A patient I'll call Sara emailed us with a spreadsheet. Three months of daily tracking. Twelve symptoms scored 1 to 5, four treatments started and stopped, sleep hours, weather, stress, the works. She had done everything the tracking guides told her to do. And she still couldn't tell what was making her brain fog worse on bad days.
She wasn't doing anything wrong. She had hit the wall every diligent tracker hits, which is that a human eye cannot read a 90-day spreadsheet of overlapping signals and pull a pattern out of it. That's what this guide is about: why pattern detection in chronic Lyme is genuinely hard, what the Compass view actually does, and what it can't do.
Why patterns are hard to see
Chronic Lyme is a high-noise environment for self-tracking. A typical patient is logging six to fifteen symptoms, two to five active treatments, and a handful of daily factors like sleep, stress, weather, activity, and diet. Even if you only track for thirty days, that's thousands of data points. Your brain isn't built to find correlations in that.
A few things make it worse than it sounds.
Most of the changes you're trying to evaluate happen in tiny sample sizes. You start a new antibiotic on Tuesday. You have one Tuesday. The herx hits day three. You have one day three. Statistically, that's a sample of one. You can't tell the medication apart from a random bad week.
The signal-to-noise ratio is brutal. Your symptoms move on their own, in cycles, for reasons no one fully understands. Some patients flare roughly every four weeks. Many flare around their menstrual cycle. Most have weather sensitivity. So when you start a treatment and feel worse, the worse could be the treatment, the cycle, the storm front, or the bad night of sleep stacked on top of all three.
Almost nothing in chronic Lyme is linear. Brain fog isn't a steady function of one input. It's loosely tied to sleep, more so when sleep drops below some personal threshold, and it gets weird when you're herxing. A "more X means more Y" mental model fails on this kind of relationship.
Then there are confounders. You sleep badly because you're flaring, and you flare more because you slept badly. Did the flare cause the bad sleep, or did the bad sleep cause the flare? In a single patient, often, you can't tell.
This is why people end up where Sara was. Months of data, and the pattern is somewhere in there, but the human brain can't pull it out by squinting.
What the Compass view actually does
The Compass view (the Insights screen in the app) is built to surface the patterns your eye can't catch in raw data. It looks across all your check-ins, all your treatments, and all your factors, and it asks a simple question for every pair: when this thing changed, did that thing change with it more often than chance would predict?
That's correlation. It's not magic, and it's not new science. It's what a researcher running an n-of-1 trial does on paper. The point of building it into the app is that you don't have to do it on paper.
For each symptom you're tracking, Compass surfaces the factors and treatments that move with it most strongly. If your brain fog severity climbs on the same days your sleep drops below six hours, that pair gets a strong correlation score. If your joint pain climbs on the days you took a particular antibiotic, same thing. Pairs are ranked by strength, so the loud signals come up first and the quiet ones don't drown them out.
Compass also flags the difference between coincidence and likely connection. A pattern that holds across thirty data points is treated differently from one that holds across three. If you've only logged a treatment four times, the app will tell you the correlation isn't reliable yet rather than pretend a confident answer exists. The worst thing a pattern-detection tool can do is hand you a false certainty.
The visualization is the part that took the longest to get right. Numbers alone don't help. What helps is being able to look at your worst brain fog week and see, on the same timeline, what your sleep was doing, what treatments you were on, and what the weather was. The HealthDayDetailScreen is the drill-down view for any single day. The Compass view is the zoom-out across weeks and months.
A walk-through with example data
To make this concrete, here's what a real four-week stretch can look like, simplified.
Week 1: A patient logs daily. Brain fog averages 2 out of 5. Sleep averages 7 hours. No new treatment.
Week 2: Starts doxycycline. Brain fog climbs to 3 out of 5 on average, with a spike to 4 on day 3 and day 4. Sleep starts dropping. Several nights under 6 hours. Stress is up because work is heavy.
Week 3: Brain fog stays elevated. Sleep is uneven. The patient starts to assume the doxy is causing the brain fog.
Week 4: Doxy continues. The patient pays more attention to sleep, gets two solid 7+ hour nights mid-week, and brain fog drops on those days even though doxy hasn't changed.
If you read that as a journal, you'd probably blame the doxycycline. The dates line up. You feel worse, you started a drug, end of story.
What Compass would show is different. When you correlate brain fog against doxy across 28 days, the relationship is weak. When you correlate brain fog against nights under 6 hours, the relationship is strong, with a higher score and more data points behind it. The doxy isn't innocent, but it's not the main driver. Sleep is. The thing to investigate first isn't the antibiotic. It's why sleep tanked.
That changes the conversation with your LLMD. Instead of "the doxy is making my brain fog worse, can we switch," you walk in with "my brain fog tracks tightly to nights under six hours, and my sleep got worse around the time I started doxy. Can we look at sleep first?" That's a more useful starting point. It might still end with switching the antibiotic, or with a sleep intervention that was the real fix all along.
What it can't do
Worth saying clearly, because tools that promise too much in chronic illness are a problem.
Small data is small data. If you've been tracking for two weeks, the Compass view will show you tentative patterns at best. Most relationships need at least four weeks of consistent data, and stronger ones often need eight to twelve. The app will tell you when a correlation isn't yet reliable. It won't pretend.
Correlation is not causation. If your headaches correlate with stress, that doesn't prove stress causes them. It might be that flaring causes both, or that the same poor sleep night causes both. The Compass view tells you what moves together. The why is still your job, and your doctor's.
Medical decisions belong to a clinician. The point of the tool is to give you and your LLMD a sharper picture to talk about, not to decide for you whether to stop a treatment, change a dose, or rule something out. We have heard from a few patients who used early Compass output to argue with their doctor about dropping a medication. That's not what it's for. Bring the patterns. Let the doctor weigh them.
The app also can't fix bad inputs. If you're scoring symptoms inconsistently, skipping factor logs, or only checking in on the bad days, the Compass view will surface patterns from that biased data, and they'll be biased too. The tool is only as good as the daily check-in feeding it.
The features that feed it
The Compass view doesn't pull patterns out of nothing. It runs on the data you give it, and the more channels you feed in, the more it can find.
The 5-step daily check-in (CheckIn1 through CheckIn5 in the app) is the primary input. It walks you through your active symptoms one at a time, with a 1 to 5 severity scale and descriptive levels for each step, so you're not guessing what a 3 means today versus last week. Consistency in the scoring is what makes the correlation math work.
The Herxheimer reaction screen is a separate channel on purpose. A herx isn't an ordinary day, and averaging it into your baseline data muddies every pattern downstream. Logging it separately means Compass can show you herx-specific patterns (which treatment triggers the worst ones, whether they're getting shorter over time) without contaminating the rest of your data.
Factor tracking is the quiet hero. Sleep hours, weather conditions, stress level, activity level, and diet notes are what let Compass distinguish "the treatment is doing this" from "the storm front is doing this." If you only track symptoms and treatments, you're missing half the variables that move your symptoms day to day. Most of the false leads we see in patient data come from skipped factor logs.
Multiple check-ins per day matter more than they sound. A lot of Lyme patients have a different morning self and afternoon self. Brain fog at 9 AM and brain fog at 4 PM aren't the same data point. Logging twice or three times a day, when it fits, gives Compass enough resolution to spot time-of-day patterns, like the afternoon crash that follows specific lunches.
Custom symptom, treatment, and factor definitions matter because Lyme doesn't fit a standard form. If your worst symptom is a buzzing in your feet that no template lists, you can add it. If your treatment is a specific herbal protocol with three components, you can name each one. The correlation engine doesn't care what you call things. It cares that you call them the same thing every time.
The HealthDayDetailScreen is the zoom-in. When Compass flags a pattern across weeks, you usually want to drill into one specific day to remember what was actually going on. It pulls every check-in, every treatment, and every factor for a single date onto one page. That's the difference between "the data says day 17 was bad" and "right, day 17 was the day after the wedding, I slept four hours and skipped the morning meds."
Getting useful insights faster
A few practical notes from watching people use this.
Track for at least four weeks before you start judging the Compass view. Three weeks of data is rarely enough for the math to settle. Eight weeks gives you something solid. If you start a new treatment, give it four weeks of data points before deciding what it's doing.
Track factors, not just symptoms. The most common reason patients don't see useful patterns is that they log a faithful daily symptom score and skip the factor screens because they feel like extra work. Factor data is what lets Compass tell symptom drivers apart.
Tag treatments precisely. "Started antibiotics" is not enough. Which one, what dose, when did you change it? If you're rotating, log the rotation. The correlation engine treats every distinct treatment as its own variable. Vague labels collapse three drugs into one bucket and ruin the math.
Be consistent on bad days. The temptation when you're flaring is to skip the check-in. That's exactly the data point Compass needs most. A two-question quick log on a hard day is more useful than a perfect log on an easy day.
Don't chase tiny correlations. If a pattern is real, it'll keep showing up week after week. The correlations worth acting on are the ones that hold across multiple separate stretches of data.
Further reading
If you want to read more about the science underneath this kind of tool:
- Patient Perceptions of Their Own Data in mHealth N-of-1 Trials for Chronic Pain. A qualitative study of patients running single-patient trials on a phone app, including what they found useful and what they found overwhelming.
- N-of-1 Trials: The Epitome of Personalized Medicine?. A readable overview of the n-of-1 trial design, which is the formal version of what self-tracking is doing informally.
- Characterising User Engagement with mHealth for Chronic Disease Self-Management. A 2024 paper on what makes patients stick with self-tracking apps long enough for the data to be useful, and where machine-learning models on patient-reported data succeed and fail.
Sara, the patient with the spreadsheet, ended up running her three months of data through Compass and found out her brain fog correlated more tightly with stress than with any of her treatments. That wasn't the answer she expected. It also wasn't the final answer. But it was the first piece of useful signal she'd had in a year, and it gave her LLMD something specific to work with. That's all the tool is supposed to do.
LymeTrack is a tracking tool, not medical advice. Talk to your LLMD or treating physician before changing a treatment plan.