Keywords: Health-Related Quality of Life, Cancer Trials, non-ignorable censoring
Health-related quality of life (HR-QoL) outcomes are important measures in evaluation of new cancer therapies and are considered both in regulatory approval and Health Technology Assessment decisions. In many oncology trials, after experiencing disease progression, patients not only discontinue study treatment but also transition to a safety-only/overall survival follow-up and therefore HR-QoL instruments are no longer assessed. Nevertheless, effect of anti-cancer therapies on time to HR-QoL deterioration or on patients’ QoL at specific time points, e.g., 1 or 2 years after treatment initiation, is of interest to all stakeholders regardless of disease status. At the same time, disease progression as well as other events such as discontinuation of study treatment due to toxicity are very likely to impact future QoL. Such events are treatment-related, and often lead to a considerable imbalance in the amount and timing of missingness between treatment arms, and may introduce an important bias into analysis conclusions if not accounted for at the analysis stage. In this presentation, we discuss several analysis approaches that deal with unavailable HR-QoL data which take into account a clinically plausible impact of reasons leading to missingness on the unobserved HR-QoL scores. We consider multiple imputation-based strategies for analysis of time to HR-QoL deterioration under clinically plausible non-ignorable censoring assumptions. We discuss how these assumptions can be formulated based on historic data from studies in similar therapies and indications or based on domain knowledge about minimum clinically important differences in QoL scores. We argue that in the oncology context, such strategies are clinically more justifiable than traditional methods based on the ignorable censoring assumption and should be considered for primary analysis of time to HR-QoL deterioration.