TITLE: Targeted Estimation of Direct and Indirect Causal Effects in Clinical Trials
SPEAKERS: Jie Chen, Merck & Co. Inc
Clinical trials establishing a therapeutic effect of a new drug on a clinical outcome often need to further investigate the magnitude of different pathways and mechanisms of action by which the drug produces the outcome. Such an investigation is called (causal) mediation analysis, which partitions the total drug effect into direct and indirect effects. Indirect effect represents drug effects through intermediate variables or mediators, whereas direct effects work throughothermechanism. For example, canagliflozin, a sodium glucose cotransporter 2 (SGLT2) inhibitor, is shown to reduce the risk of heart failure directly and indirectly through some post-randomization biomarkers (Li et al., 2020); physical activity may improve self-reported cognition of breast cancer survivors by decreasing anxiety (Hartman et al., 2019); a problem-solved educational program may help reduce the risk of depressive symptoms in low-income mothers primarily by reducing maternal stress (Silverstein et al., 2018; Lee et al., 2019). Causal mediation analysis helps usunderstand how an interventionworksand enableus to predict possible outcomes under a rich variety of conditions and interventions. Traditionally, mediation analysis is conducted through two approaches that are often referred to as the “difference method” and “product method”. This tutorial will first describe the mediation formula of Pearl (2012, 2014) using structural causal models and then introduce the targeted maximum likelihood framework of van der Laan and Rubin (2006) and van der Laan and Rose (2011) to construct semiparametric, multiple robust estimators for direct and indirect treatment effects. Examples are given throughout the tutorial for continuous, binary, and time-to-event outcome variables.
Key words: Clinical trials, intermediate variable, mediation analysis, mediation formula, targeted maximum likelihood estimation.
Hartman, S. J., L. S. Weiner, S. H. Nelson, L. Natarajan, R. E. Patterson, B. W. Palmer, A. Parker, and D. D. Sears (2019). Mediators of a physical activity intervention on cognition in breast cancer survivors: Evidence from a randomized controlled trial. JMIR Cancer 5 (2), e13150.
Lee, H., R. D. Herbert, and J. H. McAuley (2019). Mediation analysis. JAMA 321 (7), 697–698. Li, J., M. Woodward, V. Perkovic, G. A. Figtree, H. J. Heerspink, K. W. Mahaffey,
D. de Zeeuw, F. Vercruysse, W. Shaw, and D. R. Matthews (2020). Mediators of the effects of canagliflozin on heart failure in patients with type 2 JACC: Heart Failure 8 (1), 57– 66.
Pearl, J. (2012). The mediation formula: A guide to the assessment of causal pathways in nonlinear models. Wiley Online Library.
Pearl, J. (2014). Interpretation and identification of causal mediation. Psychological Meth- ods 19 (4), 459.
Silverstein, M., H. Cabral, M. Hegel, Y. Diaz-Linhart, W. Beardslee, C. J. Kistin, and
E. Feinberg (2018). Problem-solving education to prevent depression among low-income mothers: a path mediation analysis in a randomized clinical JAMA Network Open 1 (2), e180334– e180334.
van der Laan, M. J. and S. Rose (2011). Targeted learning: causal inference for observational and experimental data. Springer Science & Business Media.
van der Laan, M. J. and D. Rubin (2006). Targeted maximum likelihood learning. The International Journal of Biostatistics 2 (1).
Dr. Jie Chen is a Distinguished Scientist in Methodology Research at Merck Research Laboratories. Before rejoining Merck in February 2017 (he worked at Merck from 1995-2009), Jie worked in China for six and a half years, leading statistics and statistical programming groups for global pharma companies to support drug development globally and in China.
Jie received an M.D. in 1984 from Shanghai First College of Medicine, an MPH in 1994 in biostatistics & epidemiology from the University of Oklahoma Health Science Center, Oklahoma City, and a Ph.D. in 2003 in statistics from Temple University, Philadelphia, Pennsylvania.
Jie’s experience includes statistical methodology research and applications in non-clinical and pre-clinical research, clinical development, and post-licensure product life-cycle management. He has given short courses at FDA/Industry statistics workshop, EMA statistics symposium and many invited talks at academic institutions and statistical conferences.
Fang Liu (PhD in Statistics from Temple University), is a principal scientist at Merck Research Laboratories, Merck & Co., Inc. Dr Liu has been providing statistical support in various areas including early oncology studies, clinical pharmacology studies, PK-PD modeling and oncology biomarker statistics. She has been actively involved in statistical research and has authored/co- authored multiple scientific publications in peer-reviewed statistical and clinical journals. Her current research interests include causal inference, mediation analysis, basket trial design, umbrella trial design, missing data imputation, etc.
Yanping Liu (PhD in Statistics from Temple University), is an Associated Principal scientist at Merck Research Laboratories, Merck & Co., Inc. She works on late stage Cardiovascular and Oncology studies. Her research interests include multiple testing, survival analysis, causal inference and mediation analysis.