Unflame

Tracking · Pattern story 03

What a symptom log can tell you, and what it cannot

You wake up flaring and trace back through the last 36 hours. The tomato sauce? The email from your boss? The bad night on Sunday? One of them has to be it. The research on self-tracking says: probably not in the way you think, and here is what a log is actually good for.

By Laxmanna · Updated 2026-10-21 · 8 min read · 22 sources

5.4 vs 4.4

pain days recalled over a week versus pain days actually recorded in an electronic diary by the same people with IBS.[1]

64%

chance of at least one false 'trigger' match by luck alone if you track 20 candidate triggers at once without correction.

29%

of people with IBS who believed gluten hurt them flared on a completely inactive sham bar, versus 39% on wheat and 36% on gluten. Not significantly different.[10]

Memory is a bad recording instrument

When people with IBS kept an electronic diary for a week and were then asked to recall that same week, they remembered 5.35 pain days against 4.35 actually recorded, and rated their average pain 4.61 out of 10 against a logged 3.50.[1] Group averages for 'worst pain' matched reasonably well, but 27% of individuals misjudged their own worst pain by two or more points.[1] The longer the window, the worse it gets: in chronic-pain patients, momentary ratings averaged 42, one-day recall gave 46.7, and 28-day recall gave 53.6 on a 100-point scale.[2]

This is not a character flaw. Memory for symptoms is weighted toward the peak and the end of an episode rather than its average.[3] But it means that 'what happened before the flare' reconstructed from memory is already distorted before you start looking for patterns.

What real-time logging fixes, and what it does not

Recording in the moment removes most of the inflation. Even recording at the end of the same day produces scores 0.8 to 1.8 points higher than ratings captured throughout the day.[4] Paper diaries have a further problem: when researchers hid sensors in diary binders, people who reported 90% on-time compliance had actually written entries on time 11% of the time, filling the rest in later.[5] Time-stamped electronic entries fix the sequence of events.

What they do not fix is the gap between belief and data. In one experience-sampling study, over 81% of people with IBS believed stress drove their flares, but each person's actual stress-to-symptom slope in their own data correlated only weakly with that belief.[6] And subjective and objective measures can disagree: in 80 people with IBS wearing wrist actigraphs, self-rated poor sleep predicted next-day abdominal pain, while the actigraphy sleep metrics showed no link to next-day symptoms at all.[7]

A real-time log is still a self-report. It tells you what you felt and when, with much less distortion. It does not by itself tell you why.

Co-movement is not causation

Here is the pattern that shows up again and again. In 357 adults with IBS prompted three times a day for a week, stress and symptoms moved together strongly in the same moment, but the delayed relationships, stress at one time of day predicting symptoms later, were weak or not significant.[8] Individual-level time-series analyses show that symptoms frequently drive later changes in stress, mood and fatigue, not only the other way round.[9] A bad flare ruins your sleep and your mood; the log then shows poor sleep and low mood 'before' the next day's symptoms.

Then there is arithmetic. If you track 20 candidate triggers and test each against your flares at the usual 5% threshold, the probability of at least one false positive by chance alone is about 64%.* A single-patient psoriasis analysis that searched for the best time lag after the fact 'found' a correlation that existed only because of the search.[9]

The strongest test of a suspected trigger is to hide it. In a double-blind crossover trial of people with IBS who were sure gluten worsened their symptoms, a significant flare occurred in 39% on wheat bars, 36% on gluten bars, and 29% on inactive sham bars. The differences were not statistically significant.[10] In another trial, expecting to eat gluten raised symptom scores by 11.7 mm on a visual scale, compared with 16.6 mm for actually eating it.[11] Expectation is not imaginary; it is a measurable share of the response.

And when eight expert clinicians and dietitians were given the same 17 food-and-symptom journals to read, their agreement on which foods were the triggers was near zero.[12] If experts cannot read a diary and agree, the diary alone is not the answer.

* This is an arithmetic illustration (1 − 0.9520 ≈ 0.64), not a trial result.

Signals versus labels

Wearables add data, not necessarily clarity. Consumer sleep trackers agree reasonably well with laboratory polysomnography on total sleep time (ICC 0.77 to 0.86) but poorly on time awake (ICC 0.08 to 0.32) and sleep stages, and that validation was done on healthy young adults for a single night.[13] Smartwatch 'stress' scores track heart-rate physiology, not felt psychological stress.[14] Of ten manufacturers whose 14 'readiness' or 'recovery' scores were evaluated, none disclosed the formula, and none had been validated for predicting disease flares.[15]

The same caution applies to symptoms themselves. In eczema, felt itch intensity (Peak Pruritus NRS) correlates only weakly with clinician-scored visible rash (EASI) at baseline: they are distinct outcomes, and a log of one is not a log of the other.[16]

No consumer wearable metric, and no app, has been clinically validated to detect or predict flares in IBS, eczema or psoriasis. Results from inflammatory bowel disease, where wearable models have shown promise, do not transfer to IBS, which is a disorder of gut-brain interaction rather than an inflammatory disease.[17][18]

What a log is actually good for

None of this means tracking is useless. The evidence supports four realistic jobs for a time-stamped log:

  1. Correcting memory. It replaces an inflated reconstruction with what you actually felt, when.[1][4]
  2. Showing same-moment co-variation. It is reliable for revealing what moves together in the same hour, day or week.[8]
  3. Fixing the order of events. It tells you which came first, which is the minimum for asking whether one thing could have caused the other.[9]
  4. Turning hunches into testable hypotheses. The strongest way to test a suspected trigger in one person is a planned, repeated and ideally blinded self-experiment with a clinician, and a log is where the hypothesis comes from.[19][20]

This is the standard Unflame holds itself to. The Unflame app shows which signals in your own data have repeatedly been unusual one to three days before your flares, with the counts and a confidence rating for each, and it says plainly that this is correlation in one person. A log that tells you it has found your trigger is overstating what the method can do.

Ten questions to ask of your own data

  1. Am I recording at the time, or reconstructing events later?
  2. Am I measuring the symptom with a consistent, standard definition each time?
  3. Did I pick one or two suspects and a plausible time delay before looking, or did I search afterwards?
  4. Have I counted the days I was exposed and did not flare, and the days I flared without exposure?
  5. Which came first, and could the symptom be driving the exposure (poor sleep, comfort food, stress) rather than the reverse?
  6. What else changed at the same time: season, medication, infection, travel, hormones?
  7. Am I tracking so many variables that a false match is statistically guaranteed?
  8. Does my wearable number reflect raw physiology (heart rate, movement) or an undisclosed algorithm label?
  9. Does what I believe match what my data show?
  10. Is a structured, blinded re-challenge with a clinician the right next step?

Even a short self-experiment has limits. In a study of app-guided 12-day food experiments, participants routinely read 'no statistical evidence' as proof that a food was harmless, when the experiment was simply too short to detect a subtle or intermittent effect.[21]

When not to self-track

Self-tracking is for stable, chronic symptom variation. New, severe or changing symptoms need medical evaluation, not a diary: blood in the stool or black stools, unintended weight loss, fever, severe night pain, rapidly spreading skin lesions, or signs of skin infection (pus, crusting, severe pain). Elimination diets, challenge protocols and any change to prescribed medication belong under clinical supervision.

One honest unknown: almost no experience-sampling studies (fewer than 5%) have tested whether intensive tracking changes the symptoms it measures, so science cannot yet say whether constant monitoring breeds hypervigilance.[22] Track with a question in mind, for a defined period, and then stop and review.

Things people hear, and what the evidence says

"My tracking app found my trigger."
Observational logs show co-variation. They cannot rule out confounding, expectation or reverse causation. Only a planned, blinded re-challenge can test a trigger in one person.[19][20]
"The more I track, the more accurate the picture."
Tracking many variables inflates false positives. Twenty unadjusted comparisons give roughly a two-in-three chance of a spurious match.
"My watch's stress score shows my inflammation."
Stress scores track heart-rate physiology, not felt stress and not inflammation. No wearable label is validated to detect flares in IBS, eczema or psoriasis.[14][15][18]
"I didn't react in my two-week test, so that food is safe."
Short self-experiments are often underpowered. 'No evidence of an effect' is not evidence of no effect.[21]

Sources

Primary human studies are marked "primary"; reviews, guidelines and models are labelled. Where a source is an abstract only, that is stated.

  1. Lackner JM et al. Prospective electronic diary versus 7-day recall of IBS symptoms. Neurogastroenterol Motil 2014 — primary. https://pmc.ncbi.nlm.nih.gov/articles/PMC4247164/
  2. Broderick JE et al. The accuracy of pain and fatigue items across different reporting periods. Pain 2008 — primary. https://pmc.ncbi.nlm.nih.gov/articles/PMC2617790/
  3. Schneider S et al. Peak and end effects in patients' daily recall of pain and fatigue. J Pain 2011 — primary. https://pmc.ncbi.nlm.nih.gov/articles/PMC3012151/
  4. Bosman M et al. Experience sampling versus end-of-day diary in IBS: psychometric comparison. Am J Gastroenterol 2025 — primary. https://pmc.ncbi.nlm.nih.gov/articles/PMC12043265/
  5. Stone AA et al. Patient non-compliance with paper diaries. BMJ 2002 — primary (chronic pain population). https://www.ncbi.nlm.nih.gov/pmc/articles/PMC111114/
  6. Vork L et al. Stress, pain and belief in IBS: an experience-sampling study. Clin Transl Gastroenterol 2020 — primary (37 IBS adults). https://pmc.ncbi.nlm.nih.gov/articles/PMC7386351/
  7. Topan R et al. Sleep and next-day symptoms in IBS: experience sampling plus wrist actigraphy. Am J Gastroenterol 2024 — primary. https://pubmed.ncbi.nlm.nih.gov/37737676/
  8. Yunusova A et al. Momentary stress and IBS symptoms in 357 adults (trial baseline EMA). Clin Gastroenterol Hepatol 2026 — primary, pre-final manuscript. https://pmc.ncbi.nlm.nih.gov/articles/PMC13604563/
  9. Engel F et al. Individual time-series (vector autoregression) analyses of symptoms, stress and mood. Front Psychiatry 2022 — primary; and single-case IBS-D series, BMC Gastroenterol 2021. https://www.frontiersin.org/journals/psychiatry/articles/10.3389/fpsyt.2022.768134/full
  10. Seiler CL et al. Wheat, gluten and sham bars in IBS patients reporting benefit from gluten-free diet: double-blind sham-controlled crossover RCT. Lancet Gastroenterol Hepatol 2025 — abstract. https://pubmed.ncbi.nlm.nih.gov/40706614/
  11. de Graaf MCG et al. Expectancy versus actual gluten intake in non-coeliac gluten sensitivity: 2×2 RCT. Lancet Gastroenterol Hepatol 2024 — primary. https://www.thelancet.com/journals/langas/article/PIIS2468-1253(23)00317-5/fulltext
  12. Zia J et al. Inter-rater reliability of provider interpretation of IBS food and symptom journals. J Clin Med 2017 — primary pilot (8 raters, 17 journals). https://people.ucsc.edu/~cfchung/assets/papers/jcm-providerreview.pdf
  13. Kainec KA et al. Validation of consumer sleep wearables against polysomnography. Sensors 2024 — primary (50 healthy adults, one night). https://pmc.ncbi.nlm.nih.gov/articles/PMC10820351/
  14. Rosenbach H et al. Laboratory validation of a smartwatch stress score. Stress Health 2025 — primary. https://onlinelibrary.wiley.com/doi/full/10.1002/smi.70125
  15. Readiness, recovery and strain: a review of composite health scores from 10 wearable manufacturers. De Gruyter Brill 2025, doi 10.1515/teb-2025-0001 — narrative evaluation. https://doi.org/10.1515/teb-2025-0001
  16. Yosipovitch G et al. Peak Pruritus Numerical Rating Scale: validation in atopic dermatitis trials. Br J Dermatol 2019 — primary. https://rcastoragev2.blob.core.windows.net/4cd809d8b3ada527fd2f219d086c08b6/PMC6850643.pdf
  17. Hirten RP et al. Physiological data from wearables identify and predict inflammatory bowel disease flares. Gastroenterology 2025 — primary (IBD, not IBS). https://pubmed.ncbi.nlm.nih.gov/39826619/
  18. Veldman F et al. Wearables in disorders of gut-brain interaction: systematic review and meta-analysis. Neurogastroenterol Motil 2026 — review. https://onlinelibrary.wiley.com/doi/full/10.1111/nmo.70232
  19. Kravitz RL, Duan N et al. Design and Implementation of N-of-1 Trials: A User's Guide. AHRQ 2014 — methods guidance. https://effectivehealthcare.ahrq.gov/products/n-1-trials/research-2014-4
  20. Daza EJ. Causal analysis of self-tracked time series data using a counterfactual framework for N-of-1 trials. Methods Inf Med 2018 — methods. https://pubmed.ncbi.nlm.nih.gov/29621835/
  21. Karkar R et al. TummyTrials: a feasibility study of using self-experimentation to detect IBS food triggers. CHI 2017 — primary. https://homes.cs.washington.edu/~jfogarty/publications/chi2017-tummytrials.pdf
  22. Maher JP et al. Measurement reactivity in ecological momentary assessment: systematic review. J Meas Phys Behav 2024 — review. https://pmc.ncbi.nlm.nih.gov/articles/PMC12922600/

Written by Laxmanna · Research-led health writer. Every explainer on Unflame starts from a structured evidence dossier: primary studies retrieved, claims tabulated, contradictions logged, and wording checked against what the sources can support. How the dossiers are built.

Related