Menu Back to Poster-Presentations-Details

P225: Optimizing Site and Country Selection for Clinical Trial Planning: a Mixed Integer Linear Programming Approach

Poster Presenter

Alexandru Socolov

Senior Data Scientist
Medidata Solutions
United States

Objectives

Our approach utilizes Mixed Integer Linear Programming (MILP) to generate an optimized scenario that minimizes the number of sites and countries while respecting user-defined constraints including target enrollment, country/region preferences, and site enrollment categories (high/medium/low).

Method

The approach, coined Scenario Builder, utilizes forecasted monthly enrollment data and user inputs, including minimum/maximum site counts and country inclusion/exclusion. MILP transforms these into constraints and the open-source COIN-OR Branch and Cut (CBC) solver finds an optimal solution.

Results

To illustrate the power of this methodology, consider the following hypothetical scenario. A trial planner seeks to conduct a global Phase III clinical trial for a novel endometrial cancer drug. Using Scenario Builder, they would input specific target parameters: enrollment of 800 patients in 24 months, a requirement of at least five sites per country, and defining inclusion of the United States and Northern Europe while excluding certain countries. The planner additionally anticipates engaging user-defined tiers of 25/50/25% top/medium/low enrolling sites therefore the scenarios generated have to preserve this balance. The Scenario Builder turns the user inputs into constraints in a MILP optimization problem, leveraging forecasted enrollment data for sites experienced in endometrial cancer trials. The MILP’s objective function is to minimize the number of sites and countries in its solution. The CBC solver generates an optimal site selection, balancing enrollment capacity, geographical distribution, and user-defined constraints under the specified timeline. The result is a scenario that minimizes sites and countries while adhering to the planner’s minimum and maximum specifications. In this example, Scenario Builder’s solution identifies 412 sites across 14 countries, distributing enrollment in alignment with the trial's objectives and timelines. The United States and Northern European countries are included, while undesirable countries are excluded. The scenario preserved the 25/50/25% balance of predicted enrollment tiers among the selected sites. For Phase III endometrial cancer trials, the industry median number of sites and countries is 500 and 18, respectively. Therefore, this example shows that Scenario Builder has the potential to reduce the site and country footprint by 18% and 22%, respectively while keeping the scenario realistic for the given hypothetical study.

Conclusion

Built on the industry-leading Medidata historical clinical trial dataset, Scenario Builder presents a state-of-the-art MILP approach to enhance clinical trial planning by optimizing site and country selection. By formulating the problem as an optimization objective with constraints, the approach is a holistic framework that captures the effects of one decision on the other, e.g. the trade-off between more sites versus countries is explicitly accounted for. The approach has shown the potential to reduce the operational cost by reducing the number of sites and countries needed while keeping the scenario realistic and viable for execution. The Scenario Builder’s speed and ease of use accelerate the planning phase. The flexibility and breadth of user-defined constraints allow planners to iterate through multiple scenarios in the order of minutes and streamline the often complex process of site selection. The enrollment forecast that underpins the Scenario Builder is informed by 25,000+ clinical trials on the Medidata platform and provides a robust estimate of indication-specific site-level enrollment. Scenario Builder accounts for country and site startup times derived from site-level patterns, among other factors, but also allows manual augmentation with user-defined startup times. The next directions for development include accounting for future competition for patients due to new trial startups at the selected sites and incorporating patient diversity data to help planners meet their goals in recruiting a diverse population. Scenario Builder’s versatility is underscored by its adaptability to various user constraints. The example scenario demonstrates the method’s ability to align with specific trial requirements, ensuring a pragmatic site mix. This flexibility, coupled with the efficiency of the CBC solver, positions Scenario Builder as a transformative tool in the realm of clinical trial planning.