Exposure-safety and exposure-efficacy analyses of polatuzumab vedotin in patients with relapsed or refractory diffuse large B-cell lymphoma

Abstract

Exposure-response relationships were investigated to assess the risk/benefit of polatuzumab vedotin (pola) + bendamustine-rituximab (pola + BR) in relapsed/refractory diffuse large B-cell lymphoma (R/R DLBCL). Analyses were conducted in pivotal study GO29365 (NCT02257567; BR/pola + BR/pola + BG [BG: bendamustine-obinutuzumab]; 1.8 mg/kg pola, every 3 weeks [Q3W], six cycles), and supportive studies DCS4968g (NCT01290549) and GO27834 (NCT01691898) (pola/pola + R/pola + G; 0.1–2.4 mg/kg pola Q3W; eight-cycle landmark), separately. Exposure was characterized as simulated cycle-6 AUC and C_max for antibody-conjugated mono-methyl auristatin E (acMMAE) and unconjugated MMAE. Supportive studies showed response rate and safety risk (grade ≥2 peripheral neuropathy; grade ≥3 anemia) increased with exposure, suggesting not to dose below 1.8 mg/kg (up to eight-cycle) for balancing safety and efficacy. Pivotal study with limited exposure range showed no exposure-safety relationship and slightly positive exposure (acMMAE)-efficacy relationship for overall survival. The exposure-response analyses and the observed risk/benefit characteristics in pivotal study supported pola (1.8 mg/kg) +BR Q3W for six cycles in R/R DLBCL patients.

Keywords:

• Polatuzumab vedotin
• relapsed/refractory
• diffuse large B-cell lymphoma
• pharmacokinetics
• exposure-response
• dose justification

Introduction

Diffuse large B-cell lymphoma (DLBCL) is the most common subtype of non-Hodgkin lymphoma (NHL). It is an aggressive disease, and fatal if left untreated [1]. Rituximab (R) plus cyclophosphamide, doxorubicin, vincristine, and prednisone (R-CHOP) is the standard of care for first-line treatment of DLBCL. However, up to 30–40% of patients experience disease progression following an initial response to therapy [2].

Polatuzumab vedotin (pola) (POLIVY, Genentech, Inc., South San Francisco, CA) is an antibody–drug conjugate consisting of a humanized immunoglobulin G1 CD79b monoclonal antibody and the anti-mitotic agent, mono-methyl auristatin E (MMAE) [3,4]. In 2019, pola in combination with bendamustine (B) and rituximab (R) was approved in the United States for the treatment of relapsed/refractory (R/R) DLBCL after at least two prior therapies [5]. The safety, efficacy, and pharmacokinetics (PK) of pola have been evaluated in a number of phase 1–2 studies in patients with previously untreated or R/R NHL [6–8]. In the phase 1 open-label, dose-escalation study, DCS4968g, pola was administered as a single agent or in combination with R to assess the safety and PK of pola in R/R B-cell NHL, including R/R DLBCL, R/R follicular lymphoma (FL), and other NHL subtypes [7]. The phase 2 open-label, multi-center study, GO27834, investigated the safety and efficacy of pola + R or pola + obinutuzumab (G) in patients with R/R B-cell NHL [6,8]. GO29365 was a phase 1 b/2, open-label, international, multi-center, randomized study. As the pivotal study, it was designed to investigate the safety and efficacy of pola in combination with B plus R (pola + BR) or G (pola + BG), compared with BR in patients with R/R DLBCL and R/R FL [9,10]. The randomized phase 2 portion of the pivotal trial (N = 80, 40 patients per arm) showed that the addition of pola to BR significantly improved the independent review committee [IRC] assessed complete response (CR) rate (40.0% vs. 17.5%; p = .026) at the end of treatment and reduced the risk of death by 58% (hazard ratio for overall survival (OS), 0.42; 95% confidence interval: 0.24–0.75; p = .0023) in patients with R/R DLBCL, based on median follow-up duration of 22.3 months [10].

Exposure-response analyses have increasingly played an important role in supporting dose labeling for oncology drugs [11,12]. In the current study, we aimed to investigate the exposure-safety and exposure-efficacy relationships for pola to support the label dose of 1.8 mg/kg every 21 days (Q3W) for a fixed six-cycle treatment duration in patients with R/R DLBCL, when given in combination with BR [5].

Methods

Study design and patients

Two separate exposure-response analyses were performed for pola in R/R DLBCL patients: one for the pivotal study, GO29365 (NCT02257567), and the other for supportive studies, DCS4968g (NCT01290549) and GO27834 (ROMULUS; NCT01691898). The studies were approved by institutional review boards at each participating site and were conducted in accordance with the Declaration of Helsinki and the International Conference on Harmonization guidelines for Good Clinical Practice. All patients provided written informed consent.

In the pivotal study, the fixed duration of six-cycle Q3W dosing at 1.8 mg/kg was selected to consider the typical use of six-cycle treatment duration of BR in patients with R/R DLBCL and also to mitigate the risk of grade ≥2 peripheral neuropathy [13]. The pivotal study included BR control, pola + BR, and pola + BG. All patients with R/R DLBCL in the supportive studies were treated Q3W for varying durations: DCS4968g included pola monotherapy (0.1−2.4 mg/kg) and pola + R (2.4 mg/kg) with treatment until disease progression; GO27834 included pola + R (2.4 mg/kg) up to 1 year and pola + G (1.8 mg/kg) up to eight cycles.

Given the different treatment durations in the supportive studies compared with the pivotal study, and its potential impact on efficacy and safety outcomes, landmark data cutting was applied to the supportive studies so that the cut data could be used to justify the dose and regimen in the pivotal study. In the supportive studies, data were cut at the eight-cycle landmark (i.e. cycle 9 day 1) for patients with more than eight cycles of treatment. The eight-cycle data cut was selected to include the tumor assessment scheduled at the end of cycle 8. In this manuscript, ‘supportive studies’ therefore refers to the studies with landmark cut, unless otherwise indicated. The primary response assessment (PRA) time for the pivotal study was approximately 6−8 weeks after cycle 6 day 1 (also around cycle 8) or the last dose.

For the pivotal study, the pola + BR and pola + BG arms were included in the exposure-safety analyses (N = 69); pola + BR (as the approved combination) arm was included in the exposure-efficacy analyses (N = 44); the BR control arm was included in the Kaplan-Meier plots of the time-to-event endpoints in the efficacy analyses (N = 40). For the supportive studies, the R/R DLBCL patients in all treatment arms were included in the exposure-safety analyses (N = 119); only pola monotherapy and pola + R (N = 76) arms were included in the exposure-efficacy analyses as G was not in the approved combination.

The exposure-response analyses were conducted separately for the pivotal and supportive studies, given that B was only used in the pivotal study and that the treatment durations were slightly different even after landmark cutting of supportive studies.

The exposure-response analyses endpoints

The pivotal and supportive studies had the same safety endpoints, including the probability of key treatment-emergent adverse events of special interest (AESIs), probability of pola dose modification due to adverse event (AE), time to the first pola dose modification due to AE, and the dose intensity of pola, R, and B (pivotal study only). Dose modification included dose reduction, delay, and early discontinuation. Dose intensity (%) was computed based on the actual doses administered to each patient up to the end-of-treatment assessment relative to the planned dose, accounting for dose delay and dose reduction, but not early discontinuation. The landmark cutting was not applied to the dose intensity calculation of the supportive studies. AESIs included grade ≥3 neutropenia, infections and infestations, anemia, thrombocytopenia, diarrhea, aspartate aminotransferase (AST) increase, alanine aminotransferase (ALT) increase, and bilirubin increase, as well as grade ≥2 peripheral neuropathy (PN). Among them, endpoints with <5% incidence rate or <5 events were excluded from the analyses.

The safety endpoints were captured slightly differently in the datasets for the pivotal and supportive studies. AESIs data included in the pivotal study analyses were from first pola dose until study discontinuation, whereas AESIs included in the supportive studies analyses were from the first pola dose until 30 days (GO27834) or 90 days (DCS4968g) post-last dose of any study treatment (i.e. the safety follow-up window: FUW). The exclusion of AESIs data beyond FUW in the supportive studies only impacted patients with less than eight cycles treatment, as patients exceeding eight cycles treatment were subjected to the landmark cutting that occurred early than the last pola dose. The exclusion was not expected to impact the exposure-safety outcomes, as for the majority of AESIs, the first event (if any) would have occurred during treatment or at least within the FUW.

In the pivotal study, the efficacy endpoints included the IRC-assessed CR (IRC-CR) and objective response (IRC-OR) at the PRA by positron emission tomography-computed tomography (PET-CT), IRC-assessed best overall response (IRC-BOR) by PET-CT or CT, duration of objective response (IRC-DOR), progression-free survival (IRC-PFS), and OS. In the supportive studies, the efficacy endpoints included the probability of investigator-assessed best overall response (INV-BOR) up to the cycle 8 landmark and INV-OR at cycle 8 landmark, assessed by CT or PET-CT.

Pola exposure metrics

The exposures of antibody-conjugated mono-methyl auristatin E (acMMAE) and unconjugated MMAE were used for the exposure-response analyses. The acMMAE exposures are highly correlated with pola dose. acMMAE is considered as the key analyte driving efficacy and safety of pola [14]. Unconjugated MMAE is a catabolite of the conjugate. It is not considered as a major driver of efficacy based on the design of the molecule but might be associated with safety risk given its high potency [15].

Exposure metrics for pola were defined as the simulated cycle 6 area under the concentration−time curve (AUC) and maximum concentration (C_max) for acMMAE and unconjugated MMAE based on the nominal dose at cycle 1. Individual empirical Bayes estimates from an integrated two-analyte population PK model [16] were used for the simulation. Four patients with intra-patient dose escalation in DCS4968g were excluded from the analyses. Cycle 6 AUC and C_max for acMMAE and unconjugated MMAE were used for the exposure-safety analyses. Cycle 6 AUC for acMMAE was used for the exposure-efficacy analyses.

The details of the studies, treatment arms, exposure metrics, and endpoints for each analysis are summarized in Table 1.

Table 1. Study details, treatment arms, exposure metrics and endpoints included in the exposure-response analyses.

Exposure-response analysis	Studies	Analyte/exposure metrics	Arm included (dose)	N total (n by arm)	Endpoints
Exposure-safety analyses in R/R DLBCL	Supportive studies DCS4968g GO27834	acMMAE AUC, Cmax	Pola monotherapy (0.1–2.4 mg/kg Q3W until PD or unacceptable toxicity) Pola + R (2.4 mg/kg Q3W for up to 1 year or until PD or unacceptable toxicity) Pola + G (1.8 mg/kg Q3W × eight cycles)	119 (36/40/43)	AESI: grade ≥3 neutropenia, grade ≥2 PN, grade ≥3 infection and infestation, grade ≥3 anemia, grade ≥3 thrombocytopenia, grade ≥3 diarrhea, grade ≥3 AST increase (by lab), grade ≥3 ALT increase (by lab), and grade ≥3 total bilirubin increase (by lab) Dose intensity related measurements: pola dose modification due to AE, time to first dose modification (reduction, delay or early discontinuation) due to AE, pola dose intensity, R dose intensity, and B dose intensity (pivotal study only)
		Unconjugated MMAE AUC, Cmax
	Pivotal study GO29365	acMMAE AUC, Cmax Unconjugated MMAE AUC, Cmax	Pola + BR (1.8 mg/kg Q3W × six cycles) Pola + BG (1.8 mg/kg Q3W × six cycles)	69 (44/25)
Exposure-efficacy analyses in R/R DLBCL	Supportive studies DCS4968g GO27834	acMMAE AUC	Pola monotherapy (0.1–2.4 mg/kg Q3W until PD or unacceptable toxicity) Pola + R (2.4 mg/kg Q3W for up to 1 year or until PD or unacceptable toxicity)	76 (36/40)	INV-BOR up to cycle 8 landmark by PET-CT or CT; INV-OR at cycle 8 landmark by PET-CT or CT
	Pivotal study GO29365	acMMAE AUC	Pola + BR (1.8 mg/kg Q3W × six cycles) BR Q3W × six cycles	84 (44/40)	IRC-CR at PRA by PET-CT; IRC-OR at PRA by PET-CT (IRC-OR); IRC-BOR by PET-CT or CT; IRC-DOR up to clinical cutoff date; IRC-PFS; OS The BR control arm was included in the Kaplan-Meier time-to-event efficacy analyses (IRC-DOR; IRC-PFS; OS)

ac: antibody conjugate; AE: adverse event; AESI: adverse event of special interest; ALT: alanine aminotransferase; AST: aspartate aminotransferase;

AUC: area under the concentration − time curve; B: bendamustine; BOR: best overall response; C_max: maximum concentration; CR: complete response; CT: computed tomography; DLBCL: diffuse large B-cell lymphoma; DOR: duration of response; G: obinutuzumab; INV: investigator; IRC: independent review committee; MMAE: mono-methyl auristatin E; OR: objective response; OS: overall survival; PET: positron emission tomography; PD, progressive disease; PFS: progression-free survival; PK: pharmacokinetics; PN: peripheral neuropathy; pola: polatuzumab vedotin; PRA: primary response assessment; Q3W: every 3 weeks; R: rituximab; R/R: relapsed/refractory.

Modeling methods

Logistic regression modeling was used to assess the correlation between the probability of AESIs or pola dose modification due to AE and acMMAE or unconjugated MMAE exposures. It was also used to assess the association between the probability of response rate (INV-BOR, INV-OR, IRC-CR, IRC-OR, IRC-BOR) and acMMAE AUC. The linear model, log linear model (linear model for log of exposure), and non-linear E_max model in the logit scale were tested. Model selection was based on Akaike information criterion. Kaplan–Meier plots and Cox proportional hazards (Cox PH) models were used to assess exposure-response relationships for each time-to-event endpoint (time to first pola dose modification due to AE, IRC-DOR, IRC-PFS, OS). The BR control arm of the pivotal study was included for the Kaplan–Meier plots of efficacy endpoints only. The association between the dose intensity and acMMAE or unconjugated MMAE exposures was assessed by Locally Weighted Scatterplot Smoothing (LOWESS) regression and summarized by exposure tertiles.

Covariate analyses were performed for the logistic regression and Cox PH models in case of a positive exposure-response relationship, through the forward addition (p = .01) and backward elimination (p = .001) procedure for a list of baseline variables, including, for example, demographics, baseline laboratory measurements and disease characteristics, and treatment characteristics (Supplementary Appendix Table A). Only AUC models were used in the covariate analysis when both AUC and C_max were significant. AUC was always kept in the model during the forward addition and backward elimination process.

Results

Exposure-safety analyses

The number (and percentile) of patients with AESIs is shown in Table 2. Grade ≥3 diarrhea, AST, ALT, and total bilirubin elevation were excluded from the analyses given their low incidence (<5% or <5 events).

Table 2. Number (%) of patients with AEs in the exposure-safety analyses.

	N (%) patients with AE
	Supportive studies (N = 119)			Pivotal study
	All doses (N = 119) (pola; pola + R, pola + G)	Pola 1.8 mg/kg (N = 46) (pola; pola + G)	Pola 2.4 mg/kg (N = 67) (pola; pola + R)	Pola 1.8 mg/kg (N = 69) (pola + BR; Pola + BG)
Grade ≥3 neutropenia	39 (32.8)	12 (26.1)	26 (38.8)	35 (50.7)
Grade ≥2 PN	30 (25.2)	10 (21.7)	20 (29.9)	10 (14.5)
Grade ≥3 infections and infestations	12 (10.1)	6 (13.0)	6 (9.0)	20 (29.0)
Grade ≥3 anemia	9 (7.6)	3 (6.5)	6 (9.0)	13 (18.8)
Grade ≥3 thrombocytopenia	6 (5.0)	5 (10.9)	1 (1.5)	24 (34.8)
Grade ≥3 diarrhea	5 (4.2)	0	5 (7.5)	4 (5.8)
Grade ≥3 AST increase	2 (1.7)	0	2 (3.0)	1 (1.4)
Grade ≥3 ALT increase	3 (2.5)	1 (2.2)	2 (3.0)	1 (1.4)
Grade ≥3 bilirubin increase	1 (0.8)	1 (2.2)	0	2 (2.9)
Pola dose modification due to AE	32 (26.9)	11 (23.9)	21 (31.3)	40 (58.0)

AE: adverse event; ALT: alanine aminotransferase; AST: aspartate aminotransferase; PN: peripheral neuropathy; pola: polatuzumab vedotin.

Shading indicates AEs of low frequency (<5% or <5 events) in the supportive studies (N = 119) and pivotal study (n = 69) population that were not included in the exposure-safety analyses.

Results of the exposure-safety analyses for the supportive studies across pola 0.1–2.4 mg/kg Q3W are summarized in Table 3. The linear logistic models were selected. The incidence of grade ≥2 PN increased significantly with increasing acMMAE exposure (p = .003 and .011 for AUC and C_max, respectively) (Figure 1(A,B)). The incidence of grade ≥3 anemia increased significantly with increasing unconjugated MMAE exposure (p = .010 and .015 for AUC and C_max, respectively) (Figure 1(C,D)). Although higher unconjugated MMAE exposure was significantly correlated with an earlier time to first pola dose modification due to AE (p = .004 and .002 for AUC and C_max, respectively, from Cox PH model) (Figure 2(A,B)), the probability of the pola dose modification due to AE was not significantly impacted (p = .455 and .325). The dose intensity of pola showed a trend of slight decrease with increasing acMMAE exposure; the overall dose intensity was high (>89%, data on file) across exposure tertiles. For other safety endpoints assessed (incidence of grade ≥3 neutropenia, infections and infestations, and thrombocytopenia; dose intensity of R), there were no statistically significant correlations with exposure.

Figure 1. Logistic regression for grade ≥2 PN with acMMAE exposure [(A) AUC; (B) C_max], grade ≥3 anemia with unconjugated MMAE exposure [(C) AUC; (D) C_max], INV-BOR with acMMAE exposure [(E) AUC], and INV-OR with acMMAE exposure [(F) AUC], for R/R DLBCL patients in the supportive studies. Red solid line = logistic regression model prediction; green shaded area = 90% confidence intervals; points show exposure of individual patients with events (p = 1) and without events (p = 0); black squares and vertical green lines = observed fraction of patients with events in each exposure group and 90% confidence intervals for these fractions; dashed vertical lines = bounds of exposure groups. Note: the exposure-safety analyses of the supportive studies in A–D were based on 119 R/R DLBCL patients receiving single-agent pola (0.1–2.4 mg/kg), pola + R (2.4 mg/kg), and pola + G (1.8 mg/kg) up to the eight-cycle landmark. The exposure-efficacy analyses of the supportive studies in E–F were based on 76 R/R DLBCL patients receiving single-agent pola (0.1–2.4 mg/kg) or pola + R (2.4 mg/kg) up to the eight-cycle landmark. ac: antibody conjugate; AUC: area under the concentration − time curve; BOR: best overall response; C_max: maximum concentration; DLBCL: diffuse large B-cell lymphoma; G: obinutuzumab; INV: investigator; MMAE: mono-methyl auristatin E; OR: objective response; PN: peripheral neuropathy; pola: polatuzumab vedotin; R: rituximab.

Figure 2. Kaplan-Meier plot for time to first pola dose modification due to AE with unconjugated MMAE exposure in the supportive studies [(A) AUC; (B) C_max], time to first pola dose modification due to AE with acMMAE exposure in the pivotal study [(C) C_max], and time to death (OS) by acMMAE exposure in the pivotal study [(D) AUC]. Note: Low and high exposure were stratified by median exposure; p-value in the figure legends: p-value of the log-rank test comparing patients with low and high exposure; the BR control arm was included in the Kaplan-Meier plots of OS (D), but not the Cox PH analysis of OS. ac: antibody conjugate; AE: adverse event; AUC: area under the concentration − time curve; C_max: maximum concentration; MMAE: mono-methyl auristatin E; OS: overall survival; PH: proportional hazards.

Table 3. Summary of the exposure-safety analyses based on the supportive and pivotal studies.

Adverse event	Analysis type	N	N (%) with event	p value
				acMMAE		Unconjugated MMAE
				AUC	Cmax	AUC	Cmax
Supportive studies
Grade ≥3 neutropenia	Linear logistic regression	119	39 (32.8)	.091	.157	.338	.339
Grade ≥2 PN		119	30 (25.2)	.003 (+)	.011 (+)	.571	.755
Grade ≥3 infections and infestations		119	12 (10.1)	.615	.631	.066	.101
Grade ≥3 anemia		119	9 (7.6)	.866	.964	.010 (+)	.015 (+)
Grade ≥3 thrombocytopenia		119	6 (5.0)	.638	.928	.327	.221
Pola dose modification due to AE		119	32 (26.9)	.340	.256	.455	.325
Time to the first pola dose  modification due to AE	Cox PH model	119	32 (26.9)	.359	.601	.004 (−)	.002 (−)
Pola dose intensity	LOWESS regression	119	–	–	–	–	–
R dose intensity		40	–	–	–	–	–
Pivotal study
Grade ≥3 neutropenia	Linear logistic regression	69	35 (50.7)	.649	.269	.810	.993
Grade ≥2 PN		69	10 (14.5)	.586	.965	.309	.226
Grade ≥3 infections and infestations		69	20 (29.0)	.376	.879	.765	.930
Grade ≥3 anemia		69	13 (18.8)	.888	.390	.117	.119
Grade ≥3 thrombocytopenia		69	24 (34.8)	.781	.780	.253	.253
Pola dose modification due to AE		69	40 (58.0)	.100	.062	.448	.465
Time to the first pola dose modification due to AE	Cox PH model	69	40 (58.0)	.145	.009 (−)	.639	.541
Pola dose intensity	LOWESS regression	69	–	–	–	–	–
B dose intensity		69	–	–	–	–	–
R dose intensity		44	_	–	–	–	–

ac: antibody-conjugated; AE: adverse event; AUC: area under the concentration − time curve; B: bendamustine; C_max: maximum concentration; Cox PH: Cox proportional hazards; LOWESS, Locally Weighted Scatterplot Smoothing; MMAE: mono-methyl auristatin E; PN, peripheral neuropathy; pola: polatuzumab vedotin; R: rituximab.

Gray shaded cells indicate p value <.05; when p-value <.05, (+) indicates the endpoint of interest increases with increasing exposure; (−) sign indicates the endpoint of interest decreases with increasing exposure.

Results of the exposure-safety analyses for the pivotal study based on 1.8 mg/kg Q3W (Table 3) showed no significant association between the probability of any AESI or dose intensity of pola, R, and B with acMMAE or unconjugated MMAE exposure. The time to first dose pola modification due to AE decreased significantly (p = .009 from Cox PH model) (Figure 2(C)) with increasing acMMAE C_max; the probability of pola dose modification due to AEs increased slightly with acMMAE C_max but the relationship was not statistically significant (p = .062).

Exposure-efficacy analyses

Results of the exposure-efficacy analyses for the supportive studies are described in Table 4. The linear logistic models were selected. The probability of INV-BOR and INV-OR increased significantly with increasing acMMAE AUC, with p values of .037 and .014, respectively (Figure 1(E,F)).

Table 4. Summary of the exposure-efficacy analyses based on acMMAE AUC in the supportive and pivotal studies.

Endpoint	Analysis type	N evaluable	N (%) with event	p value
Supportive studies
INV-BOR up to cycle 8	Linear logistic regression	76	31 (40.8)	.037 (+)
INV-OR at cycle 8		76	23 (30.3)	.014 (+)
Pivotal study
IRC-CR at PRA	Linear logistic regression	44	19 (43.2)	.521
IRC-OR at PRA		44	21 (47.7)	.956
IRC-BOR		44	26 (59.1)	.852
IRC-DOR at clinical cutoff date	Cox PH model	26	11 (42.3)	.057
IRC-PFS		44	26 (59.1)	.102
OS		44	24 (54.5)	.048 (+)

ac: antibody-conjugated; AUC: area under the concentration − time curve; BOR: best overall response; DOR: duration of response; Cox PH: Cox proportional hazards; CR: complete response; INV: investigator; IRC: independent review committee; MMAE: mono-methyl auristatin E; OR: objective response; OS: overall survival; PFS: progression-free survival; PRA: primary response assessment.

Shaded cells indicate p value <.05; when p value <.05, (+) indicates the endpoint of interest increases with increasing exposure; (−) indicates the endpoint of interest decreases with increasing exposure.

Results of the exposure-efficacy analyses for the pivotal study (Table 4) indicated no statistically significant correlations between acMMAE exposure and IRC-CR, IRC-OR, IRC-BOR, IRC-DOR or IRC-PFS. The OS increased significantly (borderline significance) with increasing acMMAE AUC (p = .048 from Cox PH model); there was no statistical significance by log-rank test (p = .135) stratified by median AUC, potentially due to the small sample size (Figure 2(D)).

Covariate analysis

Patient baseline covariates were summarized for the exposure-safety population (Supplementary Appendix Tables B−C) and the exposure-efficacy population (Supplementary Appendix Tables D–E) separately. The covariate range (continuous) or percentile (categorical) were largely similar between the pivotal and supportive studies for the common covariates.

For the covariate assessment of the supportive studies, baseline lactate dehydrogenase levels and the North American region were statistically significant during the forward inclusion for grade ≥2 PN (acMMAE AUC model) and grade ≥3 anemia (unconjugated MMAE AUC model), respectively, based on the logistic analysis. For the covariate assessment of the pivotal study, baseline B-cell count, Ann Arbor stage 4 (vs. <4), and ECOG status 0 (vs. >0) were statistically significant during the forward inclusion for OS (acMMAE AUC model), based on the Cox PH model. The exposure-response relationship remained significant in the presence of those covariates identified during the forward inclusion. No covariate was retained in the final model during the backward elimination.

Discussion and conclusions

Exposure-response analyses for pola were performed in the pivotal (GO29365) and supportive studies (DCS4968g and GO27834) separately in R/R DLBCL patients. The outcome of the supportive studies over a wide dose range was considered as the main evidence for the dose justification. It mapped out the risk/benefit profile of pola, administered as monotherapy or in combination with R/G, indicating that the 1.8 mg/kg dose or higher could be considered for those combinations to maintain sufficient efficacy with acceptable safety. Due to the overlapping AEs for pola and B, a dose higher than 1.8 mg/kg was not recommended for the pola + BR combination. The 1.8 mg/kg pola dose in combination with BR has the desired risk/benefit characteristics as it demonstrated significantly improved efficacy compared with BR [10]. The pivotal-study analysis provided the description of the exposure-response relationship for the target dose, regimen and combination (pola + BR, 1.8 mg/kg, 6 cycles). As no other doses were tested, it was not considered to be the main evidence for the dose justification.

The individual empirical Bayes estimates from the population PK model [16] were used to simulate the exposure metrics, defined as cycle 6 AUC and C_max for acMMAE and unconjugated MMAE based on nominal dose (i.e. assuming patients received the planned doses at cycle 1 during the entire study). Alternatively, the average exposure based on actual dose up to the time of onset for safety or efficacy events (considering dose adjustments until event onset) can be considered [14,17]; however, for patients without an event, the actual dose would be for the entire treatment period (biased toward lower actual dose), which would cause bias in the logistic analysis, especially for early onset AEs. In addition, for the time-to-event analyses, the longer patients stayed on trial before having an event, the higher the chance of dose adjustment, resulting in a lower actual dose; this could also cause bias [18]. Thus, nominal dose was used in this study for simulation. Given the larger inter-individual variability of cycle 1 exposure [16] and the planned six cycles of treatment, the cycle 6 exposure was chosen for the exposure-response analyses.

To increase the sample size for more robust exposure-safety analyses, pola + R and pola + G arms were pooled with pola monotherapy in the supportive studies, and pola + BR and pola + BG were pooled for the pivotal study. The G-containing arms were removed from the exposure-efficacy analyses. The pooling of monotherapy and R/G combinations in the supportive studies can be justified by the statistical insignificance of G (safety analyses only) and R in the covariate analysis.

The historical time-to-event analysis using all data from the supportive studies indicated that the incidence of grade ≥2 PN increased with both acMMAE exposure and treatment duration, suggesting a fixed duration of 6–8 cycles to mitigate the risk of grade ≥2 PN [13]. The historical exposure-efficacy analysis also indicated a comparable exposure-OR relationship for data up to eight-cycle landmark and up to the disease progression [14]. The historical analyses supported the capping of the treatment duration to 6–8 cycles (depending on the combination partners) in the subsequent pola clinical studies, such as GO29365 [10] and GO29044 [19].

In the presence of landmark cutting for the supportive studies, a significant correlation of acMMAE exposure with the probability of grade ≥2 PN, and unconjugated MMAE exposure with the probability of grade ≥3 anemia was observed, suggesting an increased safety risk at the highest dose tested (2.4 mg/kg). Grade ≥3 neutropenia, an important AE related to MMAE antibody-drug conjugates, did not show statistically significant associations with acMMAE or unconjugated MMAE exposure, despite being the most common AESI (32.8%; Table 2). This was likely confounded by the administration of granulocyte colony-stimulating factor as primary prophylaxis for neutropenia in pola patients. The exposure-efficacy analyses up to the eight-cycle landmark indicated a significant correlation between acMMAE AUC and tumor response rate (INV-BOR and INV-OR), suggesting that a lower exposure may be associated with lower efficacy. Considering the aggressive nature of R/R DLBCL, a dose level below 1.8 mg/kg is not recommended from an efficacy point of view. The 1.8 mg/kg dose or higher up to eight cycles could be considered for pola monotherapy or in combination with R/G. However, for combination partners with overlapping AEs, pola doses above 1.8 mg/kg would require further evaluation to establish risk/benefit.

The dosing regimen of 1.8 mg/kg pola Q3W up to six cycles was chosen in the pivotal study to combine with BR, considering the use of a six-cycle regimen of BR in R/R DLBCL patients, and the overlapping AEs for pola and B. Based on the limited sample size and exposure range from 1.8 mg/kg alone (Table 3), no significant exposure-safety relationships were identified. A borderline significance was identified for OS and acMMAE AUC, with a trend of longer OS for higher acMMAE AUC.

The safety profiles in the pivotal and supportive studies are different (Table 2) for various reasons. The slightly lower incidence of grade ≥2 PN in the pivotal study (14.5%) might be attributed to the shorter treatment duration (up to six cycles) compared with the supportive studies (21.7% at 1.8 mg/kg; up to eight cycles). As a chemotherapeutic agent, B has been shown to be associated with hematologic AEs [20,21] and could contribute to the higher incidence of grade ≥3 neutropenia, thrombocytopenia and anemia in the pivotal study (50.7%, 34.8%, and 18.8%, respectively) than the supportive studies at 1.8 mg/kg (26.1%, 10.9%, and 6.5%, respectively). B is also associated with an increased risk of infections and prolonged time-to-onset in indolent NHL patients, compared with non-B-containing treatment regimens [22], which is consistent with our data for grade ≥3 infections and infestations (29.0% & 5.8 months in pivotal study vs. 13.0% & 1.2 months in supportive studies at 1.8 mg/kg; Table 2 and Supplementary Appendix Figure A). The higher rate of pola dose modification due to AE in the pivotal study (58.0% vs. 23.9% in supportive studies at 1.8 mg/kg) could be related to the higher incidence of AESIs mentioned above. The overlapping AEs for pola and B in pivotal study may limit the opportunity to increase the pola dose above 1.8 mg/kg, supporting the approved dosing regimen of 1.8 mg/kg (six cycles) for pola + BR in R/R DLBCL.

In summary, the exposure-response analyses over a wide dose range in the supportive studies suggested that lowering the pola dose below 1.8 mg/kg is predicted to yield decreased efficacy in patients with R/R DLBCL. To reduce the risk of grade ≥2 PN, treatment duration should be limited to 6–8 cycles, depending on the combination partners. The observed risk/benefit characteristics for pola + BR in pivotal study, shown as the increased incidence of hematologic AEs and infections (overlapping AE for pola and B) and the significantly improved efficacy (IRC-CR, IRC-PFS, OS) versus BR control [10], supported the 1.8 mg/kg dose, not higher, for pola in combination with BR. Overall, considering the exposure-response analyses outcomes for the supportive studies and the observed risk/benefit characteristics in the pivotal study, the 1.8 mg/kg pola Q3W for six cycles is appropriate for the treatment of R/R DLBCL patients in combination with BR. This manuscript provided a strategy as to how to leverage the information from early studies (with a wider dose range but different drug combinations from pivotal study) and the pivotal study (with one dose level) to justify the recommended dose and schedule for the pivotal study. Additionally, this model could be used to further our understanding of how safety and efficacy endpoints are impacted by factors affecting pola exposure (e.g. drug-drug interactions, special populations, or formulation changes).

Disclosures statement

All authors, except LG, are employees of Genentech, Inc. and stockholders of Roche. LG is a paid consultant for Roche and Genentech.

Acknowledgments

We thank the study investigators and patients of the GO29365, DCS4968g, and GO27834 studies. Medical writing support, under the direction of the authors, was provided by Angela Rogers, PhD, from Gardiner-Caldwell Communications, and was funded by F. Hoffmann-La Roche Ltd.

Data availability statement

Qualified researchers may request access to individual patient level data through the clinical study data request platform (https://vivli.org/). Further details on Roche’s criteria for eligible studies are available here (https://vivli.org/members/ourmembers/). For further details on Roche's Global Policy on the Sharing of Clinical Information and how to request access to related clinical study documents, see here (https://www.roche.com/research_and_development/who_we_are_how_we_work/clinical_trials/our_commitment_to_data_sharing.htm).

Additional information

Funding

F. Hoffmann-La Roche Ltd. and Genentech, Inc. funded the GO29365, DCS4968g, and GO27834 studies.