Annual predictions of adverse outcomes after glaucoma surgery in the United States

Abstract

Purpose: To predict the number of adverse outcomes or additional ocular surgeries after glaucoma surgery in each year from 1990 to 1994 with ordinary and hierarchical logistic regression models developed with 1989 Medicare data. Methods: We obtained data on all intraocular and laser glaucoma surgery claims (for example, laser trabeculoplasty, full-thickness and partial-thickness procedures, and cycloablation) to the Health Care Finance Administration (HCFA) in 1989 and developed an ordinary (non-hierarchical) logistic regression model and two hierarchical logistic regression models. The two hierarchical models smoothed state-level effects to different degrees in an effort to improve the precision of model predictions. Results: The hierarchical logistic regression models predicted the number of adverse outcomes or additional ocular surgeries, including repeat glaucoma surgery, from 1990 to 1994 significantly better than the ordinary model. None of the models predicted the downward trend in adverse outcomes in 1990 and 1992, although all three models were able to predict the increase in adverse outcomes in 1991 and the downward trend in adverse outcomes in 1993 and 1994. Conclusion: Although systematic changes over time in physician practice or billing patterns affected the ability of both ordinary and hierarchical regression models to predict the number of adverse outcomes in 1990 and 1992, hierarchical logistic regression provides a useful framework for analyzing adverse event rates when the model contains many covariates with imprecisely estimated coefficients. The greater accuracy of predicted adverse event rates from ophthalmic surgeries obtained with hierarchical logistic regression will be useful for future planning and budgeting purposes.