Zhou X, Sherrill B, Wu Y, Bennett L, Wang J. Evaluating effects of changing treatments in longitudinal studies. Poster presented at the ISPE 28th International Conference on Pharmacoepidemiology; September 5, 2012.

BACKGROUND: In studies evaluating treatment effects for endpoints such as overall survival, patients are expected to change treatments during follow-up due to disease progression, adverse events, or other reasons. It is often of interest to compare efficacy endpoints across the entire treatment pattern, assessing the effects of the sequence of initial treatments and second- or even third-line treatments. However, in practice, researchers often either ignore second-line treatments or simply stratify patients by whether they received second-line treatment. These simplistic approaches have a purpose but do not take full advantage of the longitudinal information. The use of time-varying covariates in Cox models can provide valuable insights into treatment sequencing, for example, in oncology research. However, these methods are not used widely even in situations where they are clearly applicable; the intent of this presentation is to highlight the usefulness of survival analyses that assess time-varying covariates.

OBJECTIVES: To appropriately account for second-line treatment when evaluating the effect of first-line treatments and to evaluate the effect of second-line treatments.

METHODS: We constructed Cox models for time-to-event endpoints with second-line treatments handled as time-varying covariates.

RESULTS: We first illustrate why commonly used methods are not always appropriate. Next, we show how the time-varying covariates for treatments work in practice when additional treatments are received after the initial treatment, focusing on interpretation of results. We show different model settings that reflect various clinical assumptions. Finally, we discuss other applications, considerations, and assumptions when using time-varying covariates.

CONCLUSIONS: The technique of incorporating time-varying covariates in analyses of time-to-event endpoints provides a flexible analysis method to evaluate treatment effects in complex situations in which patients receive a sequence of treatments. This approach has wide applications, most notably in oncology research.

Share on: