Table 3. Life Expectancies (LE)^* (years) Used in the Decision Model^**.

	Placebo	Mitoxantrone	Abiraterone	Cabazitaxel
LE_death	0.753	0.778	0.809	0.843
Range/distribution	(0.5648, 0.9413) Normal distribution	(0.5835, 0.9725) Normal distribution	(0.6068, 1.0113) Normal distribution	(0.6323, 1.0538) Normal distribution
LE_survival	1.917	2.209	2.64	2.768
Range/distribution	(1.4378,2.3963) Normal distribution	(1.6568, 2.7613) Normal distribution	(1.9800, 3.3000) Normal distribution	(2.0760, 3.4600) Normal distribution
LE overall	1.021	1.1178	1.47	1.593

^*For the deceased group, we determined the LEs by calculating the area under the curve (AUC) at 18 months normalized by the percent of the deceased population [(total AUC- percent of survival at 18 months X 18 months)/percent of deceased at 18 months]. We used the Declining Exponential Approximation of Life Expectancy (DEALE) method to approximate the overall LEs of each treatment group based on survival rates at 18 months. The DEALE model is a good approximation for our study since these patients have high mortality rates which are shown to be more accurate when using the declining exponential function. [16] , [17] . The subtraction of the weighted average of the LEs of the deceased group from the overall survival at 18 months gives the weighted average of the LE of the surviving group. From there, we derived the LEs of the surviving group by dividing by proportion surviving post 18 months.

^**We used the DEALE method to derive the overall life expectancies of the two subgroups differentiated by BPI score. We then extrapolated the life expectancies of the patients who died before 18 months and those remaining alive assuming a perfect declining exponential curve. Since median survival by baseline pain was only reported for abiraterone but not for the mitoxantrone or cabazitaxel groups, we assumed that all the treating groups had the same difference in their median survival between their individual subgroups with and without baseline pain (4.1 months). This was chosen as a more conservative estimate across all treatments than if we had used the abiraterone reported difference by baseline pain presence. With the overall survival derived from the each treatment groups median survival using the DEALE method, and the same difference of 4.1 months between the subgroups with baseline pain and no baseline pain for all treatments, we were able to derive the respective life expectancies of the subgroups with baseline pain and no baseline pain for each treatment group. From there, we further calculated the life expectancies of the patients who died before 18 months and who lived beyond that point for each subgroup as described previously. We then used these new life expectancies differentiated by baseline pain in our secondary decision tree analysis (results not shown).