A large share of reimbursement assumptions do not survive contact with payers. Comparator choice — which existing therapy the payer benchmarks against — moves the outcome more often than clinical differentiation does.
What decides a coverage decision?
Budget impact against a chosen comparator, more than clinical superiority considered in isolation. A therapy can be demonstrably better and still fail on the arithmetic of what it displaces.
That is not payers being obtuse. A formulary decision is a resource allocation across a population, and superiority per patient does not settle it.
The consequence for evidence generation is that a trial designed to demonstrate superiority may be answering a question the payer will not ask. Where the decision turns on budget impact against a specific comparator, evidence about that comparison — including cost offsets and where the therapy sits in the pathway — carries more weight than a larger effect size against a different reference.
Why does comparator choice matter most?
Because it sets the price the therapy is measured against. Benchmarked to an established generic, a new entrant faces a very different conversation from one benchmarked to a recent branded therapy — on identical evidence.
The comparator is chosen by the payer, not by the manufacturer, and manufacturers routinely model against the one they would have picked.
The comparator is also not fixed across payers. The same therapy can be benchmarked differently in two systems, or in two plans within one system, according to what each currently reimburses. That variance is knowable in advance and it is what makes early conversations valuable: the useful output is not one answer but the range of comparators a submission will have to survive.
How is budget impact assessed?
Over a defined horizon, against the population expected to receive the therapy, net of what it replaces. The horizon matters: a therapy that saves money in year four and costs money in year one is assessed on the year the payer is budgeting for.
Eligible population estimates are the other lever, and they are frequently the point where a manufacturer's model and a payer's diverge most. Sizing that population from primary work runs into the same incidence arithmetic as any narrow screen.
Horizon and population interact in a way that is easy to miss. A therapy with a favorable long-run case and an eligible population larger than the manufacturer assumes can fail on both counts at once, because the first-year cost scales with the population while the offsetting savings do not arrive until later. Testing the two assumptions separately hides that interaction.
How early can you test this?
Earlier than most launch teams do. Payers can describe how they would assess a therapy well before they formally assess one, which makes this among the cheapest assumptions to de-risk — cheaper, and more reliable, than asking prescribers what they would do.
The barrier is usually internal — market access work commissioned late because the clinical program dominates the calendar — rather than any unwillingness on the payer side.
Early conversations also have a lower access cost than teams expect. A payer being asked how they would assess a hypothetical therapy is not being asked to commit to anything, which makes the conversation easy for them to have and easy to arrange. The same person becomes considerably harder to reach once a live submission is in front of them.
Who is the right payer contact?
Formulary and health-economics staff rather than clinical leads. The clinical view informs the decision; the budget-impact model usually makes it, and the people who build it are a different population.
Former staff are frequently more candid on process than current ones — which is why a screened call reaches them first — and process is what determines how a submission is received.
The distinction between current and former staff is worth handling carefully. Current staff describe policy as it is written; former staff describe how it was applied, including the parts that never appear in guidance. Both are useful, and a program that reaches only one is either reading the rulebook or reading the anecdote.