Introduction

Immune checkpoint inhibitors (ICIs) have caused a paradigm shift in the treatment for patients with solid tumors and Food and Drug Administration (FDA) approval is rapidly expanding still.1 In non-small cell lung cancer (NSCLC), ICIs have been approved in all disease stages, either as monotherapy or combined with chemotherapy.2–8 However, these advancements come with increased costs in already stretched healthcare systems.9 Based on pharmacokinetic (PK) modeling, alternative ICI dosing strategies could reduce costs without losing efficacy.10 ICIs have a wide therapeutic range without an apparent dose-response or dose-toxicity relationship at the doses used in daily clinical practice.11 12 Initially, clinical trials investigating the PD-1 inhibitor pembrolizumab were performed with a weight-based dose of 2 mg/kg. However, based on PK modeling, this was later switched to a flat dose as reducing the reuse of single-use vials over several patients seemed more convenient and safe.13 Although a flat dose of 154 mg matched the original 2 mg/kg dosing best, surprisingly, 200 mg was selected as the preferred flat dose for subsequent clinical trials and eventually approved as the standard of care (SoC).13 14 Within the KEYNOTE-001, pembrolizumab clearance became linear at ≥0.3 mg/kg every 3 weeks suggesting maximal target saturation and this was further validated with ex vivo interleukin-2 stimulation assays at a dose of ≥1 mg/kg every 3 weeks.15 Supported by this (PK/pharmacodynamic (PK/PD)) data to optimize dosing, a ‘hybrid’ dosing scheme was introduced at the Netherlands Cancer Institute (NKI) whereby patients receive either 100, 150 or 200 mg pembrolizumab every 3 weeks based on their weight.16 The hybrid dosing regimen ensures dosing well above the theoretically minimum for maximal efficacy of ≥1 mg/kg every 3 weeks and has already been shown to provide significant institute-wide cost reduction.15–18 Despite the sound pharmacological rationale for hybrid dosing, flat dosing is still internationally recommended due to lacking clinical efficacy data for hybrid dosing.19 Therefore, we performed a bicenter, retrospective cohort study to explore whether pembrolizumab hybrid dosing is non-inferior to flat dosing in patients with advanced NSCLC.

Methods

This study was approved by the Netherlands Cancer Institute institutional review board (IRB d23-013) and was conducted according to the declaration of Helsinki. All included patients provided consent for data collection and analysis before start of treatment. Reporting was according to the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.20

Patient selection and data collection

This retrospective, observational study was conducted in two Dutch tertiary hospitals: NKI and Leiden University Medical Center (LUMC). Consecutive patients who were diagnosed with advanced NSCLC and had received at least one cycle of pembrolizumab±chemotherapy (carboplatin/pemetrexed for non-squamous cell carcinoma or carboplatin/paclitaxel for squamous cell carcinoma) in any line of treatment between August 2015 and March 2023 were identified using the hospitals electronic prescription system. Medical records were manually reviewed and baseline characteristics including patients’ age, weight, smoking history, Eastern Cooperative Oncology Group (ECOG) performance score, tumor histology, PD-L1 tumor proportion score (TPS), oncogenic driver alteration, line of treatment, and presence of brain and/or liver metastases were collected. In addition to these baseline characteristics, reason for discontinuing pembrolizumab, total number of cycles received, cumulative amount of pembrolizumab received, most recent weight at every cycle, immune-related adverse events (irAEs) and corticosteroid use were also collected. Patients were excluded if pembrolizumab was combined with any treatment other than SoC chemotherapy.

Cohort allocation

The hybrid dosing schedule under investigation in this study was introduced in the NKI early 2019, while the LUMC used flat dosing until late 2022. Therefore, most patients treated in the NKI were hybrid dosed and nearly all patients in the LUMC were flat dosed. Importantly, this was an institute-wide decision for all anti-PD-1 treated patients, the treating physician could not determine the dosing schedule. Patients were allocated by the schedule used during treatment. If the schedule did not fully adhere to the hybrid or flat dosing schedule, patients were allocated based on the dosing schedule used during the first four cycles. Unintendedly, the hybrid dosing schedule was never implemented for patients treated with the carboplatin/paclitaxel combination at the NKI. Therefore, patients treated with this combination were excluded in both cohorts since patients with squamous NSCLC have a worse overall survival (OS) and allocating these patients to the flat dose cohort would have introduced bias.21 In the hybrid dosing schedule, patients received pembrolizumab according to their weight: patients weighing <65 kg received 100 mg every 3 weeks or 200 mg every 6 weeks, patients weighing 65–90 kg received 150 mg (every 3 weeks) or 300 mg (every 6 weeks) and patients >90 kg received 200 mg (every 3 weeks) or 400 mg (every 6 weeks). Patients in the flat dose cohort received 200 mg (every 3 weeks) or 400 mg (every 6 weeks) pembrolizumab irrespective of their weight.

Outcomes

The primary outcome was OS, defined as the time between the first pembrolizumab dose administered and death. The secondary outcome was progression-free survival (PFS), defined as the time between the first pembrolizumab cycle to first progression or death. Since Response Evaluation Criteria in Solid Tumors V.1.1 measurements were not routinely reported, the first date of either reported clinical or radiological progression was used. Survival status was verified using the national Personal Records Database (BRP) and, therefore, no patients were lost to follow-up for OS. Patients were censored at time of last radiological evaluation for PFS if patients were lost to follow-up. The data cut-off date was March 1, 2024. Exploratory outcomes were OS and PFS in multiple subgroups: patients treated in first line, patients treated with pembrolizumab monotherapy or in combination with chemotherapy, patients whose dosing adhered completely to their allocated dosing scheme, flat dosed patients stratified by treating institution, patients treated at the NKI and patients stratified by weight (<65 kg, 65–90 kg or >90 kg) or by dose per kilogram (<1.8 mg/kg, 1.8–2.5 mg/kg or >2.5 mg/kg).

Calculation amount of pembrolizumab saving and cost reduction

For each cycle in the hybrid cohort, the amount of pembrolizumab saved (compared with a flat dosed cycle) was calculated. In case of an incorrectly dosed cycle, additional saving if the cycle was correctly dosed was also determined. Then, the cumulative savings for each patient and for the entire hybrid cohort were compared with a flat dose scenario. Cost reductions were calculated using US list pricing and assuming vial sharing.22

Statistical analysis

Descriptive statistics were used to describe baseline characteristics and were depicted as median and range unless otherwise stated. The Mann-Whitney U test was used to test for differences in baseline characteristics between the two cohorts. Kaplan-Meier survival curves and log-rank tests were used to estimate survival and test for differences between groups, respectively. HRs were calculated with the Cox proportional hazards model both unweighted and weighted. Inverse probability of treatment weighting (IPTW) was used in the weighted analysis in order to balance cohorts both for patient characteristics that contributed to cohort allocation and for prognostic or predictive factors. Baseline characteristics that significantly contributed to OS were selected in univariate analysis. Logistic regression was used to identify variables that significantly contributed to cohort allocation. The variables that either significantly contributed to OS or to cohort allocation were used in a final logistic model (blinded for OS and PFS) whose predicted probabilities (standard propensity score) of cohort allocation were used to generate the weights for IPTW. This final logistic model was the average of 20 models, fitted on 20 imputed datasets, imputed (missing data: PD-L1 TPS (n=8), WHO PS (n=73), smoking status (n=19) and weight (n=1)) with multiple imputations by chained equations (MICE) with ten iterations. Imputations were only used for generating the weights; the analysis of outcome (OS and PFS) was done on the original, unimputed dataset. All analyses were performed using R V.4.2.3. We set the margin of non-inferiority to 1.15, meaning that we considered hybrid dosing non-inferior to flat dosing if the entire two-sided 95% CI around the HR for OS lies below 1.15. In the OS and PFS analyses, the p value under the null hypothesis of an HR of 1.15 were computed, unless otherwise stated. For all other outcomes than OS and PFS, the standard null hypothesis of ‘no difference between groups’ were used to compute p values. P values <0.05 were considered statistically significant, were not corrected for multiplicity testing, were reported unweighted unless otherwise stated and all were two sided. Additionally, the E-value was calculated for the primary outcome OS.23 In short, this E-value estimates what effect size the unmeasured confounders in (retrospective) studies at minimum need to be to nullify the significance of the measured outcome.

ResultsBaseline characteristics

In total, 766 of 952 screened patients were included: 375 patients in the hybrid dose cohort and 391 in the flat dose cohort. Of the 186 excluded patients, 154 (83%) were not included because pembrolizumab was combined with either carboplatin/paclitaxel or with treatment not part of SoC (figure 1). Overall, baseline characteristics were well balanced between groups (table 1). In both groups, about half of the patients were male, median weight was 72 vs 73 kg for the hybrid dose and flat dose cohort, respectively, and the majority of patients were diagnosed with adenocarcinoma. Although the majority of patients were treated with pembrolizumab in first line (79%), patients in the hybrid dose cohort were more frequently treated with pembrolizumab in third or further line (online supplemental table 1) and more frequently treated with pembrolizumab combined with chemotherapy versus monotherapy compared with the flat dose cohort. More importantly, distribution of known predictive variables to immunotherapy outcomes, for example, performance score PD-L1 TPS and oncogenic driver alteration were similar between the two groups.24

Table 1

Baseline characteristics

Figure 1

CONSORT flow diagram. CONSORT, Consolidated Standards of Reporting Trials; NSCLC, non-small cell lung cancer.

Exposure to pembrolizumab

Patients in the hybrid dose cohort received a significantly higher amount of cycles (median 7 vs 6 cycles, p=0.005) and a non-significant numerical lower amount of cumulative pembrolizumab (median 1100 vs 1200 mg, p=0.44) than patients in the flat dose cohort. The amount of pembrolizumab per kilogram (median 2.0 vs 2.7 mg/kg, p<0.0001) and pembrolizumab per cycle (median 150 vs 200 mg, p<0.0001) was significantly lower in the hybrid dose cohort (figure 2, online supplemental figures 1–3 and table 2). 114 (30%) hybrid dosed patients and 37 (9%) flat dosed patients received at least one incorrect dose according to their current weight and associated dosing scheme during the course of their treatment. The incorrectly hybrid dosed patients received a median of four incorrectly dosed cycles and more rather than less cumulative pembrolizumab than their dosing schedule prescribed (62% vs 38% of incorrectly hybrid dosed patients, respectively).

Figure 2

Density plot of pembrolizumab received in mg/kg.

Pembrolizumab effectiveness

In total, 576 deaths occurred: 256 in the hybrid dose and 320 in the flat dose cohort. Median follow-up was 43.1 and 61.0 months in the hybrid dose and flat dose cohort, respectively. Median OS was 17.7 months (95% CI 14.9 to 20.9) and 11.8 months (95% CI 9.3 to 13.8) in the hybrid and flat dose cohort respectively, with an HR of 0.76 (95% CI 0.65 to 0.90, p<0.0001, figure 3A). The upper boundary of the 95% CI for the HR was smaller than 1.15 and, therefore, hybrid dosing was deemed non-inferior to flat dosing. The 1-year OS was 61% (95% CI 56% to 66%) vs 50% (95% CI 45% to 55%) for the hybrid and flat dose cohort, respectively (online supplemental table 3). Similarly, PFS in the hybrid dose cohort was non-inferior to the flat dose cohort with a median of 6.4 months (95% CI 5.7 to 7.7) vs 4.6 months (95% CI 3.9 to 5.5), respectively (HR 0.82, 95% CI 0.70 to 0.96, p<0.0001, figure 3B and online supplemental table 4). The E-value for OS was 2.0 (95% CI 1.65 to NA). This means that if the observed non-inferiority of hybrid dosing was actually due to unmeasured bias, these combined confounders would need to have a similar predictive value as the well-known biomarker PD-L1≥50% (HR: 0.60 in our cohort, vs PD-L1 negative) and these confounders would have to be 4.5× more prevalent in the hybrid dose cohort to reproduce our results.

Figure 3

Kaplan-Meier curves for overall survival (A) (hybrid vs flat dose HR 0.76, 95% CI 0.65 to 0.90, p<0.0001) and progression-free survival (B) (hybrid vs flat dose HR 0.82, 95% CI 0.70 to 0.96, p<0.0001).

Subgroup analysis

In addition, OS of hybrid dosed patients was non-inferior to flat dosed patients in the subgroups of patients treated in the first line, patients treated with pembrolizumab monotherapy, patients treated with combination chemotherapy, fully correctly dosed patients and patients weighing <65 kg and 65–90 kg (figure 4, online supplemental file 4–10). Since the vast majority of hybrid dosed patients were treated at the NKI, we stratified patients by treating institution to check whether this imbalance could have introduced (institutional) bias. Of the patients treated in the NKI, hybrid dosed patients had a non-inferior OS (HR 0.74, p<0.0001, figure 5A), but for patients treated at the LUMC, this did not reach statistical significance (HR 0.43, p=0.10, figure 5B). Additionally, OS of hybrid and flat dosed patients treated at NKI was not superior to that of hybrid and flat dosed patients treated at the LUMC (hybrid: HR 1.82, psuperiority=0.30, flat: 1.08, psuperiority=0.54, figure 5B,C). After pooling all patients, the OS of patients treated with a dose of <1.8 mg/kg (HR 1.17, p=0.89) and >2.5 mg/kg (HR 1.35, p=0.22) was not statistically non-inferior to that of patients treated with a dose 1.8–2.5 mg/kg (online supplemental figure 12). Together, these analyses confirm the non-inferiority of hybrid dosing across different subgroups and show no evidence for institutional bias.

Figure 4

Kaplan-Meier curves for overall survival of patients <65 kg (A) (hybrid vs flat dose HR 0.71, 95% CI 0.52 to 0.97, p=0.002), patients 65–90 kg (B) (hybrid vs flat dose HR 0.75, 95% CI 0.60 to 0.93, p=0.0001) and patients >90 kg (C) (hybrid vs flat dose HR 0.85, 95% CI 0.55 to 1.31, p=0.17).

Figure 5

Kaplan-Meier curves for overall survival of patients treated at the NKI (A) (hybrid vs flat dose HR 0.74, 95% CI 0.61 to 0.91, p<0.0001) and the LUMC (B) (hybrid vs flat dose HR 0.43, 95% CI 0.14 to 1.36, p=0.10) and for flat dosed patients (C) (NKI vs LUMC HR 1.08, 95% CI 0.86 to 1.33, psuperiority=0.54) and hybrid dosed patients (D) (NKI vs LUMC HR 1.82, 95% CI 0.58 to 5.56, psuperiority=0.30) stratified by treating institution. LUMC, Leiden University Medical Center; NKI, Netherlands Cancer Institute.

Univariate analysis, logistic regression and IPTW

In a univariate analysis, age, sex, PD-L1 TPS, performance score, driver mutation status, line of therapy, presence of liver metastases at start of treatment were all significantly associated with OS (online supplemental table 5). Additionally, histology, line of therapy, weight and combination with chemotherapy or monotherapy were all significantly related with cohort allocation in logistic regression analysis (online supplemental table 6). After correction for these confounding factors identified by univariate analysis and logistic regression by IPTW, OS of hybrid dosed patients remained non-inferior to flat dosed patients (HR 0.76; 95% CI 0.63 to 0.91, p<0.0001) (online supplemental table 7 and figure 17A). In addition, PFS of hybrid dosed patients was also non-inferior after adjusting for these confounding factors (HR 0.85; 95% CI 0.72 to 1.01, p<0.0001) (online supplemental figure 17B). Although both type and number of subsequent oncolytic therapies were similar between both arms, we also performed an IPTW analysis including number and type of subsequent therapy and type of progressive disease (oligoprogression or systemic progression) as a robustness check since these factors could also influence OS. In this analysis, hybrid dosing also remained non-inferior (data not shown).

Pembrolizumab saved and cost reduction

In total, 26.2% (or 52.5 mg per cycle, p<0.0001) of pembrolizumab was saved in the hybrid dose cohort and an additional saving of 2.5 mg per cycle would have been possible had all cycles been dosed adequately. This amounts to a total reduction of US$13,624,260.30 (or US$36,331.36 per patient) but could have amounted to a total of US$14,260,596.30 had all cycles been dosed correctly. If all patients in the flat dose cohort were also dosed using the hybrid dosing scheme, an additional 235.300 mg pembrolizumab or US$13,076,844.60 would have been saved.

Reason to discontinue pembrolizumab, treatment after disease progression & irAEs

Pembrolizumab was discontinued in 56% vs 60% of patients due to progressive disease or death and in 21% vs 22% due to toxicity in the hybrid and flat dose cohort, respectively. In both cohort, a similar proportion of patients had oligoprogression (29% vs 25%, p=0.32). Following progression, patients received a similar number of therapy lines (p=0.60) and types of subsequent therapy were in general comparable (online supplemental tables 1 and 8). The percentage of patients who received corticosteroids was similar between groups (23% vs 21%, p=0.48) and most frequent irAEs were pneumonitis, colitis, nephritis and hepatitis. Similarly, the percentage of patients who received second line immunosuppression and the maximum dose corticosteroids were not different between the two cohorts, suggesting a similar severity of irAEs (online supplemental table 8).

Discussion

In this large, bicenter, retrospective cohort study, we found that hybrid dosing of pembrolizumab in patients with advanced NSCLC was non-inferior to SoC flat dosing with respect to OS. After correction for well-known predictive variables and imbalances in baseline characteristics between both cohorts, OS of hybrid dosed patients remained non-inferior. Thus, our results indicate that hybrid dosing of pembrolizumab in advanced NSCLC may very well be as efficacious as flat dosing, making a significant cost saving possible.

Alternative pembrolizumab dosing strategies in NSCLC have been investigated previously. Patients treated with weight-based dosing (WBD) had non-inferior OS and safety outcomes (i.e. rate of irAEs) compared with flat dosed patients in a large retrospective cohort study (n=1413) by Chaitesipaseut et al.25 However, this study had a couple of important limitations. It reported only limited patient data and, therefore, the authors could not account for relevant confounding factors. Also, the WBD cohort was relatively small (n=138) and did not include a sufficient number of patients according to their sample size calculation (297 patients per arm). In another retrospective analysis by Low et al, 65 patients received 100 mg pembrolizumab either due to financial reasons, physicians’ choice or based on the 2 mg/kg dosing strategy.26 They compared these patients to 49 flat dosed (200 mg) patients. In this study, both PFS and OS were similar between the 100 mg and 200 mg cohorts. In another retrospective cohort study from Taiwan, patients (n=242) were stratified in two cohorts according to pembrolizumab received per kilogram: either <2.0 mg/kg or ≥2.0 mg/kg.27 After IPTW, median OS was numerically but not-significantly longer in the ≥2.0 mg/kg dose cohort, 19.3 months vs 14.3 months (p=0.15). In a third retrospective study also from Taiwan, patients with a modified pembrolizumab dose (either 2 mg/kg or 100 mg flat dose, n=64) were compared with patients with flat dose (200 mg, n=28).28 Although PFS was lower in the modified dose group, OS was similar between the two cohorts.

Our non-inferior results and outcomes of alternative dosing strategies described in the literature are supported by PK/PD modeling. These models showed linear pembrolizumab clearance at a dose of >0.3 mg/kg every 3 weeks and target saturation at a dose of ≥1 mg/kg every 3 weeks.15 Moreover, a single dose of 0.1 mg/kg resulted in >90% target engagement at minimal concentrations. Although a single dose of ≥2 mg/kg was needed to reach target saturation in a poorly vascularized tumor in another model, this model was based on uncertain assumptions, for example, pembrolizumab targets tumor infiltrating T-cells and clearance remains constant during treatment.29 Moreover, in pooled data from the KEYNOTE-001, KEYNOTE-002 and KEYNOTE-006 studies in melanoma patients, similar objective response rates (ORR) were observed in patients treated with 1 mg/kg every 3 weeks compared with higher doses.30 These data imply that a dose between 0.3 and 1 mg/kg every 3 weeks would be sufficient to achieve maximal effectiveness. Indeed, in our pooled analysis patients treated with the lowest dose (<1.8 mg/kg) had similar OS (but did not reach statistical non-inferiority) as patients treated with higher doses, indicating patients were still dosed above the minimal dose with maximal efficacy.

In addition to pembrolizumab, low-dosing strategies for another PD-1 inhibitor nivolumab have been pursued in NSCLC and other malignancies. In a retrospective cohort study, 18 patients with NSCLC were treated with low-dose nivolumab (20 or 100 mg) and compared with 29 patients treated with SoC (3 mg/kg).31 Nivolumab efficacy assessed by ORR, PFS and OS was comparable in both cohorts. In addition, low-dose nivolumab showed efficacy in two phase II trials in patients with melanoma (10 mg nivolumab) and Hodgkin lymphoma (40 mg nivolumab) and in a randomized phase III trial in patients with head and neck squamous cell carcinoma (20 mg nivolumab).32–34 Although retrospective, results of our analysis seem to be in line with these prospective trials and our cohorts comprise a much larger sample size.

These results for pembrolizumab and nivolumab suggest dose reductions are feasible without losing efficacy. Alternative dosing strategies for ICIs based on population-PK models are approved by the FDA if both mean exposure (Caverage) and minimum exposure (Ctrough) deviate <20% from the SoC without the need for clinical validation.35 In therapies with a clear dose–response relationship, this guideline is appropriate since dose reductions from the optimal SoC dose will likely result in a loss in efficacy. However, for pembrolizumab and other ICIs, this guideline may not be appropriate due to the absence of a dose–response relationship at clinically tested doses and plasma levels do not reflect target saturation.11 12 Also, the SoC flat dose of 200 mg was selected after PK modeling, but a dose of 154 mg matched the initial dosing of 2 mg/kg best thereby overdosing patients.13 Moreover, a dose of 2 mg/kg is probably still overdosing patients as the mechanism of action, that is, PD-1 receptor saturation, is already achieved with a dose of ≥1 mg/kg.15

Taken together, the FDA guidelines pursue equivalent ICI exposure of an approved flat dose that seems unnecessary high, assuming it is possible to lower the dose while preserving efficacy based on the theoretically minimal dose between 0.3 and 1.0 mg/kg. This is supported by our retrospective results and currently tested prospectively in several trials that either lower the dose or extend the dosing interval (NCT04032418 and NCT04295863).36–40

This study has three important limitations inherent to the retrospective, observational design. First, a proportion of patients in both cohorts were not fully treated according to dosing scheme of the allocated cohort. This had several reasons: patients transitioned from flat to hybrid dosing during treatment or dosing was not corrected for weight loss or gain. However, after exclusion of these incorrectly dosed patients our results and conclusions remain unchanged. Second, our retrospective analysis revealed that hybrid dosing was unintendedly not implemented for patients treated with the paclitaxel combination and therefore we excluded patients treated with this combination in both cohorts. Although all these patients received flat dosing during the first four cycles, allocating all these patients to the flat dose cohort would have introduced bias since patients with squamous NSCLC have a worse survival.21 By excluding patient treated with carboplatin, paclitaxel and pembrolizumab from the other, flat dose cohort, the distribution of patients with squamous NSCLC treated with pembrolizumab monotherapy between both cohorts was similar. Third, bias could have been introduced due to institutional differences or patient selection. To address this problem, an IPTW analysis was performed and confirmed the non-inferiority of hybrid dosing. Additionally, the E-value was calculated, an estimation of the needed strength of unmeasured confounders to abolish the measured effect in observational studies.23 We believe it is unlikely that (a combination of) confounders of this magnitude and skewed distribution (HR: 0.60 and 4.5× more prevalent in the hybrid cohort) between arms exist in our dataset. This is supported by the well-balanced known prognostic variables (both at baseline and postprogression) for OS. Moreover, outcomes of patients receiving the same 200 mg pembrolizumab in both cohorts (flat dosed patients treated in the different institutions and patients weighing >90 kg in both original cohorts) were very similar (figures 4C and 5C, online supplemental figure 10).41–43 The only notable difference between cohorts was the higher proportion of hybrid dosed patients treated with pembrolizumab plus chemotherapy. We found no explanation for this difference, such as performance score or PD-L1 TPS; however, factors like total tumor burden or patient preferences were not recorded. Importantly, hybrid dosing was also non-inferior to flat dosing in both subgroups of patients treated with pembrolizumab monotherapy and patients treated with pembrolizumab plus chemotherapy. Also, pembrolizumab combined with chemotherapy was not associated with OS or PFS in univariate analysis. In summary, the non-inferiority of hybrid dosing does not appear to be significantly influenced by specific biases in this retrospective analysis.

While our retrospective results strongly suggest non-inferiority of hybrid dosing to SoC flat dosing, the European Medicine Agency guidelines for non-inferiority trials only apply to prospective trials.44–46 Therefore, additional, prospective trials are needed to confirm our findings.

Conclusion

In this retrospective analysis of a large cohort of advanced NSCLC patients treated with pembrolizumab±chemotherapy, OS of hybrid dosed patients was non-inferior to flat dosed patients. OS remained non-inferior after correcting for possible confounding factors. This hybrid regimen resulted in significant savings of pembrolizumab and costs.