T-45: A Decision Analytic Benefit-Risk Assessment Framework to Support Portfolio Prioritization Decisions
Poster Presenter
George Quartey
Scientific Enablement Leader
Genentech, A Member of the Roche Group United States
Objectives
To introduce a Decision Analytic Framework based Benefit-Risk Assessment (DAF-BRA) tool to incorporate relevant stakeholder preferences into decisions to progress pipeline to Phase 3, ensuring that they are systematic, explicit, transparent and, reflecting the relative success of these therapies.
Method
A DAF-BRA model was constructed for the evaluation of oncology combination therapies for treating diffuse large B-cell lymphoma (DLBCL). Preferences were elicited from 2 oncologists using the swing-weight approach. A probabilistic additive model was used to assess the value of the therapies.
Results
Criteria were selected based on discussion with experts, available phase 2 data and aligning with the requirements of an additive model. The criteria include 3 benefit outcomes and 6 safety outcomes. There was no dominating combination by looking at the performance data alone, thus value judgments were required to determine which combination performed better. These were elicited via a preference elicitation workshop, which revealed heterogeneity in the preferences between the stakeholders. One of the stakeholders put more weight on improving ‘Response Sustained’ and reducing ‘Grade 3-4 AEs’. While the other put more weight on reducing ‘Fatal AEs’ and ‘Discontinuation due to AEs’. Preferences for the other endpoints are relatively similar.
The DAF-BRA which combines performance and preference data into an assessment of the overall value of a combination, show that there are clear trade-offs among the therapies when the uncertainty regarding the criteria measurements is considered. This can be seen from the rank acceptability indices that differed only in the preference rank of the efficacy criterion relative to the risk criteria and from Probabilistic Sensitivity Analysis (PSA) generated scatter plots on the benefit-risk plane.
The overall values of the 2 combinations show that the treatments have slightly different overall values for the stakeholders, but result in different ranks. This suggests the ranking of therapies is sensitive to the stakeholder weights, in particular: sustained response and fatal AEs. The PSA provides an assessment of the probability that one therapy is preferred to another given the uncertainty on the performance estimates. One stakeholder is confident about their ranking of the drugs, with a 95% probability that they prefer Therapy B. While the other is less certain, with 57% probability that they prefer Therapy A. PSA generated scatter plots on a benefit-risk plane show that both drugs have a positive benefit-risk balance.
Conclusion
The multi-criteria DAF-BRA approach allows us to formally incorporate stakeholder judgments into the assessment of which therapies should be progressed to Phase 3. Criteria weights serve the purpose of scaling the criteria performance of each therapy, enabling the comparison of overall value. By incorporating parameter uncertainty into the model, the DAF-BRA can assess the probability that one regimen is judged more likely to succeed.
Understanding how the performance of the asset on different criteria, and stakeholders’ weight on the criteria, contribute to the assessment of the overall value of an asset can support internal decisions. This enables determining where differences in value judgments have an impact on the overall preference of an asset, allowing discussion and further data collection to be focused on the factors that matter.
To implement the DAF-BRA successfully, it is recommended that: (1) The development of the DAF-BRA should be triggered during Phase 2 or beyond when the endpoints available to the analysis are known. (2) the elicitation workshop captures a diversity of perspectives, reflecting manufacturer priorities, but also patient and regulatory perspectives on the endpoints. (3) Stakeholders should inform both the criteria selection and the value judgments.
In addition, the output can be used to determine whether further evidence is needed to support the quantitative BRA. For instance, collecting patient or regulator preferences for endpoints may provide additional value. In the DLBLC case study, one stakeholder viewed the benefit-risk balance more positively than the other. In this instance, there would be value in gaining a deeper understanding of how patients perceive the benefit-risk balance by collecting their preference data alongside the Phase 3 trial. The key learnings from this study will help advance the modeling of the “patient or payer perspective” and “expert clinical judgment” to support future portfolio-level decisions.
