A Breath of Fresh Air? Firm Type, Scale, Scope, and Selection Effects in Drug Development

Service) registry number (when available) to distinguish me-too and second-generation molecules from first-in-class compounds (see Reddy 2003). it easier to improve upon existing remedies. We also expect first-in-class compounds to be drawn from an inferior distribution because there is less information about potential side effects. Unfortunately, in almost 40% of cases it is impossible to classify the molecule under study because its chemical characteristics are not available. To properly distinguish first-in-class compounds from followers, we create an indicator variable, first-in-class unknown, for when this information is missing. Through the use of these fine-grained controls at the level of indication, we can avoid using indication-fixed effects, which can bias nonlinear estimates, a possibility confirmed by (unreported) Monte Carlo simulations. Economic Characteristics of Therapeutic Area: We use information on the worldwide sales (at the ATC2 level) in the year the project is started as a proxy for V .15 We measure the degree of competition by the number of firms operating in that indication (worldwide) at the time the project is launched. We also employ the share of established pharmaceutical firms as an additional control for the intensity of competition. We use the share of projects developed with universities as a measure of the firm’s links with research institutes, and its closeness to science (closeness to science). These are hypothesized to affect only Pg∗, because they affect the profitability of a project rather than its likelihood of success. Specification and Identification of Pg∗ and : The system of equations we estimate is nonlinear. We 15 We use 1983 figures for projects started in 1980–1982 as well, because data for those years are missing. Arora et al.: Firm Type, Scale, Scope, and Selection Effects in Drug Development Management Science 55(10), pp. 1638–1653, © 2009 INFORMS 1647 Table 3 Descriptive Statistics Variable Mean SD Min Max Variable Mean SD Min Max Selection 0 48 0 50 0 1 Closeness to science 0 05 0 16 0 1 Success 0 34 0 48 0 1 Public firm 0 84 0 37 0 1 Pioneer Biotech 0 31 0 46 0 1 Market size 2 38 2 47 0 14 40 Established Pharma 0 46 0 50 0 1 Competitors (log) 3 09 1 02 0 5 23 Other Biotech 0 17 0 37 0 1 Share of established 0 28 0 18 0 1 pharma in market Other Pharma 0 06 0 24 0 1 First-in-class 0 30 0 46 0 1 Scale_project 2 67 3 53 0 31 First-in-class unknown 0 39 0 49 0 1 Scale_firm 36 23 46 19 0 586 Lethal 0 75 0 43 0 1 Scope 0 83 0 21 0 0.98 Organ damage 0 78 0 42 0 1 Presence in ATC2 0 18 0 39 0 1 Multiple causes 0 80 0 40 0 1 product market License 0 08 0 27 0 1 Chronic 0 79 0 40 0 1 preclinical only Rare 0 04 0 20 0 1 also impose three exclusion restrictions. We assume that the economic characteristics of the compound— i.e., the size of the market, the level of competition in the market, and whether the firm is private or public—only affect the selection threshold and not the distribution of the probability of success. This is how we identify selection (Pg∗) as opposed to performance ( ). Second, recall that we normalize the variance of ln Pg/ 1− Pg to unity. We do allow for interdependence across the observations for a firm by clustering standard errors at the firm level. Table 3 presents the descriptive statistics. It shows that about half the projects were selected for clinical trial, and of those, roughly one third were successful. Reflecting the dominance of established firms (see Table 1), Table 3 shows that, on average, a project was associated with over 2.5 projects in the same indication, and 8% of projects were licensed in preclinicals. 6. Empirical Results Table 4 presents simple probit estimates of the selection and success equations estimated separately. Compared to established pharmaceutical firms, biotech firms have a lower probability of selection and a lower probability of success, though the difference is not statistically significant. Firm scale increases the probability of success (though it leaves the selection probability unchanged), whereas program scale decreases both selection and success probability. Scope reduces selection but increases success. Projects for more profitable markets (larger size and lower competition) have higher probability of selection and success. The presence of downstream assets (a product already in the market) makes selection more likely but reduces success. It is tempting to try to interpret these results in terms of the various theories about differences across firms or firm scale and scope. However, as we discussed in developing our model, the probability of success will also depend on the selection threshold, and vice versa. For instance, is scope associated with higher success because it is associated with lower selection probability? If so, should not program scale (associated with lower success probability) have been associated with greater selection probability? The results in Table 4 are difficult to reconcile with a simple Heckman-selection model, in which the same underlying process drives both selection and success. In other words, we need to estimate the structural parameters of the full model described in §3, where we jointly model selection decisions and performance. Table 5 reports the results of pseudomaximum likelihood obtained using STATA. Note that Table 4 reports the drivers of the probability of selection and success, and Table 5 reports the drivers of the selection threshold and the innovative performance. For instance, the coefficients reported under probability of selection in Table 4 are − , whereas Table 5 reports (see §3). We estimate two specifications. In one, we use firm-type dummies (Established Pharma is the reference group) and disease controls only, whereas in the second we also include measures of scale and scope, and firm and market controls. 6.1. Selection Results We find that research scale at the program level reduces the selection threshold consistent with Hypothesis 1. However, firm scale and scope increase the selection threshold. Other Biotech and Other Pharma exhibit, respectively, a lower and a higher threshold than the Established Pharma. Other Pharma may have higher development costs than Established Pharma, perhaps because the former lack the strong links with reputed academic centers enjoyed by established pharmaceutical firms such as Lilly and Merck; they may also have limited production and marketing assets. By contrast, there is little difference in the selection process between Pioneer Biotech and Established Pharma. However, the coefficient of the Other Arora et al.: Firm Type, Scale, Scope, and Selection Effects in Drug Development 1648 Management Science 55(10), pp. 1638–1653, © 2009 INFORMS Table 4 Probit Estimates: Factors Affecting Selection (Pr Selection), and Factors Affecting Success for Selected Projects (Pr Success)