- This event has passed.
Informatics Research Seminar: A Primer on Propensity Score Matching with an Application to U.K. Population-Based Crohn’s Disease Data Assessing Treatment Effect
January 23 @ 4:00 pm - 5:00 pm EST
Speaker: Laura Gunn, PhD
Presented from UNC-C
Broadcast Link: Seminar
Randomized controlled trials (RCTs) are a ‘gold standard’ for estimating minimally unbiased treatment effects on health outcomes. However, RCTs are not always feasible and population-based observational studies may be more appropriate, particularly involving health analytics using big data. Assigning individuals at random between groups is not always feasible for ethical/practical concerns. Individuals ‘assign’ themselves to groups – e.g., studies about the impact of smoking cessation programs (SCPs) on blood pressure. Those who decide (not) to undertake such SCP intervention are self-selected, and biases/confounding factors (e.g., income, medical history, ethnicity, gender, age, education) may influence decisions (not) to attend SCPs, leading to potentially biased biostatistical inferences on this type of observational data. However, propensity score matching (PSM) reduces such study design biases.
PSM categorizes quantitatively individuals based on confounding factors associated with their decisions, so that samples of individuals ‘assigned’/matched to intervention(s) and control are similar/balanced across these factors. This transforms observational studies into pseudo-RCTs. For each person self-selected to the intervention, we ‘match’ M individuals with similar confounding characteristics who chose not to undertake the intervention. PSM relies on classification methods to ‘match’ individuals.
The Clinical Practice Research Datalink (CPRD) contains clinical and prescribing data for over 13 million patients in the United Kingdom; participating primary care practices are subject to regular audit to ensure data accuracy and completeness, allowing epidemiological studies of this data to be feasible. Since RCTs evaluating the impact of thiopurine treatment on Crohn’s disease patients is not practical, we used CPRD data to identify 5,640 patients with incident Crohn’s cases diagnosed over a 17-year period with at least an additional 5-year follow-up. Propensity score matching (PSM) is used to reduce bias, obtained in estimates of treatment effects as a result of confounding, between baseline factors and exposure group status. This presentation describes the PSM process, and applies optimal PSM, with a sensitivity analysis implementing additional matching techniques, using data collected from this nationally representative UK population-based study, where impact of duration and timing of thiopurine treatment on the likelihood of surgery is assessed using a Cox proportional hazards model and PSM.
Dr. Laura H. Gunn received her PhD in Statistics and Decision Sciences from Duke University, during which time she also held a research training fellowship in Biostatistics with the National Institute of Environmental Health Sciences. Laura is currently Associate Professor of Public Health Sciences and Director of Health Analytics at University of North Carolina at Charlotte (UNCC), as well as Honorary Research Fellow at Imperial College London’s (ICL) School of Public Health (SPH) within the Faculty of Medicine. Prior to UNCC, Laura was Associate Professor of Public Health in Biostatistics, Department Chair of Health Sciences, and founding Program Director of Public Health at Stetson University, where she led the development of the undergraduate minor and major in Public Health, as well as co-developed undergraduate and graduate Data/Applied Analytics programs and received the University’s Community Partnership of the Year Award for establishing partnerships with Florida Hospital and Florida Department of Health. Additional prior positions include Associate Director of the Global eHealth Unit within ICL’s SPH. During this time, she also served as Lead Biostatistician for Research Design Service London at ICL. Laura was Biostatistics Director and Interim Associate Dean of Georgia Southern University’s Jiann-Ping Hsu College of Public Health (JPHCOPH), where she was among the founding faculty to develop masters and doctoral programs primarily in Biostatistics and Epidemiology within JPHCOPH and served on the core leadership team for Council on Education for Public Health (CEPH) accreditation. She received the JPHCOPH Faculty Awards of Excellence for Teaching & Mentoring, as well as for Service. Laura served a term on the U.S. Department of Health and Human Services’ Medicare Evidence Development and Coverage Advisory Committee; and, a sample of additional service roles includes: Treasurer for the American Statistical Association Georgia Chapter; Chair and Program Committee Chair of the Biopharmaceutical Applied Statistics Symposium; and reviewer for various statistics and health journals, as well as US and UK nationally-funded grants. Her research accounts for 45 peer-reviewed journal articles, book chapters and technical reports, as well as serving as a PI or co-I on funded external grants and contracts totaling over $6.5 million, including National Institutes of Health (US) and National Institute for Health Research (UK) grants. A sample of her publications includes such journals as PLoS ONE, American Journal of Gastroenterology, Annals of Family Medicine, British Journal of Dermatology, Pediatrics, Biostatistics, Biometrics, International Journal of Integrated Care, and Cochrane Database of Systematic Reviews; and, she has over 150 invited and contributed regional, national, and international presentations.