A new drug versus current practice. Five uncertain parameters, one set of results, shown three ways.
Tornado: each parameter moved across its plausible range, one at a time
| Parameter | Low | Base | High | ICER swing |
|---|
Read the widest bar first: that is the parameter the answer actually turns on. Everything below it matters less than the room usually assumes. What one-way analysis cannot do is tell you a probability, or let parameters move together, and in reality they do. For that you need the second tab.
Every simulation, plotted on the cost-effectiveness plane
The same simulations, counted at every threshold
The two charts are the same data. Drag λ and watch the dashed line sweep the cloud while the
dot walks the curve: the height of the curve at any λ is simply the share of the cloud that falls
below the line there. That is all a cost-effectiveness acceptability curve is, and it is counted on
net monetary benefit (λ·ΔE − ΔC > 0), not on the ratio, because negative and infinite ICERs
make counting on the ratio incoherent.
What it does not tell you. Not the probability that the drug works, only that it is good value
given the model. It captures parameter uncertainty and nothing else: if the model's structure is
wrong, every one of these simulations is wrong together. A curve hovering near 50% across all
plausible thresholds is an argument for more research, not a coin toss.
Parameters here are illustrative, chosen to show the mechanism. For a real Indian example, read
Gupta N, Verma RK, Gupta S, Prinja S. Cost effectiveness of trastuzumab for management of breast
cancer in India. JCO Glob Oncol 2020;6:205–16 (PMID 32045547) — a 4% probability of being
cost-effective on one effectiveness source and 57.3% on another, rising to 90% with a 15–35% price
cut. Same drug, same country, same year.