|
Causal Inference
Transportability/Generalizability Heterogeneous treatment effect Time-dependent exposures and outcomes Non/semi-parametric theory Failure time outcomes |
Machine Learning
Convergence rate characterization A prior knowledge incorporation Structured learning Mixed integer programming High-dimensional data |
Application
Cardiovascular disease Infectious disease Mental health Oncology HIV |
Theme 1: Data fusion, transportability and machine learning in causal inference
Key words of 1.1-1.3: Targeted Learning, IPD-NMA, 31 international cohorts, 15 concurrent treatments, drug-resistance, effect heterogeneity, GPS
1.1: G. Wang, ME. Schnitzer, D. Menzies, P. Viiklepp, TH. Holtz, and A. Benedetti. Estimating treatment importance in multiple drug resistant tuberculosis using Targeted Learning: an observational individual patient data Network Meta-Analysis, Biometrics (2020) 76(3):1007-1006. Wiley
1.2: Y. Liu, ME. Schnitzer, G. Wang, E. Kennedy, P. Viiklepp, MH. Vargas, G. Sotgiu, D. Menzies, and A. Benedetti. Modeling Treatment Effect Modification in Multidrug-Resistant Tuberculosis in an Individual Patient Data Meta-Analysis, Statistical Methods in Medical Research (2021) 10.1177/09622802211046383. SAGE
1.3: AA. Siddique, ME. Schnitzer, A. Bahamyirou, G. Wang, A. Benedetti et al. Causal Inference for polypharmacy: Propensity score estimation with multiple concurrent medications, Statistical Methods in Medical Research, (2019) 28(12):3534-3549. SAGE
Keywords of 1.4-1.9: multi-cohort, transportability, causal assumptions, effect heterogeneity, doubly/multiply/rate robustness, non-parametrically efficient, software.
1.4: G. Wang, A. Levis, I. Dahabreh, Using multi-source data to estimate subgroup treatment effects in an external target population or to the populations underlying the data sources, in progress
1.5: G. Wang, A. Levis, E. Kennedy, I. Dahabreh, Using multi-source data to estimate heterogeneous treatment effects in an external target population or to the populations underlying the data sources, in progress
1.6: G. Wang*, S. McGrath*, Y. Lian*, I. Dahabreh, CMetafoR: An R package for performing causally interpretable meta-analyse, in progress
1.7: G. Wang, A. Levis, I. Dahabreh. Causal inference under transportability assumptions for conditional relative effect measures, in progress
1.8: L. Ung, G. Wang, I. Dahabreh. Combining clinical trials with external data: study design and identification of treatment effects, in progress
1.9: A massive simulation comparing 16 parametric and Bayesian methods for the use of hybrid control in early phase umbrella trials.
G. Wang, MP. Costello, H. Pang, J. Zhu, HJ. Helms, I. Reyes-Rivera, RW. Platt, M. Pang, A. Koukounari. Evaluating hybrid controls methodology in early-phase oncology trials: a simulation study based on the MORPHEUS-UC trial, Pharmaceutical Statistics, Wiley
Keywords of 1.10-1.12: Time-dependent exposures and outcomes, efficiency, variable selection
1.10: Estimate the heterogeneous treatment effects with survival outcomes and intermittent treatment on a continuous-time scale.
G. Wang, S. Liu, S. Yang. Continuous-time Structural Failure Time Models with intermittent treatment, in progress
1.11: Develop a longitudinal adaptive fused Lasso that will data-adaptively select covariates and collapse the treatment model parameters over time-points with the goal of improving the efficiency of the estimator while minimizing confounding bias.
M. Schnitzer, A. Ertefaie, D. Talbot, G. Wang, D. Berger, J. O'Loughlin, M.P. Sylvestre. Longitudinal outcome-adaptive and marginal fused Lasso for confounder selection and model pooling with time-varying treatments, in progress
1.12: A. Jaman, R. Platt, G. Wang, E. Ashkan, M. Bally, M. Schnitzer. Penalized G-estimation for effect modifier selection in the structural nested mean models for repeated outcomes, in progress
1.13: G. Wang. Review 1: ''Antibiotic Prescribing in Remote Versus Face-to-Face Consultations for Acute Respiratory Infections in English Primary Care: An Observational Study Using TMLE.'' Rapid Reviews Infectious Diseases. (2023) RR
1.14: H. Bian, M. Pang, Z. Lu, G. Wang. Non-collapsibility and Built-in Selection Bias of Hazard Ratio in Randomized Controlled Trials, in progress
Key words of 1.1-1.3: Targeted Learning, IPD-NMA, 31 international cohorts, 15 concurrent treatments, drug-resistance, effect heterogeneity, GPS
1.1: G. Wang, ME. Schnitzer, D. Menzies, P. Viiklepp, TH. Holtz, and A. Benedetti. Estimating treatment importance in multiple drug resistant tuberculosis using Targeted Learning: an observational individual patient data Network Meta-Analysis, Biometrics (2020) 76(3):1007-1006. Wiley
1.2: Y. Liu, ME. Schnitzer, G. Wang, E. Kennedy, P. Viiklepp, MH. Vargas, G. Sotgiu, D. Menzies, and A. Benedetti. Modeling Treatment Effect Modification in Multidrug-Resistant Tuberculosis in an Individual Patient Data Meta-Analysis, Statistical Methods in Medical Research (2021) 10.1177/09622802211046383. SAGE
1.3: AA. Siddique, ME. Schnitzer, A. Bahamyirou, G. Wang, A. Benedetti et al. Causal Inference for polypharmacy: Propensity score estimation with multiple concurrent medications, Statistical Methods in Medical Research, (2019) 28(12):3534-3549. SAGE
Keywords of 1.4-1.9: multi-cohort, transportability, causal assumptions, effect heterogeneity, doubly/multiply/rate robustness, non-parametrically efficient, software.
1.4: G. Wang, A. Levis, I. Dahabreh, Using multi-source data to estimate subgroup treatment effects in an external target population or to the populations underlying the data sources, in progress
1.5: G. Wang, A. Levis, E. Kennedy, I. Dahabreh, Using multi-source data to estimate heterogeneous treatment effects in an external target population or to the populations underlying the data sources, in progress
1.6: G. Wang*, S. McGrath*, Y. Lian*, I. Dahabreh, CMetafoR: An R package for performing causally interpretable meta-analyse, in progress
1.7: G. Wang, A. Levis, I. Dahabreh. Causal inference under transportability assumptions for conditional relative effect measures, in progress
1.8: L. Ung, G. Wang, I. Dahabreh. Combining clinical trials with external data: study design and identification of treatment effects, in progress
1.9: A massive simulation comparing 16 parametric and Bayesian methods for the use of hybrid control in early phase umbrella trials.
G. Wang, MP. Costello, H. Pang, J. Zhu, HJ. Helms, I. Reyes-Rivera, RW. Platt, M. Pang, A. Koukounari. Evaluating hybrid controls methodology in early-phase oncology trials: a simulation study based on the MORPHEUS-UC trial, Pharmaceutical Statistics, Wiley
Keywords of 1.10-1.12: Time-dependent exposures and outcomes, efficiency, variable selection
1.10: Estimate the heterogeneous treatment effects with survival outcomes and intermittent treatment on a continuous-time scale.
G. Wang, S. Liu, S. Yang. Continuous-time Structural Failure Time Models with intermittent treatment, in progress
1.11: Develop a longitudinal adaptive fused Lasso that will data-adaptively select covariates and collapse the treatment model parameters over time-points with the goal of improving the efficiency of the estimator while minimizing confounding bias.
M. Schnitzer, A. Ertefaie, D. Talbot, G. Wang, D. Berger, J. O'Loughlin, M.P. Sylvestre. Longitudinal outcome-adaptive and marginal fused Lasso for confounder selection and model pooling with time-varying treatments, in progress
1.12: A. Jaman, R. Platt, G. Wang, E. Ashkan, M. Bally, M. Schnitzer. Penalized G-estimation for effect modifier selection in the structural nested mean models for repeated outcomes, in progress
1.13: G. Wang. Review 1: ''Antibiotic Prescribing in Remote Versus Face-to-Face Consultations for Acute Respiratory Infections in English Primary Care: An Observational Study Using TMLE.'' Rapid Reviews Infectious Diseases. (2023) RR
1.14: H. Bian, M. Pang, Z. Lu, G. Wang. Non-collapsibility and Built-in Selection Bias of Hazard Ratio in Randomized Controlled Trials, in progress
Theme 2: Structural learning
We strongly encourage researchers to thoroughly understand and analyze the data before applying any analytical methods. Integrating this information can significantly enhance the performance of the model in use.
For example, when it comes to variable selection, incorporating specific selection rules can lead to improvements in prediction accuracy, a reduced false alarm rate, and, notably, enhanced interpretability of the selected model.
Examples of simple rules can be "if the interaction is chosen, then all or at least one of the main terms must also be selected", "if the subtopic is selected, then the overarching topic should also be chosen", "select at least one gene from each pathway", and "collectively select dummy variables representing a categorical variable."
We have developed unified methods for systematically integrating all available information into the analysis process.
2.1 Developed a formal mathematical language for expressing all possible selection dependencies that can be incorporated into structured variable selection, and provide the algorithms for identifying all permissible variable subsets that satisfying a given selection rule.
G. Wang, ME. Schnitzer, T. Chen, R. Wang, RW. Platt. A general framework for formulating structured variable selection, arXiv
2.2 Specifically, for the latent overlapping group, we develop roadmaps of grouping structure identification and apply it to identify predictors of major bleeding among hospitalized hypertensive patients using oral anticoagulants for atrial fibrillation, where seven rules are being considered.
G. Wang, ME. Schnitzer, RW. Platt, R. Wang, M. Doris, S. Perreault, Structured variable selection: an application in identifying predictors of major bleeding among hospitalized hypertensive patients using oral anticoagulants for atrial fibrillation”, arXiv
2.3 Methods for incorporating flexible selection rules into time-dependent Cox models.
G. Wang, Y. Lian, A. Yang, R. Platt, R. Wang, M. Dorias, S. Perreault, M. Schnitzer. Structured learning in Cox models with time-dependent covariates, In revision, Statistics in Medicine, R package, arXiv
2.4 Developed a unified variable selection method (based on l0-norm penalization) that can respect any arbitrary selection rule.
G. Wang, T. Chen, A. Yang, ME. Schnitzer, RW. Platt, R. Wang, Structural Best Subset Selection, in progress
2.5 Application of variable selection in HIV
A. Bouchard, F. Bourdeau, J. Roger, VT. Taillefer, N. Sheehan, ME. Schnitzer, G. Wang, IJ. Jean Batiste, and R. Therrien. Predictive Factors of Detectable Viral Load in HIV Infected Patients, accepted by AIDS Research and Human Retroviruses, (2022) 38(7):552-560, Libertpub
We strongly encourage researchers to thoroughly understand and analyze the data before applying any analytical methods. Integrating this information can significantly enhance the performance of the model in use.
For example, when it comes to variable selection, incorporating specific selection rules can lead to improvements in prediction accuracy, a reduced false alarm rate, and, notably, enhanced interpretability of the selected model.
Examples of simple rules can be "if the interaction is chosen, then all or at least one of the main terms must also be selected", "if the subtopic is selected, then the overarching topic should also be chosen", "select at least one gene from each pathway", and "collectively select dummy variables representing a categorical variable."
We have developed unified methods for systematically integrating all available information into the analysis process.
2.1 Developed a formal mathematical language for expressing all possible selection dependencies that can be incorporated into structured variable selection, and provide the algorithms for identifying all permissible variable subsets that satisfying a given selection rule.
G. Wang, ME. Schnitzer, T. Chen, R. Wang, RW. Platt. A general framework for formulating structured variable selection, arXiv
2.2 Specifically, for the latent overlapping group, we develop roadmaps of grouping structure identification and apply it to identify predictors of major bleeding among hospitalized hypertensive patients using oral anticoagulants for atrial fibrillation, where seven rules are being considered.
G. Wang, ME. Schnitzer, RW. Platt, R. Wang, M. Doris, S. Perreault, Structured variable selection: an application in identifying predictors of major bleeding among hospitalized hypertensive patients using oral anticoagulants for atrial fibrillation”, arXiv
2.3 Methods for incorporating flexible selection rules into time-dependent Cox models.
G. Wang, Y. Lian, A. Yang, R. Platt, R. Wang, M. Dorias, S. Perreault, M. Schnitzer. Structured learning in Cox models with time-dependent covariates, In revision, Statistics in Medicine, R package, arXiv
2.4 Developed a unified variable selection method (based on l0-norm penalization) that can respect any arbitrary selection rule.
G. Wang, T. Chen, A. Yang, ME. Schnitzer, RW. Platt, R. Wang, Structural Best Subset Selection, in progress
2.5 Application of variable selection in HIV
A. Bouchard, F. Bourdeau, J. Roger, VT. Taillefer, N. Sheehan, ME. Schnitzer, G. Wang, IJ. Jean Batiste, and R. Therrien. Predictive Factors of Detectable Viral Load in HIV Infected Patients, accepted by AIDS Research and Human Retroviruses, (2022) 38(7):552-560, Libertpub
