Data-Driven COVID-19 Vaccine Development for Janssen

The COVID-19 pandemic has spurred extensive vaccine research worldwide. One crucial part of vaccine development is the phase III clinical trial that assesses the vaccine for safety and efficacy in the prevention of COVID-19. In this work, we enumerate the first successful implementation of using machine learning models to accelerate phase III vaccine trials, working with the single-dose Johnson & Johnson vaccine to predictively select trial sites with naturally high incidence rates ("hotspots"). We develop DELPHI, a novel, accurate, policy-driven machine learning model that serves as the basis of our predictions. During the second half of 2020, the DELPHI-driven site selection identified hotspots with more than 90% accuracy, shortened trial duration by six to eight weeks (approximately 33%), and reduced enrollment by 15,000 (approximately 25%). In turn, this accelerated time to market enabled Janssen's vaccine to receive its emergency use authorization and realize its public health impact earlier than expected. Several geographies identified by DELPHI have since been the first areas to report variants of concern (e.g., Omicron in South Africa), and thus DELPHI's choice of these areas also produced early data on how the vaccine responds to new threats. Johnson & Johnson has also implemented a similar approach across its business including supporting trial site selection for other vaccine programs, modeling surgical procedure demand for its Medical Device unit, and providing guidance on return-to-work programs for its 130,000 employees. Continued application of this methodology can help shorten clinical development and change the economics of drug development by reducing the level of risk and cost associated with investing in novel therapies. This will allow Johnson & Johnson and others to enable more effective delivery of medicines to patients.

Course Scheduling Under Sudden Scarcity: Applications to Pandemic Planning

Problem definition: Physical distancing requirements during the COVID-19 pandemic have dramatically reduced the effective capacity of university campuses. Under these conditions, we examine how to make the most of newly scarce resources in the related problems of curriculum planning and course timetabling. Academic/practical relevance: We propose a unified model for university course scheduling problems under a two-stage framework and draw parallels between component problems while showing how to accommodate individual specifics. During the pandemic, our models were critical to measuring the impact of several innovative proposals, including expanding the academic calendar, teaching across multiple rooms, and rotating student attendance through the week and school year. Methodology: We use integer optimization combined with enrollment data from thousands of past students. Our models scale to thousands of individual students enrolled in hundreds of courses. Results: We projected that if Massachusetts Institute of Technology moved from its usual two-semester calendar to a three-semester calendar, with each student attending two semesters in person, more than 90% of student course demand could be satisfied on campus without increasing faculty workloads. For the Sloan School of Management, we produced a new schedule that was implemented in fall 2020. The schedule allowed half of Sloan courses to include an in-person component while adhering to safety guidelines. Despite a fourfold reduction in classroom capacity, it afforded two thirds of Sloan students the opportunity for in-person learning in at least half their courses. Managerial implications: Integer optimization can enable decision making at a large scale in a domain that is usually managed manually by university administrators. Our models, although inspired by the pandemic, are generic and could apply to any scheduling problem under severe capacity constraints.

A pre-registered short-term forecasting study of COVID-19 in Germany and Poland during the second wave.

Disease modelling has had considerable policy impact during the ongoing COVID-19 pandemic, and it is increasingly acknowledged that combining multiple models can improve the reliability of outputs. Here we report insights from ten weeks of collaborative short-term forecasting of COVID-19 in Germany and Poland (12 October-19 December 2020). The study period covers the onset of the second wave in both countries, with tightening non-pharmaceutical interventions (NPIs) and subsequently a decay (Poland) or plateau and renewed increase (Germany) in reported cases. Thirteen independent teams provided probabilistic real-time forecasts of COVID-19 cases and deaths. These were reported for lead times of one to four weeks, with evaluation focused on one- and two-week horizons, which are less affected by changing NPIs. Heterogeneity between forecasts was considerable both in terms of point predictions and forecast spread. Ensemble forecasts showed good relative performance, in particular in terms of coverage, but did not clearly dominate single-model predictions. The study was preregistered and will be followed up in future phases of the pandemic.