Editors' ChoiceData Analysis

Accounting for Participant Dropout in Clinical Studies

See allHide authors and affiliations

Science Translational Medicine  27 Jan 2010:
Vol. 2, Issue 16, pp. 16ec14
DOI: 10.1126/scitranslmed.3000856

Data are often collected repeatedly from clinical research subjects over months or years in order to track disease progress, detect the onset of new problems, or assess the effect of treatments. But unlike experimental animals, human subjects often choose to drop out of the study before its completion, rendering the resulting data incomplete. Failure in the analysis to appropriately account for subjects dropping out of the study for reasons unknown to the researchers can result in biased conclusions. Su and Hogan have now proposed a statistical model for longitudinal data that adjusts for these biases that are caused by drop-out. Their approach allows the estimates of change over time (that is, slope) to vary according to the time that subjects drop out of the study. They also offer a method for the summarization of varying slopes into a single estimate of change over time, making the study results easily interpretable to the medical researcher. This new approach can account for problems often encountered during real clinical experiments—such as subjects that drop out of the study between scheduled visits. Solutions to these problems do not exist in current methods.

Using this new methodology and existing data from a study on HIV-infected women and mental health, the authors show that the prevalence of depression decreases over time in women with more remaining immune cells but remains constant in those with cell counts less than 200, a result that is opposite of that found with a traditional longitudinal regression model for analysis. This dramatic example, in which the measured outcome of depression might influence the subjects' tendency to leave the study, shows why accounting for subject drop-out is critical for the accurate analysis of longitudinal studies. This new approach will make it easier to accommodate subject drop-out in analyses of clinical research data and to better understand the etiology of HIV/AIDS and other chronic diseases.

L. Su, J. W. Hogan, Varying-coefficient models for longitudinal processes with continuous-time informative dropout. Biostatistics 11, 93–110 (2010). [Full Text]

Stay Connected to Science Translational Medicine

Navigate This Article