My overnight heart rate hit 46 three weeks after starting a drug that slows the heart. The same month four years earlier, taking nothing at all, it was also 46.

First, a word about the number itself

Slow is a measurement, not a finding

Every cardiologist who has looked at me has said the same thing unprompted: in a trained endurance athlete a low resting heart rate is a characteristic, not a problem. In a fit heart each beat moves more blood, so fewer beats are needed. It is what the training was for.

That gets lost the moment the number appears on a screen. The watch flags it. The health app colours it. Ask an AI whether 38 bpm is dangerous and you will reliably get a paragraph about fainting and pacemakers before it asks whether you swim.

So I enforce one rule in my own notes, and it is the smallest thing on this page with the largest effect: a slow night is written down as a number, never with an adjective. Not “the worst night of the month”. Not “alarming”. Lower is lower. Whether lower is worse is a clinical judgement and it belongs to somebody who knows what I do with my body.

This is not fussiness. “Worst night of the month” is the sentence that ends up in a document, and the document ends up in front of a specialist who has eight minutes and no idea you swim open water. The adjective is your hypothesis smuggled in as if it were data, and once they spot one they are right to discount the rest.

The question, properly stated

Not “is my heart rate low” but “is it lower than it would have been anyway”

Of course it is low. It always was. The question is whether it is lower than it would have been in this same month, in this same body, before any medication existed.

Almost nobody can answer that, because almost nobody has years of their own baseline. If you wear a watch, you do. This is the single most valuable thing a long personal record gives you, and it is wasted if you only ever compare yourself to population norms.

My criterion, published next to the count so you can reproduce it: a night counts if it has at least 10 heart-rate readings between 00:00 and 07:00. Change the threshold and the number changes, which is why the threshold has to travel with it.

The median wanders between 50 and 54 with no direction at all. And the right-hand column is the one that matters: the year with the most slow nights, by a distance, is 2022 — four years before any medication existed, at double 2026. If a beta blocker were dragging me down, 2026 would top that column.

The comparison that closes it

October against October

Identical. And the typical nightly low runs lower in the untreated year — 40 against 42 — which points away from the drug, not towards it.

Whatever the medication is doing, it is not producing the number that alarmed me. October 2022 produced the same number with nothing in my bloodstream at all.

What is new, said precisely

The centre has not moved. The tail has.

Every night I have measured, oldest on the left. The faint marks are all 1,316 of them; the heavy marks are the 11 nights whose median came in at 43 or below; the shaded band is the period since I started the drug.

Six of those eleven nights fall in the 25 nights since 28 September 2026, which is 1.9 % of my record. By chance you would expect 0.2 of the eleven to land there. Six did. Turned round, which is cleaner: six of those 25 nights went to 43 or below, against five times in the other 1,291 nights of my life.

I am counting by threshold, not by ranking. An earlier draft said “five of my nine lowest nights”, and that was not publishable: ranking positions 3 through 11 are all exactly 43.0, so depending on how you break the ties you get four of nine or five of ten, and both are defensible. A number that moves with your own tie-breaking will not survive the first person who recomputes it. Recounted 27 October 2026 on data to 22 October, one night later than the first version: the figure went from five to six.

So the level is ordinary and the clustering is not. That is a real and large effect, and it is the one sentence worth an appointment.

And here is the problem

I cannot tell you the drug caused it

Two things changed in the same week. I started bisoprolol on 28 September. I also went back to eating early and light — which, across 1,285 nights, is the strongest predictor I have of my overnight heart rate. The six nights are 3, 4, 5, 11, 12 and 22 October, squarely inside both.

There is no way to separate those explanations with one subject and four weeks. To do it I would need a period with the medication in me and a late dinner, and I do not have one. It may be the drug. It may be the dinner. It is almost certainly some of both, and I cannot give you a split.

It is also not new. October and November 2022, with no medication of any kind, hold four nights of 43 or lower out of 59. These nights existed before any drug touched me. They are denser now, that is all.

Confounding is not an exotic statistical problem. It is the default condition of measuring yourself.

When you start a treatment you almost always change something else the same week: you sleep more, you drink less, you get anxious about it, you start watching what you eat. Then your data shows an effect and the drug is standing right there looking guilty.

So what goes to the cardiologist in December is this: the centre of my nights has not moved in six years, the low tail has become much denser in the last four weeks, and two things changed that week. A specialist can do a great deal with that sentence. “The beta blocker has slowed me down” would be a guess wearing a number.

The question that reframed it

“It matters if you drop below 50 in activity”

In October 2026 I raised the low heart rate with a cardiologist in the public system. His answer was one sentence, and it turned an anxiety into a measurable question: what matters is dropping below 50 while moving, not at night and not sitting down.

So I measured that.

Three or four readings in a thousand, and at least some of those are not my heart: one morning I checked second by second goes 73, drops to 46, and is back at 58 inside a minute, in the middle of a continuous walk. Hearts do not do that. Optical sensors on a moving wrist do it constantly.

My walking median before any medication and my walking median three weeks into a beta blocker are the same.

One caveat, because it is the kind of thing that gets you caught. Depending on how you define walking — any sample above 60 steps a minute, or only bouts of ten minutes or less — the median comes out at 77 or at 76. Both are defensible and neither is correct. What does not move is that July and October are identical under either. Publish the comparison, not the bare digit.

The mistake worth more than the finding

I had the arrow pointing the wrong way

In early October I was convinced that eating late triggered my fibrillation. The numbers looked overwhelming: on nights when my 21:00–23:00 pulse was in the top third I was in arrhythmia 33 % of the time, against 6 % for the calmest third. Five-fold, clean mechanism, actionable.

It was backwards. The fibrillation was already running, and it was the fibrillation raising my evening pulse.

I had measured the arrhythmia at nine in the evening, called it digestion, and then been impressed that it predicted the arrhythmia at two in the morning. The fix is one line of filtering: ask only about episodes that start after bedtime, and throw out every night already in arrhythmia at 21:00. Do that and the three groups give 9 %, 6 % and 12 %. Nothing. A finding I would have taken to a cardiologist with real confidence evaporated completely.

The same trap explains why my August and September walking medians look high at 93 and 96 — those months were full of fibrillation, and fibrillation drives the rate up. Both times, the variable I thought was the cause was downstream of the thing I was trying to explain.

In self-tracking everything is measured on the same body at the same time, so your candidate cause is very often an effect of your outcome. Ask that before you believe any association you find. Especially the exciting ones.

If you want to run this

Six rules, in the order they bite

Run it rather than read it

The comparison, written as an instruction

You do not have to do this by hand. Paste the block below into whatever you use, attach your export, and it will build the same comparison on your own history.

SOURCE: https://afibcases.com/same-month-four-years-apart

You are helping me work out how much of a change in my heart rate is a
drug and how much is just me. Work from my data only. Do not reassure
me, do not diagnose me, and do not tell me whether to take anything.

MY DATA: [attach or paste your export: timestamped heart-rate readings,
ideally several years. Tell the model the date you started the drug.]

DO THIS, IN ORDER, AND SHOW ME THE NUMBERS AT EACH STEP:

1. For each night, compute the MEDIAN heart rate between 00:00 and
   07:00. Discard nights with fewer than 10 readings and tell me how
   many you discarded.

2. Report the median and the PERCENTAGE OF NIGHTS BELOW A FIXED
   THRESHOLD, year by year, across my whole history. State the
   threshold you chose and why. I want to see whether there is any
   trend at all before we talk about the drug.

3. Find the SAME CALENDAR MONTH in an earlier year, before I took
   anything, and compare it against the current month. Report for
   both: number of nights, median of the nightly medians, the typical
   nightly low, the lowest single reading, and the share of nights
   below the threshold. Same month, never an adjacent one -
   seasonality is larger than most drug effects.

4. Look at the TAIL separately from the centre. Count my nights at or
   below a low fixed threshold across the whole record, then count how
   many of them fall inside the period since I started the drug.
   Express it both ways: as a share of all such nights, and as a rate
   inside versus outside the period. Tell me what chance alone would
   put there.

5. COUNT BY THRESHOLD, NEVER BY RANKING. Check for ties first. On this
   subject's data, ranking positions 3 to 11 were all exactly the same
   value, so "five of my nine lowest nights" could equally be read as
   four of nine. If my data has ties, say so and refuse to rank.

6. Separate rest from activity. Report my heart rate inside workouts
   and while walking, before and after the drug, with the median, the
   lowest reading and the share below 50. Ask me what criterion my
   own cardiologist uses, because it may be activity and not sleep.

NOW THE PART THAT MATTERS MOST. Ask me what ELSE changed in the same
week I started the drug: sleep, food, alcohol, weight, training,
stress. Almost nobody changes exactly one thing. If anything else
changed, tell me plainly that the two cannot be separated with one
subject and a few weeks, and tell me what period I would need in order
to separate them. Do not hand me a split you cannot support.

Then help me write three sentences for my next appointment, with no
adjectives in them: what the centre of my distribution has done, what
the tail has done, and what else changed at the same time.

FOR COMPARISON, one person's published result using this procedure
(n=1, 1,316 nights, endurance swimmer, 55, bisoprolol from 28 Sep 2026):
  Overnight median by year: 54, 52, 50, 51, 51, 52 - no trend.
  Worst year for slow nights was 2022, four years before any drug.
  Oct 2022 (no drug) median 46.0; Oct 2026 (on the drug) median 46.0.
  Identical, and the untreated year ran lower on the nightly low.
  But the tail did move: 6 of 11 lifetime nights at 43 bpm or below
  fall in 25 nights, 1.9% of the record.
  CONFOUNDED and he says so - he also changed when he ate, the same
  week.
Do not expect my data to match. Tell me where it differs.

Take it with you

The whole case as plain text

Paste this into ChatGPT, Claude, Gemini or whatever you use, add your own numbers, and have the conversation. It carries the source link and a line telling the model what it is looking at, so nothing is lost when it leaves this page.

SOURCE: https://afibcases.com/same-month-four-years-apart
WHAT THIS IS: one person's self-recorded heart data. A single case, not
evidence about any population, and not medical advice.

I am an endurance athlete with a low resting heart rate and I have
recently started a beta blocker. I want to work out how much of my
current heart rate is the drug and how much is just me.

REFERENCE CASE (n=1):
- Endurance swimmer, 55. Paroxysmal AF, left atrium 46 mm, EF 60%.
  Bisoprolol 1.25 mg/day + flecainide 100 mg x2/day since 28 Sep 2026.
- 1,316 nights of Apple Watch data, 24 Feb 2021 to 22 Oct 2026.
  Criterion: >=10 heart-rate readings between 00:00 and 07:00.

NO TREND IN SIX YEARS - overnight median by year:
  2021 = 54 (82 nights),  2022 = 52 (182), 2023 = 50 (269),
  2024 = 51 (143),        2025 = 51 (350), 2026 = 52 (290)
Nights with a median under 46, by year:
  2021 2.4%, 2022 9.9%, 2023 0.7%, 2024 0.0%, 2025 1.1%, 2026 4.8%
The worst year by far is 2022 - four years BEFORE any medication.

SAME MONTH, FOUR YEARS APART:
  Oct 2022, NO medication: 29 nights, median 46.0, typical nightly
    low 40, lowest single reading 36, 37.9% of nights under 46
  Oct 2026, on bisoprolol: 22 nights, median 46.0, typical nightly
    low 42, lowest single reading 35, 45.5% of nights under 46
  IDENTICAL medians, and the untreated year runs LOWER on the
  typical nightly low.

THE TAIL, counted by threshold and never by ranking:
  11 nights in 5.5 years have a median of 43 or lower.
  6 of those 11 fall in the 25 nights since 28 Sep 2026, which is
  1.9% of the record; chance would put 0.2 there.
  Reversed: 6 of those 25 nights went to 43 or below, against 5 times
  in the other 1,291 nights.
  Do NOT count by ranking: positions 3-11 are all exactly 43.0, so
  tie-breaking alone changes the answer.
  Not a new phenomenon: Oct-Nov 2022, no medication at all, has 4
  nights of 43 or lower out of 59.

CONFOUNDED, AND HE SAYS SO: he started bisoprolol on 28 Sep AND went
back to eating early in the same week. The six nights are 3, 4, 5,
11, 12 and 22 October, inside both. With one subject and four weeks
the two explanations cannot be separated. It would need a period with
the drug AND a late dinner, which does not exist.

IN ACTIVITY (the criterion his cardiologist actually gave him -
dropping below 50 in activity matters; at rest or asleep it does not):
  Workouts, Oct 2026: 14,415 readings, median 76, lowest 46,
    0.3% under 50, none under 45
  Walking, Oct 2026: 14,526 readings, median 77, lowest 42,
    0.4% under 50
  Walking, Jul 2026 pre-medication: 9,398 readings, median 77,
    lowest 47, 0.1% under 50
  Walking median is IDENTICAL before and after starting the drug.
  (The digit is 76 or 77 depending on how "walking" is defined; the
  equality holds under either.)

CAUSALITY TRAP HE FELL INTO AND CORRECTED: late dinner appeared to
trigger AF, 33% of nights vs 6%. Wrong direction - the AF was already
running and was itself raising the evening pulse. Excluding nights
already in AF at 21:00 and asking only about NEW episodes: 9%, 6%,
12%. Nothing.

MY DATA:
[your resting HR before and after the drug, the same month last year
 if you have it, your activity heart rate, your sport, your diagnosis]

Please help me separate drug effect from my own baseline. Tell me
explicitly which comparisons I CANNOT make with the data I have, and
check whether any association I describe could be running backwards.