Job Market Paper
Abstract: Estimating trends in outcomes over time typically ignores plausible confounding trends in data quality. Nominal trends may be biased, even completely spurious, due to trending misclassification error. We estimate trends in resuscitation rates following cardiac arrest in a context where improving data quality in hospital International Classification of Diseases (ICD) coding likely causes trending measurement error. We show how the partial identification approach of Horowitz and Manski (1995) can be used to estimate bounds on trend parameters when misclassified outcomes are bounded by non-random validation information. We find that a trend in successful resuscitation rates can be detected if the true outcome distribution is independent of the probability of measurement error (i.e., contaminated data). However, low measurement error hospitals have higher survival probability under independence. If independence is violated (i.e., corrupt data), worst case bounds only identify a trend for low measurement error hospitals. We introduce a pragmatic approach that narrows the identification bounds that combines partial verification, bounded corruption (plausible restrictions on proportions of false positives/negatives) and weak monotonicity restrictions that together provide sufficient identification power to detect a trend even with corrupt data. Our approach is applicable in many contexts involving survey and administrative data, including those surrounding policy changes.
Other Thesis Chapters
What Happens after Cardiac Arrest? Patterns of Care with Patient Enrollment, with Arthur Sweetman
Temporal Trends in Survival for Patients with In-hospital Cardiac Arrest in Ontario: 2003-2010, with Ahmad von Schlegell, Mathew Mercuri, Madhu K. Natarajan, and Arthur Sweetman
Peer-Reviewed Publications
Weir, S., Steffler, M., Li, Y., Shaikh, S., Wright, J. G., & Kantarevic, J. (2020). Use of the Population Grouping Methodology of the Canadian Institute for Health Information to Predict High-Cost Health System Users in Ontario. CMAJ, 192(32), E907-E912.
Li, Y., Weir, S., Steffler, M., Shaikh, S., Wright, J. G., & Kantarevic, J. (2019). Using Diagnoses to Estimate Health Care Cost Risk in Canada. Medical Care, 57(11), 875.
Sekercioglu, N., Shaikh, S., Guyatt, G., & Busse, J. (2017). Derivation and Validation of a Prognostic Model for Workers Disabled by Depression. Journal of Clinical and Analytical Medicine, 8(4), 351-356.
Sekercioglu, N., Busse, J, Sekercioglu, M., Agarwal, A., Shaikh, S., Lopes, L., Mustafa, R., Guyatt, G., & Thabane, L. (2016). Cinacalcet Versus Standard Treatment for Chronic Kidney Disease: A Systematic Review and Meta-Analysis. Renal Failure, 38(6), 857-874.