In-silico Phase 4 program using digital twins
Typically, manufacturers start to seriously think about how to set an asking price for a new treatment (drug or biologic or combination) around the time of the submission of their P3 registration data. At that point they have little real evidence of the benefits and risks of the new treatment, or the consequences of the introduction of the new treatment for payers and ACOs. The reason is that a P3 submission is the presentation of an efficacy signal extracted from an environment much simpler than the real world.
For designers of P3, a more homogeneous population (disease baseline data, “demographic” characteristics) is desirable as it favors the detection of a treatment signal, but it hurts the generalizability of the latter to the real world. Regulatory agencies take this tension into account, as best they can, in restricting the claims of the manufacturer to a population that closely resembles the P3 population. The manufacturer in turn will try to expand those claims with a “post marketing” or P4 research program.
The question is: Is it possible to do an in silico P4 program before launch that would help manufacturers and/or payers base their price negotiations? We believe it should be possible to achieve with access to P3 data and insurers health claims databases.
Information collated on P3 participants by the manufacturer (trial designer) could be used to build (elementary) digital twins which could be enriched with pathway information from their insurers in all health domains, not just the disease for which a new treatment was sought (targeted disease). These digital twins in turn could be used as templates for “corralling” all (outside of P3) digital twins with the targeted disease who were followed by insurers over a period of time (in the past) long enough to observe their completed health claim pathways during that period, including the (real world) emergence of the targeted disease, its recurrence, if any, its complications etc.. At that point, we would have assembled all P3 eligible participants in one or several geographies from one or several insurers databases using the P3 participants characteristics as seeds; let us call it the P3 replication targeted disease population or P3RTDP.
For reasons of trial design mentioned above (homogeneity, among others), insurers databases are expected to reveal that more patients who had the “targeted” disease could not have made it into a P3 trial because of stringent entry criteria; let us call it the P3 non eligible targeted disease population or P3NETDP. Its size in relation to P3RTPD is an index of generalizability of P3 data.
With these two populations in hand, it should be possible to measure the consequences of the new treatment by imputing the reduction or disappearance of costs secondary to treatment failure, re-retreatment, complications etc.. (as observed in P3) to the P3RTDP and P3NETDP over a period of several years. Tweaking of the drug efficacy (modeling) could be done to evaluate the consequences of overestimating or underestimating drug treatment effectiveness in the P3RTPD and the P3NETDP. These models could serve to establish the boundaries of health care cost reductions from putting the new drug on formularies and the expected value-based price at launch. Over time, these models would be supplanted by real world data obtained by insurers in the population of interest. These real-world data would serve as a basis of renewed price negotiations, formulary entry etc…
Chief Medical Officer and Co-founder
Pharma industry veteran with 30+ years in large Pharma & in leading small biotechs, spearheading large initiatives and securing funding, psychiatry practice and research for 10+ years.