Deep Learning for Pharmacoepidemiologists: An Accessible Introduction to Neural Networks

Background: Deep learning methods are increasingly being applied to problems in medicine and healthcare. With the abundance of research in this domain, it is important that epidemiologists, biostatisticians, and other health researchers can critically engage with and contribute to research using deep learning methods.

Objectives: Participants will leave the workshop with a basic understanding of deep learning methods, with the intention of empowering inter-disciplinary collaborations. Particular emphasis will be placed on deep learning methods for time series data (including electronic health records and insurance claims) and free-text language. The workshop is introductory and is specifically targeted towards participants with little to no prior knowledge of deep learning. It will build on concepts presented in the Machine Learning for Pharmacoepidemiologists pre-conference course; however, the course is not a mandatory prerequisite.

Description: Deep learning presents real opportunities for improving the quality of care provided to patients, as well as numerous challenges requiring subject-matter and epidemiologic expertise, including assessing fairness of systems or designing real-world evaluations of utility and harms. The goal of this workshop is to equip participants with the information necessary to facilitate a dialogue between computer scientists and epidemiologists, allowing for collaborations that will improve the safety and efficacy of applications of this technology. Emphasis will be placed on exploring the down-stream consequences of modeling decisions.
This workshop, endorsed by the ISPE Digital Epidemiology SIG, offers an accessible introduction to the basics of deep learning from an epidemiologic perspective. It will provide explanations of common deep learning model architectures: feed-forward networks, convolutional neural networks, recurrent neural networks, and transformer models. Use cases, practical considerations, and limitations for each of these models will be discussed. The workshop will include dedicated time for interactive small group discussions centered around important topics in the application of deep learning in pharmacoepidemiology: fairness and bias, interpretability, and real-world evaluation of safety and effectiveness.