13 In both trials, compliance with questionnaire completion was high over the duration of treatment in each trial (LUX-Lung 1, 65–100%; LUX-Lung 3, >90%), which helps to reduce concern of bias due to missing data; however, attrition was greater in the control arms, with the main cause being disease progression, potentially resulting in bias. We Ganetespib STA-9090 do not consider missing data due to attrition an issue in these analyses,
because we explicitly compare HRQoL in patients with and without progression at each assessment time (ANCOVA analysis), as well as assessing change in HRQoL due to progression within patients (longitudinal model); therefore, the effect of attrition should only be to reduce sample size at each assessment. Furthermore, both studies extensively evaluated the impact of missing data through sensitivity analyses and found that differences in HRQoL questionnaire completion were unlikely to bias the findings of either study. A limitation associated with all statistical methods that estimate the effect of progression is that the comparison is non-randomised (as in an observational study) leading to potential bias. This potential bias was limited in the ANCOVA analysis by using covariate adjustment, while within patient comparisons in the longitudinal model avoided bias as long
as the piecewise linear model is correct. For ANCOVA as well as longitudinal analyses, data from active and control treatment arms were pooled, which assumes that the effect of progression on HRQoL is independent of treatment. While this may be a potential source of bias, the ANCOVA model included a term for treatment as a covariate, and estimates of treatment-specific effects of progression from mixed-effects longitudinal models did not suggest that this was the case. It should be considered that these findings
are specific to the type of patients with NSCLC enrolled in LUX-Lung 1 and LUX-Lung 3 and may not generalise to other patient types. Finally, adverse events associated with afatinib treatment have the potential to impact on specific HRQoL items11 Dacomitinib 13 and thus have a confounding effect on the results reported here. However, there were few grade 3/4 toxicities, which were confounded with assessments of progression and when these effects were included in longitudinal models the effects of progression on HRQoL were only slightly reduced (data not shown). Additionally, the HRQoL measures used in these analyses (EORTC Global Health/QoL, EQ-5D UK Utility and EQ VAS) measure global health and thus would likely reflect the effects of drug toxicity. Taking these points into consideration, we do not believe drug toxicity is an important confounding factor in our analyses. The demonstration of a relationship between PFS and HRQoL in patients with lung cancer has important implications for healthcare policy decision-making, among others, in patients with NSCLC.