Aerobic decoupling: what cardiac drift tells you, and what it doesn't

Late in a long ride your heart rate creeps up while your output does not. That divergence has a name, a formula and one widely quoted threshold — and it also has confounds serious enough that most tools showing you the number are quietly overselling it. Here is the metric, and where it stops being trustworthy.

The metric

Aerobic decoupling — also called cardiac drift, or Pw:Hr decoupling — was popularised by Joe Friel and TrainingPeaks. The procedure is simple:

• Split a steady aerobic effort into a first and a second half.
• For each half, compute an efficiency factor: EF = output ÷ heart rate.
• The decoupling percentage is how far the second half's EF has fallen below the first's.

decoupling % = (EF_first_half − EF_second_half) / EF_first_half × 100

A positive number means you produced less output per heartbeat as the effort wore on: the heart is working harder to sustain the same work. That is the signature of the aerobic base giving out.

Friel's 5%

The rule of thumb: 5% or more decoupling marks an effort as "coupled poorly", a sign the aerobic endurance base needs work; below about 5% the effort is well coupled.

That 5% is worth stating precisely because it is the only genuinely sourced number in this whole area. Everything else you will read about decoupling — including the rest of this page — is someone's defensible choice about how to apply it.

The confound that makes this hard on a bike

EF is defined on power. The speed substitution (Pa:Hr) is possible, and the same sources that describe it warn, in the same breath, that on a bike it is confounded by hills, wind and drafting.

Think about what that means for a real ride. A rider crawling uphill at 9 km/h and descending at 55 km/h has a wildly swinging speed that reflects the road, not their heart. Compare the second half of a hilly ride against the first on speed alone and you are mostly comparing the terrain of the two halves. The number will look like physiology and be mostly topography.

How tracé computes it without a power meter

A GPX file contains position, elevation, time and sometimes heart rate. There is no power meter and no rider weight, and the app does not ask for either. So the naive route — decouple on speed — is exactly the one the research rules out.

Instead the app substitutes something better than speed and worse than a power meter: Ferrari's gradient-normalised estimate of relative power, computed per climb.

est. W/kg ≈ VAM / (200 + 10 × gradient%)

It is a sourced formula, it needs no rider weight, and — the reason it belongs here — it is normalised for gradient, which is precisely the confound that makes raw speed useless for this. See the companion guide on VAM and estimated W/kg for how that conversion works and what it ignores. The metric then becomes:

• Efficiency per climb = estimated W/kg ÷ average heart rate.
• Climbs are split into first-half and second-half by the ride-distance midpoint — the honest analog of Friel's first-and-second-half-of-effort split.
• Decoupling % is the drop in median efficiency from the first-half climbs to the second-half climbs. Median rather than mean, so one exceptional climb does not redefine the day.

Where it is weakest, stated plainly

• Ferrari's estimate is weakest on shallow gradients, where aerodynamic drag dominates and a headwind or a fast group changes the answer. Detected climbs average at least 3%, which is the low end of where the formula is usable at all.
• It is climbs-only. VAM, and therefore estimated W/kg, is defined only on sustained ascent, so there is no continuous whole-ride efficiency curve to plot. The read is a handful of discrete points, not a trace.
• It needs heart rate. Without it the feature is absent rather than shown empty.
• It needs enough climbs on both sides of the midpoint. A read built on one early climb and one late climb is not a read.
• The heat and dehydration confound remains. Disclosed, not modelled.

The verdicts, and the thresholds behind them

A negative decoupling means efficiency rose in the second half. That usually indicates a conservative or not-yet-warm first half — common, and emphatically not a fatigue finding.

Two more rules, both ours:

• Two qualifying climbs per half, minimum. Fewer than two on either side of the ride-distance midpoint and no number is given at all. Two per half is the floor at which a median is more than a single sample; "faded from one climb to one climb" is not a finding.
• Interrupted climbs are set aside. A climb whose stopped time exceeds 10% of its elapsed time is excluded from both medians — its average heart rate describes a café, not a rider. It is still shown, greyed out, so you can see that it exists and was set aside. Numbers stand; judgement does not.

How to read your own number

Decoupling is a comparison within one ride, on one day, under one set of conditions. Used that way it is genuinely informative: a ride where your late climbs cost noticeably more heartbeats per watt than your early ones is a ride your aerobic base was stretched by. Used as a fitness score, or compared against someone else's figure, it is noise dressed as physiology — and on a hot day, it may be telling you to drink more rather than to train more.

Sources

Aerobic decoupling, the efficiency-factor definition and Friel's 5% threshold: TrainingPeaks — Aerobic Endurance and Decoupling.