Comparison of the impact of different structural assumptions on the decision process, represented by the EIB and the EVPI. For models $\mathcal{M}{1}$ and $\mathcal{M}{3}$ the break-even point is nearly identical, while it is slightly larger for model $\mathcal{M}{2}$. is modelled as $$ \begin{aligned} &d{i} \sim \operatorname{Bernoulli}\left(\pi_{i}\right) \
&\operatorname{logit}\left(\pi_{i}\right)=\alpha_{0}+\sum_{j=1}^{J} \alpha_{j} x_{i j}
\end{aligned}
$$
where $\pi_{i}$ is the individual probability of producing positive costs and $\mathbf{x}{i}=$ $\left(x{i 1}, \ldots, x_{i, J}\right)$ is a set of $J$ relevant covariates (e.g. individual age, sex, comorbidities, etc.).
The second part models the distribution of costs for the subjects in $\mathcal{D}{\text {pos }}$ only. For example, for each of the $i=1, \ldots, n{\text {pos }}$ subjects with positive costs, we could define
$$
\begin{aligned}
&c_{i} \mid c_{i}>0 \sim \log \operatorname{Normal}\left(\mu_{i}, \sigma^{2}\right) \
&\mu_{i}=\beta_{0}+\sum_{l=1}^{L} \beta_{l} z_{i l},
\end{aligned}
$$
STAT0019 COURSE NOTES ：
As for the absorbent state TP, the associated cost is computed as a weighted average of the cost produced by patients in the other states under the status quo $(t=0)$, where the weights are given by the total number of individuals present in each
$$
c_{5 j}=\frac{\sum_{s=1}^{4} m_{j s}^{(0)} c_{s}^{(0)}}{\sum_{s=1}^{4} m_{j s}^{(0)}}
$$
Slightly abusing the notation, we can then define the total cost associated with each treatment as
$$
c_{t}=\sum_{j=1}^{J} m_{j S}^{(t)} c_{5 j}+\sum_{j=1}^{J} \sum_{s=1}^{S-1} m_{j s}^{(t)} c_{s j}^{(t)}
$$