Making sense of your headache patterns — what AI sees that you can't

You've been tracking for a few weeks. You have entries — headaches with timing and severity, sleep logs, stress scores, meals, medications. And if you look at the data yourself, you might see some things: "I had a headache on Tuesday, and I also slept badly Monday night." That feels like a pattern.

But is it? You also slept badly on Thursday and didn't get a headache. And you had a headache on Saturday after sleeping well. So sleep alone isn't the story.

This is where most self-investigation stalls. You see a potential connection, test it informally, find exceptions, and conclude your headaches are unpredictable. They're almost certainly not. The issue is that you're looking for single-cause explanations in a multi-cause system.

Why your brain can't do this analysis

The threshold model means your attacks are driven by combinations, not individual triggers. To properly analyze your data, you'd need to simultaneously compare: the conditions present on every attack day, the conditions present on every non-attack day, every possible two-factor and three-factor combination, and the frequency with which each combination appears before attacks versus on non-attack days.

That's hundreds of comparisons. Research published in Headache found that multi-factor predictive models consistently outperformed single-factor models in predicting individual migraine attacks — but building those models requires computational analysis that human working memory simply can't do.

You also have confirmation bias working against you. You remember the headache that followed red wine. You don't remember the four times you drank red wine and felt fine. AI counts both equally.

What the analysis actually examines

When you ask Iris to analyze your headache patterns, the Data Analyst runs a structured comparison across your entire dataset. Here's what that involves:

Individual variable screening. For each tracked factor — sleep quality, stress level, specific foods, caffeine, weather changes, menstrual cycle phase, medication timing — the analysis calculates the attack rate when that factor is present versus absent. This identifies which individual variables have any statistical relationship to your attacks, and filters out the ones that don't.

Combination analysis. This is where the real signal lives. For every pair and triple of variables that showed individual correlation, the analysis calculates the attack rate when they co-occur. A single factor might only slightly increase your risk, but two or three factors together might be highly predictive. Research in Cephalalgia found that trigger combinations were consistently more predictive than any single trigger alone.

Confounding check. High-stress days might also tend to be poor-sleep days. If two variables always co-occur in your data, it's hard to know which one actually matters — or whether both do. The analysis needs to identify whether you have enough instances of one without the other to distinguish their independent effects. If not, it flags that as an open question requiring more targeted tracking.

Timing analysis. When does the trigger appear relative to the attack? A food eaten 30 minutes before a headache has a different implication than a food eaten 18 hours before. Sleep disruption from the previous night is a different signal than stress from three days ago. The timing patterns help distinguish direct triggers from background sensitizers.

Protective factors. Not everything makes headaches more likely. Some factors make them less likely. Regular exercise, consistent sleep schedules, or specific dietary patterns might show up as protective in your data. These are as valuable to identify as triggers — they're part of managing your threshold.

What the results mean

The output is a ranked list of your strongest trigger combinations, with numbers. Something like: "The combination of fewer than 6 hours of sleep plus skipping a meal appeared before 7 of your 9 tracked attacks. Poor sleep alone appeared on 12 non-attack days. Skipped meals alone appeared on 8 non-attack days. The combination has a much higher attack rate than either factor individually."

Some common patterns that emerge:

Sleep + stress is the dominant combination. This is the most commonly identified combination in headache research and in practice. If your data shows this, the intervention strategy focuses on sleep hygiene and stress management as the highest-leverage changes.

Hormonal + any other factor. If attacks cluster around menstruation or ovulation, hormonal timing sets the baseline and other triggers push you over the threshold during vulnerable windows. This means different management strategies at different points in your cycle.

Stress transition patterns. Not just high stress, but the drop after high stress — the let-down effect. If your worst attacks follow stress resolution rather than peak stress, that changes your approach: gradual de-escalation matters more than stress reduction.

Medication timing patterns. If attacks cluster around specific medication schedules or consistently appear on days when you take acute medication, medication overuse headache needs to be considered. This is one of the most actionable findings because it has a clear (if uncomfortable) treatment path.

No clear pattern yet. This happens. It usually means one of three things: not enough data (keep tracking), a missing variable (something you're not tracking is important), or your threshold is highly variable (which itself is useful information that points toward nervous system sensitization as the primary issue). Ambiguity after a thorough analysis isn't failure — it narrows the investigation.

What to do with findings

The analysis shifts your investigation from pattern-finding to hypothesis testing. Take your strongest combination and test it.

If your top combination is poor sleep plus skipped meals, spend 3-4 weeks prioritizing sleep consistency and meal regularity while tracking everything else normally. If attacks decrease, those factors are confirmed as major drivers. If attacks stay the same, either the combination isn't causal (correlation isn't always causation) or a third variable you haven't isolated is involved.

This iterative approach — identify, test, confirm, refine — is how you move from "I have unpredictable headaches" to "I know what drives my attacks and which factors I can manage."

And "manage" is the operative word. You probably won't eliminate every trigger. But if you can lower your baseline threshold by addressing two or three key factors, you make the remaining triggers less likely to push you over the edge. That's how the threshold model works in your favor.

References

1. Migraine trigger interaction and the threshold theory — The Journal of Headache and Pain, 2018. Multi-factor threshold model evidence.
2. Perceived trigger factors of migraine: a comprehensive review — Cephalalgia, 2019. Combinations outperforming individual triggers.
3. Predicting individual migraine attacks using machine learning — Headache, 2021. Multi-factor predictive models for migraine.
4. Stress and the onset of migraine attacks — Neurology, 2014. Let-down effect and stress transition triggers.
5. Medication overuse headache: a critical review — The Journal of Headache and Pain, 2014. Medication patterns in chronic headache.