Conventional quality-of-life data analyses underestimate gains in Mantle Cell Lymphoma clinical trials: A re-examination of enrich study data

Abstract Introduction ENRICH compared Ibrutinib + Rituximab (IR) with immunochemotherapy (R-CHEMO) as first line treatments for mantle cell lymphoma (MCL) in older people. A key secondary outcome was quality of life (QOL), measured using the EORTC QLQ-C30 (QLQ-C30), a patient-reported outcome measure (PROM) used widely in cancer clinical trials. PROMs have fixed and restricted measurement ranges. These can constrain the magnitude of observed changes when samples are not well aligned to PROMs. We demonstrate, using ENRICH study QLQ-C30 physical functioning subscale (PF-5) data, that the PF-5 constrained measurement. When this constraint was reduced, remarkably different results were found. Patients and methods People aged ≥60ys with untreated, stage II-IV MCL were randomised to IR or R-CHEMO. The treatment arms were 6-8 cycles of chemotherapy or daily Ibrutinib. Both arms received Rituximab; 6-8 cycles then maintenance for 2ys. Ibrutinib participants continued daily treatment until disease progression or unacceptable toxicity. QLQ-C30 data were collected at baseline (BL), end of treatment (EOT) and end of maintenance (EOM). PF-5 data were analysed using the modern psychometric method Rasch measurement theory (RMT). Unlike conventional PROM data analyses, RMT provides forensic, context-specific information enabling measurement problems to be diagnosed and managed. Specifically, changes in PF-5 physical functioning estimates from BL to EOT in the total sample (TS), where measurement of many people was constrained (see below), were compared with changes in the subset of people whose physical functioning measurement was unconstrained (UC sample). PF-5 changes were compared at the group-level (p-values from paired samples t-tests, effect sizes [ES]) and individual-level (percent of sample with a significant improvement in PF-5 estimate, SI%). Results ENRICH randomized 397 patients (IR=199; R-CHEMO=198). QLQ-C30 data were available for n=376 at BL, and n=297 at BL and EOT. In the people with data at BL and EOT (TS; n=297), PF-5's measurement range was poorly aligned to the BL physical functioning of ENRICH participants. The distribution of PF-5 estimates was artificially skewed, and the ceiling effect was 30%, indicating constrained measurement as many people could not improve their PF-5 estimate. Only n=151 (50.8%) participants had unconstrained measurement as they could have a significant change from their BL estimate (UC sample) The group-level and individual-level changes for the two treatment arms, in the TS and UC samples are below. They show: A) group level changes were statistically and clinically trivial in the total sample, but significant in the UC sample. B) more individuals had significant changes in NC than TS sample. C) IR was superior to R-CHEMO, albeit not significantly. R-CHEMO [p (ES), SI%] TS: p=0.816, ES=-0.02, SI%=4.3%; UC: p=0.026, ES=0.25, SI%=7.4%. I-R [p (ES), SI%] TS: p=0.111, ES=0.13, SI%=8.3%; UC: p=<0.001, ES=0.43, SI%=17.1%. The Fatigue and Global Health Status QLQ-C30 subscales were also examined fully. The same pattern of results was found. The other six multi-item QLQ-30C subscales were examined partially. All had evidence of constrained measurement implying the same pattern of results would be found. Conclusion ENRICH was associated with statistically and clinically significant gains in QoL that conventional total sample analyses underestimated and misrepresented. IR was consistently superior to R-CHEMO. Results imply clinical trial PROM data should be examined pro-actively for constrained measurement. This supports the routine use of modern psychometric methods like RMT, to better capture patient-perceived benefits. .