Assessing Learning Effects and Nonrandom Dropout in a Contraceptive Device Trial

Abstract

When there is a new device involved in a randomized trial, participants in the new device group might need more time to become skilled at using the new device than those who are in the control group. In addition, participants who are not used to the new device might drop out from the study at a different rate than participants in the control group. Such learning effects and possible nonrandom dropout can substantially influence the evaluation of treatment effects. Without considering these factors, the conclusion of the main study result is vulnerable to potential bias. In this paper we propose a transition model under a Markov process to assess the effects of learning and nonrandom dropout in a randomized control device trial. The transition probabilities of the proposed model can be obtained by maximum likelihood approaches, and then fit by weighted local regressions to describe patterns of learning and participant dropout. A generalized estimating equation approach is applied to evaluate the impact of prognostic factors on those patterns. We use data from a randomized contraceptive device trial to illustrate the usefulness of the proposed method. The potential impact of learning effects and nonrandom dropout on the primary study objective is discussed.

Key Words:

• Device trial
• Learning effect
• Markov chain
• Nonrandom dropout

ACKNOWLEDGMENTS

The authors would like to thank Dr. Taylor and two anonymous referees for their critical reading of the manuscript and useful comments and suggestions. Their comments have led to considerable improvement in the presentation. Support for this study was partially provided by the United States Agency for International Development (USAID) through the Cooperative Agreement, CCP-A-00-95-00022-02, with Family Health International (FHI). The views expressed in this article do not necessarily reflect those of USAID or FHI.