migraine Attacks: Here Are the 7 Triggers Nobody Talks About

Most published lists of migraine triggers were assembled from small clinical trials, patient self-reports, and survey data collected months or years after the fact. Memory is unreliable, surveys are sparse, and most studies track fewer than a dozen factors simultaneously. We had the opportunity to do something different.

Over 18 months, Haven users generated 12,847 documented migraine attacks across 2,341 active users who tracked at least 30 consecutive days. With their anonymized, aggregated data, we ran a cross-factor Relative Risk analysis across 37 environmental, behavioral, and physiological variables. Here is what we found — and what the standard trigger lists have been getting wrong.

Our Methodology

Every attack was logged with at minimum: onset time, pain intensity (1-10), duration, accompanying symptoms, and rescue medication used. Alongside attack data, users completed a daily check-in capturing sleep timing and quality, hydration, caffeine and alcohol intake, stress level, screen time, meal timing, exercise, and menstrual cycle phase. Weather data — including barometric pressure, humidity, temperature, and 6-hour and 24-hour deltas — was automatically pulled from meteorological APIs based on each user's location.

For each factor, we calculated Relative Risk (RR) — the ratio of migraine occurrence when that factor is present versus when it is absent. We excluded users with fewer than 60 days of consistent tracking, which filtered the cohort to 2,341 users with high-quality data. Only factors reaching statistical significance (p < 0.05) across at least 30% of the cohort are reported here.

Relative Risk > 1.5 means a factor is consistently associated with higher migraine occurrence. An RR of 2.0 means twice the risk. All seven triggers reported here reached RR > 1.4 in the aggregate cohort.

The 7 Triggers

1. Barometric Pressure Drops — 68% Correlation

Barometric pressure has appeared in migraine research before, but most studies report it as a binary: did pressure drop or not? Our data is more specific. A drop of more than 5 hPa within a 6-hour window produced a 68% higher attack rate across the cohort compared to days with stable pressure. Smaller drops (1-3 hPa) showed no significant association. The threshold matters — and most clinical studies have missed it by averaging pressure changes over 24-hour windows.

What this means for you: it is not just low-pressure weather that is the issue — it is rapid low-pressure weather. A slow-moving front is far less triggering than a fast-moving one. Monitoring 6-hour pressure deltas, not daily pressure readings, gives you the actionable warning window.

2. Sleep Timing Shifts — 73% Correlation

Sleep duration gets nearly all the attention in sleep-migraine research. Our data tells a different story. Users who slept their usual duration but shifted their bedtime by more than 90 minutes from their personal average showed a 73% higher attack rate the following day. Duration-matched nights with consistent timing showed no significant association. This pattern was symmetric — sleeping early and sleeping late both triggered at equivalent rates.

What this means for you: the weekend sleep-in habit is a genuine risk factor even if you are technically getting enough total sleep. Your circadian rhythm appears to be a stronger predictor than sleep quantity alone. Anchor your wake time, not just your total hours.

3. Meal Gap Duration — 61% Correlation

Skipped meals appear on every trigger list. What that framing misses is the mechanism. Our data shows it is not which meal is skipped — it is the duration of the fasting gap. Gaps exceeding 7 hours between any two eating events were associated with a 61% higher attack rate. Gaps of 5-6 hours showed no significant association. Skipping breakfast but eating within 5 hours of waking was far less risky than eating dinner at 6pm and not eating again until 11am the next day.

What this means for you: the 7-hour threshold is more useful than any meal label. Track the time between your last bite and your next, not whether you had a conventional breakfast, lunch, or dinner.

4. Hormonal Phase Transitions — 78% Correlation

Among female users who tracked menstrual cycle data, attacks concentrated overwhelmingly at one specific transition: the 48-hour window straddling the end of the late luteal phase and the beginning of the menstrual phase. This transition — characterized by the steepest drop in estrogen and progesterone — produced a 78% higher attack rate than all other cycle phases combined. Mid-cycle attacks (around ovulation) appeared in a subset of users but at lower magnitude.

What this means for you: if you menstruate, knowing the precise cycle day matters far more than knowing the phase name. Tracking day-by-day rather than by loose phase (follicular, luteal) gives you a 48-hour predictive window you can use to preemptively optimize other factors.

5. Cumulative Screen Fatigue — 54% Correlation

Total daily screen time showed no significant correlation with migraine attacks in our data. What did show correlation was consecutive screen exposure without a break. Users logging more than 6 continuous hours of screen time (a single session, or two sessions with less than 20 minutes between them) had a 54% higher attack rate the same day or the following morning. Equivalent total screen time spread across shorter sessions was not significant.

What this means for you: it is not how much you look at a screen — it is how long you look without a break. The 20-20-20 rule (every 20 minutes, look at something 20 feet away for 20 seconds) likely matters more than total screen time restrictions.

6. Combined Dehydration and Caffeine — 47% Correlation

This is the finding that surprised our team most. Dehydration alone (logged hydration below 60% of personal target) showed a weak and non-significant association with attacks. Caffeine intake above personal average alone was similarly non-significant. But when both occurred on the same day — lower-than-usual water intake and higher-than-usual caffeine — the combined factor showed a 47% correlation with attacks within 18 hours.

What this means for you: caffeine is a diuretic that compounds dehydration. On high-caffeine days, hydration becomes more critical, not less. The synergistic risk is not additive — it appears to be multiplicative.

7. Humidity-Temperature Swing Combination — 52% Correlation

High humidity alone and large temperature swings alone both showed modest but sub-threshold associations in our data. The combination, however, cleared the significance threshold: days where humidity exceeded 80% AND temperature changed by more than 8°C within a 24-hour period showed a 52% higher attack rate. Neither factor alone reached 20% correlation. This compound weather pattern correlates with rapidly moving thunderstorm systems — the kind of weather that often coincides with multiple physiological stressors.

What this means for you: checking the humidity forecast alongside the temperature swing forecast gives you a more complete weather-risk picture than either data point alone. The combination is a red-flag signal worth tracking.

The Trigger That Surprised Us

Exercise. Or rather, the absence of it as a trigger. Popular migraine discourse frequently lists intense exercise as a trigger, and clinical studies have reported exercise-induced migraine in a subset of patients. In our cohort, exercise — at any logged intensity — showed only a 12% correlation with attacks. More notably, users who exercised regularly had a lower baseline attack frequency than sedentary users, after controlling for other factors.

The 12% correlation was driven almost entirely by users who rarely exercised and then logged a single intense session. In regular exercisers, exercise showed no significant trigger effect at all. The likely explanation: deconditioning creates vulnerability to exercise-induced physiological stress that regular training eliminates.

Why Cross-Factor Analysis Matters

The most significant finding in our study is not any individual trigger — it is the interaction effect. Users who experienced two or more high-risk factors on the same day had attack rates 2.3 times higher than users experiencing a single high-risk factor. The threshold model of migraine — where multiple small inputs combine to cross a susceptibility threshold — is strongly supported by this data.

This means that single-variable trigger avoidance is incomplete advice. Telling someone to avoid barometric pressure drops (which they cannot control anyway) without addressing the modifiable factors that could lower their threshold on a high-risk weather day misses the point entirely. The actionable insight is: on days when one uncontrollable trigger is present, aggressively manage all the controllable ones.