Select Projects

Select Ongoing Projects

A clinical decision tool to optimize the selection of antibiotics for patients with drug-resistant tuberculosis in low-resource, high-burden settings

This project, which is funded by an R01 award, has two primary goals: 1) to develop a systematic approach to translate surveillance, clinical, and epidemiological data into individualized, evidence-based, equitable, and cost-effective treatment recommendations at the point of care; and 2) to create a prototype of the proposed clinical decision support (CDS) tool, which is based on user-centered design principals, for implementation in practice. As part of this project, we develop methods that combine spatiotemporal machine learning and decision models to optimize the treatment of patients with multidrug-resistant tuberculosis (MDR-TB) in response to patient characteristics, the local epidemiology of drug-resistant TB, and the cost and toxicity of anti-TB drugs. In contrast to the current practice of treating MDR-TB according to standardized regimens, which are endorsed at the global level, these pioneering methods allow clinicians to optimize the treatment for patients with MDR-TB at the point of care and are based on each patient’s basic demographic and clinical information and the latest data from national surveillance systems of drug-resistant TB.

Measuring the disutility of non-pharmaceutical interventions for controlling epidemics in diverse populations

The standard approach to resource allocation in health care relies on cost-effectiveness analysis (CEA). Applying CEA to guide the use of nonpharmaceutical interventions (NPIs), however, is limited. CEA requires estimates for the long-term financial and health consequences of alternatives under consideration, which are not available for NPIs, especially among groups that are economically and socially marginalized. To address these limitations, we are conducting discrete-choice experiments (funded by an R21 award) among a sample of people of varying sociodemographic, clinical vulnerability, and economic vulnerability backgrounds to quantify the disutility of various combinations of NPIs, including school and business closures. We are recruiting a statistically valid representation of the U.S. population as well as many under-researched and harder-to-reach populations, such as minorities, people with disabilities, parents of teens and young children, and rural communities. This will allow us to better understand the disparity in how different population groups experienced the use of NPIs during the COVID-19 pandemic.

Improving the surveillance systems for antimicrobial-resistant gonorrhea

As part of this project, which is funded by an R01 award, we aim to develop methods to optimize surveillance systems for antimicrobial-resistant (AMR) gonorrhea and maximize the utility of the data provided by these surveillance systems. The national surveillance systems for AMR gonorrhea in the U.S. (namely, the Gonococcal Isolate Surveillance Project (GISP)) include only 30–40 surveillance sites. As such, for most areas, there is no information about the burden of AMR gonorrhea. Our current research seeks to answer two fundamental questions: 1) How should surveillance sites be added or removed over time to provide accurate information about the spatial and temporal changes in the prevalence of AMR gonorrhea, and 2) How could the data provided by existing surveillance sites be used to project the burden of AMR gonorrhea in cities with no surveillance sites? 

Select Completed Projects

Predicting local surges in COVID-19 hospitalizations

Early in the COVID-19 pandemic, predictive models of COVID-19 hospitalizations have focused almost exclusively on national- and state-level predictions during the COVID-19 pandemic. This left local policy-makers in need of tools that can provide early warnings of the possibility that COVID-19 hospitalizations might increase to levels exceeding local capacity. Some risk metrics, such as the CDC’s Community Levels, were developed to predict the impact of COVID-19 on community-level health care systems using routine surveillance data. However, they had limited utility, as they were not routinely updated with accumulating data and were not directly linked to specific outcomes, such as surges in COVID-19 hospitalizations beyond local capacities. In collaboration with the Council of State and Territorial Epidemiologists (CSTE), we developed machine learning models and a webtool to identify U.S. counties at high risk of surpassing hospitalization capacity due to COVID-19. These predictions were publicly available and updated weekly by our team throughout 2023 (https://yaesoubilab.shinyapps.io/predicting_covid_hosp_surges/). We showed that these classifiers maintained their performance temporally (i.e., over the duration of the pandemic) and spatially (i.e., across U.S. counties). 

Methods for adaptive decision-making during pandemics 

The cost-effectiveness of specific control interventions may change markedly over time. Physical distancing measures, for example, may play an important role in reducing disease transmission during the early stages of pandemics, although the benefit diminishes as the pool of susceptible individuals is depleted (either because of the deployment of a vaccine or unabated progression of the epidemic). As such, these interventions should be employed only when their health benefits outweigh their high social and economic costs. To facilitate a decision-making process that can be optimally responsive to new data accumulating over the course of an epidemic, we developed a dynamic decision model to define and optimize adaptive policies that make recommendations using the latest epidemic data. In a series of papers, we demonstrated how these adaptive policies can be characterized computationally efficiently, via reinforcement learning and simulation optimization methods.

Informing the cost-effective use of polyvalent meningococcal vaccines in the African meningitis belt

The introduction of a conjugate vaccine for serogroup A Neisseria meningitidis has dramatically reduced disease in the African meningitis belt. In this context, important questions remain about the performance of different vaccine policies that target remaining serogroups. In this project we developed mathematical and agent-based models that describe the key characteristics of meningococcal epidemics within districts of Burkina Faso and Niger. Our findings indicated that these novel polyvalent conjugate vaccines can be used in a cost-effective manner if they are adopted within routine immunization programs and used for catch-up nationwide vaccination.

Evaluating the effectiveness and cost-effectiveness of tuberculosis control interventions targeted to previously treated people in South Africa

In high-incidence settings, recurrent tuberculosis among previously-treated individuals contributes substantially to the burden of incident and prevalent tuberculosis. Using a transmission-dynamic model of tuberculosis and HIV in a high-incidence setting in South Africa, we project the population-level effect of control interventions targeted to individuals with a history of previous tuberculosis treatment in a high-incidence setting. We showed that the use of targeted active case finding in combination with secondary isoniazid preventive therapy in previously treated individuals could accelerate decreases in tuberculosis morbidity and mortality. In a follow-up, cost-effectiveness analysis study, we showed a strategy combining targeted follow-up examinations and secondary isoniazid preventive therapy for 1 year after treatment completion is expected to be cost effective with respect to the status quo.