Principal Investigator: Dr Andre Ribeiro
Department: Harvard University
Cambridge, Massachusetts, USATags: 54287, causal inference, effect heterogeneity, experimental design, genetics/genotyping, Machine Learning, treatment effects
Diseases are caused by genetic and/or environmental factors. Random Controlled Trials remain the gold-standard for causal effect estimation of genes and environmental factors on phenotypes. Current techniques are still unable to make use of large datasets such as the UK Biobank to make causal claims. We search for biobank subsamples that can approximate statistically experimental samples. We use recent techniques to estimate selection and confounding biases for distinct subsamples, when studying effects on phenotypes. In particular, we consider which environmental or genetic variables can be deemed causes of medical conditions of interest (such as cancer), and which procedures or drugs in medical records can be deemed causes for recovery.
Revealing causal associations between genetic or environmental factors and common diseases is of great importance for public health. Given trial costs and the growing availability of observational data, understanding experimental and non-experimental methods constraints and relationship is also of great practical relevance for patients, scientists and policy-makers.