Risk prognostication in prolapse and incontinence following childbirth
State of the art lecture - Wednesday 4 September
J Eric Jelovsek, MD, MMEd
Professor Department of Obstetrics, Gynecology and Reproductive Sciences
2nd degree connection2nd/Vice Chair, Education for OBGYN, Associate Professor at Duke University School of Medicine/Durham, North Carolina
Summary: Statistical modeling and machine learning have already transformed online retail, web searching, advertising, finance, politics and more. The prediction modeling vortex continues to attract healthcare into the mix and will likely impact many aspects of day to day patient care ranging from prevention, diagnosis, prognosis, choice of therapy, health care delivery and population health. This talk is for those interested in gaining a broader understanding of how modern statistical and machine learning methods are used for understanding patterns in data and in making clinical predictions. To accomplish this, clinicians should understand answers to a few basic questions: Are their differences between developing prediction models in medicine compared to other industries such as finance or retail? What are the different assumptions when modeling with the primary purpose of understanding patterns in data, performing inference and prediction and how does that primary purpose affect judgment of a model’s performance? Answers to these fundamental questions will be demonstrated using examples of data from women around the time of childbirth and among those with pelvic floor disorders from electronic health records, federal and international datasets. At the end, clinicians should be motivated to recognise that many clinical questions may be accurately addressed by considering a predictive approach and have a better understanding of how modern analytic tools may assist in future research.
Read full biography