Introduction. A growing body of evidence from both solid and hematologic malignancies has demonstrated a significant center effect on treatment outcomes, observed in both real-world settings and clinical trials. This phenomenon may be driven by a combination of patient-level and center-level factors. Key contributors include treatment center volume, geographic location, socioeconomic context, and institutional expertise. While prospective phase II and III trials attempt to mitigate such disparities through strict inclusion/exclusion criteria and stratification based on geographic region or prognostic factors such as International Prognostic Index (IPI), residual center effect may still influence trial outcomes. In this study, we examine the impact of center effect on treatment outcomes among patients with newly diagnosed large B-cell lymphoma treated with frontline immunochemotherapy across consecutive LYSA clinical trials.

Patients and Methods. A total of 3,149 patients enrolled in 11 consecutive prospective LYSA trials (LNH01-5B, LNH03-2B, LNH03-39B, LNH03-3B, LNH03-6B, LNH03-7B, LNH07-3B, LNH09-1B, LNH09-7B, REMARC, and GAINED) were included in the analysis. To assess the impact of center effect on treatment outcomes, specifically progression-free survival (PFS) and overall survival (OS), we employed a mixed-effects (frailty) model, incorporating treatment center as a random effect. The model was adjusted for relevant fixed effects including age, sex, IPI, and trial cohort. Empirical Bayes Estimates (EBEs) of the random center effect were calculated to quantify the correlation between individual centers and outcomes. Centers were then dichotomized: centers with EBEs below zero (the assumed mean value of the random effect distribution) were associated with longer PFS, while those above were associated with shorter PFS.

Results. Age, sex, IPI, and trial cohort were all significantly associated with both PFS and OS. Incorporating a random center effect significantly improved the model fit compared to a fixed-effect model without center adjustment (P=9×10-14 for PFS; P=2×10-7 for OS), indicating that despite strict eligibility criteria, trial stratification, and multivariable adjustment for potential confounders, a highly significant residual center effect on outcomes remained.

To explore potential explanations for this effect, we evaluated imbalances in both patient-level and center-level characteristics. Patients treated in centers associated with longer PFS were significantly more likely to be treated in tertiary care institutions rather than secondary care centers (P<1×10-5), and in geographic areas not considered medical deserts (P<1×10-3).

Importantly, the diagnosis-to-treatment interval (DTI) —a known prognostic factor correlated with disease severity and a potential confounder— was not longer in centers associated with better PFS (median DTI was 0.76 months versus 0.89 months in centers with longer PFS versus those with shorter PFS, respectively), suggesting that earlier treatment initiation did not account for the observed center effect.

The center effect remained consistent when analyzing disease progression and non-relapse mortality as competing events, suggesting that both contributed to disparities in PFS outcomes between centers.

Finally, when PFS was included as a time-dependent covariate in the mixed-effects model for OS, the center effect was greatly reduced but remained statistically significant, indicating that part of the center-related disparity in OS persists even after relapse or progression (P=3×10-4).

Conclusion. These findings suggest that, despite rigorous inclusion/exclusion criteria and statistical stratification, a significant and underrecognized center effect persists in large B-cell lymphoma trials. Imbalances in the type of treatment center across study arms may introduce unintentional bias, potentially confounding trial outcomes. To mitigate this risk, future trials should consider incorporating randomization stratified not only by traditional prognostic factors such as IPI and geographic region, but also by type of treatment center (e.g., tertiary vs. secondary care) aggregating unmeasured confounders like socioeconomic determinants, primary care access in larger cities, or opportunity to enter early-phase trial at relapse. Such measures could help minimize the impact of center-related variability on key efficacy endpoints.

This content is only available as a PDF.
2025
Sign in via your Institution