Estimating causal effects from observational data is a fundamental problem in many fields that face challenges (e.g., expensive, time-consuming, or even unethical) in running randomized control trials. Classical causal inference methods often underperform with high-dimensional data (where estimation bias increases with dimensionality) or non-tabular data (which leads to information loss during feature engineering for text, images, EHRs, etc.). To overcome these challenges, I have developed new methods that integrate deep learning with causal inference. First, I have developed a deep learning-based propensity score model for representing high-dimensional data and adjusting for confounding bias, and leveraged the estimated average treatment effects for emulating clinical trials for drug development. Second, I have developed a deep learning-based balancing-matching method for controlling time-dependent and hidden confounders, and leveraged the estimated individual treatment effects for clinical decision-making in antibiotic stewardship for sepsis. Finally, I have developed the first foundation model for causal inference, which is pre-trained on large-scale patient data to learn contextual patient representations and fine-tuned on small-scale patient data for estimating causal effects. In future work, I aim to continue exploring this interdisciplinary field by developing and deploying end-to-end deep learning and causal inference approaches in real-world clinical scenarios yielding new theoretical advances and practical impacts, with the goal of improving patients' well-being and extending their lifespans.
Ruoqi Liu is a Ph.D. candidate in the Department of Computer Science and Engineering at The Ohio State University, advised by Dr. Ping Zhang in the Artificial Intelligence in Medicine lab. Her research focuses on the intersection of deep learning and causal inference, with the overarching goal of enhancing accurate causal effect estimation and enabling reliable decision-making in healthcare and biomedicine. Her work has been published in top-tier data mining and machine learning venues, such as Nature Machine Intelligence and ACM SIGKDD. Her research has been featured in Nature Online, MIT Technology Review, Drug Discovery News, etc.