19 2

share), oxaliplatin + 5-FU/LV (price of $25,400 and 17.8 percent market share), and 5-FU/LV (price of $75 and 13.6 percent market share).

To derive a naive price index that does not account for the changing attributes of the regimens, one would divide the mean prices in Figure 2 by $127, the mean price in the first quarter of 1993. The shape of the naive index is identical to that of Figure 2, with the only difference that it is indexed to one in the first quarter of 1993 and reaches 286 in the third quarter of 2005. That is, prices increased by 28,600 percent between the first quarter of 1993 and the third quarter of 2005.

Coefficient estimates from the hedonic regression are reported in Table 2. The dependent variable is the logarithm of the price a customer paid for regimen j in quarter t. Regimen attributes are included as well as a full set of quarter indicator variables. We also interact the second line therapy indicator with a regimen’s response rate.^{19 }

The coefficients on two of the three efficacy measures are positive and signifi- cant. An increase of one month in the median patient survival is associated with a 64.2 percent increase in the price of a regimen.^{20 }Evaluated at the mean regi- men price in the sample ($21,113), this implies an increase of $13,555. Physicians are implicitly valuing an expected year of life saved at $162,700. Regimens with relatively high response rates are also priced higher, with the effect smaller for second-line therapies.

The coefficient on time to progression is negative, which seems to indicate that physicians assign a negative valuation to that attribute. Pakes (2005) shows that coefficients in hedonic regressions will not necessarily have their expected signs. His insight is that the degree of competition will differ across the attribute space. In the situation of colorectal cancer, for example, there may be greater differenti- ation and less competition with the survival and response rate attributes relative to time to progression. Pharmaceutical and biotech firms design their products

We did not include other second-line attribute interactions due to multi-colinearity. 0.642 = exp(0.496) - 1.

22