How do I convert the SURMOUNT-OSA hazard ratio into an absolute risk reduction?

This is a question about interpretation, not about whether to act — I will take action questions elsewhere.

I want the working, not the result — I need to be able to redo it with different numbers.

I care about the precision as well as the value — I want to know how many figures are real.

How many significant figures are actually justified here?

Answer first: the cardiovascular signal in this class is a reduction in major adverse cardiovascular events in populations already at elevated risk, not a general cardioprotective claim for everyone.

Benefit appears to track exposure duration rather than switching on at a threshold, which is what you would expect from a mechanism working partly through weight, blood pressure and glycaemia rather than instead of them.

Baseline risk decides how much the relative reduction is worth. The same twenty per cent applied to a one per cent annual risk and a five per cent annual risk produce very different absolute numbers.

SELECT is the trial to read for primary-prevention-adjacent populations without diabetes; SUSTAIN-6 and LEADER are the diabetes-population outcome trials for semaglutide and liraglutide respectively.

Ask for the absolute risk reduction and the number needed to treat. If a source will not give you both, it is selling something.

Answering this properly needs the absolute risk reduction rather than the relative one, because the relative figure is stable across risk strata and the absolute figure is not.

Systolic blood pressure falls by about three to six millimetres of mercury at the doses studied, most of it early and much of it attributable to weight loss rather than to a direct vascular effect.

Resting heart rate rises by roughly two to four beats per minute across the class. The mechanism is not fully settled, the magnitude is consistent, and it has not translated into an adverse outcome signal in the trials that looked.

These are trial populations on licensed product. Research-grade material of unverified content is not the thing that was studied.

Read SELECT for the without-diabetes population and the earlier outcome trials for the with-diabetes one. They are not the same result.

edited 21 May 2024 by u100_marks — corrected a unit error in the worked example

The mechanism question and the outcome question are separate. The outcome data stand whether or not the mechanistic story is settled, and the mechanistic story is not settled.

The heart-failure question is separate again, and the evidence there is largely in the preserved-ejection-fraction population with obesity, where the trials measured symptoms and function rather than mortality.

Where a cardiovascular claim is made for an agent with no completed outcome trial, the honest statement is that the class evidence is suggestive and the agent-specific evidence does not yet exist.

A relative risk reduction quoted without the baseline risk is close to meaningless, and it is how most of these numbers travel.

Heart rate up a few beats, blood pressure down a few millimetres, events down about a fifth in high-risk populations. That is the honest summary.

Blood pressure falls by a few millimetres of mercury in this class, which is small individually and not small across a population.

SELECT randomised people with established cardiovascular disease and overweight or obesity but without diabetes to semaglutide 2.4 mg, and reported roughly a twenty per cent relative reduction in the primary composite. The absolute difference over about three years was on the order of one and a half percentage points.

Nothing here is medical advice. If cardiovascular risk is the actual question, it is a conversation for a clinician with your numbers in front of them.

Population, baseline risk, endpoint definition. In that order, then the effect size.