Challenges of Guarantee-Time Bias

TLDR

Guarantee‑time bias occurs when survival analyses compare groups defined by events that happen during follow‑up, such as disease response or toxicity, and its presence is often difficult to detect. This article defines guarantee‑time bias, illustrates it with published examples, and reviews three analytic methods to eliminate the bias. The authors discuss conditional landmark analysis, extended Cox models, and inverse probability weighting, evaluate their strengths and limitations, and demonstrate their application to bisphosphonate use in the BIG 1‑98 trial. A naive analysis suggested a disease‑free survival benefit from bisphosphonates, but all three bias‑removing methods found no evidence of risk reduction.

Abstract

The potential for guarantee-time bias (GTB), also known as immortal time bias, exists whenever an analysis that is timed from enrollment or random assignment, such as disease-free or overall survival, is compared across groups defined by a classifying event occurring sometime during follow-up. The types of events associated with GTB are varied and may include the occurrence of objective disease response, onset of toxicity, or seroconversion. However, comparative analyses using these types of events as predictors are different from analyses using baseline characteristics that are specified completely before the occurrence of any outcome event. Recognizing the potential for GTB is not always straightforward, and it can be challenging to know when GTB is influencing the results of an analysis. This article defines GTB, provides examples of GTB from several published articles, and discusses three analytic techniques that can be used to remove the bias: conditional landmark analysis, extended Cox model, and inverse probability weighting. The strengths and limitations of each technique are presented. As an example, we explore the effect of bisphosphonate use on disease-free survival (DFS) using data from the BIG (Breast International Group) 1-98 randomized clinical trial. An analysis using a naive approach showed substantial benefit for patients who received bisphosphonate therapy. In contrast, analyses using the three methods known to remove GTB showed no statistical evidence of a reduction in risk of a DFS event with bisphosphonate therapy.