How 14 Days of Symptom Logging Unlocks Hidden Triggers
You feel wrecked on Tuesday and fine on Thursday, and nothing in your routine explains the difference. Sleep was the same. Lunch was the same. Work was the same amount of annoying.
The reasonable next step is to write it down. That instinct is right, and a symptom log is one of the few genuinely useful things you can do about a vague recurring problem. What follows is how to keep one that will still look sound in three months, rather than one that produces a satisfying answer you later have to abandon.
The good days are the data
Most people record health information only when something is wrong. That produces a list of bad days, and a list of bad days cannot show you a pattern.
Here is why, plainly. Suppose you log five headaches, and four of them fall on days when PM2.5 was elevated. That looks convincing until you ask the question the log can’t answer: how many elevated-PM2.5 days did you have without a headache? If the answer is thirty, your pattern is meaningless. If the answer is one, you have something.
The comparison is the whole thing. Days when you felt fine are not empty rows in the log, they are the control group, and they carry as much information as the bad ones. This is what the “I’m fine now” button is for and it is the single most important habit to build. One tap on a good day is worth more than a detailed entry on a bad one.
Log before you look
This one is specific to using an app that shows you the air quality.
Research on environmental symptom reporting is consistent and slightly uncomfortable. In a population study combining modelled PM10 levels with survey responses, perceived pollution and health risk perception predicted symptoms in their own right, alongside the modelled exposure. Experimental work goes further: people report symptoms during sham exposures they believe are real, and the belief that an exposure is hazardous generates more symptoms than the belief that the same exposure is benign.
None of that means air pollution doesn’t affect you. Everything in this blog about particulates and ozone is real physiology. It means that if you read “PM2.5: 48 µg/m³, unhealthy” and then rate your fatigue, you have contaminated your own data, and you will find the correlation you went looking for whether or not it exists.
The fix costs nothing:
Rate how you feel first. Check today’s conditions after.
Log your entry from the notification without reading the day’s numbers, then look at the air. Over a few weeks this is the difference between a record that could persuade a sceptical doctor and one that only persuades you.
If you have already been logging the other way round, don’t discard the data. Just start a clean stretch and treat the earlier entries as less reliable.
How long it actually takes
Two weeks is the right first commitment and the wrong place to draw conclusions.
In single-person research designs, statistical power comes from the number of repeated measurements taken from that one person over time. There is no other source of it. And consecutive days are not independent observations, because weather is autocorrelated, sleep debt accumulates, and a bad Tuesday makes Wednesday more likely to be bad. That autocorrelation shrinks your effective sample below the number of rows in the log.
Fourteen days might contain three or four genuinely bad days. Three or four points cannot distinguish a real effect from ordinary variation. A pattern that looks striking at day 14 will often dissolve by day 40, and that is not a failure, it is what the method is supposed to do.
A more honest schedule:
- Days 1–14. Build the habit and rule out the boring explanations. Sleep, alcohol, workload, hydration, menstrual cycle. Most recurring symptom patterns resolve here, and this is a good outcome, not a disappointing one.
- Weeks 3–8. Enough range in conditions for a pattern to be worth taking seriously. Aim to cover hot days and cool ones, still days and windy ones.
- Across seasons. Distinguishing an ozone pattern from a particulate pattern needs summer and winter, because the two pollutants run on opposite schedules.
Four ways a log will mislead you
Confounding. A hot, still, sunny day is a high-ozone day. It is also a hot day, a poor-sleep night before, and often a dehydrated day. Your log cannot separate these because they arrive together. Anything you conclude about ozone from summer data is partly a conclusion about heat.
Multiple comparisons. Count what you’re testing. Five symptoms against PM2.5, ozone, NO2, pollen, pressure, temperature and humidity is thirty-five comparisons. At conventional thresholds, one or two will look impressive purely by chance. If you go hunting across every combination until something lines up, you will always find something. Decide in advance which symptom and which pollutant you’re interested in.
Regression to the mean. People start logging during a bad stretch. Bad stretches end on their own. Almost any intervention started at your worst moment will appear to work.
Recall drift. Rating yesterday from memory is not the same measurement as rating today. Log same-day or not at all.
An illustrative example
The following is a constructed example, not a real user.
Suppose you log every afternoon for four weeks and notice that your five worst fatigue days all fell on days when ozone exceeded 65 ppb. That looks like an answer.
Before accepting it, three questions.
Were those also the hottest days? In most climates, yes. High ozone is manufactured by sunlight and heat. So you have not found an ozone pattern, you have found a hot-weather pattern, and heat alone produces fatigue. To separate them you need hot low-ozone days or cool high-ozone days, which do occur, and which you may need to wait for.
Did the pattern exist in the first half of the data too? Split your log in half and check each separately. A real effect shows up in both. A chance pattern usually lives in one.
Can it predict? This is the strongest test available to an amateur and it is genuinely rigorous. Write down the prediction before the day happens: “tomorrow’s forecast is hot and still, so I expect a 4.” Then log as usual and check. Ten predictions made in advance are worth more than a hundred correlations found in hindsight, because hindsight always finds something.
If the pattern survives all three, you have something worth acting on. If it doesn’t, you have saved yourself from reorganising your life around a coincidence.
The strongest thing you can do: change one thing
Correlation from observational logging is weak evidence. Changing something and watching what happens is much stronger, and it is available to you.
Pick one intervention. Run an air purifier in the bedroom. Move your run from afternoon to morning. Close windows overnight. Then:
- Keep logging exactly as before, same time, same scale.
- Change one thing at a time. Two changes at once tells you nothing about either.
- Give it at least three weeks, ideally longer than the natural cycle of whatever you’re chasing.
- If you can bear it, change it back and see whether the effect reverses. Reversal is the closest thing to proof a single person can generate.
This is the informal version of a single-case experimental design, which is a real methodology used in clinical research precisely because group averages don’t tell an individual what works for them.
How to log so the entries stay comparable
Same time every day. Consistency matters more than which time you pick. Set a notification and use it.
Anchor the scale to yourself. A 0 to 5 rating only works if your 3 means the same thing in week six as it did in week one. Write down, once, what each number means for you. Something like: 3 is when I stop being able to concentrate on hard tasks. Then hold to it.
Log the good days. Repeating this because it is the step everyone skips and the one the whole method rests on.
Add context, briefly. The optional profile covers time outdoors, indoor or outdoor location, activity level, mitigation taken, sleep, stress, hydration. These are what let you rule out the boring explanations before reaching for the interesting one. Fill in the ones you’ll actually maintain and ignore the rest. A sparse log you keep beats a thorough one you abandon in week two.
What this is really for
Three honest payoffs, in order of how likely you are to get them.
Ruling things out. Most people who start logging discover their pattern tracks sleep, workload or alcohol. That is a real finding and an actionable one, and it arrives in the first fortnight.
A record for your doctor. A dated symptom log with severity ratings and environmental context is genuinely useful clinical material, far better than trying to reconstruct three months from memory in a ten-minute appointment. This is the highest-value output of the whole exercise and it needs no correlation analysis at all.
Testing whether things help. Once you’ve made a change, the log tells you whether it worked. This is where the habit pays off long-term.
Discovering a clean environmental trigger is possible and it does happen. It is also the least likely of the four, and an article that promised it as the standard result would be selling you something.
One caveat worth stating plainly. Persistent fatigue, breathlessness or headaches have many causes that have nothing to do with air, and some of them need a blood test rather than a chart. If symptoms are severe, worsening, or lasting weeks, take them and your log to a doctor rather than continuing to collect data.
SharedSky is not a medical device and does not provide medical advice. See our Medical Disclaimer.
Sources
- The role of perceived air pollution and health risk perception in health symptoms and disease: a population-based study combined with modelled levels of PM10, International Archives of Occupational and Environmental Health (2018). https://link.springer.com/article/10.1007/s00420-018-1303-x
- Modeled and perceived RF-EMF, noise and air pollution and symptoms in a population cohort, Science of the Total Environment (2018). https://www.sciencedirect.com/science/article/abs/pii/S0048969718316395
- De Carvalho et al., N-of-1 trials in clinical research: methodological foundations, statistical approaches and implementation challenges, British Journal of Clinical Pharmacology (2026). https://bpspubs.onlinelibrary.wiley.com/doi/10.1002/bcp.70382
- Analysing N-of-1 observational data in health psychology and behavioural medicine, Health Psychology and Behavioral Medicine (2020). https://www.tandfonline.com/doi/full/10.1080/21642850.2019.1711096
- AHRQ, Statistical Design and Analytic Considerations for N-of-1 Trials. https://effectivehealthcare.ahrq.gov/products/n-1-trials/research-2014-1
Ready to uncover your hidden triggers? Open SharedSky and complete Day 1 of your personal baseline today.