Model for the Analysis of Binary Longitudinal Pain Data Subject to Informative Dropout through Remedication

• Published in: Journal of the American Statistical Association Vol. 93; no. 442; pp. 438 - 450
• Main Authors: Pulkstenis, Erik P., Ten Have, Thomas R., Landis, J. Richard
• Format: Journal Article
• Language: English
• Published: Alexandria, VA Taylor & Francis Group 01.06.1998; American Statistical Association; Taylor & Francis Ltd
• Subjects: Applications; Applications and Case Studies; Binary system; Biology, psychology, social sciences; Complementary log-log link; Data analysis; Discrete time survival; Drug design; Exact sciences and technology; Generalized estimating equation; Inference; Insurance, economics, finance; Log-gamma random effects; Mathematical models; Mathematics; Maximum likelihood; Medical sciences; Medicine; Missing data; Modeling; Multilevel models; Multivariate analysis; Pain; Parametric models; Pharmaceuticals; Placebos; Probability and statistics; Reliability, life testing, quality control; School dropouts; Sciences and techniques of general use; Simulation; Standard error; Statistical analysis; Statistics; Numerical approximation; Gamma distribution; Binary data; Data analysis; Medical application
• ISSN: 0162-1459; 1537-274X
• DOI: 10.1080/01621459.1998.10473693

We address the problem of accounting for informative dropout in the form of rescue medication when comparing pain relievers with respect to longitudinal binary pain-relief outcomes. We present a selection model approach for binary longitudinal data that accommodates informative dropout. The relationship between dropout or remedication and the binary pain-relief response is assumed to be characterized by a random effect. That is, conditional on this random effect, response and dropout are independent. Unlike previous approaches to this problem, which rely on numerical or approximation methods, we obtain a closed-form expression for the marginal log-likelihood of response and dropout by specifying a complementary log-log link function for both components and a conjugate log-gamma random effect distribution. A data analysis supported by simulation results suggest that the model fits reasonably well. Results are compared to those obtained from conventional, but somewhat inappropriate analyses.