The use of proportional hazards regression in investigating dropout rates in a longitudinal study

One difficulty with the interpretation of data from longitudinal studies is the bias associated with those who do not complete the study. In a 12-month study on the occurrence of musculoskeletal problems in 966 runners (583 of whom completed the study), a proportional hazards model with time-dependent covariates was used to assess factors associated with dropout at the various stages of the study. This approach allowed examination of baseline factors as well as the effect of change in mileage, the occurrence of a musculoskeletal problem, or the occurrence of another health problem on the rate of dropout. Those most likely to drop out of the study were younger and heavier at baseline and, prior to drop out, were less likely to experience general health problems and more likely to show a 40% decline in weekly running mileage in the month before dropout. Examination of factors associated with dropout is important since factors influencing dropout may also affect the study outcome for the risk factor analysis (a musculoskeletal problem severe enough to be seen by a physician). The results of the dropout analysis can be used to guide in the choice of analytic methods and to aid in the interpretation of the risk factor analyses.