What 97% looks like in real prescribing data

In January 2021, English GP practices dispensed 138,682 GLP-1 receptor agonist items at a cost of £12.0m. By June 2025 that had reached 393,871 items and £38.5m a month. The class grew 184% by volume and 221% by cost in four and a half years.

That is the number most forecasts capture. It is also the number that hides everything interesting. Underneath a smoothly growing class, the molecules were changing places.

Three things happened at once

Liraglutide collapsed. It opened the period at 35,528 items a month. Through 2021 and 2022 it drifted gently downward, the sort of decline a mature molecule shows. Then from October 2023 it fell off a cliff: 24,089 items in September, 14,780 in October, 7,518 by December, and 1,049 by June 2025. A 97% loss of volume, most of it inside nine months.

Tirzepatide arrived and took the market. Effectively absent before February 2024, it reached 73,147 items by month six, 145,149 by month twelve, and 215,103 by June 2025, making it the single largest GLP-1 in English primary care, ahead of semaglutide.

Semaglutide stopped growing. After climbing steadily to about 135,000 items a month by mid-2024, it went sideways: 127,258 in June 2025, essentially flat for a year while the class around it kept expanding. Dulaglutide, meanwhile, peaked at 111,259 items in June 2023 and has since fallen 56%.

The same class, a different market

Run the same analysis on US Medicaid and the pattern rhymes but does not repeat. Liraglutide is down 69% since early 2023 and dulaglutide 32%, the same generational displacement. But the leadership position is reversed.

In England, tirzepatide overtook semaglutide within fifteen months of launch. In US Medicaid it is still closing a gap, growing faster (2.09m scripts a quarter, up 18-fold in three years) but from further behind, against an incumbent that is itself still growing. Two reimbursement systems, two adoption curves, one molecule.

Why this matters for a budget impact model

Most budget impact submissions handle a new entrant with a market-share assumption: the product takes x% of the class by year three, and the class grows at y%. Both numbers are usually defended with analogues chosen from memory or from a competitor's published forecast.

The data above supplies three things that assumption sets normally lack:

Real uptake curves, not smooth ones. Tirzepatide reached 67% of its eventual peak by month twelve. Daridorexant, launched into a different therapy area in the same window, reached 45%. Inclisiran reached 13%. These are three genuinely different shapes, and picking the wrong one moves a three-year budget impact by a wide margin.

Displacement, not just addition. Liraglutide's 97% fall is money that left one line of a formulary and appeared on another. A model that grows the class without modelling where the switching comes from will overstate net budget impact, sometimes severely.

Market-specific behaviour. The England and US curves for the same molecule differ enough that a single global uptake assumption is indefensible. Analogues have to be drawn from the market you are submitting to.

None of this requires proprietary data. Every figure on this page comes from national prescribing datasets that any organisation can download: England's under the Open Government Licence, the US Medicaid file in the public domain. What it requires is the work of assembling them, mapping products to molecules, and aligning launches on a common clock.