336,951 hours of headache

4,375 Japanese people logged their headaches hour by hour for a whole year. What the machine found in the data wasn't what people believed.

336,951. That's how many times a group of people logged their headache and the weather around it, hour by hour, over the course of a whole year.

4,375 users of a Japanese headache app were behind the number. From December 2020 to November 2021, they logged whether they had a headache, while an artificial intelligence linked each entry to weather data for that exact moment. 89.2% were women, and the average age was 34.

What the machine found that you wouldn't have found yourself

Try the task without a computer. Think back over the last year, and try to work out: was it the air pressure right now that triggered the headache? Was it the humidity? Was it what the pressure had done six hours earlier? Was it the pattern in the pressure over the past six days?

The Japanese study could ask all these questions at once, because it had 336,951 data points to work with instead of a single human brain's memory of a year. The answer wasn't one number, but a whole list: lower air pressure, higher humidity, more rainfall, a marked pressure drop six hours earlier, higher morning pressure, lower pressure the next day — and two specific patterns in the pressure over six days. All of them were statistically significantly linked to headache, in both of the statistical models the researchers used.

The fact that something holds up in two different models on the same dataset is an extra layer of assurance. It's also a luxury that requires a certain number of entries to even build into the analysis at all — you need enough data to split it and test it twice, and still find the same thing.

No single person could have worked this out by thinking back. There are too many overlapping variables, and too many days where nothing in particular happened, for a memory to keep track of them.

The cost of the 336,951 hours

Data of that kind costs something to build. Not money — people. Another study, this time from the UK, shows how hard it actually is to get people to keep logging.

The app in that study was downloaded by 13,207 users over a year of recruitment, from all 124 UK postcode areas. 10,584 got as far as filling in baseline data and at least one entry. But only 6,850 — 65% — were still in it after the first week. After a month it was 4,692, or 44%.

The participants were predominantly women, 81%, with an average age of 49, and most already had a chronic pain diagnosis: arthritis, rheumatoid arthritis, or fibromyalgia. That's exactly the kind of people you most need to keep hold of over time — and exactly the kind of study that shows "holding on" is the hard part.

More than half of those who started had stopped within a month. That's not a one-off for this particular study. It's the reality every attempt to collect weather and pain data over time runs into: people log eagerly at the start and drop off as the novelty wears off.

Why it makes a difference for you specifically

The 336,951 hours tell you what holds true across 4,375 people. They don't tell you what holds true for you. The bigger the dataset, the clearer the pattern becomes for the group — and the more it can still conceal about the individual.

But without that scale, no one could have seen the pattern at all. It's not something a single headache diary could have revealed.