Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Discovery Grants Program - Individual

Université de Montréal

$27,000.00

Consistent estimation of causal effects under irregular visit times

FY 2024-25

Summary
Non-experimental longitudinal data, like those from electronic health records, are widely available for drawing inferences about treatment effects on outcomes. However, some features of these data must be considered for correctly inferring on a causal treatment effect, i.e., the effect truly due to the treatment (or exposure) on the outcome, not to third variables creating associations in the data. First, data in electronic health records (and other sources of observational data) are recorded sporadically at times that typically depend on patients' conditions and other factors. Suppose that we only observe patients on days when they are sick. We might not be able to assess their outcomes on other days when they are not observed, which can affect our inference. Second, treatments in electronic health records are not randomized. In comparing two groups who have received different treatments in this non-experimental setting, therefore, we might find associations between patient characteristics (age, sex) and the treatment they received. This leads to spurious associations in the data called confounding. In the past, I have proposed novel statistical methods to infer the causal effect of an exposure on a longitudinal outcome, for settings with irregularly observed outcome data and confounding. In this research program, I aim to improve these methods by making them more robust to the assumptions they rely upon and by constructing useful extensions that apply to common data problems met in practice. First, I will build a novel multiply robust estimator for the causal effect of treatment that requires weaker assumptions on the visit and the treatment models to be unbiased. Then, I will extend that estimator to new settings that are commonly met in practice. Currently, there exists no statistical method to infer the causal effect of a treatment in complex settings where multiple variables (e.g., the treatment, the outcome, and some confounders) are measured irregularly, all at different points in time and as a function of different predictors (a common setting in practice). As an example, inferring the causal effect of air pollution on asthma requires data on both longitudinal processes (asthma and air pollution), which are not measured by the same person nor at the same times. There also exists no method to consistently estimate an optimal dynamic treatment regime (a sequence of treatment rules, also based on causal effects) when patients are measured irregularly at times that depend on variables such as mediators. I will propose new methods for such estimation problems.  This research is important for scientific fields and engineering, as causal inference is a central scientific topic. The proposed developments will fill important methodological gaps in statistics and will be used in several other areas such as biology, physics, and engineering to improve the causal inferences drawn from irregularly observed longitudinal data.
Program
Discovery Grants Program - Individual
Location
Quebec, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
RGPIN-2023-03946
Name nom
Coulombe, Janie
Department département
Department of Mathematics and Statistics
Organization id
63
Institution établissement
Université de Montréal
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
Consistent estimation of causal effects under irregular visit times
Keywords
causal inference; irregular monitoring times; confounding; efficiency; robustness; dynamic treatment regimes; observational data; missing data; adaptive treatment strategies; semiparametric theory
Application summary
Non-experimental longitudinal data, like those from electronic health records, are widely available for drawing inferences about treatment effects on outcomes. However, some features of these data must be considered for correctly inferring on a causal treatment effect, i.e., the effect truly due to the treatment (or exposure) on the outcome, not to third variables creating associations in the data. First, data in electronic health records (and other sources of observational data) are recorded sporadically at times that typically depend on patients' conditions and other factors. Suppose that we only observe patients on days when they are sick. We might not be able to assess their outcomes on other days when they are not observed, which can affect our inference. Second, treatments in electronic health records are not randomized. In comparing two groups who have received different treatments in this non-experimental setting, therefore, we might find associations between patient characteristics (age, sex) and the treatment they received. This leads to spurious associations in the data called confounding. In the past, I have proposed novel statistical methods to infer the causal effect of an exposure on a longitudinal outcome, for settings with irregularly observed outcome data and confounding. In this research program, I aim to improve these methods by making them more robust to the assumptions they rely upon and by constructing useful extensions that apply to common data problems met in practice. First, I will build a novel multiply robust estimator for the causal effect of treatment that requires weaker assumptions on the visit and the treatment models to be unbiased. Then, I will extend that estimator to new settings that are commonly met in practice. Currently, there exists no statistical method to infer the causal effect of a treatment in complex settings where multiple variables (e.g., the treatment, the outcome, and some confounders) are measured irregularly, all at different points in time and as a function of different predictors (a common setting in practice). As an example, inferring the causal effect of air pollution on asthma requires data on both longitudinal processes (asthma and air pollution), which are not measured by the same person nor at the same times. There also exists no method to consistently estimate an optimal dynamic treatment regime (a sequence of treatment rules, also based on causal effects) when patients are measured irregularly at times that depend on variables such as mediators. I will propose new methods for such estimation problems.  This research is important for scientific fields and engineering, as causal inference is a central scientific topic. The proposed developments will fill important methodological gaps in statistics and will be used in several other areas such as biology, physics, and engineering to improve the causal inferences drawn from irregularly observed longitudinal data.