How late is too late for coffee?

Ultrahuman members' sleep window was shortest after a last caffeinated drink under 4 hours before usual bedtime: 14 minutes shorter (likely between 6 and 22 minutes) than their own early-caffeine nights. The shortening eased to 8 minutes, then 3.6 minutes, in the next two groups as the drink came earlier. At 8 to 10 hours out, the change was about 1 minute, no clear difference. When only nights with similar logged amounts counted, the gaps for drinks 4 to 10 hours before bed were smaller or reversed, so extra caffeine may explain part of them.

When does late coffee start to shorten sleep?

Sleep ran shorter when the last caffeinated drink came within about 8 hours of a member's usual bedtime, and most when it came under 4 hours before bed. A review pooling 24 studies found that caffeine shortened total sleep time by 45 minutes on average, and advised having a standard coffee at least 8.8 hours before bed [1]. This group's pattern points the same way, though the shortening here was smaller than in those trials, which measured total sleep time.

A drink counted when the member logged it in the Caffeine Window with at least 20 mg of caffeine, so decaf did not count. Every caffeinated drink counted, not only coffee: tea, cola, energy drinks and matcha too. A night's drinks were those timed between waking from the night before and falling asleep, so a drink just after midnight still belonged to that evening. A member's usual bedtime was the middle of the times they fell asleep over the 14 nights before, from at least 10 valid nights on the same ring and firmware version. It never used the night being measured, so a late night after caffeine could not move its own drink into another group.

Each night was grouped by how long before that usual bedtime the last logged caffeinated drink came: 10 hours or more, 8 to 10, 6 to 8, 4 to 6, or under 4 hours, which includes drinks after the usual bedtime. The night's sleep was timed from the start to the end of the main sleep as the ring detected them, so brief awakenings in between count toward it.

Each member's nights in a later group were compared with the same member's nights whose last caffeine came 10 hours or more before bed, so only members with both kinds of night counted in a comparison. Nights with no logged caffeine were left out, because a member may simply not have logged. Each member counted equally, however many nights they logged; the uncertainty ranges resampled whole members.

In a stricter version that kept only nights whose drink log was last changed before the member fell asleep, the gap for drinks under 4 hours before bed was smaller. Drinks under 8 hours before bed still came with shorter sleep when the usual bedtime needed a valid night on all 14 nights before. The same held when only drinks from the app's library counted.

Does the amount of caffeine matter as much as the timing?

Timing still mattered for caffeine close to bedtime, even at similar amounts. When only nights with 100 to 250 mg of logged caffeine counted, the sleep gap for drinks under 4 hours before bed was larger. Days with later last drinks also carried more logged caffeine, so timing and amount moved together in this group. Days in the latest timing group averaged 205 mg of logged caffeine (likely between 193 and 217 mg). Each logged amount came from the app's drink library or the member's own entry.

The chart above shows that every later timing group averaged more logged caffeine than the early-caffeine days, with each member counted once. The step up came between the earliest group and the rest, and the later groups sat close together.

Are late coffee drinkers different sleepers anyway?

Lasting differences between late and early coffee drinkers do not explain the gaps, because each member was compared only with their own early-caffeine nights. Differences between days could still play a part. A busy day can bring both a late coffee and a late night, and a bad night can be followed by a late coffee. Comparing weeknights only with weeknights, and weekends with weekends, made the gaps larger.

What this analysis can't tell you

Late and early caffeine days can differ in other ways, so the gaps cannot show that caffeine caused shorter sleep. The last logged drink is the last one a member logged, so an unlogged later coffee would put a night in the wrong timing group. Members can edit drink times later, and only the latest edit is kept. Caffeine amounts from the app's drink library are typical values for a drink and size. Travel across time zones could put a drink in the wrong timing group. Members chose to wear the Ultrahuman Ring and log their drinks, so this group likely pays closer attention to caffeine than people who do not track their drinks.

Frequently asked questions

Does late coffee raise your heart rate during sleep?

Sleeping heart rate showed no clear rise when the last caffeinated drink came under 4 hours before bed. On those nights, the difference from the same members' early-caffeine nights was −0.4 bpm (likely between −0.8 and 0.0 bpm), which is no clear change.

Does tea or an energy drink count, or only coffee?

Tea and energy drinks counted, along with every other caffeinated drink logged with at least 20 mg. On average, coffee was the last logged drink on 70.7% of each member's counted nights (likely between 69.1% and 72.3%). In a stricter version that kept only nights whose last drink was coffee, the gaps were larger. The Caffeine Window explainer describes how the app sets its window.

Who does this not apply to?

This finding does not cover teenagers, because the group was adult members. Each member also logged caffeine on at least 20 days, so people who drink caffeine only now and then are not represented. People whose bedtime moves from week to week, such as shift workers, fit the comparison less well, because each usual bedtime came from recent nights.

References