Start with what this is. A Markov cohort model runs a hypothetical population through decades of simulated health states and adds up the costs and the quality-adjusted years. It is a structured argument with numbers attached, not an observation of anyone [1]. This desk has treated other models the same way in the cost per quality-adjusted life year.
With that established, the interesting part is not the headline ratio. It is what the model had to assume to get there. Japan’s Optimal Use Promotion Guidelines restrict who is eligible for semaglutide 2.4 mg and cap how long they can stay on it. So the authors modeled zero, one or two retreatment courses, each followed by a fixed year off. The two published ratios divide out at roughly 1.34-fold: adding type 2 diabetes to the modeled cohort raises the cost per quality-adjusted year by about a third.
The modeled population was people with a BMI of 35 or more plus at least one obesity-related comorbidity, or a BMI of 27 or more plus at least two obesity-related disorders — a narrower group than the sellers on this desk will accept. The intervention was semaglutide 2.4 mg with diet and exercise, against diet and exercise alone. Complication costs came from Japanese sources, which is the right choice and also the reason the ratio does not travel — the same boundary this desk draws in the cost-effectiveness threshold question.
On that point, this desk will not convert the yen figure to dollars. A cost-effectiveness ratio is only meaningful against the willingness-to-pay threshold of the system that produced it, and Japan’s threshold, prices and complication costs are its own. An exchange rate applied to a health-economic ratio produces a number that looks informative and is not.
The authors flag the uncertainties themselves: retreatment efficacy and long-term discontinuation are the weak joints in the model, and both are exactly the places where real data is thin. For a buyer, the transferable lesson is about structure rather than price. Any value case for these drugs assumes a pattern of use over years, and the pattern most people actually follow is the one in the dose people stay on — which is not the pattern any model assumes.