Grant: Discovery Grants Program - Individual
Université Laval
$27,000.00
Causal inference methods for adaptive treatment strategies
FY 2024-25
Summary
Background: Adaptive treatment strategies (ATSs) aim to determine the best treatment decision as a function of previous data. Estimation of the optimal ATS using observational data is a challenging task that generally requires estimating treatment-covariates interactions. Unfortunately, the statistical power to detect such interactions is often low. In addition, ATSs that involve multiple tailoring covariates can be impractical and costly to implement. Finally, the validity of the results depends on the adequate identification and modeling of confounders.
Objectives: 1) Compare the statistical power of different methods for ATSs, 2) Develop data-driven methods for selecting tailoring variables in ATSs, 3) Develop methods to improve adjustment for confounding in ATSs.
Methods: Most methods for ATSs involve specifying for the outcome according to treatment and confounders. The terms related to the treatment and to treatment-covariates interactions is called the blip and is the quantity of interest, since the optimal ATS is identified through it. Many methods require the blip to be correctly specified. In practice, misspecifying the blip may be desirable, for example to focus on a subset of the existing treatment-covariates interactions to yield a simplified decision rule. I have proposed an approach, called partially adaptive treatment strategies (PATS), that circumvents this challenge. Objective 1 involves comparing the statistical power to detect treatment-covariates interactions between PATS and ATS methods using both theoretical results and simulation studies. Objective 2 will consist in developing methods for tailoring-variables selection. The methods will aim to select a practical subset of the most important tailoring variables and are adapted to high dimensional (e.g., genes) data. I plan to develop a method that optimizes the statistical properties (e.g., statistical power) of the estimated optimal treatment strategy. I will also develop methods for selecting the most cost-effective treatment strategy. Bayesian methods will be used to account for the variable selection step in producing inferences. Objective 3 concerns the improvement of the statistical adjustment for confounders. Double robust estimators of ATSs have been proposed. These methods involve estimating a model for the treatment in addition to a model for the outcome and are consistent if either model is correctly specified. Machine learning may provide further benefits through data-adaptive modeling. However, using machine learning for the outcome model may yield complicated ATSs. As such, I will first focus on using machine learning only to model the treatment. In a second step, I plan to use machine learning to model both models and define the parameter of interest as a projection of the true model onto a simplified parametric blip.
Impact: My research will advance statistical theory in a novel direction and provide methods to improve applied research in priority areas.
Program
Discovery Grants Program - Individual
Recipient
Location
Quebec, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
RGPIN-2023-04911
Name nom
Talbot, Denis
Department département
Département de médecine sociale et préventive
Organization id
51
Institution établissement
Université Laval
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
27000
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
1508
Committee name en
Mathematics and Statistics
Committee name fr
Mathématiques et statistique
Area of application code
1211
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Medical and health sciences
Area of application fr
Sciences médicales et de la santé
Research subject code
3005
Research subject group en
Statistics and probability
Research subject group fr
Statistique et probabilité
Research subject en
Biostatistics
Research subject fr
Biostatistiques
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
Causal inference methods for adaptive treatment strategies
Keywords
causal inference; observational data; bayesian statistics; model selection; double robust; semiparametric
Application summary
Background: Adaptive treatment strategies (ATSs) aim to determine the best treatment decision as a function of previous data. Estimation of the optimal ATS using observational data is a challenging task that generally requires estimating treatment-covariates interactions. Unfortunately, the statistical power to detect such interactions is often low. In addition, ATSs that involve multiple tailoring covariates can be impractical and costly to implement. Finally, the validity of the results depends on the adequate identification and modeling of confounders.
Objectives: 1) Compare the statistical power of different methods for ATSs, 2) Develop data-driven methods for selecting tailoring variables in ATSs, 3) Develop methods to improve adjustment for confounding in ATSs.
Methods: Most methods for ATSs involve specifying for the outcome according to treatment and confounders. The terms related to the treatment and to treatment-covariates interactions is called the blip and is the quantity of interest, since the optimal ATS is identified through it. Many methods require the blip to be correctly specified. In practice, misspecifying the blip may be desirable, for example to focus on a subset of the existing treatment-covariates interactions to yield a simplified decision rule. I have proposed an approach, called partially adaptive treatment strategies (PATS), that circumvents this challenge. Objective 1 involves comparing the statistical power to detect treatment-covariates interactions between PATS and ATS methods using both theoretical results and simulation studies. Objective 2 will consist in developing methods for tailoring-variables selection. The methods will aim to select a practical subset of the most important tailoring variables and are adapted to high dimensional (e.g., genes) data. I plan to develop a method that optimizes the statistical properties (e.g., statistical power) of the estimated optimal treatment strategy. I will also develop methods for selecting the most cost-effective treatment strategy. Bayesian methods will be used to account for the variable selection step in producing inferences. Objective 3 concerns the improvement of the statistical adjustment for confounders. Double robust estimators of ATSs have been proposed. These methods involve estimating a model for the treatment in addition to a model for the outcome and are consistent if either model is correctly specified. Machine learning may provide further benefits through data-adaptive modeling. However, using machine learning for the outcome model may yield complicated ATSs. As such, I will first focus on using machine learning only to model the treatment. In a second step, I plan to use machine learning to model both models and define the parameter of interest as a projection of the true model onto a simplified parametric blip.
Impact: My research will advance statistical theory in a novel direction and provide methods to improve applied research in priority areas.