The clinical trial cost/decision-making model described above requires numerous data points, including phase durations, success probabilities, expected revenues, and a discount rate, as well as a full range of itemized costs associated with clinical trials, broken down by phase and therapeutic area. The model uses a real annual discount rate of 15 percent based on input from interviews conducted with drug sponsors as default, and we were able to obtain some of the other data needed from the available clinical research literature. Phase durations were one such parameter. Though they are not differentiated by therapeutic area, DiMasi, Hansen, & Grabowski (2003) provide mean phase lengths of 21.6 months (1.8 years) for Phase 1, 25.7 months (2.1 years) for Phase 2, and 30.5 months (2.5 years) for Phase 3. The NDA/BLA review time, as we are defining it,6 includes the time from first submission of an NDA/BLA to regulatory marketing approval, and comes from DiMasi, Grabowski, & Vernon (2004). Trial phase times generally do not reflect differences between therapeutic areas; however, therapeutic-area-specific NDA/BLA review times were available and used for a select list of therapeutic areas.
Clinical trial success probabilities are available from two recent studies, one conducted by DiMasi and colleagues (Tufts University) in 2010 (DiMasi, Feldman, Seckler, & Wilson, 2010), and another one conducted by BioMedTracker in 2011(Hay, Rosenthal, Thomas, & Craighead, 2011). The two studies, however, provide different success rate estimates—for example, DiMasi, et al. (2010) found an overall success rate of 19 percent, while Hay and colleagues (2011) arrived at nine percent. The differences in the two studies can be attributable to the fact that they were drawing from different pools of data. DiMasi, et al. (2010) collected data on 1,738 drugs that entered Phase 1 between 1993 and 2004 and were developed by the 50 largest pharmaceutical companies. The BioMedTracker study covered 4,275 drugs from biotechnology and pharmaceutical companies of all sizes. The drugs included were in any phase of development between October 2003 and December 2010 (Hay, Rosenthal, Thomas, & Craighead, 2011).
As the BioMedTracker study was more recent and included more drugs and a broader range of companies, we opted to use the success probabilities reported by BioMedTracker in our model. These success probabilities were broken down by clinical trial phase and, for Phase 2 and Phase 3, by therapeutic area as well. For Phase 1, we used 67 percent for all therapeutic areas. For Phases 2 and 3 and the NDA/BLA review phase, we used therapeutic-area-specific percentages where available and general success probabilities (41, 65, and 83 percent, respectively) for therapeutic areas for which no specific probabilities were reported. All probabilities used in the model were for lead indications.
In order to construct the model’s “baseline scenario,” we obtained itemized clinical trial cost data from Medidata Solutions (hereafter “Medidata”), which compiles data from a portfolio of CRO contracts, investigator grants/contracts, and clinical trial protocols. Medidata Grants Manager’s database, PICAS® , and CRO Contractor’s database, CROCAS® , contain numerous data elements derived from actual negotiated contracts, and these resources are widely used by pharmaceutical companies, CROs, and academic researchers to identify prevailing rates for trial planning, budget development, and grant negotiation (Medidata Solutions, 2012). We obtained the number of clinical investigator sites per study/protocol from Medidata Insights™, based on 7,000 study protocols that allows numerous views of study performance metrics on demand, by therapeutic area, study phase, geography and more.
The custom tabulation received from Medidata contained means and variances for a wide range of clinical trial cost elements, including study-level costs (such as IRB approvals and source data verification (SDV) costs), patient-level costs (such as recruitment and clinical procedure costs), and sitelevel costs (such as monitoring and project management). Number of planned patients per site and number of sites per study were also provided. A complete list of these data elements can be found in Appendix B, along with more detailed descriptions of each field, unit specifications, and sources. The data are from 2004 and later and have not been adjusted for inflation by Medidata. As the data points represent averages across this range of time and cannot be assigned specific years, we were unable to adjust them for inflation, which is one of the study limitations.
Medidata provided means and variances of costs by trial phase (Phases 1 through 4), geographic region (U.S., global, and rest of world), and therapeutic area. For the purposes of this analysis, we focused on the data points specific to U.S. trials. The therapeutic areas for which Medidata provided data were: anti-Infective, cardiovascular, central nervous system, dermatology, devices and diagnostics7 , endocrine, gastrointestinal, genitourinary System, hematology, immunomodulation, oncology, ophthalmology, pain and anesthesia, pharmacokinetics8 , and respiratory system. To the extent possible, we attempted to match the success probabilities by therapeutic area (from BioMedTracker) to the therapeutic area categories used by Medidata. Some additional data cleaning steps were performed using the statistical software STATA; these are outlined in Appendix E.
On the revenue side, we used an estimate from a study by DiMasi, Grabowski, & Vernon, (2004), which reports worldwide sales revenues over the product life cycle for new drugs approved in the United States during the period from 1990 to 1994. Figures were available for some specific indications; for the others, we used the reported figure for “All Drugs.” The numbers reported by DiMasi, Grabowski, & Vernon (2004) are NPVs, discounted at 11 percent to the launch year; however, they are in year 2000 dollars. Therefore, we inflated the revenue figures to 2008 dollars (the midpoint between 2004 and 2012, the range covered by the itemized cost data) using the producer price index for commodities in the category “Drugs and Pharmaceuticals” from the Bureau of Labor Statistics (BLS) (series WPU063).
6 From FDA’s perspective, each submission has a set time period (priority or non-priority review) that does not include time between submissions; however that time is included in our definition of the NDA/BLA review phase time for the purposes of this analysis.
7 The “Devices and Diagnostics” category includes any industry-sponsored studies where a device or drug delivery system is being studied instead of a drug. Among the devices included in this category are stents, implants, joint replacements, inhalers, and blood sugar monitoring devices.
8 Pharmacokinetic (PK) studies are often conducted at the discovery or candidate selection stages of a development program. These studies look at the mechanisms of absorption and distribution of a drug candidate as well as the rate at which a drug action begins and the duration of this effect.