Association of Anticancer Immune Checkpoint Inhibitors With Patient-Reported Outcomes Assessed in Randomized Clinical Trials

This systematic review and meta-analysis of randomized clinical trials involving patients with solid tumors assesses the association of immune checkpoint inhibitors with patient-reported quality of life.


Meta-analysis of difference in mean changes
We performed two separate meta-analyses, considering as effect size of interest the difference between the intervention group I and control group C of the mean change of PROs score from baseline to 12 weeks, and the difference from baseline to 24 weeks.
We extracted, for each study, the mean change at week w in the intervention group d I(w) and in the control group d C(w) , along with the standard deviations of the individual changes SD (d I(w) ) and SD (d C(w) ). The standard error of the mean change in the intervention group was calculated as SE(d I(w) )=SD(d I(w) )/√n I(w) , with n I(w) the number of respondents in the intervention arm at the week w of interest; SE(d C(w) ) was similarly derived. When in the intervention group or in the control group n w was not reported, we imputed it multiplying the number of patients n 0 evaluated at baseline (usually reported in the paper) by the average proportion of respondents at time w calculated from all the other studies reporting n w .
For each study, we estimated the effect size as δ w =(d I(w) -d C(w) ), and its standard error as ( ) = √ ( ( ) ) 2 + ( ( ) ) 2 . For studies reporting only the mean μ w and the standard deviation SD w of the PRO scores for each arm separately, we calculated the mean change at week w as d w = μ w -μ 0 . The standard deviation of the individual changes was calculated taking into account the correlation between the paired measures, as and SE(d C(w) ) were imputated. The imputed value was calculated as the weighted mean of the two values relative to the previous and the subsequent weeks, with weights inversely proportional to the distance from the week of interest being estimated and the two available ones.
To estimate the pooled difference in mean changes at 12 and 24 weeks, we used a random-effects model, weighting each study estimate by the inverse of its variance. A pooled difference in mean change greater than 0 indicated a greater benefit in PROs score for the immunotherapy-containing arm. We also calculated the I 2 statistics, which express the percentage of the total observed variability due to heterogeneity between studies' results, and the Q statistic to test the null hypothesis of homogeneity between studies.
To adjust the overall pooled difference in mean changes at 12 and 24 weeks for potential baseline imbalances in PRO scores between treatment and control groups, we used a two-stage meta-analytical approach based on pseudo individual patient data (IPD), as described by Papadimitropoulou et al. (21) This approach proceeded as follows: for each included study, we first constructed the pseudo IPD using the means and the standard deviations of PRO scores reported at baseline and at the follow-up week of interest. Then, in the first meta-analytical stage, we fitted, for each of the N included studies, a linear model with the follow-up score as dependent variable and the treatment and the baseline score as independent variables. This yields N treatment effects δ i with standard error SE i , adjusted for baseline imbalance. At the second stage, a random-effects meta-analysis was run on the estimated study specific δ i to estimate the pooled adjusted treatment effect.
Meta-analysis of the time to deterioration hazard ratio.