Prediction of 5‐year overall survival of diffuse large B‐cell lymphoma on the pola‐R‐CHP regimen based on 2‐year event‐free survival and progression‐free survival

Abstract This study aimed to predict the 5‐year overall survival (OS) benefit of pola‐R‐CHP versus R‐CHOP in the POLARIX trial based on the 2‐year event‐free survival (EFS) and progression‐free survival (PFS) rates in diffuse large B‐cell lymphoma (DLBCL). We identified randomized controlled trials (RCT) published before 31 May 2023. The correlation between the logarithmic (log) hazard ratio (HR) for EFS (HREFS) or PFS (HRPFS) and the HR for OS (HROS) was estimated at the trial‐level. Correlation analysis was performed between 2‐year PFS or EFS and 5‐year OS rates at the treatment arm‐level. Linear regression models were used to calculate the 5‐year OS of pola‐R‐CHP and R‐CHOP. In the included 20 RCTs, a linear correlation between HREFS (r = 0.765) or HRPFS (r = 0.534) and HROS was observed at the trial‐ level. Two‐year EFS (r = 0.918) or 2‐year PFS (r = 0.865) correlated linearly with 5‐year OS. Linear regression analysis between 2‐year EFS/PFS and 5‐year OS gave estimated 5‐year OS rates between pola‐R‐CHP and R‐CHOP of 6.4% and 6.3%, respectively. Two‐year EFS and PFS are feasible early endpoints in patients with DLBCL treated primarily with immunochemotherapy. The pola‐R‐CHP regimen is expected to improve 5‐year OS.


| INTRODUCTION
2][3][4] The prognosis of previously untreated patients with DLBCL has improved, with a 5-year overall survival (OS) rate of 60%-95%.However, approximately 40% of patients develop disease progression or relapse. 5he unfavorable outcomes of patients with refractory/relapsed DLBCL demand more effective systemic regimens in the first-line treatment setting.
][38][39] Another early study also demonstrated that improvements in 3-year EFS/PFS are highly correlated with improvements in 5-year OS in aggressive non-Hodgkin lymphoma. 40Furthermore, improved EFS and PFS are associated with prolonged OS at the treatmentarm level in DLBCL. 41Given that EFS and PFS was effective early efficacy endpoints in DLBCL, we hypothesized that 2-year EFS and PFS might also serve as predictors of long-term OS in the immunochemotherapy era, and allow for precise assessment of long-term OS benefit.The aim of this study was to evaluate the correlations between EFS/PFS and OS and then predict the 5-year OS benefit of pola-R-CHP versus R-CHOP in the POLARIX trial (Tilly 2022). 32

| Data sources and searches
All RCTs published before 31 May 2023 were included via a systematic literature search of PubMed, EMBASE, and the Cochrane Library, mainly using the search terms "DLBCL AND rituximab".Simultaneously, we conducted a thorough literature search using all relevant synonyms to avoid missing any relevant publications.Formal publications and meeting abstracts were included.Two investigators (W.R.Z. and S.N.Q.) independently searched the relevant studies, and any discrepancies were settled in collaboration with a principal investigator (Y.X.L.).

| Inclusion and exclusion criteria
After eliminating duplicate publications, we included articles and abstracts that met the following criteria: (1) Phase III RCTs reporting the long-term survival of patients with DLBCL who received first-line rituximab-containing immunochemotherapy; (2) R-CHOP or R-CHOP-like regimens as a controlled treatment arm in RCTs; (3) data available for 2-year EFS/PFS and 5-year OS rates extracted from studies directly or the Kaplan-Meier survival curve; and (4) patients with DLBCL consisted of >80% of the whole-sample size.Studies were excluded if they met any of the following criteria: Phase I/II or retrospective studies; transformed or relapsed/refractory DLBCL; inadequate survival data; studies on serum positivity for HIV, hepatitis B/C virus, or Epstein-Barr virus; or a sample size of <90 patients per arm.

| Quality assessment
The quality of the included RCTs were assessed using the Cochrane risk of bias tool with six domains: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, and selective reporting (Figure S1).Ultimately, 20 RCTs were included in the analyses (Table 1). 6-25,29

| Data extraction
The detailed data of all included studies were extracted using a standardized form, including first author's name, year of publication, inclusion criteria, median follow-up time, number of patients in each treatment arm, 2-year EFS/PFS rate, 5-year OS rate, and hazard ratios (HRs).Survival rates and HRs were extracted from the included studies directly (labeled "*" in Table 1) or the Kaplan-Meier survival curve using Engauge Digitizer software. 43or a repeatedly reported RCT, the latest results with the longest follow-up time were selected.

| Statistical analysis
The correlation between the logarithmic (log) HR for EFS (HR EFS ) or PFS (HR PFS ) and the log HR for OS (HR OS ) was estimated using the Pearson correlation coefficient, r, in weighted linear regression at the triallevel.The correlation between the 2-year EFS or PFS rate and the 5-year OS rate was also evaluated using r in weighted linear regression (with the weight equal to sample size) at the treatment arm-level.A value of r close to 1 indicated a strong correlation.The 5-year OS rates of the two treatments in the POLARIX trial were calculated using these two linear regression models.To assess the consistency and stability of the developed predictive models in different treatment settings, sensitivity analyses were performed by leaving out each subgroup of trials in turn.All statistical analyses were performed using SPSS (version 27.0,IBM Corp., Armonk, NY, USA).Data visualization was performed using the gg-plot2 package in R software (version 4.2.3,R Foundation for Statistical Computing).

| Baseline characteristics of the included studies
The median follow-up time ranged from 3.3 to 9.3 years (Table 1).Eleven RCTs (55%) reported one pair of HR EFS and HR OS ; one RCT reported two pairs of HR EFS and HR OS .Sixteen RCTs (80%) reported one pair of HR PFS and HR OS ; one RCT reported three pairs of HR PFS and HR OS .Moreover, 2-year EFS rates were extracted from 11 RCTs with 24 treatment arms, whereas 2-year PFS rates were extracted from 18 (95%) RCTs with 39 treatment arms.

| Correlation between HR EFS or HR PFS and HR OS at the trial-level
We first determined the treatment effects of EFS and PFS on OS in RCTs in the setting of immunochemotherapy.At the trial level, there was a strong correlation between 13 pairs of log HR EFS and log HR OS (r = 0.765; 95% CI: 0.754-0.775;Figure 1A).The analysis of 19 pairs of log HR PFS and log HR OS demonstrated a moderate correlation (r = 0.534; 95% CI: 0.521-0.547;Figure 1B).This finding indicated that the treatment benefit in terms of EFS or PFS correlated with OS benefit at the trial-level.Sensitivity analyses showed good consistency in most subgroups (Table 2).

| Linear correlation between 2-year PFS or EFS rates and 5-year OS rates at the treatment arm level
We further analyzed the correlation between 2-year PFS or EFS rates and 5-year OS rates, and established two linear regression models.At the treatment arm level, 2year EFS (r = 0.918; 95% CI: 0.913-0.922;Figure 2A) or 2-year PFS (r = 0.865; 95% CI: 0.861-0.870;Figure 2B) correlated linearly with 5-year OS.The linear regression models of EFS/PFS and OS were as follows: 5-year OS (%) = 0.897 × 2-year EFS + 14.082% (Figure 2A), and 5-year OS (%) = 0.866 × 2-year PFS + 12.465% (Figure 2B).As estimated from the EFS/PFS predictive models, an absolute gain of 10% 2-year EFS and PFS provides 8.97% and 8.66% improvements in 5-year OS, respectively.This finding indicated that improvement in 2-year EFS or PFS is associated with increased 5-year OS probabilities.Sensitivity analyses showed good consistency in all subgroups.There was always a strong correlation between 2-year PFS or EFS and 5-year OS at the treatment armlevel, regardless of included subgroups (Table 2).

| Predicted 5-year OS rates for the POLARIX trial using linear regression models
Using the EFS linear regression model, the predicted 5-year OS rates for treatment with pola-R-CHP versus treatment with R-CHOP for the intention-to-treat population were 77.4% versus 71%, respectively (Table 3).Similarly, based on the PFS linear regression model, patients treated with pola-R-CHP (74.7%) showed better 5-year OS than those treated with R-CHOP (68.4%) (Table 3).The absolute differences in the predicted 5year OS rates between the two treatment groups were 6.4% and 6.3% using the EFS and PFS linear regression models, respectively.

F I G U R E 1 Trial-level correlation between treatment effects on EFS or PFS and OS in RCTs.
Trial-level correlations between HR for EFS and HR for OS (A), and HR for PFS and HR for OS (B).The circle size is proportional to the number of patients in each comparison.The solid blue line indicates the fitted weighted linear regression line; the light gray zone represents its 95% CI.CI, confidence interval; EFS, event-free survival; HR, hazard ratio; OS, overall survival; PFS, progression-free survival; r, correlation coefficient; RCTs, randomized controlled trials.
Our study demonstrated the feasibility of EFS and PFS as early endpoints in patients with DLBCL primarily treated with immunochemotherapy at the trial-level.Moreover, the linear association of 2-year EFS/PFS with 5-year OS was confirmed at the treatment arm-level, providing two predictive models.The predicted 5-year OS differences were consistently ~6% with pola-R-CHP versus R-CHOP in the POLARIX trial.The results of our study provide a practically and valuable suggestion for the application of pola-R-CHP to treat newly diagnosed DLBCL.
After the introduction of rituximab, which showed improved results for DLBCL, a ceiling effect of first-line rituximab-based immunochemotherapy has been observed over the last decade,  until the emergence of polatuzumab vedotin. Currntly, two antibody drug conjugates (polatuzumab vedotin and loncastuximab tesirine) have been approved for previously untreated or relapsed/ refractory DLBCL.[44][45][46] The pilot study from the POLARIX trial demonstrated 2-year EFS and PFS benefits of pola-R-CHP compared with R-CHOP in patients with intermediate-and high-risk DLBCL.32 The potent therapeutic efficacy of pola-R-CHP might be attributed to the characteristics of the highly specific targeting ability and potent killing effect of polatuzumab vedotin. Howevr, with a median follow-up time of 2.4 years, the OS benefit was not confirmed in the POLARIX trial.32 Although 2-year EFS and PFS rates are widely used as surrogate endpoints in RCTs for DLBCL, it remains unclear whether early efficacy endpoints could predict long-term OS.In this study, we further demonstrated the linear correlations of 2-year EFS/PFS with 5-year OS at both the trial-and treatment arm-level, indicating the association between improved 2-year EFS or PFS and increased long-term OS probabilities.Based on the EFS/PFS predictive models, the pola-R-CHP regimen is expected to result in a 5-year OS improvement of approximately 6% compared with that of  R-CHOP.This finding provides additional data supporting the recommendation of pola-R-CHP in the current NCCN guidelines. 35The strengths of this study include strict inclusion criteria with uniform control treatment arms, a relatively large sample size from the RCTs, and an innovative approach to predicting OS rates.First, the data were obtained from high-quality RCTs with strict screening processes according to our inclusion criteria.Our study enrolled a relatively large cohort (>12,000) of patients with newly diagnosed DLBCL treated with rituximabcontaining immunochemotherapy.The enrolled RCTs on immunochemotherapy with R-CHOP or R-CHOPlike regimens as a controlled treatment arm for analysis made the linear regression models more applicable in the modern treatment era.Second, applying linear regression models to predict OS rates is unique in a specific RCT.If the predictive 5-year OS rate and absolute difference between the two treatment arms are proven in the subsequent individual data analysis, the established 2-year PFS and EFS linear regression models for DLBCL could be widely used in clinical practice.Any attempt at applying predictive model to predict OS rate of a specific RCT has the potential to assist in the design of future clinical trials and the interpretation of interim trial data.

OS predicted by early EFS OS predicted by early PFS
Our study has several limitations.First, the definition of PFS and EFS events and the requirement for follow-up were inconsistent across the included RCTs.In addition, the exact date of disease progression was dependent on the interval between two consecutive follow-up visits, which could result imprecise dates in clinical practice.Such an inherent heterogeneity cannot be removed nor quantified in the linear regression analysis.Second, the establishment of the linear regression models in this study did not take into account information regarding post-progression treatment because of the lack of data on salvage treatment in these RCTs.8][49][50][51] The OS improvements with novel salvage therapy strategy may alter the effect of first line therapy in DLBCL, leading to the decreased predictive ability of predicted models for OS.Upon the enhanced efficacy of salvage treatment or the availability of additional RCTs reporting salvage treatment data, the linear regression models will be further modified and optimized.However, it would be still very important to cure DLBCL patients with first-line therapy as these salvage therapies are more complex and expensive, and have potentially serious side effects.Third, because this is a literature-based predictive study without individual patient data, it is impossible to make a final conclusion until the long-term OS in the POLARIX trial is published.
In conclusion, using the established linear regression models of 2-year EFS/PFS and 5-year OS in the modern treatment era, it is suggested that the pola-R-CHP regimen for intermediate-and high-risk DLBCL would improve the 5-year OS by approximately 6% compared with using R-CHOP.

T A B L E 1
Summary of 20 Phase III randomized controlled trials on immunochemotherapy in trial-and treatment arm-level analyses.(like) versus R + intensified/de-escalated chemotherapy (n = 11) LNH03-2B

F I G U R E 2
Rituximab immunochemotherapy armlevel correlation between 2-year EFS or PFS and 5-year OS in RCTs.The rituximab immunochemotherapy arm-level associations between 2-year EFS and 5-year OS (A), and 2-year PFS and 5-year OS (B).The circle size is proportional to the number of patients in each treatment arm.The solid blue line indicates the fitted weighted linear regression line; the light gray zone represents its 95% CI.CI, confidence interval.EFS, event-free survival; OS, overall survival; PFS, progression-free survival; r, correlation coefficient; RCTs, randomized controlled trials.T A B L E 3 Predicted 5-year OS rates by 2-year EFS and PFS rates.