Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Discovery Grants Program - Individual

McGill University

$37,000.00

Data-driven methods for simulating population heterogeneity in decision-analytic models

EF 2024-25

Résumé
Decision-analytic models can inform policymaking in health, criminal justice, energy, and other sectors. Models traditionally focus on aggregate policy impacts, but policymakers concerned about equity are increasingly interested in the distributional impacts across heterogeneous populations. Modeling heterogeneity with traditional methods-in which a cohort model is parameterized from second-hand literature reports-is challenging, particularly when one aims to explore how multiple dimensions of heterogeneity interact (e.g., age, sex, race, and geography). My research program investigates integrating data-driven methods (Bayesian statistics, machine learning) into decision-analytic models to improve estimation of how policies impact subgroups within heterogenous populations, enabling equity-informative decision analysis. Aim 1. Develop a two-step method for efficient compositional calibration of compartmental cohort models Due to state space explosion and the curse of dimensionality, it is difficult to stratify cohort models by multiple dimensions of individuals' attributes. We will develop an alternative method, where a compartmental model is first developed at the "population level" and then Bayesian calibration is used to stratify each compartment into subpopulations at the "compositional level". This approach separates modeling of population-level and compositional dynamics into separate stages, improving tractability. We will fully develop this potential novel approach, analyzing its properties and limitations. Aim 2. Estimate population heterogeneity in decision-analytic models with Bayesian multilevel regression Bayesian multilevel models estimate outcomes based on the combination of covariates at the individual level (e.g., age, sex) and group level (e.g., geographic region or specific study site). We will explore using multilevel models to estimate relevant model parameters (e.g., costs, health outcomes) for subgroups in a cohort dataset based on their relevant covariates for use in decision-analytic models. Bayesian methods can produce posterior distributions that can be directly integrated into probabilistic sensitivity analysis, the preferred method of uncertainty propagation in decision analysis. Aim 3. Integrate machine learning-based estimates of heterogeneous treatment effect into decision analytic models Not all individuals in a population respond to a treatment in the same way. The use of machine learning to estimate heterogeneous treatment effects (treatment effect conditional upon observed covariates) has been a productive area of research in econometrics over the last 15 years. We will develop and assess methods to integrate these methods into decision analytic models, including cost-effectiveness analyses. Together, these projects will produce new methods for integrating data-driven approaches into decision-analytic models, enabling equity-informative decision analysis.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
Quebec, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data

Tel que publié

Application id
RGPIN-2023-04160
Name nom
Russell, W. Alton
Department département
School of Population and Global Health
Organization id
61
Institution établissement
McGill University
Province en
Québec
Province fr
Québec
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2023
Award amount
37000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1509
Committee name en
Civil, Industrial and Systems Engineering
Committee name fr
Génie civil, industriel et des systèmes
Area of application code
405
Area of application group en
Environment
Area of application group fr
Environnement
Area of application en
Modelling and mathematical simulation of natural processes
Area of application fr
Modélisation et simulation mathématique des processus naturels
Research subject code
1600
Research subject group en
Industrial engineering
Research subject group fr
Génie industriel
Research subject en
Industrial engineering
Research subject fr
Génie industriel
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Data-driven methods for simulating population heterogeneity in decision-analytic models
Keywords
decision analysis; data-driven modeling; simulation; heterogeneity; bayesian modeling; modal calibration; synthetic data; equity-informative decision analysis
Application summary
Decision-analytic models can inform policymaking in health, criminal justice, energy, and other sectors. Models traditionally focus on aggregate policy impacts, but policymakers concerned about equity are increasingly interested in the distributional impacts across heterogeneous populations. Modeling heterogeneity with traditional methods-in which a cohort model is parameterized from second-hand literature reports-is challenging, particularly when one aims to explore how multiple dimensions of heterogeneity interact (e.g., age, sex, race, and geography). My research program investigates integrating data-driven methods (Bayesian statistics, machine learning) into decision-analytic models to improve estimation of how policies impact subgroups within heterogenous populations, enabling equity-informative decision analysis. Aim 1. Develop a two-step method for efficient compositional calibration of compartmental cohort models Due to state space explosion and the curse of dimensionality, it is difficult to stratify cohort models by multiple dimensions of individuals' attributes. We will develop an alternative method, where a compartmental model is first developed at the "population level" and then Bayesian calibration is used to stratify each compartment into subpopulations at the "compositional level". This approach separates modeling of population-level and compositional dynamics into separate stages, improving tractability. We will fully develop this potential novel approach, analyzing its properties and limitations. Aim 2. Estimate population heterogeneity in decision-analytic models with Bayesian multilevel regression Bayesian multilevel models estimate outcomes based on the combination of covariates at the individual level (e.g., age, sex) and group level (e.g., geographic region or specific study site). We will explore using multilevel models to estimate relevant model parameters (e.g., costs, health outcomes) for subgroups in a cohort dataset based on their relevant covariates for use in decision-analytic models. Bayesian methods can produce posterior distributions that can be directly integrated into probabilistic sensitivity analysis, the preferred method of uncertainty propagation in decision analysis. Aim 3. Integrate machine learning-based estimates of heterogeneous treatment effect into decision analytic models Not all individuals in a population respond to a treatment in the same way. The use of machine learning to estimate heterogeneous treatment effects (treatment effect conditional upon observed covariates) has been a productive area of research in econometrics over the last 15 years. We will develop and assess methods to integrate these methods into decision analytic models, including cost-effectiveness analyses. Together, these projects will produce new methods for integrating data-driven approaches into decision-analytic models, enabling equity-informative decision analysis.