Three things before the number. This is China’s health system, not a US buyer’s. It is liraglutide against sitagliptin, not the molecules most people are choosing between. And it runs thirty years, roughly ten times longer than anyone has actually taken one of these — a horizon which is the whole reason a cost-per-event figure can look so different from a monthly bill.
What the model said
A Markov model simulated costs and quality-adjusted life-years for patients with type 2 diabetes not controlled on metformin alone. [1] Adding liraglutide rather than sitagliptin came out at an incremental cost-effectiveness ratio of 267,985.96 RMB per QALY gained.
That is below the willingness-to-pay threshold the authors used, three times China’s 2025 per capita gross domestic product, so within this framework the combination counts as cost-effective.
A third of the runs said no
The probabilistic sensitivity analysis re-ran the model thousands of times with the inputs varying across their plausible ranges. Liraglutide came out cost-effective in 63.6% of them.
Which means it did not in 36.4%. The authors write that it is “likely to be cost-effective”, which is the accurate phrasing for a result that close to the line, and a long way from the flat claim a marketing page would make of it — the difference between a number and the certainty it gets quoted with.
The line that matters to a buyer
The scenario analysis found the economic advantage became more pronounced as the price of liraglutide fell.
That is obvious and it is also the point. Cost-effectiveness is a ratio with price in the numerator, so the same drug is cost-effective at one price and not at another, and none of it is a fact about the molecule. On this market the price moves constantly between sellers, which is why the spread between them does more work than any modeled ratio, and why who is paying decides more than what the model concludes.