The Cost-Effectiveness Case, and What Is Inside It

A cost-effectiveness analysis reports that tirzepatide saves $41,688 per patient and adds half a quality-adjusted life year compared with semaglutide.^[1] Before that figure gets used to argue anything, it is worth understanding what kind of number it is: nobody measured it. It was computed, from assumptions, and one of those assumptions is doing most of the work.

What a cost-effectiveness model does

It takes efficacy from clinical trials, assumes a level of adherence, projects the consequences across years or decades, attaches costs to each event avoided, and discounts it all back to a present value. The output is a single number that looks like a measurement and is a simulation.

That is a legitimate and necessary exercise — payers have to decide something, and waiting decades for outcome data is not an option. It also means the result is only as good as its inputs, and the inputs are choices.

A quality-adjusted life year is a modeled quantity. Nobody has ever observed one.

The assumption that undoes it

Models like this are fed trial efficacy. We published, days ago, what these drugs achieve outside a trial: 5.4% to 9.1% weight loss at one year, against the 14.9% and 20.9% their pivotal trials reported. That is in what these drugs do outside a trial.

And we published why: only 58% of people reach maintenance dosing, 21% never leave the starting dose, and discontinuation runs from 37% to 56% depending on the drug — in most people never reach the dose.

A model that assumes trial adherence computes savings that ordinary care does not generate. The events avoided are avoided in the simulation because the simulated patients kept taking the drug at the studied dose. Roughly half of real ones do not.

What is inside the total

The analysis includes reductions in absenteeism — days lost from work — and presenteeism, being at work while unwell and less productive.^[1] Both are legitimate societal costs and neither appears on a payer’s balance sheet.

That matters for how the figure is used. From a societal perspective, productivity losses are real money. From an insurer’s perspective, they are somebody else’s money, and an analysis that includes them will always look more favorable than the one a payer actually runs. Whenever you see a savings figure, the perspective it was computed from is the first thing to establish.

How to read any of these

• Whose costs? Societal analyses include productivity; payer analyses do not.
• What adherence was assumed? Trial adherence produces trial savings.
• Over what horizon? Lifetime models bank decades of avoided events from a year or two of data.
• Who funded it? Economic models are commissioned, and the commissioning party chooses the comparator.
• Does it beat doing nothing, or beat the alternative? Those are different claims and get conflated.

None of this means tirzepatide is not more cost-effective than semaglutide. It may well be — it produces more weight loss in trials and in ordinary care, which is the substance underneath the model. It means the specific dollar figure is a projection with a large adherence assumption inside it, and should be quoted as one.