Trends in Successful Resuscitation after Cardiac Arrest under Trending Misclassification Error: Estimating Bounds for Partially Verified Data
Arthur Sweetman, McMaster University
Date: Friday, March 16th
Time: 11 AM – 1:00 PM
Location: HSB 100 (155 College Street)
Arthur Sweetman is a professor in the Department of Economics where he holds the Ontario Research Chair in Health Human Resources and is also a member of CHEPA. Previously he was Director, School of Policy Studies at Queen’s University where he also held the Stauffer-Dunning Chair in Policy Studies. He obtained his PhD in economics at McMaster University. His research focuses primarily on empirical (econometric) approaches to economic policy issues, and he has an interest in quantitative program evaluation. Prior to returning to McMaster in 2010, his research areas were extremely broad involving, among other topics, labour market, social policy and health topics. At McMaster his primary focus is on economic and policy issues related to health human resources.
Abstract
Estimating trends over time, including those surrounding policy changes, typically does not address the plausible confounding issue of trends in data quality, leading to a non-classical measurement error problem. This may be a concern with either survey or administrative data, especially when measurement quality improves with time. Our application is to administrative health data, which is often used in epidemiological studies. We address the detection of a trend in a binary outcome – successful resuscitation following cardiac arrest – allowing for trending misclassification error. Employing a mixture model, we compute bounds on the outcome following under contaminated and corrupt data assumptions. Using validation information from a non-random subsample of the data, we also consider how identification can be improved with monotonicity assumptions, bounded variation assumptions, and subgroup specific verification rates. We show evidence of a trend in the successful resuscitation rate for the population of reported cardiac arrests in Ontario under assumptions that are weaker than those in the existing literature.