Low end-tidal carbon dioxide predicts death, and not through the blood pressure

Design

retrospective single-centre cohort study, multivariable logistic regression with a prespecified interaction test and sensitivity analyses

Population

185,455 adults having non-cardiac surgery under general anaesthesia with mechanical ventilation, 2010–2020

Primary outcome

in-hospital mortality (0.85% overall)

Effect

adjusted odds ratio 1.63 (95% CI 1.36–1.86) per 5 mmHg decrease in mean end-tidal carbon dioxide below the median, independent of minute ventilation and hypotension; no interaction with hypotension (P = 0.19)

Low intraoperative end-tidal carbon dioxide has been linked to postoperative mortality before, and the objection has always been obvious: it could simply be a marker of hypotension, which we already know predicts death, or of how the patient was ventilated. A retrospective cohort of 185,455 adults having non-cardiac surgery under general anaesthesia with mechanical ventilation at one tertiary centre between 2010 and 2020 was designed to test exactly that.

In-hospital mortality was 0.85%. Lower mean intraoperative end-tidal carbon dioxide was associated with mortality in a non-linear way, with an adjusted odds ratio of 1.63 (95% confidence interval 1.36 to 1.86) per 5 mmHg fall below the cohort median — after adjustment for minute ventilation and for both the severity and duration of intraoperative hypotension. There was no interaction between end-tidal carbon dioxide and hypotension (P = 0.19), meaning the two carry separate information rather than one explaining the other. The result held across sensitivity analyses.

What this does not do is tell you to raise the capnograph reading. A low value in a well-ventilated, normotensive patient most plausibly reflects reduced pulmonary blood flow or dead space — that is, a sicker patient — and treating the number by changing the ventilator would address the reading rather than the physiology. The usable action is interpretive: a persistently low end-tidal carbon dioxide during a case, unexplained by ventilation and unaccompanied by hypotension, is information about that patient's risk that the arterial trace is not giving you, and it belongs in the handover and in the postoperative destination decision.

• Read the capnograph as a perfusion signal, not only as a ventilation check.
• Look for a cause when the value is persistently low: hypovolaemia, embolism, reduced cardiac output, raised dead space.
• Do not correct the number by adjusting minute ventilation and consider the problem addressed.
• Record the intraoperative end-tidal carbon dioxide trend in the handover, alongside pressure and blood loss.
• Let an unexplained low value influence level-of-care decisions after surgery, not just intraoperative management.

Why it matters

It says the capnograph carries prognostic information the arterial pressure does not, which is not how most of us read it.

Don't overread it

This is an association in retrospective data — there is no evidence that raising the end-tidal carbon dioxide changes the outcome.

The statistics, in plain English

An odds ratio of 1.63 per 5 mmHg applies to an event with a baseline rate of 0.85%, so the absolute risk change is small for any individual patient — this is a risk marker, not a death sentence. The association being non-linear means the relationship is steeper at some values than others, so a single odds ratio flattens something more complex. Most importantly, this is a retrospective single-centre cohort: adjustment for minute ventilation and hypotension removes two obvious explanations but cannot remove the possibility that low end-tidal carbon dioxide is simply a marker of illness that was not otherwise recorded. That it survives adjustment makes it a useful signal; it does not make it a target.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.