Abstract
Purpose: Despite the expansion of immune checkpoint inhibitor (ICI) indications, the relationship between ICI dose and toxicity or response is not well established. To understand this correlation, we performed a meta-analysis of ICI trials that used dose escalation.
Experimental Design: We searched PubMed and abstracts presented at (inter)national meetings for trials using FDA-approved ICIs. The reported rates of grade 3–5 adverse events (G3–5 AE), immune-related adverse events (irAE), and response were correlated with doses within each ICI using marginal exact generalized linear models.
Results: A total of 74 trials (7,469 patients) published between January 2010 and January 2017 were included. For ipilimumab, the incidence of G3–5 AEs was 34% with a significant 27% reduced risk in lower doses (P = 0.002). However, no relationship was observed between dose and irAEs or response. For nivolumab, the incidence of G3–5 AEs was 20.1% which was lower in non–small cell lung cancer (NSCLC) compared with renal cell carcinoma (RCC) or melanoma (P ≤ 0.05) with no dose-toxicity relationship. In melanoma and NSCLC, a dose–response association was observed, which was not observed in RCC. For pembrolizumab, the incidence of G3–5 AEs was 13.3%, which was lower in melanoma compared with NSCLC (P = 0.03) with no dose-toxicity relationship. In melanoma, lower dose levels correlated with decreased odds of response (P = 0.01), a relationship that was not observed in NSCLC.
Conclusions: Our analysis shows a lack of consistent dose-toxicity or dose–response correlation with ICIs. Therefore, dose escalation is not an appropriate design to conduct ICI studies. Here we present an innovative trial design for immune-modulating agents.
Translational Relevance
The phase I dose-escalation trial design has been used in immune checkpoint inhibitor (ICI) development. Given the unique ICI mechanism of action and toxicities (immune-related adverse events), the dose-escalation trial design might not be the appropriate one to conduct ICI studies. In this meta-analysis, we report the lack of consistent correlation between ICI dose and clinical response or immune toxicities for CTLA-4 and PD-1/PD-L1 antibodies. Thus, we suggest a novel design, multiple dosing response seeking design, to determine the optimal ICI dose by enrolling patients concurrently at multiple dose levels and choose the lowest dose with the highest response rate while being vigilant for toxicities. This work provides the clinical investigators with a comprehensive understanding of the relationship between ICI dose and clinical response and toxicity and a new tool to conduct the early phases of the next generation of ICI clinical trials.
Introduction
The use of immune checkpoint inhibitors (ICI) has led to a paradigm shift in cancer treatment over the past decade demonstrating impressive clinical activity in a variety of malignancies (1, 2). As part of the FDA Safety and Innovation Act, the agency has expedited approval of many of these agents and granted breakthrough designation to others. This rapid pace of oncology drug development is unprecedented and will affect many future immune therapeutic agents entering clinical trials (3).
ICIs are known for their relatively improved safety profile compared with chemotherapy and targeted therapy. However, with more patients treated, there has been heightened awareness of a novel class of toxicities defined as immune-related adverse events (irAE), such as colitis and pneumonitis that can be fatal (4, 5). Agents targeting the PD-(L)1 and CTLA-4 pathways have similar toxicity profiles with different incidence and patterns. Twenty percent of patients receiving the anti-CTLA-4 agent ipilimumab report grade 3/4 irAEs (6) compared with 13%–14% of patients treated with anti-PD-1 [pembrolizumab (7) or nivolumab (8)] and 55% of those treated with nivolumab plus ipilimumab (9). Contrary to chemotherapy, irAEs generally do not occur within the first 2 weeks of ICI administration and while most AEs occur within 6–8 weeks they can occur any time after administration (10, 11).
Mirroring chemotherapy early clinical development, ICI clinical trials have followed the FDA-recommended standard dose-escalation design to determine target dosing. This assumed that higher dose is associated with greater toxicity and potentially increased efficacy. But contrary to chemotherapy, ICIs induce their clinical outcome through enhancing the immune system by blocking inhibitory checkpoints and breaking immune tolerance rather than direct cytotoxic targeting of cancer cells (12). This leads to increased activation and frequency of preexisting T cells that recognize both tumor and normal host (13). Accordingly, and as has been seen with other immune-activating agents (14), ICI dosing may not be a determinant for the activation level of T cells and hence clinical outcome.
Our understanding of ICI toxicity and clinical response kinetics is very limited and based on few early-phase trials with contradictory outcomes (7, 15–17). Such understanding of the relationship between dose and clinical outcome is crucial for proper drug development including clinical trial design and more importantly, for proper patient management. We conducted a meta-analysis of the relationship between dose and outcome to better understand the ideal methods for evaluating ICIs and to establish the need for a novel design in ICI drug development.
Materials and Methods
Study inclusion and literature search criteria
To establish our database, PubMed and abstracts presented at (inter)national meetings were queried from January 1, 2010 to January 1, 2017 for studies of FDA-approved ICIs including ipilimumab, atezolizumab (anti-PD-L1), nivolumab, and pembrolizumab. Over the interval of data query, there were insufficient studies to include the FDA-approved PD-L1 inhibitor durvalumab in our analysis. A total of 3,739 publications were screened for inclusion. All phase clinical trials and pilot studies were included. Screened publications were excluded if they did not include data on efficacy and/or toxicity for single-agent ICI therapy. In addition, arms of trials were excluded if they included combination therapies. A total of 3,650 publications were excluded on the basis of these criteria (Fig. 1). For the dose-toxicity analysis, studies were excluded if toxicities were not quantified and separated by dose. For the dose–response analysis, studies were excluded if response was not quantified for an individual dose.
CONSORT diagram.
Dose-toxicity analysis
We reported the frequency of systemic grade 3–5 (G3–5) AEs for all ICI therapies queried to examine the effect of different doses given within similar intervals of time. Dose intervals of every 2 weeks and every 3 weeks were grouped together, and each drug was analyzed separately. Weights for each dose cohort (DC) in the analysis were directly proportional to the sample size of the individual DCs. Incidence rates are displayed in forest plots with 95% confidence intervals (CI) estimated using exact binomial methods. The overall estimate for each set of forest plots was based on the average incidence from 1,000 samples bootstrapped from the subset of studies defined by therapy (i.e., pembrolizumab, nivolumab, etc.). The estimated CIs were based on the percentile method, which identified the 2.5% and 97.5% percentile values in the list of 1,000 estimated incidence rates from the bootstrap sampling. Comparisons of overall incidence rates between doses were based on marginal, exact generalized linear models. Because the DCs could be correlated by study, estimation used generalized estimating equations. Dose-toxicity relationships for each drug are expressed as OR with 95% CIs. Comparisons of the incidence of G3–5 AEs were also investigated by disease type and dose for each therapy to evaluate relationships between underlying pathology and risk for AEs; results are reported as OR with 95% CI. Reported P values for multiple comparisons within an ICI analysis were adjusted using the Holm–Bonferroni method to preserve an overall type-I error rate of 0.05 for each analysis.
Dose–response analysis
We reported the overall response rates (ORR) for each drug according to disease type. Analysis methods paralleled those of the dose-toxicity analysis.
Results
Nivolumab
Dose-toxicity analysis
The investigation of the dose-toxicity relationship included 34 DCs from 23 different manuscripts. Disease areas included melanoma (11 DCs), renal cell carcinoma (RCC; 10 DCs), non–small cell lung cancer (NSCLC; 6 DCs), and multiple disease types (7 DCs). Nineteen cohorts were dosed at 3 mg/kg every 2 weeks, 5 were dosed at ≤1mg/kg every 2 weeks, 3 were dosed at ≥10 mg/kg every 2 weeks, 3 were dosed at ≥10 mg/kg every 3 weeks, and 2 each were dosed at 2 mg/kg every 3 weeks or ≤1 mg/kg every 3 weeks. The bootstrap estimate of the overall incidence of G3–5 AEs was 20.1% (95% CI, 15.8–25.3; Fig. 2A). No relationship between dose and incidence of G3–5 AEs or irAEs was found (Table 1; Supplementary Table S1; Supplementary Fig. S1A). Tumor type was associated with risk of toxicity from nivolumab. The incidence of G3–5 AEs was significantly lower for patients with NSCLC than for any other tumor type, with a statistically significant 24% reduction in odds of G3–5 AEs compared with RCC (OR 0.76; 95% CI, 0.62–0.94; P = 0.004) and a 38% reduction in odds of G3–5 AEs compared with melanoma (OR 0.62; 95% CI, 0.42–0.93; P = 0.01; Table 2).
A–C, Bootstrap analyses for G3–5 AEs across different treatments.
Comparison of G3–5 AEs across dose cohorts.
Comparison of G3–5 AEs across tumor types.
Dose–response analysis
A total of 30 DCs were included in the efficacy analysis. Disease areas included melanoma (9 DCs), NSCLC (9 DCs), and RCC (12 DCs). The bootstrap estimates of ORR in these groups were 37.2% (95% CI, 29.5–41.7), 18.5% (95% CI, 15.6–20.9), and 22.7% (95% CI, 19.0–24.8), respectively. In melanoma, 3 cohorts were dosed at ≤1 mg/kg every 2 weeks, 4 at 3 mg/kg every 2 weeks, 1 at 10 mg/kg every 2 weeks, and 1 at 3 mg/kg with varied week dosing (Fig. 3A). When compared with 10 mg/kg, patients treated at both ≤1 mg/kg and 3 mg/kg had increased odds of response (OR 1.35; 95% CI, 1.25–1.47; P <0.0001; OR 1.64; 95% CI, 1.44–1.86; P < 0.0001, respectively; Table 3). Response rates appeared best in patients treated at 3 mg/kg, with the 1 mg/kg dose having 17% reduced odds of response compared with 3 mg/kg (OR 0.83; 95% CI, 0.71–0.96; P = 0.001). Thus, in melanoma, response rates appeared to level off at 3 mg/kg without increased efficacy in doses exceeding this.
A–C, Bootstrap analysis of overall response rates for melanoma.
Comparison of overall response rates with checkpoint therapy expressed as OR with 95% CI.
In NSCLC, 7 cohorts were dosed at 3 mg/kg every 2 weeks, and 1 cohort each was dosed at 10 mg/kg every 2 weeks and 1 mg/kg every 2 weeks (Supplementary Fig. S2A). When compared with the 3 mg/kg dose, patients treated at 1 mg/kg had 64% decreased odds of response (OR 0.36; 95% CI, 0.34–0.39; P < 0.0001; Table 3). There was no difference in observed odds of response between 3 mg/kg and 10 mg/kg (OR 0.96; 95% CI, 0.89–1.04; P = 0.23). Therefore, in NSCLC, as was observed in melanoma, there appeared to be a flattening of response at doses exceeding 3 mg/kg.
In RCC, 5 cohorts were dosed at 10 mg/kg, 4 cohorts were dosed at ≤1 mg/kg, 2 were dosed at 2 mg/kg, and 1 was dosed at 3 mg/kg (Supplementary Fig. S3). There was no difference in odds of response observed between dose levels in RCC (Table 3).
Pembrolizumab
Dose-toxicity analysis
The investigation of toxicity included 23 DCs from 17 published manuscripts. Disease areas covered were melanoma (6 DCs), NSCLC (5 DCs), and “mixed/multiple types” (12 DCs). A total of 8 cohorts were dosed at 200 mg every 3 weeks, 7 at 10 mg/kg every 2 weeks, 5 at 10 mg/kg every 3 weeks, and 3 at 2 mg/kg every 3 weeks. The bootstrap estimate of overall incidence of G3–5 AEs for pembrolizumab was 13.3% (95% CI, 10.9–15.8; Fig. 2B). No relationship between dose and incidence of G3–5 AEs or irAEs was observed (Table 1; Supplementary Table S1; Supplementary Fig. S1B). Tumor type was associated with risk of toxicity from pembrolizumab. The odds of G3–5 AEs were 17% lower in melanoma compared with NSCLC (OR 0.83; 95% CI, 0.70–0.99; P = 0.03; Table 2).
Dose–response analysis
A total of 23 DCs were included in the efficacy analysis. Disease areas analyzed included melanoma (13 DCs) and NSCLC (9 DCs). The bootstrap estimates of ORR in these groups were 30.7% (95% CI, 27.2–33.7) and 20.9% (95% CI, 18.4–27.0), respectively. In melanoma, 4 cohorts were dosed at 2 mg/kg every 3 weeks, 5 at 10 mg/kg every 3 weeks, and 4 at 10 mg/kg every 2 weeks (Fig. 3B). When compared with 10 mg/kg every 2 weeks, patients dosed at 2 mg/kg every 3 weeks had 22% lower odds of response (OR 0.78; 95% CI, 0.63–0.96; P = 0.01; Table 3). There was also a trend toward lower odds of response with less frequent dosing at the 10 mg/kg dose level (OR for 10 mg/kg every 3 weeks vs. every 2 weeks, 0.86; 95% CI, 0.75–1.00; P = 0.05). Thus, in our analysis, it appears that increased dose combined with increased frequency of dosing contribute to increased odds of response to pembrolizumab in patients with melanoma. In NSCLC, 1 cohort was dosed at 2 mg/kg every 2 weeks, 3 at 2 mg/kg every 3 weeks, 1 cohort at 200 mg every 3 weeks, and 4 at 10 mg/kg every 3 weeks (Supplementary Fig. S2B). There was no difference in ORRs between the 2 mg/kg group compared with the 10 mg/kg group (OR 0.95; 95% CI, 0.88–1.02; P = 0.10), nor was there a difference when 200 mg flat dose was included with the 2 mg/kg group (OR 1.05; 95% CI, 0.85–1.30; P = 0.63; Table 3). Therefore, increasing dose of pembrolizumab was not associated with improved efficacy in NSCLC.
Ipilimumab
Dose-toxicity analysis
Six published studies with 6 DCs were included in the toxicity analysis, all in patients with melanoma (Fig. 2C). Three DCs used 3 mg/kg every 3 weeks dosing and 3 DCs used 10 mg/kg every 3 weeks dosing. The bootstrap estimate of the overall incidence of G3–5 AEs for ipilimumab was 34% (95% CI, 25.8–43.2). Patients with melanoma treated with ipilimumab 3 mg/kg every 3 weeks had 27% reduced odds of G3–5 AEs compared with patients treated with 10 mg/kg every 3 weeks (OR 0.73; 95% CI, 0.59–0.89; P = 0.002; Table 1). However, when the analysis was performed specifically for irAEs in the same studies, the overall incidence of G3–5 irAEs was 33.7% (23.3–51.7) with no difference in irAE rate between 3 mg/kg every 3 weeks and 10 mg/kg every 3 weeks (OR 1.32; 95% CI, 0.70–2.50; P = 0.39; Supplementary Fig. S1C; Supplementary Table S1). Accordingly, increased dose of ipilimumab was associated with increased toxicity; however, this association was not present when analyzing irAEs.
Dose-efficacy analysis
Six DCs from three trials were included in the efficacy analysis, all in patients with melanoma. Observed doses were 0.3 mg/kg every 3 weeks (1 DC), 3 mg/kg every 3 weeks (2 DCs), and 10 mg/kg every 3 weeks (3 DCs). The ORR was 6.5% (4.0–9.4; Fig. 3C). No difference in ORR was observed in comparing 3 mg/kg every 3 weeks with 10 mg/kg every 3 weeks (OR 0.80; 95% CI, 0.55–1.15; P = 0.15; Table 3). Therefore, increased dose of ipilimumab in melanoma was not associated with increased response.
Discussion
Despite many ongoing trials using ICIs as a backbone for drug development, there remains a paucity of data assessing dose-effect relationships for these therapies. Here, we conducted a large meta-analysis of 74 trials reported between 2010 and 2017 treating an aggregate of 7,469 patients. We found that, unlike chemotherapy or targeted therapies, the relationship between dose and clinical outcome in ICIs is not predictable since both irAEs and response were not dose-dependent with some exceptions for PD-1 antibodies dose–response based on disease type. Accordingly, our data indicate a need to rethink early clinical trial design for immune-modulating agents.
Our findings were consistent with other studies including a newly reported trial for ipilimumab in advanced melanoma by Ascierto and colleagues which compared ipilimumab 10 mg/kg to 3 mg/kg and found no difference in overall response (15). A recent FDA analysis of safety and response of nivolumab exposure in 13 clinical trials also demonstrated no difference in AE rate across doses (0.3–10 mg/kg) and a plateau of dose response at 3 mg/kg in melanoma and NSCLC (18).
Finally, a systematic review and meta-analysis of 36 phase II–III trials comparing ICIs with conventional or combination therapy (19) reported no safety difference between doses of the same ICI except ipilimumab.
Our meta-analysis has inherent limits relating to toxicity reporting and its evolution during the era in which these trials were conducted. We purposely excluded inter-ICI combinations or with chemotherapy, to allow a clear analysis of dose safety and dose response not influenced by additional therapy. Also, some comparisons made between dose levels included only one cohort at that dose level, which is another potential limitation. Finally, because this study’s intention is to address the limitation in dose-escalation design, our analysis concentrated on short-term toxicities that are relevant to the outcome of such studies. Hence, the study did not consider delayed immune toxicities.
In a previous study, we also have shown a lack of dose-toxicity correlation in cancer vaccines except for very specific vaccines derived from pathogenic bacterial vectors, and proposed an alternative design to the traditional 3+3 dose escalation design in cancer vaccine drug development (14). Both studies emphasize the unique mechanism of immune-related toxicities.
On the basis of the results presented in our report, future clinical trial design of ICIs should not be based on the standard dose-escalation method with a primary objective of identifying a safe dose. Our analysis clearly shows this as a futile objective. However, knowing that higher dose level is important for response, whether with or without a plateau, requires a way to determine an optimum dose. Hence dose determination should be driven by response while remaining vigilant for potential toxicities.
Multiple dosing response seeking design
We suggest selecting several dose levels/schedules to start then enrolling 1 patient per dose level at the same time and performing evaluation with each starting dose. When a response at a dose level is achieved, add (n) patients to that dose level. The process should be applied for all doses at which a response is seen. At the conclusion, the selected dose should be the lowest dose at which the response is estimated to be at least an acceptable value r. To be vigilant for toxicities which may be different for each novel molecule, we suggest setting prespecified rules to stop development if, for example, two grade 4 toxicities occur at any dose level. In addition, once the dose is expanded, safety stopping rules should be applied in a similar fashion. An example of dose selection is provided in Table 4. The development starts with selecting four starting doses, for example, 1 mg/kg, 10 mg/kg, 40 mg/kg, and 100 mg/kg. For this example, no responses are seen in the starting doses of 1 or 10 mg/kg but grade 4 toxicity was observed at the 10 mg/kg dose. For the starting dose of 40 mg/kg, a response is seen. Therefore 5 more patients are treated at that dose and 4 more patients responded. At the 100 mg/kg starting dose, the first patient treated had a response and 2 of 5 additional patients treated at that dose also had a response. Hence, the response at 100 mg/kg is 3/6 and the response rate at 40 mg/kg is 5/6. Accordingly, the dose 40 mg/kg should be selected for future development. Similar approach can be used for combinational therapy with non-ICIs including chemotherapy or targeted therapy to identify the optimal ICI dose for the combination.
Example of MDRSD.
Almost all ICIs developed post–PD-1 and CTLA-4 have not demonstrated meaningful clinical activity as single agents and therefore combinational approach may be needed The multiple dosing response seeking design (MDRSD) is adaptable to this approach. If a tested agent (A) fails to show any clinical activity despite optimizing the dose to preplanned maximum dose, that dose could be combined with another immune-modulating agent of interest (B). The combination will be tested in different cohorts by changing the dose of drug B using the same MDRSD method.
Furthermore, the MDRSD does not consider the variability in host immune response as some patients may develop stronger immune response that may translate to clinical activity and may not be dose related but rather host related. Personalized designs may be developed in the future to factor in the host immunity as predictive biomarkers continue to evolve.
In summary, we demonstrated a lack of consistent dose-toxicity or dose–response correlation for PD-1/CTLA-4 and presented a novel, practical, and efficient design to test novel immune-modulating agents emphasizing clinical response as main target while keeping high vigilance for toxicities. Our efforts encourage both academia and industry to change the traditional way of drug development and represent a step forward for future new designs to conduct those studies.
Authors’ Disclosures
O.E. Rahma reports personal fees from Imvax, GSK, Bayer, Gennentch, Puretech, Sobi, Maverick Therapeutics, Merck, Celgene, and Five Prime outside the submitted work; in addition, O.E. Rahma has a patent for methods of using pembrolizumab and trebananib pending. S.N. Khleif reports grants and other from Syndax, IOBiotech, BiolineRX, AstraZeneca, and Lycera; grants from MedImmune; and other from Northwest Biotherapeutics, PDS Biotechnology, Advaxis, EMD Serono, GSK, UbiVac, McKinsey, Georgiammune, KAHR Medical, CytomX, NewLink Genetics, AratingaBio, CanImGuide, aMoon, Aummune, Incyte, Adaptive Biotech, Israel Biotech Fund, Livzon Mabpharm USA, Tessa Therapeutics, AgonOx, Autolus, Bayer, and Rheo Medicine outside the submitted work. No disclosures were reported by the other authors.
Authors' Contributions
O.E. Rahma: Conceptualization, resources, supervision, writing-original draft, writing-review and editing. J.E. Reuss: Conceptualization, resources, data curation, writing-original draft, writing-review and editing. A. Giobbie-Hurder: Data curation, formal analysis, methodology, writing-review and editing. G. Shoja E Razavi: Data curation, investigation, writing-review and editing. O. Abu-Shawer: Investigation, writing-original draft, writing-review and editing. P. Mehra: Data curation, investigation, writing-review and editing. S. Gupta: Data curation, investigation, writing-review and editing. R. Simon: Conceptualization, software, investigation, methodology, writing-review and editing. S.N. Khleif: Conceptualization, supervision, investigation, writing-original draft, project administration, writing-review and editing.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Footnotes
Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).
Clin Cancer Res 2021;27:485–91
- Received July 15, 2020.
- Revision received September 21, 2020.
- Accepted October 16, 2020.
- Published first October 20, 2020.
- ©2020 American Association for Cancer Research.
References
Volume 27, Issue 2
