OBJECTIVES

It becomes complicated to estimate treatment effect, when an event impinges on the measurement of a patient reported outcome (PRO). We explore how estimates of treatment effect can vary, depending on how censoring is taken into account.

METHODS

We used KCCQ scores from the TOPCAT (NCT00094302) and HF-ACTION (NCT00047437) heart failure studies, where the KCCQ was censored by death in 8-10% of subjects. The event of death was accounted for by a) the standard repeated measures approach; b) a death=“worst score” approach analysed via quantile regression; and c) a principal stratum approach, which identifies a “stratum” of subjects who would be expected to survive to the end of the study irrespective of treatment group assigned.

RESULTS

For both studies, differences in survival rates between the treatment groups was relatively small; nevertheless, the pattern of results was helpful in understanding the interplay of death vs. quality of life in the studies. For TOPCAT (N=1767), where the experimental treatment was estimated to have lower hazard of death in the period analysed, the effect on change from baseline (CFB) in KCCQ was estimated to be slightly higher in the principal stratum of survivors, than in the equivalent “naïve” observed survivor population (4.3 vs. 3.7). For the HF-ACTION study (N=2331), the survivors stratum was also estimated to have higher CFB (1.2 vs. 1.09) in the second half of the study where the separation of the survival estimates was slightly larger.

CONCLUSIONS

It is challenging to obtain an universally acceptable estimate of treatment effect when the primary objective is to assess changes in PROs but censoring occurs. Whatever the approach is chosen, it is recommended to provide more than one estimate and thus allow readers of the study report to get a more complete picture regarding the interplay of benefit and risk for subjects.