Comparative analysis of the phase III clinical trials of anti-PD1 monotherapy in head and neck squamous cell carcinoma patients (CheckMate 141 and KEYNOTE 040)

Two phase III clinical trials (CheckMate 141 and KEYNOTE 040) have independently demonstrated that overall survival (OS) in recurrent and/or metastatic head and neck squamous cell carcinoma (R/M HNSCC) patients, who have failed platinum-based therapy, can be improved with anti-PD1 monotherapy. Treatment with nivolumab or pembrolizumab in R/M HNSCC patients led to an improved OS with a hazards ratio (HR) of 0.70 (95%CI 0.51–0.96; p = 0.01) and HR of 0.80 (95%CI 0.65–0.98, p = 0.0161), respectively, as compared to standard of care (SOC) chemo monotherapy regimens (specifically, cetuximab, docetaxel, or methotrexate). The gain in OS was similar in both studies, underscoring the role of anti-PD1 drugs in R/M HNSCC patients. One of the striking discrepancies between CheckMate 141 and KEYNOTE 040 was the OS observed in the control SOC arms (6.9 months median in KEYNOTE 040 versus 5.1 months in CheckMate 141), which inadvertently set a higher threshold in the bio-statistical analysis of KEYNOTE 040 so that the clinical outcome of every patient was influential in the analysis. We perform a comparative analysis of the two studies to identify potential factors in the control arm that can impact clinical trial bio-statistical outcomes and which may have implications for future immunotherapy clinical trial designs.


Factors impacting overall survival in the control arms
One of the main differences between these two clinical trials was the OS observed in the control SOC arms (6.9 months median in KEYNOTE 040 versus 5.1 months in CheckMate 141). The 1.8-month difference in median OS in the control arms may have had significant bio-statistical implications for the endpoint analysis.

Patient selection
There were two important differences in the selection of patients between these two trials. Although both trials included patients with platinum refractory R/M HNSCC, KEYNOTE 040 excluded patients who recurred or progressed within 3 months of previous multimodal therapy containing platinum for locally advanced disease. Thus, KEYNOTE 040 was excluding the rapidly progressing patient population from the trial. In addition, in KEY-NOTE 040, only 1.2% (N = 6 of 495) of the HNSCC patients had received ≥ 3 prior lines of therapy as compared to 19.9% (N = 72 of 361) in CheckMate141. Thus, in KEYNOTE 040, the population was less heavily pre-treated and aggressive tumor growth characteristics were excluded from the trial and these exclusion criteria most likely contributed to the improved OS observed in the control arm. This inadvertently set a higher threshold in the bio-statistical analysis of KEYNOTE 040 so that the clinical outcome of every patient was influential in the analysis.

Differential distribution of SOC treatment regimens
In addition, even though the mono chemotherapy treatment options in the SOC arms were the same in both clinical trials, there were differences in the dosing and the overall distribution of the patients in the SOC treatment regimen received. Specifically, there were differences in the dosing of docetaxel. In the KEYNOTE 040 trial, docetaxel was given at 75 mg/m 2 q3weeks whereas in CheckMate 141 the docetaxel dose was 30 mg/m 2 qweek. Whether this difference in dosing of docetaxel makes a difference is unclear but we may speculate that the q3week dosing may be reserved for more robust patients in KEYNOTE040. In addition, docetaxel q3 weeks has been reported as slightly more efficient in terms of response rate or survival than the weekly schedule in other sensitive tumor types, such as breast and prostate cancer [5,6]. There was also a higher percentage of patients who were treated with docetaxel (42% versus 21%) and cetuximab (30% versus 11%) in KEYNOTE 040 as compared to CheckMate 141, respectively. In contrast, in CheckMate 141, there was a higher number of patients who received methotrexate (38% versus 27% in KEYNOTE 040). While the CheckMate 141 trial was not designed to compare the three regimens used in the SOC arm, docetaxel appeared to result in a slightly improved OS as compared to patients receiving methotrexate and cetuximab, although the overall numbers were small to be able to make definitive conclusions. Thus, the difference in the number of R/M HNSCC patients receiving docetaxel in KEYNOTE 040 as compared to methotrexate in CheckMate 141 may have also contributed to the improved OS in the control arms (6.9 months versus 5.1 months).

Subsequent treatment with immune checkpoint inhibitors in the SOC arm
Both phase III trials had a subset of patients who were allocated to the SOC arm but did not receive SOC therapy. In

Patient selection based on PD-L1 expression
These trials also highlight the importance of patient selection in clinical trial design and raises the question whether patient selection based on expression of a prognostic biomarker, such as PD-L1, should be considered as an eligibility criteria for future PD-1/PD-L1 targeted clinical trials. Both of the phase III clinical studies provide insight into this question through a sub-analysis performed on OS based on PD-L1 expression > 1% within the tumor microenvironment. In KEYNOTE 040, using a PD-L1 combined positive score (CPS) of ≥ 1, which is defined by PD-L1 expression on both immune and tumor cells, the HR was 0.74 (95% CI: 0.58-0.93, p = 0.0049) with a median OS of 8.7 months with anti-PD1 treatment. Similarly, in CheckMate 141, a PD-L1 tumor cell expression score of ≥ 1% resulted in a HR of 0.55 (95% CI: 0.36-0.83) with a median OS of 8.7 months. If PD-L1 expression serves as a true prognostic biomarker of response for PD-1 therapy, an increasing percentage of PD-L1 expression within the tumor microenvironment should correlate with improved OS and this is indeed what is observed. In KEYNOTE 040, a PD-L1 tumor proportion score (TPS) of ≥ 50% improved the HR to 0.53 (95% CI: 0.35-0.81, p = 0.0014) with a median OS of 11.6. Correspondingly, the complete response (CR) rates also increase with PD-L1 expression. In KEYNOTE 040, the overall CR rate in the pembrolizumab treated group was 1.6% (N = 4 of 247). However, in patients with a CPS ≥ 1 the CR was 2.0% (N = 4 of 196), and in patients with a TPS ≥ 50%, CR was 4.7% (N = 3 of 64). Furthermore, PFS and RR also increased with increasing PD-L1 expression (Table 1).

Conclusion
In the initial reported protocol-specified final analysis, the KEYNOTE 040 study did not meet their bio-statistical primary endpoint of OS [2]. However, the survival data of a pending 12 patients, with no change to the duration of the follow-up, made the difference in the positive outcome reported in the protocol-specified final analysis of KEYNOTE 040 [3,4]. Although inter-study comparisons must be put in the context of hidden biases which always exist in trials, confounding factors which may influence the clinical outcomes of the control arm will continue to be a challenge as we move toward combinatorial immunotherapy trials and calls upon innovative bio-statistical approaches and clinical trial design. The observed clinical benefit with anti-PD1 therapies in patients expressing PD-L1 within the tumor microenvironment, may call for paradigm shifts to include biomarker expression as part of primary or secondary bio-statistical endpoints in future clinical trials, rather than as exploratory endpoints as traditionally done. This is particularly important, since an increasing number of CRs is being observed in the biomarker expressing patient populations. Furthermore, given the impact on costs and tolerability, patient selection has to take place in order to provide individualized treatment, sparing toxicities and maximizing clinical outcome. Immunotherapy when applied in the appropriate setting has the potential to change the lives of cancer patients and affords us further opportunities to explore personalized treatment plans for our patients.

Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.