CADTH Reimbursement Review

Pembrolizumab (Keytruda)

Sponsor: Merck

Therapeutic area: Metastatic microsatellite instability-high/mismatch repair-deficient colorectal cancer

This multi-part report includes:

Clinical Review

Pharmacoeconomic Review

Clinical Review

Executive Summary

An overview of the submission details for the drug under review is provided in Table 1.

Introduction

Colorectal cancer begins in the colon or rectum and is characterized by cancerous cells that grow into and destroy nearby tissue. In metastatic disease, the tumour spreads to and damages other parts of the body.¹ About 70% to 90% of colorectal cancers are diagnosed after symptom onset, although population-based screening is increasing the number of asymptomatic cases identified.²

The Canadian Cancer Society expects colorectal cancer to be the third most commonly diagnosed cancer in Canada in 2020 and the second and third leading cause of death from cancer in men and women, respectively.³ Approximately 20% to 25% of newly diagnosed colorectal cancers are metastatic at the time of diagnosis, where untreated patients with metastatic disease survive only 6 to 10 months.^4-6 Despite advances in treatment, less than 20% of patients are alive at 5 years, and unfavourable prognostic factors (e.g., mutations in BRAF, proximal primary tumour location) are associated with survival under 18 months.⁷

For most patients with metastatic disease, therapy is often palliative and focuses on systemic chemotherapy. The recommended chemotherapeutic regimen varies by jurisdiction and is individualized based on the patient’s health, patient’s ability to tolerate chemotherapy, mutation status, presence of mismatch repair-deficient (dMMR) proteins or microsatellite high instability, location of primary tumour, aggressiveness of disease, and goals of therapy. Monoclonal antibodies are often used in addition to combination chemotherapy to target specific genetic subtypes of colorectal cancer. The microsatellite instability-high (MSI-H)/dMMR subtype may be present in 15%⁸ of colorectal cancer patients and in about 3.5% to 6.5% of stage IV patients. There is some evidence of favourable prognosis in early-stage disease and a hypothesis that such tumours may have lower metastatic potential.⁹ However, these patients may benefit less from adjuvant chemotherapy than other patients.^9-12

Pembrolizumab is an IV-administered monoclonal antibody that binds PD1 receptors on T cells, reversing T-cell suppression and inducing antitumour responses.

This report’s objective is to summarize a systematic review of the beneficial and harmful effects of pembrolizumab 200 mg administered every 3 weeks intravenously for the first-line treatment of adults with MSI-H/dMMR metastatic colorectal cancer (mCRC).

Stakeholder Perspectives

The information in this section is a summary of input from patient and clinician groups who responded to CADTH’s call for input, from clinical experts consulted by CADTH, and from drug programs that provided input to CADTH for the purpose of this review.

Patient Input

With the assistance of Colorectal Cancer Canada (CCC) and Colorectal Cancer Resource and Action Network (CCRAN), CADTH received feedback from 81 patients, 19 caregivers, and 5 individuals who were both patients and caregivers. Feedback was collected through online surveys, focus groups, and phone interviews. Patients noted the most common symptoms were bloody stools, diarrhea, fatigue, and pain. Most felt that their symptoms affected their daily lives, work, ability to exercise, and quality of life and had a psychological impact leading to various mental health issues. In general, patients hoped therapies could improve physical health and quality of life. Patients were willing to tolerate increasingly significant side effects for improved survival. In addition to improved health and quality of life, treatment funding, symptom improvement, simple administration, and limitation of side effects were important.

Clinician Input

Input From Clinical Experts Consulted by CADTH

Among patients with unresectable or mCRC, the experts identified an unmet need among MSI-H/dMMR patients who would benefit from pembrolizumab. In these patients, the experts felt pembrolizumab could be used as any line of therapy, consequently shifting the current treatment paradigm for these patients. Given the severity and type of disease, a patient with tolerable or manageable adverse events (AEs), preservation of quality of life, tumour response, survival, and stable disease would be considered a responder, while the opposite would be considered treatment failure and likely result in treatment discontinuation.

Clinician Group Input

Three registered clinician groups (the Ontario Health Gastrointestinal Cancer Drug Advisory Committee, the Canadian Gastrointestinal Oncology Evidence Network [CGOEN], and the Medical Advisory Board of the CCC), consisting of 16 clinicians, provided input for this review.

Drug Program Input

Input was obtained from all 9 provinces (ministries of health and/or cancer agencies) participating in CADTH reimbursement reviews. The Provincial Advisory Group identified the following as clinical factors that could impact the implementation:

• numerous clinical eligibility criteria

• barriers to diagnosing MSI-H/dMMR colorectal cancer and monitoring treatment effects.

The clinical experts consulted by CADTH provided responses that can be found in the Drug Program Input section.

Clinical Evidence

Pivotal Studies and Protocol Selected Studies

Description of Studies

One open-label randomized controlled trial (RCT) (KEYNOTE-177) was included in this review; the trial compared pembrolizumab 200 mg, intravenously, every 3 weeks to standard of care (SOC) to assess the efficacy and safety in adult patients with unresectable or metastatic MSI-H/dMMR colorectal cancer. The study randomized 307 patients across 120 sites from 23 countries. Notably, patients in the SOC arm were permitted to cross over to the pembrolizumab arm. Further, all patients were permitted to receive subsequent anticancer medications.

Efficacy Results

Median progression-free survival (PFS) was higher in the pembrolizumab group (16.5 months; 95% confidence interval [CI], 5.4 to 32.4) than in the SOC group (8.2 months; 95% CI, 6.1 to 10.2), resulting in a statistically significant hazard ratio (HR) of 0.60 (95% CI, 0.45 to 0.80). However, because the HR may be biased due to a violated proportional hazards assumption (PHA), the restricted mean survival time (RMST) was calculated. The RMST after 24 months of follow-up suggested pembrolizumab survival (13.7 months; 95% CI, 12.0 to 15.4) was higher than SOC survival (10.8 months; 95% CI, 9.4 to 12.2), indicating that patients receiving pembrolizumab for 24 months would survive for an average of 13.7 months, compared to 10.8 months for patients receiving SOC, a difference of 2.9 months (95% CI, 0.7 to 5.1). Because crossovers or use of subsequent anticancer therapies could bias these results, 2 sensitivity analyses were conducted. These analyses yielded similar HRs (HR = 0.61; 95% CI, 0.48 to 0.79, in the first analysis, and HR = 0.63; 95% CI, 0.45 to 0.88, in the second), suggesting the benefit observed in the primary analysis is generally robust. However, the first sensitivity analysis suggested only a small difference in median survival (7.0 months; 95% CI, 4.2 to 12.4, for pembrolizumab, versus 6.3 months; 95% CI, 4.6 to 8.2, for SOC).

Median overall survival (OS) was not reached; however, a statistically non-significant difference in OS was reported (HR = 0.77; 95% CI, 0.54 to 1.09). Three methods leading to 3 sensitivity analyses were conducted to adjust for any bias introduced by crossovers or subsequent use of anti-PD1/PDL1 treatment (but not subsequent use of other anticancer medications). The inverse probability of censoring weighting method (HR = 0.59; 95% CI, 0.32 to 1.24), the rank preserving structural failure time method (HR = 0.72; 95% CI, 0.47 to 1.11) and 2-stage analysis (HR = 0.89; 95% CI, 0.76 to 1.04) led to a wide range of point estimates, suggesting some bias and uncertainty in the primary analysis. While each point estimate suggests pembrolizumab may improve OS, due to variable point estimates, wide CIs, and the lack of statistical significance testing, the exact magnitude of any benefit is uncertain.

The overall response rate (ORR) was higher in the pembrolizumab arm (43.8%; 95% CI, 35.8 to 52.0) than in the SOC arm (33.1%; 95% CI, 25.8% to 41.4%). The between-group difference (10.7%; 95% CI, –0.2% to 21.3%) was statistically insignificant. The exploratory outcome of median time to response was similar in the pembrolizumab and SOC arms (2.2 and 2.1 months, respectively). Another exploratory outcome, median duration of response, could not be compared as there was no reported value in the pembrolizumab arm, but the percentage of patients with an extended response duration was higher in the pembrolizumab arm at 6, 9, 12, 18, and 24 months. Another exploratory analysis, quality of life, was measured by the European Organisation for Research and Treatment of Cancer Quality of Life Questionnaire Core 30 (EORTC QLQ-C30) at baseline and at 18 weeks. Scores in the pembrolizumab arm improved from baseline by 3.33 points (95% CI, –0.05 to 6.72), while scores decreased by 5.63 points (95% CI, –9.32 to –1.94) in the SOC arm. This difference (8.96; 95% CI, 4.24 to 13.69) exceeds the minimal important difference (MID) of 5 for colorectal cancer patients. The final exploratory analysis of quality of life was also measured by the EuroQol Visual Analogue Scale (EQ VAS) and EuroQol 5-Dimensions 3-Levels questionnaire (EQ-5D-3L) utility score at baseline and 18 weeks. Visual analogue scale scores in the pembrolizumab arm increased from baseline by 4.50 points (95% CI, 1.16 to 7.83) and decreased by 2.88 points (95% CI, –6.46 to 0.69) in the SOC arm for a difference of 7.38 (95% CI, 2.82 to 11.93) in favour of pembrolizumab, exceeding the MID of 6 for cancer patients. Finally, EQ-5D-3L utility scores in the pembrolizumab arm increased from baseline by 0.04 points (95% CI, 0.00 to 0.08) and decreased by 0.01 points (95% CI, –0.05 to 0.02) in the SOC arm for a difference of 0.05 (95% CI, 0.00 to 0.10) in favour of pembrolizumab, meeting the MID of 0.05 for cancer patients.

Harms Results

The proportion of individuals experiencing an AE was slightly lower in the pembrolizumab arm (97.4%) than in the SOC arm (99.3%). The most common AEs in both groups were gastrointestinal events and fatigue. The percentage of patients who reported serious AEs (SAEs) was lower in the pembrolizumab arm (40.5%) than in the SOC arm (52.4%). The frequency of AEs resulting in treatment discontinuation, death, or death due to an AE were similar between both groups. The frequency of immune-mediated AEs (which were identified as harms of interest for the review) were higher in the pembrolizumab arm (30.7%) than in the SOC arm (12.6%). The most common immune-mediated AEs were hypothyroidism, colitis, hyperthyroidism, pneumonitis, adrenal insufficiency, hepatitis, and infusion reactions. However, the frequency of infusion-related reactions was higher in the SOC arm.

Critical Appraisal

The major limitations on the internal and external validity of KEYNOTE-177 stem from the open-label nature of the study and the ability for patients in the SOC arm to cross over to the pembrolizumab arm and patients in either arm to receive subsequent anticancer medications, including other anti-PD1/PDL1 medications. While no patients crossed over from the pembrolizumab arm to the SOC arm, 56 patients (36.4%) in the SOC arm crossed over to the pembrolizumab arm. Further, 44 patients (28.6%) and 44 patients (28.8%) in the SOC and pembrolizumab arms, respectively, received subsequent anticancer therapy. The differential crossover and subsequent use of anticancer medications could have introduced bias. Consequently, 2 sensitivity analyses were conducted to assess the impact of crossovers and the use of subsequent anticancer medications on PFS. These sensitivity analyses did not alter the HR from the primary analysis substantially and indicated that the primary analysis is robust. Notably, 1 analysis suggested there may be similar median PFS in the pembrolizumab arm (7.0 months; 95% CI, 4.2 to 12.4) and SOC arm (6.3 months; 95% CI, 4.6 to 8.2); however, this is likely an overly conservative estimate of the treatment effect. Three sensitivity analyses were conducted to adjust for the impact of crossovers but not the subsequent use of other anticancer medications. Each analysis differed from the primary analysis, which suggested some bias in the primary analysis. However, despite this bias, all estimates also suggest some benefit from pembrolizumab. Unfortunately, due to variations in the point estimates, lack of significance testing, and wide CIs, the exact magnitude of any benefit is uncertain. While crossovers and use of subsequent anticancer therapies in such a trial is clinically and morally reasonable, real-world application of these data should acknowledge their contributing role in the results achieved in the study. Moreover, the trial did not enrol patients with an Eastern Cooperative Oncology Group Performance Status (ECOG PS) score greater or equal to 2. Finally, the PHA was violated in the PFS analysis; thus, the RMST is likely a more accurate estimate of any PFS benefit from pembrolizumab.

Indirect Comparisons

Description of Studies

The sponsor-submitted indirect treatment comparison (ITC) evaluated the comparative efficacy and safety of pembrolizumab versus competing interventions for the treatment of patients with MSI-H/dMMR mCRC. Pembrolizumab was compared with CAPOX, panitumumab plus FOLFOX, and SOC (which consisted of FOLFOX or FOLFIRI with or without cetuximab, as administered in KEYNOTE-177) for OS, PFS, ORR, and safety outcomes. Bayesian network meta-analysis (NMA) methods were used for the comparisons. The base-case NMA for OS and PFS consisted of all included studies and used a subgroup of patients from the KEYNOTE-177 trial who did not receive treatment with bevacizumab (termed “ITT-Bev”).¹⁴

Efficacy Results

Five studies (mostly phase II or III, open-label RCTs) were included in the NMA, including the pembrolizumab pivotal study, KEYNOTE-177. The base-case analysis of OS showed no difference between pembrolizumab and SOC or other interventions. Results from other analyses (e.g., time-varying analysis, adjustment for different crossover methods) generally showed similar results.

The base-case analysis for PFS indicated that pembrolizumab was favoured over all other treatments. Pembrolizumab was favoured for ORR compared to CAPOX for ORR.

Harms Results

Pembrolizumab was favoured over comparator treatments with lower odds of AEs of grade 3 or higher, treatment-related AEs of grade3 or higher, and treatment-related SAEs. No treatments were favoured over the others for discontinuations due to AEs (DAEs).

Critical Appraisal

Few inferences can be made from the results of the NMA because of important limitations with the included studies and the methods and assumptions made in the NMA.

There were several clinical assumptions made in the NMA and several notable differences in the characteristics of patients across studies, which impact the interpretation of the NMA results. In addition, the small size (5 studies) and structure of the network prevented the use of random-effects models in most outcomes of interest and the ability to model between-study heterogeneity and to assess consistency between direct and indirect comparisons as there were no closed loops.

The systematic review did not sufficiently follow accepted guidelines for reporting and conduct: Information was not provided on the search criteria or the search methodology, the methods used for study selection and data extraction were omitted, and no critical appraisal assessment of the included studies was reported.

Bevacizumab was excluded from the list of comparators, and so studies that included it as a treatment were excluded, including a subset of patients who received bevacizumab in the SOC group of KEYNOTE-177. Bevacizumab was identified as a relevant comparator for this review, and its exclusion limits the generalizability of the results. Furthermore, base-case networks were constructed using subgroups of patients from KEYNOTE-177, which would potentially bias the results in favour of pembrolizumab because this would break randomization of the trial(s) and because a comprehensive set of relevant comparators that were evaluated as part of the SOC group in the trial would not be included. As well, this approach led to imprecision in the results of these comparisons due to the smaller sample size for the subgroup of patients who did not receive bevacizumab, as evidenced by the wider credible intervals (CrIs).

The clinical characteristics of the populations of patients were heterogeneous. For example, 1 study enrolled patients with MSI-H/dMMR mCRC, 1 looked at patients with KRAS mutations, and 3 evaluated the general mCRC population. It was assumed that there would be no difference in treatment effect estimates between patients with MSI-H/dMMR, KRAS wild type (other than panitumumab), or general mCRC. However, these variations in population are expected to impact the relative treatment effects, as these are potentially important treatment effect modifiers, as evidenced by accumulated literature, like many other genetic biomarkers in oncology therapies. In addition, baseline ECOG PS and durations on treatment or on trial also varied considerably across trials.

Study design was another key source of heterogeneity. KEYNOTE-177 allowed patients to cross over from SOC to pembrolizumab following progression during the study. Statistical methods were used to account for crossover from SOC to pembrolizumab. Given the immature results on OS in the KEYNOTE-177 trial, the sponsor-submitted NMA was compromised on this outcome.

The authors noted that the networks were constructed based on availability of outcome data. Besides the comparison between SOC and CAPOX for OS, comparisons for almost all competing interventions were based on single trials. The available trials formed networks with no closed loops; therefore, it was not possible to validate the transitivity assumption of the NMA and check for consistency of results between direct and indirect comparisons. Random-effects models were attempted and determined not to be feasible due to the small number of studies for all outcomes except for OS. Consequently, as connections were only described by a single trial for PFS, ORR, and safety, results for these outcomes are based on fixed-effects model, as random-effects analyses were not meaningful.

The variable length in follow-up across trials means there is potential for bias related to underestimating the frequency of AEs, as some may occur rarely or may only become apparent long after the beginning of treatment.

Conclusions

Pembrolizumab improved PFS versus SOC in a single RCT of patients with MSI-H/dMMR mCRC. The difference between treatment groups was considered clinically meaningful based on input from the clinical experts consulted by CADTH. Although fewer patients died in the pembrolizumab group than in the SOC group, no statistically significant benefit in OS was found. The study was not adequately designed to evaluate treatment effects on OS, and estimates are uncertain due to crossovers and use of subsequent anticancer medications. Overall response rate, duration of response, and health-related quality of life (HRQoL) were also numerically improved relative to SOC. Although benefits aside from PFS could not be supported by statistical significance in all outcomes, the efficacy results in totality suggest pembrolizumab improves outcomes over SOC in a difficult-to-treat patient population whose cancer is often unresponsive to SOC. Pembrolizumab does not appear to be associated with more AEs than the SOC regimens evaluated. Known AEs of interest, such as immune-mediated AEs, occurred more frequently in the pembrolizumab group and remain a concern.

Introduction

Disease Background

Colorectal cancer is a potentially fatal disease that begins in the colon or rectum and is characterized by a group of cancerous cells (a tumour) that grow into and destroy nearby tissue. In metastatic disease, the tumour spreads to and damages other parts of the body.¹ About 70% to 90% of colorectal cancers are diagnosed after symptom onset, although population-based screening is increasing the number of asymptomatic cases identified.² Common early symptoms include a change in bowel habits, rectal bleeding, a rectal or abdominal mass, iron deficiency anemia, and abdominal pain. In metastatic disease, right upper-quadrant pain, abdominal distention, early satiety, supraclavicular adenopathy, and periumbilical nodules can be observed. Colorectal cancers are often diagnosed by histologic examination of a biopsy. Colonoscopies are accurate and versatile diagnostic tests, while flexible sigmoidoscopy and CT colonography are appropriate diagnostic techniques in specific circumstances. The Canadian Cancer Society expects colorectal cancer to be the third most commonly diagnosed cancer in Canada in 2020 and the second and third leading cause of death from cancer in men and women, respectively.³ Colorectal cancer is also expected to account for 12% of new cancer cases and 12% of new cancer deaths, leading to an additional 26,900 Canadians being diagnosed with colorectal cancer and 9,700 additional deaths due to the disease.³

Approximately 20% to 25% of newly diagnosed colorectal cancers are metastatic at time of diagnosis, and untreated patients with metastatic disease survive only 6 to 10 months.^4-6 With the development of 5-fluorouracil therapy, survival is extended to 11 to 12 months and up to 3 years with additional therapeutic innovations.¹⁵ Yet despite these innovations, less than 20% of patients are alive at 5 years, and unfavourable prognostic factors (e.g., mutations in BRAF genes, proximal primary tumour location) are associated with survival under 18 months.⁷

Standards of Therapy in mCRC

For most patients with metastatic disease, therapy is palliative and focuses on systemic chemotherapy. Several therapeutic options are available, with the backbone of therapy consisting of a combination of 5-fluorouracil and leucovorin (or capecitabine) plus either irinotecan or oxaliplatin. The most recommended combinations are FOLFOX (fluorouracil-leucovorin-oxaliplatin), FOLFIRI (fluorouracil-leucovorin-irinotecan), FOLFOXIRI (fluorouracil-leucovorin-irinotecan-oxaliplatin) or CAPOX (capecitabine-oxaliplatin). The recommended combination of therapy varies by jurisdiction and is individualized based on the patient's health, the patient’s ability to tolerate chemotherapy, the goals of therapy, plus the tumour’s molecular profile (e.g., mutation status in RAS and BRAF genes, presence of dMMR or MSI-H) and the location of the primary tumour. Monoclonal antibodies such as bevacizumab, cetuximab, and panitumumab are often used in addition to combination chemotherapy. Cetuximab and panitumumab target the epidermal growth factor receptor and are used in patients with tumours with wild-type RAS and BRAF genes. Bevacizumab, which targets vascular endothelial growth factor, is also often considered for use in the first-line setting. Although the treatment of patients with mCRC continues to evolve, it is now best thought of as a continuum of care, where the evidence-based treatment options are administered with consideration of biomarkers and primary tumour location, in combination and in sequence with the intent to carefully balance a patient’s quality of life with the treatment’s life-prolonging effects. In patients with incurable distribution of metastatic disease, the primary goals of therapy may be to delay tumour progression to ensure quality of life and prolong survival.¹⁶

Ultimately, the diversity of therapeutic approaches is a function of the heterogenous pathophysiology of mCRC. The MSI-H/dMMR subtype may be present in 15%⁸ of colorectal cancer patients and in about 3.5% to 6.5% of stage IV patients.¹⁵ There is some evidence of favourable prognosis in early disease and a hypothesis that such tumours may have lower metastatic potential.⁹ However, these patients may benefit less from adjuvant therapy with 5-fluorouracil or capecitabine.^9-12 In brief, dMMR patients accumulate errors throughout their genome, specifically the accumulation of repeating sequences of nucleotide bases called microsatellites. Accumulation of microsatellites in critical growth regulatory genes may lead to abnormalities in cell growth control.

Drug

Pembrolizumab falls under the class of checkpoint inhibitors or anti-PD1 medications. It is a monoclonal antibody that binds PD1 receptors on T cells, reversing T-cell suppression and inducing antitumour responses. It is administered intravenously and is indicated in Canada for classical Hodgkin lymphoma, mediastinal B-cell lymphoma, adult urothelial carcinoma, endometrial cancer, melanoma, non–small cell lung carcinoma, renal cell carcinoma, and head-neck squamous cell carcinoma. Pembrolizumab is also indicated for but not universally funded for dMMR/MSI-H colorectal cancer whose tumours have progressed following treatment with fluoropyrimidine, oxaliplatin, and irinotecan.¹⁷

Pembrolizumab received a Health Canada Notice of Compliance in March 2021 as first-line therapy in adult patients with unresectable or metastatic dMMR/MSI-H colorectal cancer. The sponsor has requested CADTH to review pembrolizumab in line with the indication. Pembrolizumab has been approved for this indication by the FDA¹⁸ and the European Medicines Agency.¹⁹

Stakeholder Perspectives

Patient Group Input

This section was prepared by CADTH staff based on the input provided by patient groups.

About the Patient Groups and Information Gathered

CADTH received 2 patient input submissions from CCC and CCRAN for this review. Both groups are national, not-for-profit patient advocacy groups. CCC posted a survey on its social media platforms as well as on those of international colorectal cancer organizations between October 30, 2020, and January 1, 2021. Information from 8 patients and 2 caregivers from Canada (n = 6), the US (n = 3), and Malaysia (n = 1) made up the responses. All responses were from or on behalf of patients receiving pembrolizumab, 50% of whom were females and 70% of whom had stage IV cancer at the time of survey. At the time of diagnosis, 2 patients were younger than 40 years old, 3 were between 40 and 49, 2 were between 50 and 59, and 3 were between 60 and 69. CCRAN collected information through 3 separate processes for its submission and advertised with the help of its support group members and members of its Medical Advisory Board. The first was a national online survey run from December 6 to 30, 2020, during which time the organization received input from 63 colorectal cancer patients, 17 caregivers, and 5 patients who were also caregivers, all of whom were living in Canada. Almost 59% of respondents were female, and patient age varied from 31 to 90 years old, although most were between 41 and 70 years old. Later stages of cancer were more heavily represented, with most respondents in either stage III or IV. The second process was a focus group discussion conducted over Zoom (teleconference platform) on November 15, 2020, to gain insight into the experiences and symptoms of 7 patients who had mCRC. The third process was phone interviews made between December 8 and 21, 2020, with 3 patients who had specifically received pembrolizumab treatment.

Disease Experience

Patients from both CCC and CCRAN surveys noted that some of their common cancer symptoms included bloody stools, diarrhea, fatigue, and pain. Most felt that their symptoms affected their daily lives, work, and ability to exercise; interfered with their quality of life; and had a psychological impact leading to various mental health issues. Respondents from the CCRAN survey reported that they were unable to work (51%) and unable to fulfill family obligations (42%). Patients also felt that their treatments made them forgetful and fatigued and made it difficult to plan for the future. One respondent stated, “Currently afraid to make future plans as historically, I have done so many times and had to stop and restart my life over and over again. Anxiety is high and not inappropriate for someone in my position. The less I talk or have to think about cancer and the more I engage with people and life pursuits outside of a cancer focus, the healthier I get, so it is a self-care balance.” The CCRAN focus group identified pain in various locations, breathing issues, debilitating fatigue, and diminished appetite as key burdens resulting from the spread of the disease to other organs. Moreover, respondents noted that, from their experience, there were often no warning signs before the cancer had advanced and was more difficult to treat. “I felt fine. You would never know that I have stage four disease. I guess that’s why they call it the silent killer.”

Caregivers also shared feelings of helplessness, challenges with frequent travel over long distances for treatments, difficulties of managing side effects, loss of lifestyle, and loss of work and income. From the CCRAN survey, caregivers expressed feeling the psychological impacts that come with supporting someone with cancer, and some had sought out help from support groups (n = 20), from psychologists (n = 16), and through online resources (n = 9).

Experiences With Currently Available Treatments

Most patients from the CCC survey had accessed previous therapies, such as chemotherapy (7), radiation therapy (3), and surgery (4), to treat their cancer. Two respondents had not received pembrolizumab but were on other immunotherapies (nivolumab and ipilimumab) at the time of the survey. Six patients who had previous treatments felt that symptoms were either not controlled or only partially controlled. Respondents identified diarrhea, nausea, hair loss, vomiting, fatigue, and pain as the most difficult symptoms to tolerate. On a scale from 1 to 10, all respondents ranked at least an 8 the importance of having a choice in treatment based on known side effects.

When asked about accessing cancer treatments, 4 out of 10 patients expressed difficulties, while 4 individuals reported that the therapies they had been recommended were based on what was funded in their region of residence. For the 2 who were receiving immunotherapies, 1 had to pay out of pocket for the drug and 1 had to go to the US to gain access. Three patients had received financial assistance that covered from 20% to 100% of their treatment cost. In addition to treatment costs, patients and caregivers reported financial stress related to travel expenses, medical visits and tests, and lost time from work. Most patients (70%) felt that paying out of pocket for access to new therapeutics would depend on the cost.

Six patients from the CCC survey who received treatment in Canada felt that access to treatment was appropriate or fair in their region. One patient shared their concerns with accessing drugs in Canada:

I don’t think colorectal patients should have to pay for treatments with proven efficacy. There are studies that now recommend immunotherapy treatment for MSI- High/mismatch repair deficient colorectal cancers. In other countries these treatments are now first-line treatments. Canadians should have access to these treatments and they should be covered. If new treatments arise that haven’t had proven efficacy through peer reviewed studies, but look promising, I would consider paying out of pocket if the cost wasn’t exorbitant and some financial help was available on a compassionate basis.

Patients from the CCRAN survey listed a variety of therapies they had received, including FOLFOX (72%), FOLFIRI (34%), capecitabine (40%), bevacizumab/MVASI (21%), panitumumab (15%), cetuximab (6%), pembrolizumab (6%), trifluridine-tipiracil (4%), regorafenib (2%), and encorafenib (2%). Most respondents noted that diarrhea, hand and foot syndrome, and neuropathy were common with their current medication, but that fatigue and nausea were the 2 side effects most difficult to tolerate. Aside from the previous list of drugs, respondents had also had surgery, radiation therapy, and other forms of chemotherapy.

When asked about access to treatments, CCRAN responses (all respondents were from Canada) were variable from very limited or restrictive (4% of responses) to very appropriate and/or fair (24%). When asked if the treatments they were offered were based solely on what was publicly funded in their province, 14 patients responded yes, 12 responded no, and many were unsure (26). Twelve individuals reported receiving financial assistance for their treatments, while another 41 had not received any. For those who had assistance, the amounts received ranged from 10% to 100% of their treatment costs. Furthermore, 19 patients noted that in addition to the cost of the drugs, they had expenses for travel, accommodations, medical tests, supplies, and procedures. Forty respondents stated that they would be willing to pay out of pocket to access new therapies through a private clinic, while another 10 responded they would not. Reasons for the latter were often due to high costs or to avoid creating a financial burden on their family.

Improved Outcomes

Respondents from both CCC and CCRAN surveys felt it was very important that new therapies would improve both physical condition and quality of life. Furthermore, nearly all patients and caregivers expressed interest in a treatment that was proven to provide a better quality of life even if it did not extend OS, since this would allow them to engage in social activities and return to daily life without worrying about side effects. When asked about the severity of side effects patients would be willing to tolerate for extended survival, both survey results were variable from no side effects to significant ones, though the general trend was a tolerance toward more significant side effects in exchange for improved survival. Patients from the CCRAN survey chose the top 3 outcomes they would like to see with cancer treatments: provides a cure, if possible (94%); prolongs life by a substantial amount of time (86%); and promotes good quality of life (82%). Other key outcomes included limited side effects, treatment funding, improvement in symptoms, and simple administration.

Experience With Drug Under Review

Five patients from the CCC survey confirmed receiving pembrolizumab, 2 as first-line treatment and 1 each as second, third, and fourth line. Patients were able to access pembrolizumab via clinical trials, special access programs, and private insurance. At the time of the survey, most patients had been treated for 18 months to 24 months.

With regard to trialling different treatments before finding 1 that worked, 1 individual stated the following, “a lot of time was wasted on other treatments,” and she felt she “could have avoided brutal, ineffective chemo [had she] started with Keytruda instead of other treatments.” While receiving pembrolizumab, patients experienced hair loss, fatigue, constipation, fever, and muscle, bone, and joint pain but felt these side effects were somewhat tolerable and relatively minor. However, 2 respondents stated that diarrhea, bloody stools, and constipation were managed less effectively compared to other therapies they had received. In general, on a scale from 1 to 10, the side effects had a low (1 to 3) impact on their lives and their overall experience was much better (10) than with other treatments. When asked specifically about immunotherapies, patients felt that the psychological impacts like feelings of isolation, emotional drain, anxiety, and fatigue from long medical appointments were the most difficult to deal with. When asked about treatment response, most patients stated that their cancer was completely gone, shrunken, or controlled.

Patients and caregivers hope pembrolizumab will increase OS, maintain or improve quality of life, delay onset of symptoms, delay need for chemotherapy, be easy to use, and reduce the side effects from other medications. Importantly, nearly all patients felt that their quality of life was high or normal and that they had been able to return to all their daily activities. One patient emphasized, “Without Keytruda, I probably wouldn’t be here. I’ve endured this battle for four years now. I’ve been able to switch to the 6-week treatment making it even easier to enjoy a ‘normal’ life.” Another recalled her experience, “Tumour Board deemed me ‘inoperable.’ I underwent 25 radiation, 25 oral chemo. Nothing. I was told to get my affairs in order. Two months later Keytruda became an option. Here I am 45 treatments later and still kicking...working full-time, working part-time, two teenage sons, a single mom taking care of the household. Without Keytruda, I would be dead.”

CCC respondents noted additional travel costs related to accessing treatment, with 1 respondent citing the lack of provincial coverage as a barrier; all agreed that pembrolizumab should be funded in their regions. CCRAN interviewed patients who had experience with pembrolizumab, 1 of whom had received the immunotherapy as a first-line treatment during a clinical trial (in combination with cyclophosphamide and Survivac vaccine). The other 2 patients used pembrolizumab as second-line therapy, 1 through a clinical trial (in combination with cyclophosphamide and Survivac vaccine) and 1 as monotherapy through private insurance at a private infusion clinic. None travelled or paid out of pocket for testing. They also did not wait long for their test results, though 1 patient experienced some anxiety while waiting, knowing it would determine if they could receive pembrolizumab.

When asked about their lives since beginning treatment with pembrolizumab, 1 reported no side effects, a second stated they only had elevated lipase enzyme levels, and a third was uncertain since their experience was a combination therapy. All 3 were enthusiastic about the improvements in quality of life compared to side effects related to either chemotherapy or cancer symptoms. One patient stated, “My life did a total turn around. I have more energy and feeling more like myself today than ever, cancer notwithstanding.” The patients all reported their tumours had shrunken, some to the point that they were unidentifiable. The individual who reported having elevated lipase levels had to stop pembrolizumab treatment in May 2020 after 9 cycles and had not resumed treatment at the time of the interview, though they noted that their cancer was still controlled despite stopping immunotherapy. A second patient reported having 42 cycles and only had to stop for a month while receiving radiation treatment in March 2018, while the last individual had received 9 cycles and did not have to stop at any time. Patients indicated that not only were infusion times shorter than for traditional chemotherapy, but they were also able to receive pembrolizumab every 3 weeks — rather than every 2 weeks — with minimal side effects. When asked if accessing pembrolizumab was worth it, 1 patient responded, “Oh, yes, yes, yes, it was very worth it. It saved my life! It brought me back to the woman I used to be who used to play and love life. I felt like nothing was going to happen for me and my cancer and then, bang, Keytruda came along and pulled off a miracle for me. It gave me a second chance at life. So, it was very worth it for me to have accessed it.” Furthermore, patients have been able to return to their normal lives, work again, and spend time with family.

Companion Diagnostic Test

Patients listed the various methods that were used to diagnose their colorectal cancer, such as reporting symptoms, colonoscopy, CT scan, incidental finding or physical examination, biopsy, and blood work. Most patients from the CCC survey received positive biomarker testing after diagnosis, and all were tested before being prescribed pembrolizumab. For individuals who received biomarker testing, 2 received chemotherapy as their first treatments, another received immunotherapy right away, and a fourth could choose between the 2 options but had to pay out of pocket for immunotherapy. CCRAN asked patients if there were additional costs for testing such as tumour genomic profiling and, if so, what the costs were. Nine responses confirmed the supplementary costs, and while not all patients followed through with the tests, for those who did, expenses were more than $2,500.

Additional Information

Both patient groups emphasized the need for new and effective treatments for patients with MSI-H/dMMR cancer. They highlighted that patients should have equitable access without financial restrictions to a medication like pembrolizumab that has led to improvements in quality of life and survival outcomes, has few side effects, and can be easily administered.

Clinician Input

Input From Clinical Experts Consulted by CADTH

All CADTH review teams include at least 1 clinical specialist with expertise in the diagnosis and management of the condition for which the drug is indicated. Clinical experts are a critical part of the review team and are involved in all phases of the review process (e.g., providing guidance on the development of the review protocol, assisting in the critical appraisal of clinical evidence, interpreting the clinical relevance of the results, and providing guidance on the potential place in therapy). The following input was provided by 3 clinical specialists with expertise in the diagnosis and management of colorectal cancer.

Unmet Needs

The clinical experts highlighted the lack of a current therapy that exploits the distinct profile observed in MSI-H/dMMR patients. This mutational profile can be unresponsive to the largely chemotherapy-based conventional treatments — treatments that are associated with important adverse effects. Thus, a treatment that better targets these tumours with improved outcomes and fewer adverse effects is needed.

Place in Therapy

The clinical experts stated that pembrolizumab would be used alone for any line of therapy, including first line, in patients with MSI-H/dMMR colorectal cancer.

Patient Population

The clinical experts suggest that only patients with MSI-H/dMMR confirmed mCRC would benefit from pembrolizumab. Any eligibility criteria for first-line use should adhere to the enrolment criteria of the KEYNOTE-177 study. In short, this would include patients with an incurable distribution of mCRC, an MSI-H/dMMR profile, an ECOG PS of 0 or 1, willingness to adopt an adequate method of contraception, and adequate organ function. Patients should be free of active autoimmune disease (requiring disease-modifying agents, corticosteroids, or immunosuppressive therapies); uncontrolled metastases to the central nervous system; prior therapy with an immune checkpoint inhibitor (i.e., another anti-PD1/PDL1 or anti-CTLA-4 agent); unsuppressed HIV, hepatitis B, or hepatitis C; interstitial lung disease or active non-infectious pneumonitis; active infection requiring systemic therapy; pregnancy; or breastfeeding. It is considered SOC to assess tumours for MSI-H/dMMR status through immunohistochemistry and polymerase chain reaction–based analysis; thus, potentially eligible patients would be identified by the patient’s oncologist through standard care.

Assessing Response to Treatment

The experts suggested that assessing treatment response carefully balances survival, HRQoL, and AEs. Benefits related to survival and HRQoL (including a delay in deterioration in quality of life), while minimizing SAEs, would be considered a response to treatment. In some circumstances, disease stability or surrogate end points, such as PFS as determined by radiologic imaging, will be assessed. Typically, a 2- to 3-month trial of therapy is offered before reassessment of therapy.

Discontinuing Treatment

Disease progression (as determined by radiologic imaging or unequivocal clinical findings), intolerable or unmanageable toxicities, and a change in the patient status taking them outside of standard eligibility criteria would preclude ongoing treatment.

Prescribing Conditions

Treatment should be provided in a multidisciplinary centre with expertise in the use of such medications. This includes patient selection and management by an experienced medical oncologist, drug preparation in an accredited oncology-specific pharmacy, and administration by trained nurses in an environment where infusion-related and other emergencies can be promptly addressed. A gastroenterologist, surgeon, and/or interventional radiologist would need to obtain a biopsy or surgical specimen. A pathologist would be required to confirm the histologic diagnosis and to interpret the necessary molecular studies. Ultimately, a medical oncologist would ensure that the relevant criteria are met and that the patient is involved in the decision-making process.

Clinician Group Input

This section was prepared by CADTH staff based on the input provided by clinician groups.

Three registered clinician groups provided input for this review. One submission was by the Ontario Health (Cancer Care Ontario) (OH-CCO) Gastrointestinal Cancer Drug Advisory Committee (GI DAC), which included 4 physicians. OH-CCO GI DAC provides evidence-based clinical and health system guidance for the Provincial Drug Reimbursement Programs and Systemic Treatment Program. CGOEN, the Medical Advisory Board of the CCC, and other gastrointestinal cancer-treating clinicians provided a separate submission. The group consisted of 12 medical oncologists and gastroenterologists who practice in Alberta, British Columbia, Nova Scotia, Ontario, Quebec, and Saskatchewan. CGOEN consists of Canadian gastrointestinal oncology clinicians who work to advance the knowledge and treatments of gastrointestinal cancers, while the Medical Advisory Board of the CCC acts as a link between patient groups and the medical community, helping to ensure that activities and health information are relevant and communicated between the 2. More information about the latter group can be found at https://www.colorectalcancercanada.com/about/staff-board-medical-advisory.

Unmet Needs

Both GI DAC and CGOEN noted a lack of targeted treatment available for the MSI-H/dMMR colorectal cancer population, as well as for patients who have more aggressive disease, are unresponsive to standard chemotherapy, or have chemotherapy-related toxicities. Of these groups, patients diagnosed with MSI-H/dMMR colorectal cancer — who make up approximately 5% of the unresectable or mCRC population — were identified as having the greatest unmet needs.

Place in Therapy

GI DAC anticipated that if patients are treated with pembrolizumab as a single agent for first-line use, therapies that are current SOC will continue to be funded and shifted downstream. It was also suggested that patients who experience disease progression after being treated with chemotherapy could receive pembrolizumab next.

Patient Population

Patients with MSI-H/dMMR advanced or mCRC who also have good performance status and who align with the criteria outlined in the KEYNOTE-177 clinical trial would be most suitable and have the greatest need for treatment with pembrolizumab. The patients who are best suited for pembrolizumab would be identified through companion diagnostics, although GI DAC acknowledged that reflex testing is not universally available. The group also emphasized that testing should be available to all patients with colorectal cancer and performed in a timely manner to prevent treatment delays. Likewise, CGOEN highlighted that MSI-H/dMMR cancers are readily identifiable with locally available assays such as immunohistochemistry and/or polymerase chain reaction testing and suggested that next-generation sequencing could also be an option for detecting multiple genetic aberrations such as MSI. Conversely, patients who are not MSI-H/dMMR, patients who have contraindications to immunotherapy or poor performance status, or patients who meet the exclusion criteria used in KEYNOTE-177 would not be candidates to receive pembrolizumab.

Assessing Response to Treatment

To assess treatment response, GI DAC listed the following outcomes that would be used in clinical practice: carcinoembryonic antigen testing, biochemistry results, standard imaging, and clinical improvement. CGOEN stated that patients will normally have a CT scan performed every 2 to 3 months to assess response and that this will continue until disease progression or treatment-limiting toxicities occur or until the patient decides to stop. Furthermore, the group confirmed that the outcomes used in clinical trials are consistent with those used in clinical practice. Overall survival and quality of life outcomes were identified as the most important outcome measures when considering a clinically meaningful response. Both groups noted that other measures of meaningful response include improvement or stabilization of symptoms; reduction in the frequency or severity of symptoms such as abdominal pain, rectal bleeding, anemia, and constipation; and ability to perform daily activities. These outcomes would be assessed per current SOC or every 3 to 4 cycles.

Discontinuing Treatment

When deciding to discontinue treatment, both groups stated that evidence of disease progression, such as deterioration in clinical status or objective progression on imaging, would be important. GI DAC would also consider the occurrence of specific AEs and the need for additional treatment as major factors, while CGOEN identified immune-mediated adverse reactions and grade 3 or 4 toxicities as reasons for stopping treatment.

Prescribing Conditions

Pembrolizumab should be administered in an approved oncology infusion clinic in an outpatient hospital setting. GI DAC felt that it would be ideal if reflex testing were in place at the time of reimbursement and if all current downstream chemotherapy continued to be funded.

Drug Program Input

The drug programs provide input on each drug being reviewed through CADTH’s reimbursement review process by identifying issues that may impact their ability to implement a recommendation. The implementation questions and corresponding responses from the clinical experts consulted by CADTH are summarized in Table 4.

Clinical Evidence

The clinical evidence included in the review of pembrolizumab is presented in 2 sections. The first section, the systematic review, includes pivotal studies provided in the sponsor’s submission to CADTH and Health Canada, as well as those studies that were selected according to an a priori protocol. The second section includes indirect evidence from the sponsor and indirect evidence selected from the literature that met the selection criteria specified in the review.

Systematic Review (Pivotal and Protocol Selected Studies)

Objectives

To perform a systematic review of the beneficial and harmful effects of pembrolizumab 200 mg administered every 3 weeks intravenously for the treatment of unresectable or metastatic MSI-H/dMMR colorectal cancer in adults.

Methods

Studies selected for inclusion in the systematic review included pivotal studies provided in the sponsor’s submission to CADTH and Health Canada, as well as those meeting the selection criteria presented in Table 5.

Outcomes included in the CADTH review protocol reflect outcomes considered to be important to patients, clinicians, and drug plans.

The literature search for clinical studies was performed by an information specialist using a peer-reviewed search strategy according to the PRESS (Peer Review of Electronic Search Strategies) checklist available on the CADTH website.²⁰

Published literature was identified by searching the following bibliographic databases: MEDLINE All (1946–) via Ovid and Embase (1974–) via Ovid. The search strategy comprised both controlled vocabulary, such as the National Library of Medicine’s MeSH (Medical Subject Headings), and keywords. The main search concepts were Keytruda/pembrolizumab and colorectal cancer. Clinical trials registries were searched: the US National Institutes of Health’s clinicaltrials.gov, WHO’s International Clinical Trials Registry Platform search portal, the Canadian Cancer Trials Database, Health Canada’s Clinical Trials Database, and the European Union Clinical Trials Register. No filters were applied to limit the retrieval by study type. Retrieval was not limited by publication date or by language. Conference abstracts were excluded from the search results. See Appendix 1 for the detailed search strategies. The initial search was completed on January 5, 2021. Regular alerts updated the search until the meeting of the CADTH pan-Canadian Oncology Drug Review Expert Committee on May 13, 2021.

Grey literature (literature that is not commercially published) was identified by searching relevant websites from the Grey Matters: A Practical Tool for Searching Health-Related Grey Literature checklist available on the CADTH website.²¹ Included in this search were the websites of regulatory agencies (FDA and European Medicines Agency). Google was used to search for additional internet-based materials. See Appendix 1 for more information on the grey literature search strategy.

Two CADTH clinical reviewers independently selected studies for inclusion in the review based on titles and abstracts, according to the predetermined protocol. Full-text articles of all citations considered potentially relevant by at least 1 reviewer were acquired. Reviewers independently made the final selection of studies to be included in the review, and differences were resolved through discussion.

Findings From the Literature

One study was identified from the literature for inclusion in the systematic review (Figure 1). The included study is summarized in Table 6. A list of excluded studies is presented in Appendix 2.

Figure 1: Flow Diagram for Inclusion and Exclusion of Studies

Description of Studies

KEYNOTE-177 is an ongoing, open-label randomized trial comparing pembrolizumab 200 mg every 3 weeks with SOC, as defined in Table 6, in adults with unresectable or metastatic MSI-H/dMMR colorectal cancer. The submitted information constitutes the final analysis of 1 co-primary end point (PFS) and the second interim analysis of another co-primary end point (OS). The study recruited patients from 120 sites across 23 countries, including Canada (4 sites), the US, Asia, Europe, and Australia. It randomized 307 adult patients in total. The first patient visit was on November 30, 2015; the last visit was on February 19, 2020, and the database was locked on March 16, 2020.

Populations

Inclusion and Exclusion Criteria

The inclusion and exclusion criteria for KEYNOTE-177 are described in Table 6. KEYNOTE-177 recruited adults with locally confirmed MSI-H/dMMR stage IV colorectal cancer. Patients needed to be 18 years or older, have an ECOG PS of 0 or 1 established 10 days before initiation of therapy, and have a life expectancy greater than 3 months.

Baseline Characteristics

Baseline characteristics for the intention-to-treat (ITT) analysis are summarized in Table 7. Between the pembrolizumab and SOC arms, there are small differences in the proportion of female participants (53.6% versus 46.8%), the number of patients recruited from outside Asia, North America, and Western Europe (14.4% versus 9.7%), the proportion of patients with an ECOG PS of 0 (49.0% versus 54.5%), the proportion of patients with an ECOG PS of 1 (51.0% versus 45.5%), the proportion of patients with hepatic or pulmonary metastases (56.2% versus 47.4%), and the proportion of patients with other metastases (43.8% versus 52.6%) or other mutation status (34.0% versus 24.7%).

Interventions

Patients in KEYNOTE-177 were centrally randomized in a 1:1 ratio to pembrolizumab 200 mg administered intravenously every 3 weeks or to SOC every 2 weeks. SOC was determined by the treating clinician and could consist of any of the following:

• mFOLFOX (oxaliplatin, 85 mg/m² of body surface area on day 1; leucovorin, 400 mg/m² of body surface area on day 1; and 5-fluropyrimidine, 400 mg/m² of body surface area on day 1, then 2,400 mg over 46 to 48 hours)

• FOLFOX (as above) plus bevacizumab (5 mg/kg on day 1)

• FOLFOX (as above) plus cetuximab (400 mg/m² of body surface area on day 1 then 250 mg/m² weekly)

• FOLFIRI (irinotecan, 180 mg/m² of body surface area on day 1; leucovorin, 400 mg/m² of body surface area on day 1; and 5-fluropyrimidine, 400 mg/m² of body surface area on day 1, then 2,400 mg over 46 to 48 hours)

• FOLFIRI (as above) plus bevacizumab (as above)

• FOLFIRI (as above) plus cetuximab (as above).

Pembrolizumab and SOC regimens were provided for a maximum of 35 treatments, until disease progression, development of unacceptable AEs, or decision by the physician or patient to withdraw. Those in the SOC arm were permitted to cross over to the pembrolizumab arm, and those in either arm could receive subsequent anticancer therapy. Crossovers were allowed if several criteria were met, notably including the following: progressive disease per Response Evaluation Criteria in Solid Tumors version 1.1 (RECIST 1.1), not having received other anticancer therapy since the last study dose, adverse effects from study medications at grade 1 or better, and having maintained an ECOG PS of 0 or 1. There is no information regarding the criteria used to initiate subsequent anticancer therapy, when it was initiated, or why it was initiated.

Outcomes

A list of efficacy outcomes identified in the CADTH review protocol that were assessed in the clinical trials included in this review is provided in Table 8. A detailed discussion and critical appraisal of outcome measures is provided in Appendix 4.

Progression-Free Survival

Progression-free survival was a co-primary outcome and defined as the time from randomization to the first of either disease progression per RECIST 1.1 based on a blinded central imaging vendor or death due to any cause. Patients were censored at the last disease assessment if there was no progression, death, or use of subsequent anticancer treatment. If a new anticancer treatment was initiated, patients were censored at the last disease assessment before treatment initiation. A sensitivity analysis was conducted that treated the crossovers or the addition of anticancer medications as death or progression events. Additionally, a second sensitivity analysis defined PFS as the time from randomization to progression on next line of therapy or death from any cause and thus provided no consideration for the crossovers or addition of anticancer medications. The clinical experts consulted suggest any survival beyond 8 to 10 months is clinically significant.

Overall Survival

Overall survival was a co-primary outcome and defined as the time from randomization to death from any cause. Patients with no known death were censored at the point of last contact. The clinical experts consulted suggested any incremental survival above the SOC arm would be considered clinically significant.

Overall Response Rate

Overall response rate was a secondary outcome and was defined as the proportion of patients who had a partial or complete response per RECIST 1.1 based on a blinded central imaging vendor. No MID was identified.

Duration of Response

Duration of response was an exploratory outcome and defined as the time from first response to disease progression or death for any patient with a demonstrated partial or complete response, per RECIST 1.1, based on a blinded central imaging vendor. No MID was identified.

Time to Response

Time to response was not defined in the documentation provided.

Health-Related Quality of Life

HRQoL was measured by the EORTC QLQ-C30, the EORTC Quality of Life Questionnaire Colorectal Cancer Module (QLQ-CR29), and the EQ-5D-3L. The EORTC QLQ-30 is a widely used cancer-specific HRQoL instrument consisting of 30 items measuring 5 functional dimensions (physical, role, cognitive, emotional, and social), 3 symptoms dimensions (fatigue, nausea/vomiting, and pain), 6 additional items (dyspnea, sleep disturbance, appetite loss, constipation, diarrhea, and financial impact), and a global HRQoL measure. The EORTC QLQ-CR29 includes scales to assess functional issues and symptoms such as urinary frequency, fecal seepage, stool consistency, body image, and other common problems associated with treatment of colorectal cancer. EQ-5D-3L is another standard instrument to measure health outcomes and is particularly useful in developing economic models. EQ-5D-3L measures mobility, self-care, usual activities, pain/discomfort, and anxiety/depression using a 3-point ordinal scale. These measurements can be pooled into a single utility score. Further, the EQ-5D-3L contains a Visual Analogue Scale (EQ VAS) ranging from 0 to 100 so that participants may rate their general health state.

Each of the HRQoL questionnaires was conducted at baseline and at 18 weeks so that the change could be calculated and compared between groups. The MID for colorectal cancers on the EORTC QLQ-C30 is 5,²² the MID for cancer patients on the EQ VAS is 6 to 10, and the MID for US patients with cancer on the EQ-5D utility scale is 0.05 to 0.08.²³ No MID on the EQ-5D-3L scale was identified for colorectal cancer patients.

Harms

AEs were defined as any unfavourable or unintended sign, symptom, disease, or worsening of a pre-existing condition temporally associated with a study treatment. Serious AEs were those that resulted in death; were life-threatening; resulted in persistent or significant disability or incapacity; resulted in or prolonged existing inpatient hospitalization; were a congenital anomaly or birth defect, an important medical event, or a new cancer; or were associated with an overdose. Immune-mediated AEs were defined as SAEs characterized by inflammation and included infusion-related reactions.

Statistical Analysis

The study was a group sequential trial with PFS and OS as co-primary end points and ORR as a secondary end point. All analyses were conducted on an ITT basis. PFS and OS were analyzed using log-rank tests for Kaplan–Meier plots and simple Cox proportional hazards models using the Efron method of tie handling to assess the magnitude of treatment differences between the treatment arms.

In the PFS analysis, the PHA was violated; hence, the RMST was calculated instead. Like an HR, RMST measures long-term survival profiles relative to median survival, better accounts for long-term survivors, and is a more stable measure. The RMST is typically calculated over a specific follow-up period; thus, an RMST of 20 months over a 24-month follow-up period would suggest that if an individual is followed for 24 months, they would survive an average of 20 months. The sponsor also calculated piecewise HRs and weighted HRs. In the former, HRs were calculated before and after 3 months, and in the latter, Fleming-Harrington weights as described by León et al. were derived and applied to the Cox model.²⁴ In this analysis, 3 additional HRs were calculated with heavier weights on the early, late, or middle time points. Again, patients were censored at the last disease assessment if there was no progression, death, or use of other anticancer treatment. If any new anticancer treatment was initiated, patients were censored at the last disease assessment before treatment initiation. A sensitivity analysis was conducted to assess the impact of this phenomenon on PFS, and crossovers or the addition of any anticancer medications were treated as death or progression events. Additionally, a second sensitivity analysis defined PFS as the time from randomization to progression on the next line of therapy or death from any cause, thus providing no consideration for crossovers or subsequent use of any anticancer medications.

To adjust for patients crossing over or receiving subsequent anti-PD1/PDL1 (but not other anticancer) therapies in the OS analysis, the inverse probability of censoring weighting, rank preserving structural failure time, and 2-stage analysis methods were used. In brief, the inverse probability of censoring weighting method adjusts for crossovers by differentially weighting all patients in the control arm. Those who cross over receive a lower weight, while those who do not cross over but have characteristics that suggest they are likely to cross over receive a higher weight. However, these weightings rely on a comprehensive collection and analysis of all variables correlated with crossing over. The rank preserving structural failure time method assumes all patients would receive equal benefit from identical interventions. This method compares time on and off treatment to estimate survival times without treatment and a treatment effect adjusted for crossovers. Finally, 2-stage adjustment assumes crossovers only occur following disease progression and uses this point to establish a “secondary” baseline. Within the control group, the treatment effect is estimated between those who do and do not cross over and adjusts for “secondary” baseline characteristics. The incremental treatment effect between these groups is then used to discount the treatment effect observed in those who cross over; these effects are then compared with those of the experimental group to estimate the treatment effect adjusted for crossovers. Additional information on these methods can be found in Latimer et al.²⁵

Overall response rate was analyzed using the Miettinen and Nurminen method, and duration of response and time to response were analyzed using the Kaplan–Meier method. A partial or complete response or progression was defined according to the RECIST 1.1 guidelines. Health-related quality of life measures were analyzed using mixed-effects models with an assumption that missing data were random. No statistical testing or models were used to analyze harms data.

Power calculations in the PFS Cox model required 209 events within 24 months for 98% power to detect a 0.55 HR and assumed a 30-month enrolment period and 13 months of follow-up. In the OS analysis, 190 events were required for 85% power to detect an HR of 0.62 and assumed a 30-month enrolment period and 33.5 months of follow-up. The ORR analysis assumed a 50% response rate in the SOC arm would result in 92% power to detect an effect size of 19%.

The study had originally planned for 2 interim analyses and a final analysis of PFS and OS. The first interim analysis was performed after approximately 162 PFS events had occurred and 6 months after the last patient was randomized. The statistical plan stipulated that the second interim analysis should be performed after approximately 209 PFS events had occurred or 24 months after the last patient was randomized, whichever occurred first. This report constitutes the final analysis for PFS, but the second interim analysis for OS. At the interim analysis, the O’Brien (Lan-DeMets) alpha-spending function was used to construct group sequential boundaries to control the type I error rate across both primary end points and interim analyses at a 1-sided alpha level of 2.5%. By using the graphical approach of Maurer and Bretz, if 1 primary hypothesis is rejected, the alpha will be shifted to the other hypotheses. Initially, 1.25% was allocated each to the PFS and OS analysis and 0% to the ORR analysis. The statistical analysis plan stipulated that should the PFS analysis null hypothesis fail to be rejected, the ORR analysis would then be tested at 0.0125% and the OS analysis at 0.249%. An identical method was used should the OS analysis fail to reject the null hypothesis.

Analysis Populations

Efficacy data were analyzed using an ITT approach, while only those who received at least 1 dose of pembrolizumab or SOC were included in the safety analysis.

Results

Patient Disposition

KEYNOTE-177 screened 852 patients, of whom 307 met the pre-specified eligibility criteria. These 307 were randomized in a 1:1 ratio (153 and 154 patients to the pembrolizumab and SOC arms, respectively) and were analyzed in the efficacy analyses. All patients in the pembrolizumab arm received at least 1 dose of treatment and were analyzed in the safety analyses, while 143 patients (92.8%) in the SOC arm received at least 1 dose of treatment and were analyzed in the safety analyses. Fifty-eight pembrolizumab patients (37.9%) and 75 SOC patients (48.7%) discontinued the trial. An additional 44 patients (28.8%) and 44 patients (28.6%) in the pembrolizumab and SOC arms, respectively, received other anticancer therapy. No patients from the pembrolizumab arm crossed over to the SOC arm, but 56 SOC patients (36.4%) crossed over to the pembrolizumab arm. All efficacy end points were analyzed with an ITT approach, while only those receiving the medication were assessed for safety outcomes.

Exposure to Study Treatments

Median exposure to treatment was higher in the pembrolizumab arm (11.1 months) than in the SOC arm (5.7 months). Median follow-up was similar between the pembrolizumab arm (28.4 months) and SOC arm (27.2 months). No patient in the pembrolizumab arm crossed over to the SOC arm, but 56 patients (36.3%) in the SOC arm crossed over to receive pembrolizumab. Similarly, about 28% in each arm received subsequent anticancer therapies. The class of medication used subsequently differed in each arm, with use of anti-PD1/PDL1 higher in the SOC arm and use of chemotherapy, epidermal growth factor receptor inhibitors, folic acid derivatives, and vascular endothelial growth factor inhibitors higher in the pembrolizumab arm.

Efficacy

Only those efficacy outcomes identified in the review protocol are reported below. See Appendix 3 for detailed efficacy data.

Progression-Free Survival

In the pembrolizumab and SOC arms, 82 patients (53.6%) and 113 patients (73.4%) experienced an event, respectively. Median survival was higher in the pembrolizumab arm (16.5 months; 95% CI, 5.4 to 32.4) than in the SOC arm (8.2 months; 95% CI, 6.1 to 10.2), yielding a statistically significant HR of 0.60 (95% CI, 0.45 to 0.80). However, because the PHA was violated, complicating the interpretation of the HR, the RMST after 24 months of follow-up was reported. Pembrolizumab survival (13.7 months; 95% CI, 12.0 to 15.4) was higher than SOC survival (10.8 months; 95% CI, 9.4 to 12.2), indicating that patients who received pembrolizumab for 24 months would survive an average of 13.7 months compared to 10.8 months for patients receiving SOC. The RMST results are consistent with those of the median survival and HR in favour of pembrolizumab. The mean difference in RMST was 2.9 months (95% CI, 0.7 to 5.1). Progression-free survival at specific time points is also displayed in Table 11. To further investigate this issue, piecewise and weighted HRs are presented in Appendix 3. Appendix 3 also outlines 2 sensitivity analyses to examine of the effect of crossovers and the use of subsequent anticancer medications on PFS.

Figure 2: Kaplan-Meier Estimates of Progression-Free Survival in KEYNOTE-177

SOC = standard of care.

Overall Survival

Fewer deaths occurred in the pembrolizumab arm (36.6%) than in the SOC arm (44.8%). No median survival was reported for the pembrolizumab arm, but median survival was 34.8 months (95% CI, 26.3 to not reached) in the SOC arm. The HR was 0.77 (95% CI, 0.54 to 1.09; P = 0.0694). Survival rates at specific time points are displayed in Table 12. To adjust for the impact of crossovers and the use of subsequent anti-PD1/PDL1 therapies, the sponsor also conducted 3 sensitivity analyses, which are available in Appendix 3.

Figure 3: Kaplan-Meier Estimates of Overall Survival in KEYNOTE-177

SOC = standard of care.

Overall Response Rate

The ORR was higher in the pembrolizumab arm (43.8%; 95% CI, 35.8% to 52.0%) than the SOC arm (33.1%; 95% CI, 25.8% to 41.1%). After adjustment for multiplicity, the between-group difference of 10.7% (95% CI, –0.2% to 21.3%) was not statistically significant. Full results are displayed in Table 13.

Duration of Response

Duration of response was evaluated in KEYNOTE-177 as an exploratory outcome without adjustment for multiplicity, but no median duration of response could be calculated for the pembrolizumab arm. The median duration of response in the SOC arm was 10.6 months. The proportion of patients with an extended duration of response at discrete time points is displayed in Table 14.

Time to Response

The exploratory outcome mean time to response was similar between the pembrolizumab arm (4.0 months; standard deviation [SD] = 3.7) and SOC arm (3.6 months; SD = 4.1). Similarly, the median times to response were similar between the pembrolizumab arm (2.2 months; range, 1.8 to 18.8) and SOC arm (2.1 months; range, 1.7 to 24.9). Full results are presented in Appendix 3.

Health-Related Quality of Life

Health-related quality of life was evaluated in KEYNOTE-177 as an exploratory outcome without adjustment for multiplicity. Based on the EORTC QLQ-C30, HRQoL improved between 18 weeks and baseline in the pembrolizumab arm (3.33; 95% CI, –0.05 to 6.72) and decreased in the SOC arm (–5.63; 95% CI, –9.32 to –1.94). The least squares mean difference between these scores was 8.96 (95% CI, 4.24 to 13.69). Results for the EORTC QLQ-CR29 symptom scale were only provided as figures. Of note, those in the SOC arm saw more improvements in buttock pain, hair loss, and taste relative to those in the pembrolizumab arm, while bloating improved more in the pembrolizumab arm. Health-related quality of life as measured by the EQ VAS determined that HRQoL also improved in the pembrolizumab arm between 18 weeks and baseline (4.50; 95% CI, 1.16 to 7.83) and decreased in the SOC arm (–2.88; 95% CI, –6.46 to 0.69). The least squares mean difference between these scores was 7.38 (95% CI, 2.82 to 11.93). The EQ-5D-3L utility score increased in the pembrolizumab arm (0.04; 95% CI, 0.00 to 0.08) and decreased in the SOC arm (–0.01; 95% CI, –0.05 to 0.02), with a least squares mean difference of 0.05 (95% CI, 0.00 to 0.10).

Figure 4: Least Squares Mean Change in EORTC QLQ-CR29 Symptom Scale From Baseline to Week 18

EORTC = European Organisation for Research and Treatment of Cancer; LS = least squares; QLQ-CR29 = Quality of Life Questionnaire Colorectal Cancer Module; SOC = standard of care.

Figure 5: Least Squares Mean Change in EORTC QLQ-CR29 Functional Scale From Baseline to Week 18

EORTC = European Organisation for Research and Treatment of Cancer; LS = least squares; QLQ-CR29 = Quality of Life Questionnaire Colorectal Cancer Module; SOC = standard of care.

Harms

Only those harms identified in the review protocol are reported below. See Table 16, Table 17, and Table 18 for detailed harms data.

Adverse Events

The frequency of any AE was similar in both arms of KEYNOTE-177. The most common AEs in the pembrolizumab arm were diarrhea, fatigue, nausea, abdominal pain, decreased appetite, and vomiting. The most common AEs in the SOC arm were diarrhea, fatigue, nausea, decreased appetite, vomiting, constipation, stomatitis, abdominal pain, anemia, decreased neutrophil count, peripheral sensory neuropathy, asthenia, neutropenia, and alopecia.

Serious Adverse Events

The rate of SAEs was lower in the pembrolizumab arm (40.5%) than in the SOC arm (52.4%). The most common SAEs in the pembrolizumab arm were abdominal pain, diarrhea, pyrexia, acute kidney injury, colitis, and pneumonia. The most common SAEs in the SOC arm were diarrhea, febrile neutropenia, small intestinal obstruction, and pulmonary embolism.

Withdrawals Due to Adverse Events

Withdrawals due to AEs were similar between the pembrolizumab (13.7%) and SOC (11.9%) arms.

Mortality

The proportion of individuals with an AE resulting in death was lower in the pembrolizumab arm (3.9%) than in the SOC arm (4.9%). Deaths attributed to an AE were similar between the arms (0% versus 0.7%).

Immune-Mediated Adverse Events

The frequency of immune-mediated AEs is summarized in Table 18. Such AEs were higher in the pembrolizumab arm (30.7%) than in the SOC arm (12.6%). The most common immune-mediated AEs in the pembrolizumab arm were hypothyroidism, colitis, hyperthyroidism, and pneumonitis. However, the rate of infusion reactions was higher in the SOC arm.

Critical Appraisal

Internal Validity

Patients in KEYNOTE-177 were randomized 1:1 centrally using an interactive voice response system and integrated web response system. The baseline patient characteristics were generally balanced. A steeper drop in survival among pembrolizumab patients was observed early in the trial. The clinical experts suggested that pembrolizumab’s onset of action may be slower than SOC, leading to this phenomenon; however, the time to response was equal in each arm. The experts suggested, alternatively, that there may be a subgroup of patients, yet to be identified, who may not benefit from pembrolizumab. While the trial was open-label, clinical outcomes such as PFS, ORR, and duration of response were assessed by blinded central independent reviewers per RECIST 1.1 and are unlikely subject to bias.

Patients in the SOC arm were permitted to cross over to the pembrolizumab arm, and patients in either arm were eligible to receive subsequent anticancer medication, which confounds the study’s ability to ascertain the size of the effect between pembrolizumab and SOC. Fifty-six patients (36.4%) originally randomized to SOC crossed over to the pembrolizumab arm. No patients randomized to the pembrolizumab arm crossed over to the SOC arm. An additional 44 patients (28.8%) and 44 patients (28.6%) in the pembrolizumab and SOC arms, respectively, subsequently received other anticancer therapy. Notably, use of anti-PD1/PDL1 therapies was higher among patients in the SOC arm, and use of chemotherapy, epidermal growth factor inhibitors, folic acid derivatives, and vascular endothelial growth factor inhibitors was higher among patients in the pembrolizumab arm. No additional information is provided regarding the criteria that triggered the use of subsequent anticancer medications. Also, discontinuation from the trial and treatment were higher in the SOC arm.

Crossing over or receiving subsequent anticancer therapy was considered a censoring event in the primary PFS analysis; however, crossing over or receiving subsequent anticancer medications may not be independent of pembrolizumab or SOC use, an assumption of this analysis. Thus, to supplement the ITT method in the primary analysis, PFS sensitivity analysis 1, presented in Appendix 3, was conducted; this analysis considered subsequent use of another anticancer medication or a crossover as a death or progression event instead of a censoring event. This analysis yielded an HR of 0.61 (95% CI, 0.48 to 0.79) and similar median PFS between the pembrolizumab arm (7.0 months; 95% CI, 4.2 to 12.4) and SOC arm (6.3 months; 95% CI, 4.6 to 8.2), but these are likely conservative estimates of the true median survival. As a corollary, PFS sensitivity analysis 2 was conducted. This analysis treated death or disease progression on the subsequent line of therapy as an event and, thus, did not account for crossovers or subsequent use of new anticancer medications. This analysis did not report a median survival for pembrolizumab patients but reported a similar treatment effect (HR = 0.63; 95% CI, 0.45 to 0.88) as the primary PFS analysis (HR = 0.60; 95% CI, 0.45 to 0.80). Thus, the similar HRs in each analysis suggest that the benefit observed in the primary PFS analysis is robust to crossovers and subsequent use of other anticancer medications.

In the primary OS analysis, crossing over to pembrolizumab or subsequent use of another anticancer therapy was not treated as a censoring event. A sensitivity analysis adjusting for those who were randomized to SOC and later crossed over to receive pembrolizumab or subsequently received another anti-PD1/PDL treatment (but not other treatments) yielded statistically non-significant results whether using the inverse probability of censoring weighting method (HR = 0.54; 95% CI, 0.27 to 1.39), the rank preserving structural failure time method (HR = 0.72; 95% CI, 0.47 to 1.11), or 2-stage analysis (HR = 0.89; 95% CI, 0.76 to 1.04). These point estimates vary from the primary analysis (HR = 0.77; 95% CI, 0.54 to 1.09), suggesting that crossover and subsequent use of another anti-PD1/PDL treatment introduced some degree of bias. Moreover, each sensitivity analysis possesses wide CIs, suggesting further uncertainty regarding the true effect size and significance testing was not applied. However, in each analysis, the point estimates favour pembrolizumab, and while the exact effect size and statistical significance is uncertain, the direction of the benefit is likely in favour of pembrolizumab.

The primary and secondary outcomes were analyzed using an ITT analysis. Appropriate statistical tests were used; however, the PHA was not met in the PFS analysis. To address the issue, the sponsor performed the RMST analysis at 24 months, which suggests a benefit in the use of pembrolizumab over SOC (13.7 months; 95% CI, 12.0 to 15.4 versus 10.8 months; 95% CI, 9.4 to 12.2); this difference (2.9 months; 95% CI, 0.7 to 5.1) was deemed clinically significant by the clinical experts. Further sensitivity analyses to address this issue using differing methods to create weighted HRs yielded HRs ranging from 0.32 (95% CI, 0.19 to 0.54) to 0.81 (95% CI, 0.57 to 1.16). Piecemeal HRs were also reported, with an estimate of 1.88 before 3 months and 0.22 beyond 3 months; unfortunately, 95% CIs were not reported in this analysis. These analyses are available in Appendix 3 and, in general, suggest a benefit in the use of pembrolizumab on PFS. Finally, it is unclear if the PHA was met in the OS analysis.

The data presented constitute a final analysis of PFS but only an interim analysis for OS. The trial for the final analysis of OS was intended to be conducted when 190 deaths occurred; however, only 125 deaths are presented, further complicating the interpretation of OS. The independent data-monitoring committee reviewed the issue and recommended the final analysis still be conducted at 190 events, or 12 months from this second interim analysis, which should be reached in February or March 2021.

External Validity

The chosen SOC regimens generally align with the recommended guidelines in Canada and contain the most commonly used regimens, but some regimens were not studied in KEYNOTE-177, such as FOLFOX with panitumumab, FOLFIRI with panitumumab, FOLFOXIRI with or without bevacizumab, CAPOX with or without bevacizumab, capecitabine with or without bevacizumab, 5-fluorouracil and leucovorin with or without bevacizumab, and single-agent irinotecan. Cetuximab is not funded in Canada in the first-line setting, and panitumumab is a reasonable substitution for those with RAS wild-type disease and distal primary tumours, which compose about 40% of this patient population. Regardless, the clinical experts consulted suggest the KEYNOTE-177 results are sufficiently compelling to conclude that pembrolizumab is likely beneficial over alternative comparators not studied. No unreasonable co-interventions were identified. Study participants were representative of the population in question, and the inclusion and exclusion criteria likely did not intentionally exclude any populations of interest, except for those with an ECOG PS of 2 or higher who may have a more severe disease. The outcomes analyzed were of interest to patients and clinicians and were studied for an appropriate duration. As discussed previously, some patients crossed over to the opposite arm or received subsequent anticancer medications. Sensitivity analyses were provided to investigate the confounding effect, but clinicians should be aware that residual confounding may remain that could limit generalizability to patients who are not subject to similar crossover conditions or are unlikely to receive similar anticancer medications, if any. The clinical experts consulted also suggest that a longer follow-up and more information regarding the number of patients who required re-treatment would be helpful. Table 19 summarizes the generalizability of the evidence.

Indirect Evidence

Objectives and Methods for the Summary of Indirect Evidence

A focused literature search for NMAs dealing with colorectal cancer was run in MEDLINE All (1946–) on January 8, 2021. No limits were applied to the search. Forty-eight citations were identified, but no indirect evidence met the inclusion criteria of the CADTH review protocol, as none included pembrolizumab as a treatment of interest.

The objective of this section is to summarize and critically appraise the methods and findings of the sponsor-submitted review of the literature and NMA comparing pembrolizumab with comparators relevant for the treatment of MSI-H/dMMR mCRC.¹⁴

Description of Indirect Comparisons

The sponsor-submitted ITC first conducted a systematic literature review (SLR) to evaluate the comparative efficacy and safety of pembrolizumab versus competing interventions for the treatment of patients with MSI-H/dMMR mCRC, following which it was of interest to conduct an NMA. The feasibility of performing an NMA regarding OS, PFS, ORR, and safety outcomes for pembrolizumab, CAPOX, panitumumab plus FOLFOX, and SOC (which consisted of FOLFOX or FOLFIRI with or without cetuximab, as administered in KEYNOTE-177) was assessed, and results were synthesized by means of Bayesian NMA.¹⁴

Table 20 summarizes the study selection criteria and method for the SLR in the sponsor-submitted ITC and NMA. From the 13 studies identified, studies were subselected for inclusion in the feasibility assessment based on network geometry constraints and availability of outcome data. The selection criteria specific to the ITC and NMA were not provided.

Methods of Sponsor-Submitted NMA

Objectives

The objective of the sponsor-submitted report was to conduct a feasibility assessment via an SLR and, if possible, conduct an NMA evaluating the relative efficacy and safety of pembrolizumab versus competing interventions relevant to the UK perspective for the treatment of MSI-H/dMMR mCRC patients.¹⁴

Study Selection Methods

An SLR was conducted to identify relevant clinical evidence on pembrolizumab and competing interventions. Study selection was conducted in accordance with the predefined inclusion and exclusion criteria and summarized in the population, intervention, comparator, outcome, and study (PICOS) framework in Table 21. Eligible studies included randomized and nonrandomized clinical trials in previously untreated adult patients with unresectable or metastatic MSI-H or dMMR CRC, as well as the general mCRC population. Treatments of interest included FOLFOX or mFOLFOX6 (plus cetuximab or panitumumab), FOLFIRI (plus cetuximab or panitumumab), and capecitabine plus oxaliplatin. For the SLR and NMA, patients taking bevacizumab combinations before randomization were excluded, as bevacizumab is not a relevant comparator in the UK. Key outcomes of interest for the SLR included efficacy (OS, PFS, duration of response, time to second objective disease progression, ORR, and surgical conversion rate), safety (AEs, DAEs, SAEs), and HRQoL (patient-reported outcomes, such EQ-5D-3L, EORTC QLQ-C30, or EORTC QLQ-CR29). Outcomes assessed in the NMA included OS, PFS, ORR, and safety (DAEs, SAEs, and grade ≥ 3 and serious treatment-related AEs).¹⁴

Citations were identified through searches of Embase, MEDLINE, and CENTRAL databases. From the studies that met the inclusion criteria in the SLR, studies were reselected for inclusion in the NMA based on network geometry constraints and availability of outcome data.¹⁴ Information regarding key steps of study selection for screening and data extraction and for quality assessment, including how many reviewers were involved and how differences of opinion were settled, was not provided for the SLR or the NMA. No list of studies excluded at the full-text stage was provided, and therefore it was not possible to assess whether potentially eligible studies may have been excluded.

Feasibility Assessment

An assessment was conducted to determine the feasibility and appropriateness of performing an ITC and NMA with the eligible trials, which included (1) determination of whether the RCT evidence for the interventions of interest form 1 evidence network for each population and outcome of interest and (2) assessment of the distribution of treatment, outcomes, studies, and patient characteristics that may affect treatment effects across direct comparisons of the evidence networks.¹⁴

ITC Analysis Methods

The NMA methods are briefly summarized in Table 22. The analyses were conducted within a Bayesian framework. The outcomes included in the NMA included OS, PFS, ORR, and safety. Both fixed-effects and random-effects models were considered for the analysis of the OS, PFS, ORR, and safety outcomes. Random-effects models were fitted within the OS outcome for all networks, as connections were described by more than 1 trial; however, this was not the case for all other outcomes. For the remaining outcomes, because almost all connections in the network were only described by a single trial, stable estimates of between-study heterogeneity could not be obtained, and results of random-effects analyses were not meaningful. Therefore, only the results of fixed-effects analyses are shown for PFS, ORR, and safety outcomes. For all outcomes analyzed, no closed loops were present in the evidence base, which prevented the evaluation of consistency.¹⁴

For binary outcomes, ORR, DAEs, AEs of grade 3 or higher, treatment-related AEs of grade 3 or higher and serious treatment-related AEs, the NMA was performed (based on the proportion of patients experiencing the event of interest) using a regression model with a binomial likelihood and logit link. The NMA of reported HRs in terms of OS and PFS, assuming proportional hazards between treatments, was performed using a regression model with a contrast-based normal likelihood for the log HR of each trial in the network. Additionally, a time-varying HR NMA in terms of OS and PFS was carried out where feasible. Normal non-informative prior distributions for the parameters were used with a mean of 0 and a variance of 10,000. Relative treatment effects were expressed as odds ratios (ORs). No sensitivity or subgroup analyses were conducted to assess the impact of the priors.¹⁴

The model introduced by Jansen et al., was used for the NMA of OS and PFS.^26,27 Multiple survival distributions for OS and PFS were considered and included the following: Weibull, Gompertz, and second-order fractional polynomials (FPs), including p₁ = 0 or 1 and p₂ = –0 or 1. For the relative treatment effects in the second-order FP framework, we assessed models that assumed (1) treatment only has an impact on 2 of the 3 parameters describing the hazard function over time (i.e., 1 scale and 1 shape parameter), and (2) treatment has an impact on all 3 parameters describing the hazard function over time (i.e., 1 scale and 2 shape parameters). A test of proportionality suggested there were violations of the PHA for several trials (TREE-1, KEYNOTE-177).¹⁴

The deviance information criterion (DIC) was used to compare the goodness-of-fit of competing survival models and to inform model selection. A difference in DIC between models of approximately 5 points was considered clinically meaningful. The plausibility of the obtained HR was also considered. The parameters of the different models were estimated using a Markov chain Monte Carlo method implemented in the OpenBUGS software package. A first series of iterations from the OpenBUGS sampler were discarded as “burn-in,” and the inferences were based on additional iterations using 2 chains.¹⁴

The results of the NMA were presented with estimates for treatment effects of each intervention relative to the reference treatment. The posterior distributions of estimated relative treatment effects are summarized by the median and 95% CrIs, which were constructed from the 2.5th and 97.5th percentiles of the posterior distributions. The results of the NMA are presented in terms of cross-tables with relative treatment effect estimates between all interventions of interest, along with 95% CrIs for all outcomes, except for time-to-event outcomes based on Kaplan–Meier outcomes. For time-to-event outcomes based on reported HRs (PFS and OS), the results are also presented as HRs. The HR with the shortest follow-up was used as reported in primary publications, where possible. For time-to-event outcomes based on Kaplan-Meier curves, results are presented in terms of HRs up to 40 months. In cases where no HR was reported but the trial reported Kaplan-Meier curves, HRs were derived using the Guyot algorithm.²⁸

Assumptions

Comparisons of study design, inclusion and exclusion criteria, and patient characteristics were used to assess transitivity, homogeneity, and consistency; however, no information on this process was provided. The authors noted the following assumptions to construct the network¹⁴:

1. Networks constructed with mFOLFOX or FOLFIRI with or without cetuximab regimens being considered a single SOC node. The authors noted that 6 trials comparing mFOLFOX or FOLFIRI to mFOLFOX or FOLFIRI with or without cetuximab — treatments with no other UK-relevant comparator — were excluded from the analysis as they did not contribute any indirect evidence to the desired comparisons.^29-34 As these regimens were considered a single SOC node, the studies were considered single-arm studies in the context of the network. One additional trial was excluded as it was single arm,³⁵ and 1 was excluded as it only included 4 MSI-H/dMMR patients in a single trial arm.³⁶

2. Applicability of treatment effect estimates in mCRC patients to MSI-H/dMMR patients. To construct a network and make requested comparisons to SOC, the authors assumed that relative treatment effects do not differ between MSI-H/dMMR patients and mCRC patients.

3. Equivalence of FOLFOX to FOLFIRI and equivalence of different FOLFOX regimens. The authors noted that several studies have shown the similarity between FOLFOX and FOLFIRI as well as different FOLFOX regimens.

4. Lack of effect of adding cetuximab to FOLFOX or FOLFIRI. Due to data availability, the authors noted that results for KEYNOTE-177 were not available for patients just receiving FOLFOX or FOLFIRI; these results were combined with the patients receiving FOLFOX or FOLFIRI with or without cetuximab to form the SOC arm. In 1 study, there was evidence of advantage in PFS in unselected and KRAS wild-type patients receiving cetuximab.³⁰ In the other study, there was no difference in response rate or PFS between patients receiving cetuximab plus FOLFOX4 versus FOLFOX4 alone, but an advantage in KRAS wild-type patients.³² Therefore, caution should be taken in interpreting results that include cetuximab in the SOC arm, particularly for KRAS wild-type patients. Results for OS may be less biased than those involving response or PFS.

5. Lack of effect modification in KRAS patients for interventions other than panitumumab. Authors noted that a systematic review found that while KRAS was an effect modifier for patients receiving anti–epidermal growth factor receptor treatment, there was no evidence of effect modification on chemotherapy regimens. Within KEYNOTE-177, KRAS wild-type patients on pembrolizumab had significantly improved PFS versus SOC (HR = 0.44; 95% CI, 0.29 to 0.67), while KRAS mutant patients on pembrolizumab showed evidence of worse PFS than those on SOC (HR = 1.19; 95% CI, 0.68 to 2.07). This indicates that conclusions concerning KRAS patients should be regarded with caution.

Results of Sponsor-Submitted NMA

Summary of Included Studies

Thirty-one publications from 13 unique studies were identified from the SLR and evaluated in a feasibility assessment to determine the possibility of conducting an NMA in the study population. No list of studies excluded at the full-text stage was provided, and therefore it was not possible to assess whether potentially eligible studies may have been excluded. Five studies were subselected for inclusion in the NMA based on network geometry and availability of outcome data: KEYNOTE-177, PRIME, NO16966, Porschen 2007, and TREE-1.^8,37-40 Key study characteristics and inclusion criteria for these studies are summarized in Table 23.

All studies were phase II or III, open-label RCTs, except for the NO16966 study, which was double blind; masking was not reported for the Porschen 2007 study. Three studies were multinational, while the Porschen 2007 study was conducted in Germany and Austria, and the TREE-1 study was conducted in the US. All trials enrolled adult patients (≥ 18 years). Porschen 2007 and PRIME allowed patients with an ECOG PS of 0 to 2; the remaining trials allowed only patients with an ECOG PS of 0 or 1. No trials allowed prior chemotherapy. The median duration of follow-up differed across trials, ranging from 15 months in TREE-1 to 28.4 months in KEYNOTE-177; it was not reported in the NO16966 study. Additionally, KEYNOTE-177 only enrolled patients who had MSI-H or dMMR mCRC, while PRIME only enrolled patients with KRAS wild-type mCRC. The 3 remaining studies enrolled mCRC patients without selecting for any specified biomarkers, and so the assumption of homogeneity may not hold, given that the treatment effect may vary across different mCRC patients, particularly between subgroups of patients with MSI-H or dMMR versus those with general mCRC. Moreover, the included studies spanned more than 18 years during which the treatment paradigms evolved rapidly in the management of cancer patients.

Some key differences between the KEYNOTE-177 trial and other included studies had the potential to bias the results. The KEYNOTE-177 study was the only included study that allowed crossover from the SOC arm to receive pembrolizumab after confirmed disease progression. The authors noted that this may introduce bias to the NMA and therefore incorporated 3 statistical methods to adjust for crossover in KEYNOTE-177. In the first, a 2-stage parametric survival model that accounted for important covariates was developed to estimate the effect of switching in the control arm. The adjusted survival times were analyzed with a proportional hazards model in the second stage. The second method was the inverse probability of censoring weighting model, which addressed treatment switching by introducing artificial censoring at the time of switch for each patient and by weighting the observations before switching. Observations after switching treatment were ignored (weight = 0), while the times during which a patient is not at risk of switching were unweighted (weight = 1). The third statistical method was the rank preserving structural failure time method, which estimates the survival time gained or lost by receiving active treatment. The model assumes that active therapy is acting on longevity by multiplying survival by a given factor once a patient starts receiving the treatment. In the case of KEYNOTE-177, since only patients in the SOC arm crossed over to active treatment, the model would assume that the crossover patients would present a similar effect on OS had they continued in the SOC arm, by comparing to those patients who remained on the SOC with similar baseline and disease characteristics. Other studies either did not allow crossover or did not report it.

All included trials had an SOC arm. However, there was substantial heterogeneity in the regimens of SOC across trials, as chemotherapy regimens were made up of different combinations and doses of FOLFOX or FOLFIRI, with or without cetuximab. FOLFOX was administered in all included trials. Administration of FOLFOX varied across the trials with respect to dose, cycle, and treatment duration. KEYNOTE-177 was the only trial to include FOLFIRI regimens, and the authors noted that while the inclusion of FOLFIRI is unlikely to introduce substantial bias to the analysis, the addition of cetuximab in the SOC arm would likely reduce the estimated treatment effect of pembrolizumab compared to FOLFOX or FOLFIRI alone. Doses of oxaliplatin varied from 50 mg/m² to 85 mg/m²; leucovorin doses varied from 20 mg/m² to 500 mg/m², and 5-fluorouracil doses varied from 400 mg/m² to 2,000 mg/m². The KEYNOTE-177 trial was the only included study that administered the FOLFIRI and pembrolizumab regimens. Pembrolizumab monotherapy was administered at a dose of 200 mg on day 1 of each 21-day cycle for up to 35 cycles. The CAPOX regimen was administered in 3 trials (NO16966, Porschen 2007, and TREE-1), and panitumumab was only administered in the PRIME trial at a dose of 6 mg/kg until disease progression.

Both the KEYNOTE-177 trial and the NO16966 study evaluated bevacizumab in the SOC arms at a dose of 5 mg/kg and 8 mg/kg, respectively. These treatment arms, inclusive of bevacizumab, were included for completeness but were not of interest to the base case of the UK NMA. The authors noted that the use of subgroup data causes smaller sample sizes than the use of the ITT population, which may reduce precision, and that the use of the sample sizes that do not include bevacizumab from KEYNOTE-177 would result in more uncertainty and wider CrIs for comparisons involving pembrolizumab, which limits the ability to draw statistically meaningful conclusions. Bevacizumab is included in an alternate analysis that includes the full ITT population of KEYNOTE-177.

Table 24 summarizes the reported primary outcomes and definitions for the included studies. KEYNOTE-177 reported progression measured by RECIST 1.1, while NO16966 and PRIME reported RECIST. KEYNOTE-177 used version 4.0 of Common Terminology Criteria for Adverse Events to measure AEs, while NO16966 and PRIME used version 3.0. Porschen 2007 and TREE-1 did not report a measure of disease progression or a measure for reporting of AEs.

Baseline characteristics of the included studies are summarized in Table 25. The KEYNOTE-177 study only enrolled MSI-H/dMMR mCRC patients, while the PRIME study enrolled mCRC patients with wild-type KRAS mutations. All other studies included mCRC patients, with no specific biomarkers. Overall, baseline characteristics were well balanced, as the median age of patients in studies included in the NMA and ITC ranged from 60 years to 66 years and the proportion of males ranged from 46% to 67%. Patients with an ECOG PS of 0 ranged from 49% (KEYNOTE-177) to 61% (TREE-1) and patients with an ECOG PS of 1 ranged from 39% (TREE-1) to 51% (KEYNOTE-177). Only Porschen 2007 enrolled patients with an ECOG PS of 2 (9%).

No information on the quality assessment of the studies included in the SLR was reported.

Results

The overall network for the 5 trials that were included in the ITC and NMA — based on the assumptions outlined in Table 26 — is shown in Figure 6, irrespective of outcomes or dose. In KEYNOTE-177, the 6 comparator regimens were combined into a single SOC arm to connect the network. In the primary analysis of the NMA, patients assigned to the bevacizumab combinations before randomization in KEYNOTE-177 were excluded, as bevacizumab was not a relevant comparator in the UK context. No breakdown of SOC regimens was available for KEYNOTE-177, and networks were constructed with mFOLFOX or FOLFIRI with or without cetuximab considered a single SOC node. Three trials compared CAPOX to SOC of (m)FOLFOX or FOLFIRI with or without cetuximab (NO16966, Porschen 2007, and TREE-1),^38-40 and 1 trial compared (m)FOLFOX with panitumumab to SOC of (m)FOLFOX or FOLFIRI with or without cetuximab (PRIME).³⁷

The base-case NMA consisted of all included studies and used a subgroup of patients from the KEYNOTE-177 trial who did not receive treatment with bevacizumab (termed “ITT-Bev”). The base-case ITT-Bev population was only used for NMAs of OS and PFS. The scenario analyses of the NMA included all ITT populations from each trial and were conducted for all outcomes (OS, PFS, ORR, safety). Additional scenario analyses using crossover-adjusted results were conducted for OS using both populations.

As previously mentioned, no closed loops were shown in any network, and random-effects models were only conducted within the OS outcome, as situations where network connections were described by more than 1 trial existed.

Figure 6: Network of Evidence for All Included Randomized Controlled Trials in Feasibility Assessment: All Outcomes

Cet = cetuximab; Pan = panitumumab; Pembro = pembrolizumab; SOC = standard of care; XELOX = CAPOX (capecitabine and oxaliplatin).

Source: Sponsor-submitted indirect treatment comparison and network meta-analysis.¹⁴

Overall Survival

The network for OS is the same as that presented for all included trials in the NMA, shown in Figure 6. All 5 RCTs were included in the base-case OS analysis, and 4 interventions (pembrolizumab, CAPOX, mFOLFOX plus panitumumab, and SOC) were evaluated. Network meta-analyses for the base-case analyses were deemed feasible for OS since KEYNOTE-177 reported subgroup data (ITT-Bev). Constant and time-varying HR analyses were both performed because some of the Kaplan-Meier curves crossed, which indicates that the PHA may have been violated.¹⁴

The pairwise results of the fixed-effects NMA are shown for each trial in Table 27 for both the base-case and scenario analysis. The results showed that pembrolizumab was no different from SOC or any other intervention of interest with regard to OS in the ITT-Bev or ITT populations. Of the included interventions, only panitumumab plus FOLFOX was favoured over SOC (HR = 0.83; 95% CrI, 0.70 to 0.98, for both ITT-Bev and ITT populations). This combination also showed statistical benefit over CAPOX (HR = 0.80; 95% CrI, 0.66 to 0.98, for both ITT-Bev and ITT populations).¹⁴ In the random-effects model, no treatment was favoured over another (Table 28).

Results of the time-varying analysis are summarized in Table 29, where HRs at selected time points are presented. In both the fixed- and random-effects analyses, the best-fitting model was the second-order FP model, with p₁ = 0, p₂ = 1 (fixed effects: ITT-Bev DIC = 1,861.18 and ITT DIC = 1,938.24; random effects: ITT-Bev DIC = 1,862.36 and ITT DIC = 1,939.14). Point estimates generally decreased over time for all interventions.

In the fixed-effects NMA of the ITT-Bev population, pembrolizumab alone and panitumumab plus FOLFOX showed no benefit compared to SOC at any time point. CAPOX was favoured over SOC at 4 months and all time points at 28 months and beyond.

In the ITT population, results for CAPOX versus SOC were the same as in the ITT-Bev group, with CAPOX being favoured at the 4-month and the 28-month and beyond time points. Pembrolizumab was favoured over SOC at 12 months and onward until 40 months. Panitumumab plus FOLFOX was no different than SOC at any time point.

No treatments were favoured over SOC in the random-effects analysis for the ITT-Bev or ITT groups.

Results of the different crossover adjustment methods are summarized in Table 30; the methods generally yielded similar results. For constant HR analysis, the point estimate favoured pembrolizumab compared to all other interventions; however, results were also not statistically meaningful. In all crossover-adjusted analyses, panitumumab plus FOLFOX was the only regimen favoured over SOC, as well as over CAPOX, but was no different to pembrolizumab.

For time-varying Kaplan-Meier analysis, pembrolizumab showed a much more significant result compared to the base-case scenario, which had the lowest point HR versus SOC compared to other interventions from 4 months throughout the follow-up and was favoured over SOC starting from 8 months (Table 31 and Table 32).

Progression-Free Survival

For PFS, a connected network included 3 trials (KEYNOTE-177, Porschen 2007, and PRIME; see Figure 7) evaluating 4 interventions (pembrolizumab, CAPOX, mFOLFOX plus panitumumab, and SOC). Network meta-analyses for the base-case analyses were deemed feasible for PFS since KEYNOTE-177 reported subgroup data (ITT-Bev). Visual assessment (curves are non-parallel) of Schoenfeld residuals and P values (P < 0.05) of the Grambsch and Therneau test for proportional hazards in KEYNOTE-177 indicated these interventions violated the PHA. Network connections were only described by a single trial; therefore, between-study heterogeneity could not be estimated, and the results of random-effects analyses were not meaningful. As a result, only the fixed-effects analyses are shown.

Figure 7: Network of Evidence for Progression-Free Survival — Base Case

CET = cetuximab; Pembro = pembrolizumab; Pan = panitumumab; SOC = standard of care; XELOX = CAPOX (capecitabine and oxaliplatin).

Note: Trials in red evaluate patients in the microsatellite instability-high/mismatch repair-deficient population. Trials in blue were conducted in RAS wild-type patients.

Source: Sponsor-submitted indirect treatment comparison and network meta-analysis.¹⁴

Pairwise results of the fixed-effects NMA are summarized in Table 33. In the ITT-Bev analysis, pembrolizumab was favoured over all other treatments. Panitumumab plus FOLFOX was also favoured over SOC and CAPOX. For the ITT population, both pembrolizumab alone and panitumumab plus FOLFOX were favoured over SOC and CAPOX.

Results of the time-varying analysis are summarized in Table 34. In the ITT-Bev population, the best-fitting model was the second-order FP model with p₁ = 0, p₂ = 1 (DIC = 869.35), while in the ITT group, the best-fitting model was the second-order FP model with p₁ = 0, p₂ = 0 (DIC = 996.24). In the ITT-Bev and ITT populations, pembrolizumab was favoured over SOC starting, respectively, at 4 months and 8 months and — for both populations — throughout the entire follow-up period. Panitumumab plus FOLFOX showed benefit over SOC from 4 months to 16 months, and from 4 months to 12 months in the ITT-Bev and ITT groups, respectively. The regimen of CAPOX was no different from SOC at any time point beyond 4 months in either group.

Overall Response Rate

For ORR, a connected network was available for ITT analysis, which included 4 studies (Figure 8). The population of KEYNOTE-177 for the ORR NMA was the ITT population, which included bevacizumab patients. Stable estimates of between-study heterogeneity could not be obtained, and results of random-effects analyses were not meaningful; therefore, only results of fixed-effects analyses are shown.

Figure 8: Network of Evidence for Overall Response Rate: ITT Population

Cet = cetuximab; ITT = intention to treat; Pan = panitumumab; Pembro = pembrolizumab; SOC = standard of care; XELOX = CAPOX (capecitabine and oxaliplatin).

Note: Trials in red evaluate patients in the microsatellite instability-high or mismatch repair-deficient population. Trials in blue were conducted in RAS wild-type patients.

* The ITT population of KEYNOTE-177 was used.

Source: Sponsor-submitted indirect treatment comparison and network meta-analysis.¹⁴

The results of the fixed-effects NMA are summarized in Table 35. Pembrolizumab was favoured for ORR compared to CAPOX (OR = 1.98; 95% CrI, 1.12 to 3.47), and panitumumab plus FOLFOX was favoured over SOC (OR = 1.47; 95% CrI, 1.08 to 2.00) and CAPOX (OR = 1.84; 95% CrI, 1.18 to 2.90).

Safety

The results of safety NMAs were based on the KEYNOTE-177 ITT population. The authors noted that this could bias results in favour of trials that did not include bevacizumab as it has a worse safety profile than cetuximab. Figure 9 to Figure 12 show the networks of evidence for DAEs, AEs grade of 3 or higher, treatment-related AEs of grade 3 or higher, and treatment-related SAEs, respectively.

Discontinuations Due to Adverse Events: For DAEs, a connected network including 2 trials (KEYNOTE-177 and Porschen 2007) and evaluating 3 interventions (pembrolizumab, CAPOX, and SOC) is shown in Figure 9.

Figure 9: Network of Evidence for Discontinuations Due to AEs

AE = adverse event; CET = cetuximab; Pembro = pembrolizumab; SOC = standard of care; XELOX = CAPOX (capecitabine and oxaliplatin).

Note: Trials in red evaluate patients in the microsatellite instability-high or mismatch repair-deficient population. ITT population of KEYNOTE-177 was used.

* Discontinuation due to serious AEs were used.

Source: Sponsor-submitted indirect treatment comparison and NMA.¹⁴

Discontinuations due to AEs were reported in the 2 included trials. The results of the fixed-effects NMA are summarized in Table 36. No treatments were favoured over any other with respect to discontinuations due to AEs.

Adverse Events of Grade 3 or Higher: For AEs grade 3 or higher, a connected network including 4 trials (KEYNOTE-177, PRIME, NO16966, and TREE-1) evaluating 4 interventions (SOC, pembrolizumab, CAPOX, panitumumab plus FOLFOX) is shown in Figure 10.

Figure 10: Network of Evidence for AEs of Grade 3 or Higher

AE = adverse event; Cet = cetuximab; ITT = intention to treat; Pan = panitumumab; Pembro = pembrolizumab; SOC = standard of care; XELOX = CAPOX (capecitabine and oxaliplatin).

Note: Trials in red evaluate patients in the microsatellite instability-high or mismatch repair-deficient population. Trials in blue were conducted in RAS wild-type patients.KEYNOTE-177 reported grade 3 to 5 AEs; the other trials reported grade 3 to 4 AEs.

* The ITT population of KEYNOTE-177 was used.

Source: Sponsor-submitted indirect treatment comparison and network meta-analysis.¹⁴

The KEYNOTE-177 study reported AEs of grade 3 or higher (grade 3 to 5), whereas NO16966, TREE-1, and PRIME reported grade 3 to 4 AEs. The authors noted that although grade 5 AEs are unlikely to occur in patients who did not also experience a grade 3 to 4 AE, the inclusion of grade 3 to 4 and grade 3 or higher AEs in this analysis may bias results against KEYNOTE-177. The results of the fixed-effects NMA are shown in Table 37. Pembrolizumab was favoured over all other treatments with lower odds of AEs of grade 3 or higher (SOC: OR = 0.37; 95% CrI, 0.22 to 0.61; CAPOX: OR = 0.47; 95% CrI, 0.26 to 0.83; and panitumumab plus FOLFOX: OR = 0.16; 95% CrI, 0.08 to 0.30). Additionally, CAPOX was associated with lower odds of AEs grade 3 or higher compared to SOC. Panitumumab plus FOLFOX was associated with increased odds of grade 3 or higher AEs compared to all other treatments.

Treatment-Related Adverse Events of Grade 3 or Higher: For treatment-related AEs of grade 3 or higher, a connected network including 3 trials (KEYNOTE-177, PRIME, and TREE-1) evaluating 4 interventions (SOC, pembrolizumab, CAPOX, and panitumumab plus FOLFOX) is shown in Figure 11.

Figure 11: Network of Evidence for Treatment-Related AEs of Grade 3 or Higher

AE = adverse event; Cet = cetuximab; Pan = panitumumab; Pembro = pembrolizumab; SOC = standard of care; XELOX = CAPOX (capecitabine and oxaliplatin).

Note: Trials in red evaluate patients in the microsatellite instability-high or mismatch repair-deficient population. Trials in blue were conducted in RAS wild-type patients.KEYNOTE-177 reported grade 3 to 5 AEs; the other trials reported grade 3 to 4 AEs.

* ITT population of KEYNOTE-177 was used.

Source: Sponsor-submitted indirect treatment comparison and network meta-analysis.¹⁴

The results of the fixed-effects NMA are shown in Table 38. Pembrolizumab was favoured over all other treatments (OR = 0.14; [95% CrI, 0.08 to 0.24], OR = 0.06 [95% CrI, 0.03 to 0.15], and OR = 0.05 [95% CrI, 0.03 to 0.10] versus SOC, CAPOX, and panitumumab plus FOLFOX, respectively), showing reduced odds of treatment-related AEs of grade 3 or higher. Conversely, both panitumumab plus FOLFOX and CAPOX were also associated with increased odds of treatment-related AEs of grade 3 or higher over SOC (OR = 2.68 [95% CrI, 1.87 to 3.92] and OR = 2.22 [95% CrI, 1.09 to 4.73], respectively).

Treatment-Related Serious Adverse Events: For treatment-related SAEs of grade 3 or higher, a connected network including 2 trials (KEYNOTE-177 and PRIME) evaluating 3 interventions (SOC, pembrolizumab, and panitumumab plus FOLFOX) is shown in Figure 12.

Figure 12: Network of Evidence for Treatment-Related SAEs

Cet = cetuximab; ITT = intention to treat; Pan = panitumumab; Pembro = pembrolizumab; SOC = standard of care; TRAE = treatment-related adverse event.

Note: Trials in red evaluate patients in the microsatellite instability-high or mismatch repair-deficient population. Trials in blue were conducted in RAS wild-type patients.

Source: Sponsor-submitted indirect treatment comparison and network meta-analysis.¹⁴

The results of the fixed-effects NMA for serious treatment-related SAEs are summarized in Table 39. Pembrolizumab was favoured over panitumumab plus FOLFOX (OR = 0.41; 95% CrI, 0.21 to 0.77), as well as SOC (OR = 0.48; 95% CrI, 0.27 to 0.83), with reduced odds of serious TRAEs.

Critical Appraisal of Sponsor-Submitted NMA

The key limitations of the ITC and NMA include several major clinical assumptions made and several notable differences in the characteristics of patients across studies, which impact the interpretation of the NMA results, as well as the small size and structure of the network, which prevented the use of random-effects models in most outcomes of interest and the ability to model between-study heterogeneity and to assess consistency between direct and indirect comparisons as there were no closed loops.

The ITC and NMA was based on an SLR that identified studies according to pre-specified inclusion criteria. Insufficient information was provided on the search criteria and search methodology, and therefore it is unclear if all relevant published or unpublished studies were included. The methods of study selection and data extraction were omitted, so it is unclear whether appropriate methods were followed. Moreover, no risk of bias assessment was included in the report. The key limitation of the SLR, and therefore the ITC and NMA, was that it was conducted for the UK context and only assessed interventions relevant to the UK. As such, key PICOS criteria related to Canadian practice, such as bevacizumab as a treatment option, were not included in the pre-specified PICOS framework, and publications related to these criteria were likely missed. Therefore, the relevance of the SLR and NMA to the Canadian context is unclear. A full list of excluded publications and reasons for exclusion was not provided, and therefore it was not possible to assess whether potentially relevant studies may have been excluded. The SLR search aimed at identifying UK-relevant interventions in the broader mCRC population, as opposed to that of the funding request in Canada. The populations of patients from the included publications for the SLR and NMA varied, with 1 study looking at MSI-H/dMMR mCRC, 1 evaluating KRAS mutants, and 3 evaluating the general mCRC population. Other key differences in study characteristics include the study design of the KEYNOTE-177 trial, where patients could cross over from SOC during the study. The sponsor noted that this may bias the results, and therefore statistical methods were implemented to account for crossover from SOC to pembrolizumab. Results from the crossover analyses for OS for 3 models used yielded identical results (HR = 0.83; 95% CrI, 0.7 to 0.98), indicating that the crossover adjustments, as compared to the primary unadjusted analyses, did not show significant impact on the results of OS and that both of these sets of results were statistically non-significant. However, it is unclear whether the applied adjustment models would have effectively accounted for the crossover from SOC to pembrolizumab, which highly likely biased the effect toward null, provided that patients who shifted to receive pembrolizumab would have had responded to the treatment in the subsequent short observation period. Nevertheless, given the immature results on OS in the KEYNOTE-177 trial, the sponsor-submitted NMA was significantly compromised on this outcome.

Overall, the outcomes assessed were appropriate; however, other important outcomes, such as HRQoL, were not considered, despite HRQoL being an outcome of interest in the pre-specified PICOS framework for the SLR. It is unclear whether the authors chose not to assess HRQoL or whether no information was provided in the included studies. Based on the NMA report, the outcomes assessed in the included trials appeared similar with respect to the definitions used. Assessment methods (blinded independent review or investigator assessed) were not reported for any of the studies. The authors also did not report the criteria of the trials used to determine response. Any differences in outcome assessment and the criteria used for response across included studies may be a source of heterogeneity between studies and have the potential to influence relative treatment effect estimates.

Several differences in study characteristics across the trials are noted as potential sources of heterogeneity: namely, characteristics of the populations in the studies. As previously mentioned, the populations of included studies varied from the general mCRC population to populations with specific biomarkers. One of the main clinical assumptions in the NMA was that there was no difference in effect estimates between patients with MSI-H/dMMR, KRAS wild-type (other than panitumumab), or general mCRC; however, these variations in population are expected to impact the relative treatment effects as they are potentially important treatment effect modifiers, as evidenced by accumulated literature, like many other genetic biomarkers in oncology therapies. Therefore, any results of the NMA should be interpreted with caution with respect to population. In addition to the differences in the baseline characteristics of the patients (particularly the site and stage of metastasis), ECOG PS and durations on treatments or on trial had considerable variations. The ECOG important characteristics of age and gender were relatively similar across trials and study arms; however, there appeared to be a lower proportion of males in the KEYNOTE-177 trial than in the other studies. The primary tumour site proportions varied somewhat among trials, although NO16966 and Porschen 2007 trials did not report this distribution. The Porschen 2007 study was also the only study that enrolled patients with an ECOG PS of 2, likely underestimating the effectiveness of CAPOX. Lastly, although follow-up time was reported, there were considerable differences in median follow-up of patients in the included trials, ranging from 15 months to 28.4 months, which may impact time-to-response outcomes and safety outcomes.

The authors made some clinical assumptions around equivalence of treatments, as well as assumptions around the addition and removal of treatments in the SOC regimens, which may impact the results and generalizability of the NMA. As previously mentioned, the comparator of bevacizumab was not of interest in the NMA as it is not a relevant comparator in the UK. Therefore, the authors constructed networks for the base case that consisted of mFOLFOX or FOLFIRI, with or without cetuximab, and did not include patients from KEYNOTE-177 who received bevacizumab in the SOC arm (ITT-Bev), which may bias the results in favour of pembrolizumab because a comprehensive set of relevant comparators was not evaluated. Moreover, there is uncertainty and imprecision in the results of these comparisons due to the smaller sample size for the subgroup of patients who did not receive bevacizumab, as evidenced by the wider CrIs. Although not drastically different from the overall ITT population, results of the NMA for OS and PFS using the ITT-Bev analysis should be interpreted with caution. It was assumed that there was equivalence of different FOLFOX regimens since various FOLFOX regimens were used by different trials, as well as equivalence of FOLFOX to FOLFIRI, although these assumptions are not expected to drastically affect the results of the NMA. Additionally, the authors assumed a lack of effect of adding cetuximab to SOC of FOLFOX or FOLFIRI regimens, given the lack of available data from KEYNOTE-177 on patients who did not receive cetuximab. The authors noted that conclusions based on the analysis of the ITT-Bev population assumed similarity between FOLFOX or FOLFIRI plus bevacizumab and FOLFOX or FOLFIRI plus cetuximab regimens. It was also noted that this may impact the results of the NMA, particularly in those with KRAS wild-type mCRC, as the authors noted there was evidence of advantage in terms of PFS in such patients who were receiving cetuximab. Moreover, cetuximab is not a routinely funded treatment in all Canadian provinces and, therefore, may not be considered a relevant comparator.

The authors noted that the networks were constructed based on availability of outcome data. Besides the comparison between SOC and CAPOX for OS, comparisons for almost all competing interventions were based on single trials. The available trials formed networks with no closed loops; therefore, it was not possible to validate the transitivity assumption of NMA and check for consistency of results between direct and indirect comparisons. Random-effects models were attempted and determined not to be feasible due to the small number of studies for all outcomes except OS. Consequently, as connections were only described by a single trial for PFS, ORR, and safety, results for these outcomes are based on a fixed-effects model, as random-effects analyses were not meaningful.

The authors noted that the PHA in the KEYNOTE-177 trial was violated for OS and PFS, as shown in Kaplan-Meier curves, and therefore both constant and time-varying analyses were conducted. In this case, results from the time-varying analyses are preferred; however, the authors noted that uncertainty in time-varying estimates of HRs often leads to wider CrIs, which can make it more difficult to draw statistically meaningful conclusions. In general, point estimates and CrIs of OS and PFS for interventions compared to SOC in the ITT population were not significantly different to the NMAs using constant HRs. Results tended to favour pembrolizumab in the time-varying analyses, with pembrolizumab exhibiting statistically different results on PFS compared to SOC after the first few months, while other treatments were only favoured over SOC at the beginnings of treatment, but not long-term. Interestingly, in the constant HR analysis, panitumumab plus FOLFOX was the preferred treatment compared to SOC but was not favoured over SOC in the time-varying analysis of OS and PFS. It was noted by the authors that constant HR analyses are considered to be appropriate compared to time-varying HR analyses due to the overlap of CrIs and uncertainty as a result of short follow-up times and low numbers of events; however, violations to the PHA need to be acknowledged, and therefore results for the time-varying analyses are preferred.

Safety results show that pembrolizumab was favoured over all competing interventions for AEs of grade3 or higher and treatment-related AEs of grade 3 or higher, and over panitumumab plus FOLFOX for treatment-related SAEs in the ITT population. The results of the ITC and NMA, focusing on overall AEs reported in the trials, should be interpreted with caution given the differences in study design, definitions, and reporting of AEs. Given the variable length in follow-up across trials, there is potential for biased results in the frequency of AEs, as some may occur rarely or may only become apparent long after the beginning of treatment. Therefore, results of the NMAs focusing on DAEs, AEs of grade 3 or higher, treatment-related AEs of grade 3 or higher, and treatment-related SAEs of grade3 or higher should be interpreted with caution. For a more meaningful interpretation of AEs, the authors should have focused on special AEs of interest.

Overall, the results of the ITC and NMA may not be generalizable to the Canadian context. The exclusion of bevacizumab as a comparator in the SLR and NMA may have biased the results, as relevant studies may not have been identified. Moreover, the results of the NMA include a heterogeneous patient population made up of MSI-H/dMMR mCRC, KRAS wild type mCRC, and general mCRC patients, who were not assessed as subgroups, and therefore it is uncertain how the different subgroups of patients would respond to treatment with pembrolizumab.

Summary

The sponsor-submitted NMA did not clearly demonstrate that pembrolizumab was favoured, not favoured, or similar to the comparators included in the analyses. Regardless, the significant limitations associated with the NMA would have precluded drawing definitive conclusions on the comparative efficacy and safety of pembrolizumab. Key limitations of the NMA were the exclusion of bevacizumab as a comparator, the small and sparsely populated networks, and the considerable clinical and methodological heterogeneity, which was not adequately addressed.

Discussion

Summary of Available Evidence

The systematic review protocol identified 1 study of interest, KEYNOTE-177. This study was a phase III, open-label trial consisting of 307 adults with unresectable or metastatic MSI-H/dMMR colorectal cancer randomized to receive pembrolizumab 200 mg every 3 weeks or SOC, which was defined as 1 of FOLFOX, FOLFOX plus bevacizumab, FOLFOX plus cetuximab, FOLFIRI, FOLFIRI plus bevacizumab, or FOLFIRI plus cetuximab. The trial’s co-primary outcomes were PFS and OS, and it contained a secondary outcome of ORR. In KEYNOTE-177, 50.2% of patients were female; average patient age was 61.2 years; 74.6% of patients were White, 16.3% Asian, 4.6% Black, and 6.8% Hispanic or Latino; and 51.8% had an ECOG PS of 0 and 48.2% had an ECOG PS of 1.

Interpretation of Results

Efficacy

The primary analysis of PFS indicates that pembrolizumab increases median PFS by approximately 8 months, a clinically significant amount as determined by CADTH’s clinical experts. The HR was statistically significant, but because of the violated PHA, the estimate may be biased. Instead, RMST, a commonly used alternate metric of long-term survival for such situations, suggests a 2.9-month improvement in PFS, which was also considered clinically significant by the clinical experts. Additionally, PFS was higher in the pembrolizumab arm at 9, 12, 18, and 24 months. Two sensitivity analyses investigating the impact of crossovers and the use of subsequent anticancer medications on the results were conducted. In both analyses, the HRs only changed slightly, suggesting that pembrolizumab’s benefit as observed in the primary analysis is robust. However, 1 sensitivity analysis suggests the median survival in the pembrolizumab arm (7.0 months; 95% CI, 4.2 to 12.4) and SOC arm (6.3 months; 95% CI, 4.6 to 8.2) may not differ substantially. However, in this analysis, the survival rate at 12 months was higher in the pembrolizumab arm (44.1%; 95% CI, 36.0% to 51.8%) than in the SOC arm (30.2%; 95% CI, 23.0% to 37.6%). Survival rates at other time points were not provided. Ultimately, the totality of evidence suggests pembrolizumab improves PFS to a clinically meaningful degree.

Pembrolizumab’s impact on OS is less certain. Median survival could not be reported in the pembrolizumab arm, and the HR was statistically insignificant (HR = 0.77; 95% CI, 0.54 to 1.09). However, the OS at 9, 12, 18, and 24 months was higher in the pembrolizumab arm than in the SOC arm. As crossovers and subsequent use of anti-PD1/PDL1 therapy could bias the primary OS analysis, 3 sensitivity analyses were conducted. The results from each sensitivity analysis differ from the primary analysis, suggesting some bias in the latter. However, the point estimates from the primary and sensitivity analyses indicate some benefit from pembrolizumab in OS, but due to the variation in point estimates, wide CIs and lack of significance testing, the exact magnitude of any benefit from pembrolizumab on OS is uncertain. Although PFS is a surrogate outcome, there is evidence that it is correlated with OS in mCRC. Correlation between PFS HRs and OS HRs has been estimated to be moderate to very strong (r ≥ 0.4) and may be strongest in the setting of first-line phase III trials of targeted drugs.⁴¹ Nonetheless, the lack of robust and final OS results from KEYNOTE-177 preclude drawing concrete conclusions regarding the effects of pembrolizumab on OS as a first-line treatment in adults with unresectable or metastatic MSI-H/dMMR colorectal cancer.

In the PFS and OS analyses, pembrolizumab patients experienced worse outcomes for the first 6 months to 8 months. In some circumstances, this may indicate an imbalance in prognostic factors at baseline, but the table of baseline characteristics did not identify such imbalances. Instead, the clinical experts suggest that pembrolizumab’s onset of action may be slower than that of SOC, leading to poor outcomes in the early stages of the study. Alternatively, there may be a subgroup, yet to be identified, which may not benefit from pembrolizumab.

The ORR was higher in the pembrolizumab arm and, while it was not statistically significant, it was considered clinically significant by the clinical experts. A median duration of response could not be calculated for the pembrolizumab arm, but more pembrolizumab patients experienced a response extending beyond 6, 9, 12, 18, and 24 months. Mean and median time to response was similar between both arms. Finally, while statistical significance testing adjusted for multiplicity was not applied, HRQoL was higher to a clinically significant degree in the pembrolizumab arm when measured by the EORTC global health–quality of life score or EQ-5D visual analogue or utility scores. Although the HRQoL results are only considered supportive because they were analyzed as exploratory outcomes in KEYNOTE-177, the different directions of the within-group changes from baseline in the 2 treatment groups are notable.

The totality of the evidence from KEYNOTE-177, improved PFS and HRQoL, suggests that patients with unresectable or metastatic MSI-H/dMMR colorectal cancer will have better outcomes with pembrolizumab than with the SOC regimens used in the trial.

The sponsor provided an NMA that compared pembrolizumab versus other treatments for unresectable or metastatic MSI-H/dMMR colorectal cancer. However, the geometry of the sparsely populated networks and considerable clinical and methodological heterogeneity that could not be fully accounted for made the results of the NMA difficult to interpret. It could not be concluded whether pembrolizumab was favoured, not favoured, or similar to the comparators included in the analysis of comparative efficacy.

Harms

The frequency of overall AEs was similar in both treatment groups, but patients in the pembrolizumab group reported fewer SAEs. Treatment discontinuation and deaths following an AE were similar in each arm. Compared to the sponsor’s cumulative running safety dataset, these rates did not differ. No new AEs were identified in KEYNOTE-177 beyond the ones typically observed with pembrolizumab. Unsurprisingly, immune-mediated AEs, which were pre-specified in the review protocol as events of interest, were more frequent in the pembrolizumab arm. Relative to the sponsor’s cumulative running safety dataset for pembrolizumab, the number of such AEs were higher in KEYNOTE-177.

The sponsor-provided NMA reported that pembrolizumab was favoured over comparators with regard to grade 3 and higher AEs and grade 3 and higher treatment-related AEs. Pembrolizumab was also favoured over panitumumab plus FOLFOX for treatment-related SAEs. However, as described for the efficacy outcome analyses, the results of comparative harms may not be valid because of the significant limitations of the NMA.

Patient groups in their input to CADTH for the review of pembrolizumab identified safety as an important consideration. It was suggested that patients were willing to tolerate more SAEs for additional survival time, delayed progression, and improved quality of life. Likewise, clinicians noted that treatment maybe limited with SOC regimens because of associated serious adverse effects. Although the comparative safety data are not robust, there is a suggestion that patients may experience fewer SAEs with pembrolizumab than with SOC regimens.

Implementation Considerations

In 2016, CADTH published an Optimal Use Report titled DNA Mismatch Repair Deficiency Tumour Testing for Patients With Colorectal Cancer: Recommendation.⁴² The gold standard for detecting MSI-H/dMMR patients is through germline genetic testing; however, polymerase chain reaction to test for MSI-H status or immunohistochemistry to assess dMMR status may be considered before genetic testing. Both tests have high sensitivity and moderate-to-high specificity, though the latter provides additional information regarding the nature of the mutation, which can guide further diagnostics, is cheaper, and may be preferred by clinicians. The report recommended universal dMMR tumour testing for patients with colorectal cancer followed by reflex tumour testing for MLH1 promoter hypermethylation, which would be cost-effective if the willingness to pay ranges from $28,902 to $387,330.

Conclusions

Pembrolizumab improved PFS versus SOC in a single RCT of patients with unresectable or metastatic MSI-H/dMMR colorectal cancer. The difference between treatment groups was considered clinically meaningful based on input from the clinical experts consulted by CADTH. Although fewer patients died in the pembrolizumab group than in the SOC group, no statistically significant benefit in OS was found. The study was not adequately designed to evaluate treatment effects on OS, and estimates are uncertain due to crossovers and use of subsequent anticancer medications. The ORR, duration of response, and HRQoL were also numerically improved relative to SOC. Although benefits aside from PFS could not be supported by statistical significance in all outcomes, the efficacy results in totality suggest pembrolizumab improves outcomes over SOC in a difficult-to-treat patient population whose cancer is often unresponsive to SOC. Pembrolizumab does not appear to be associated with more AEs than the SOC regimens evaluated. Known AEs of interest such as immune-mediated AEs occurred more frequently in the pembrolizumab group and remain a concern.

References

1.Canadian Cancer Society. Colorectal cancer. What is colorectal cancer? 2021; https://www.cancer.ca/en/cancer-information/cancer-type/colorectal/colorectal-cancer/?region=on. Accessed 2021 Mar 8.

2.Macrae FA, Bendell J. Clinical presentation, diagnosis, and staging of colorectal cancer. In: Post TW, ed. UpToDate. Waltham (MA): UpToDate; 2021: www.uptodate.com. Accessed 2021 Mar 8.

3.Canadian Cancer Society. Colorectal cancer. Colorectal cancer statistics. 2021; https://www.cancer.ca/en/cancer-information/cancer-type/colorectal/statistics/?region=on. Accessed 2021 Mar 8.

4.Pestana C, Reitemeier RJ, Moertel CG, Judd ES, Dockerty MB. The natural history of carcinoma of the colon and rectum. Am J Surg. 1964;108:826-829. PubMed

5.Bengmark S, Hafström L. The natural history of primary and secondary malignant tumors of the liver. I. The prognosis for patients with hepatic metastases from colonic and rectal carcinoma by laparotomy. Cancer. 1969;23(1):198-202. PubMed

6.Rodriguez-Bigas MA, Grothey A. Overview of the management of primary colon cancer. In: Post TW, ed. UpToDate. Waltham (MA): UpToDate; 2021: www.uptodate.com.

7.Heinemann V, von Weikersthal LF, Decker T, et al. FOLFIRI plus cetuximab versus FOLFIRI plus bevacizumab as first-line treatment for patients with metastatic colorectal cancer (FIRE-3): a randomised, open-label, phase 3 trial. Lancet Oncol. 2014;15(10):1065-1075. PubMed

8.André T, Shiu KK, Kim TW, et al. Pembrolizumab in microsatellite-instability-high advanced colorectal cancer. N Engl J Med. 2020;383(23):2207-2218. PubMed

9.Venderbosch S, Nagtegaal ID, Maughan TS, et al. Mismatch repair status and BRAF mutation status in metastatic colorectal cancer patients: a pooled analysis of the CAIRO, CAIRO2, COIN, and FOCUS studies. Clin Cancer Res. 2014;20(20):5322-5330. PubMed

10.Boland CR, Goel A. Microsatellite instability in colorectal cancer. Gastroenterology. 2010;138(6):2073-2087. PubMed

11.Innocenti F, Ou FS, Qu X, et al. Mutational analysis of patients With colorectal cancer in CALGB/SWOG 80405 identifies new roles of microsatellite instability and tumor mutational burden for patient outcome. J Clin Oncol. 2019;37(14):1217-1227. PubMed

12.Tougeron D, Sueur B, Zaanan A, et al. Prognosis and chemosensitivity of deficient MMR phenotype in patients with metastatic colorectal cancer: an AGEO retrospective multicenter study. Int J Cancer. 2020;147(1):285-296. PubMed

13.Clinical study report: P177V01MK3475. A phase III study of pembrolizumab (MK-3475) vs. chemotherapy in microsatellite instability-high (MSI-H) or mismatch repair deficient (dMMR) stage IV colorectal carcinoma (KEYNOTE-177) [internal sponsor's report]. Kenilworth (NJ): Merck Sharp & Dohme; 2020 May 13.

14.Systematic literature review and network meta-analysis for treatments of untreated, unresectable or metastatic, microsatellite instability-high (MSI-H) or mismatch repair deficient (dMMR) for colorectal cancer [internal sponsor's report]. In: Drug Reimbursement Review sponsor's submission: Keytruda (pembrolizumab) powder for solution for infusion 50 mg, solution for infusion 100 mg/4 mL vial. Kirkland (QC): Merck Canada; 2020.

15.Clark JW, Grothey A. Systemic therapy for nonoperable metastatic colorectal cancer: selecting the initial therapeutic approach. In: Post TW, ed. UpToDate. Waltham (MA): UpToDate; 2021: www.uptodate.com. Accessed 2021 Feb 13.

16.Clark JW, Grothey A. Systemic chemotherapy for metastatic colorectal cancer: general principles. In: Post.T.W., ed. UpToDate. Waltham (MA): UpToDate; 2021: www.uptodate.com. Accessed 2021 Feb 13.

17.PrKeytruda® (pembrolizumab): powder for solution for infusion 50 mg, solution for infusion 100 mg/4 mL vial [product monograph]. Kirkland (QC): Merck Canada; 2020 Jul 20.

18.FDA approves pembrolizumab for first-line treatment of MSI-H/dMMR colorectal cancer. Silver Spring (MD): US Food and Drug Administration; 2020: https://www.fda.gov/drugs/drug-approvals-and-databases/fda-approves-pembrolizumab-first-line-treatment-msi-hdmmr-colorectal-cancer. Accessed 2021 Mar 8.

19.European Public Assessment Report (EPAR): Keytruda (pembrolizumab). Amsterdam (NL): European Medicines Agency; 2021: https://www.ema.europa.eu/en/medicines/human/EPAR/keytruda. Accessed 2021 Mar 8.

20.McGowan J, Sampson M, Salzwedel DM, Cogo E, Foerster V, Lefebvre C. PRESS Peer Review of Electronic Search Strategies: 2015 guideline statement. J Clin Epidemiol. 2016;75:40-46. PubMed

21.Grey matters: a practical tool for searching health-related grey literature. Ottawa (ON): CADTH; 2019: https://www.cadth.ca/grey-matters. Accessed 2020 Dec 18.

22.Musoro JZ, Sodergren SC, Coens C, et al. Minimally important differences for interpreting the EORTC QLQ-C30 in patients with advanced colorectal cancer treated with chemotherapy. Colorectal Dis. 2020;07:07.

23.Sinnott PL, Joyce VR, Barnett PG. Guidebook: preference measurement in economic analysis. Menlo Park (CA): Health Economics Research Center; 2007: https://www.herc.research.va.gov/files/BOOK_419.pdf. Accessed 2020 Oct 13.

24.León LF, Lin R, Anderson KM. On weighted log-rank combination tests and companion Cox model estimators. Stat Biosci. 2020;12:225-245.

25.Latimer NR, Henshall C, Siebert U, Bell H. Treatment switching: statistical and decision-making challenges and approaches. Int J Technol Assess Health Care. 2016;32(3):160-166. PubMed

26.Jansen JP, Cope S. Meta-regression models to address heterogeneity and inconsistency in network meta-analysis of survival outcomes. BMC Med Res Methodol. 2012;12(1):152. PubMed

27.Jansen JP. Network meta-analysis of survival data with fractional polynomials. BMC Med Res Methodol. 2011;11(1):61. PubMed

28.Guyot P, Ades AE, Ouwens MJNM, Welton NJ. Enhanced secondary analysis of survival data: reconstructing the data from published Kaplan-Meier survival curves. BMC Med Res Methodol. 2012;12(1):9. PubMed

29.Ye LC, Liu TS, Ren L, et al. Randomized controlled trial of cetuximab plus chemotherapy for patients with KRAS wild-type unresectable colorectal liver-limited metastases. J Clin Oncol. 2013;31(16):1931-1938. PubMed

30.Van Cutsem E, Köhne C-H, Hitre E, et al. Cetuximab and chemotherapy as initial treatment for metastatic colorectal cancer. N Engl J Med. 2009;360(14):1408-1417. PubMed

31.Tveit KM, Guren T, Glimelius B, et al. Phase III trial of cetuximab with continuous or intermittent fluorouracil, leucovorin, and oxaliplatin (Nordic FLOX) versus FLOX alone in first-line treatment of metastatic colorectal cancer: the NORDIC-VII study. J Clin Oncol. 2012;30(15):1755-1762. PubMed

32.Bokemeyer C, Bondarenko I, Makhson A, et al. Fluorouracil, leucovorin, and oxaliplatin with and without cetuximab in the first-line treatment of metastatic colorectal cancer. J Clin Oncol. 2009;27(5):663-671. PubMed

33.Personeni N, Rimassa L, Verusio C, et al. Prognostic factors in KRAS wild-type (wt) metastatic colorectal cancer (mCRC) patients (pts) treated with biweekly cetuximab (C) plus irinotecan, fluorouracil, and leucovorin (FOLFIRI): a phase II study [abstract]. J Clin Oncol. 2013;31(15 Suppl):e14611.

34.Qin S, Li J, Wang L, et al. Efficacy and tolerability of first-line cetuximab plus leucovorin, fluorouracil, and oxaliplatin (FOLFOX-4) versus FOLFOX-4 in patients with RAS wild-type metastatic colorectal cancer: the open-label, randomized, phase III TAILOR trial. J Clin Oncol. 2018;36(30):3031-3039. PubMed

35.des Guetz G, Mariani P, Cucherousset J, et al. Microsatellite instability and sensitivitiy to FOLFOX treatment in metastatic colorectal cancer. Anticancer Res. 2007;27(4c):2715-2719. PubMed

36.Müller CI, Schulmann K, Reinacher-Schick A, et al. Predictive and prognostic value of microsatellite instability in patients with advanced colorectal cancer treated with a fluoropyrimidine and oxaliplatin containing first-line chemotherapy. A report of the AIO Colorectal Study Group. Int J Colorectal Dis. 2008;23(11):1033-1039. PubMed

37.Douillard JY, Siena S, Cassidy J, et al. Randomized, phase III trial of panitumumab with infusional fluorouracil, leucovorin, and oxaliplatin (FOLFOX4) versus FOLFOX4 alone as first-line treatment in patients with previously untreated metastatic colorectal cancer: the PRIME study. J Clin Oncol. 2010;28(31):4697-4705. PubMed

38.Cassidy J, Clarke S, Díaz-Rubio E, et al. XELOX vs FOLFOX-4 as first-line therapy for metastatic colorectal cancer: NO16966 updated results. Br J Cancer. 2011;105(1):58-64. PubMed

39.Porschen R, Arkenau HT, Kubicka S, et al. Phase III study of capecitabine plus oxaliplatin compared with fluorouracil and leucovorin plus oxaliplatin in metastatic colorectal cancer: a final report of the AIO Colorectal Study Group. J Clin Oncol. 2007;25(27):4217-4223. PubMed

40.Hochster HS, Hart LL, Ramanathan RK, et al. Safety and efficacy of oxaliplatin and fluoropyrimidine regimens with or without bevacizumab as first-line treatment of metastatic colorectal cancer: results of the TREE Study. J Clin Oncol. 2008;26(21):3523-3529. PubMed

41.Sidhu R, Rong A, Dahlberg S. Evaluation of progression-free survival as a surrogate endpoint for survival in chemotherapy and targeted agent metastatic colorectal cancer trials. Clin Cancer Res. 2013;19(5):969-976. PubMed

42.DNA mismatch repair deficiency tumour testing for patients with colorectal cancer: recommendations. (CADTH optimal use report vol.5, no.3d). Ottawa (ON): CADTH; 2016: https://www.cadth.ca/sites/default/files/pdf/OP0522_dMMR_Report_Recommendations.pdf. Accessed 2021 Mar 8.

43.Giesinger JM, Kieffer JM, Fayers PM, et al. Replication and validation of higher order models demonstrated that a summary score for the EORTC QLQ-C30 is robust. J Clin Epidemiol. 2016;69:79-88. PubMed

44.EORTC Quality of life: FAQs. Brussels (BE): EORTC: https://qol.eortc.org/faq/. Accessed 2021 Mar 8.

45.Aaronson NK, Ahmedzai S, Bergman B, et al. The European Organization for Research and Treatment of Cancer QLQ-C30: a quality-of-life instrument for use in international clinical trials in oncology. J Natl Cancer Inst. 1993;85(5):365-376. PubMed

46.EORTC QLC-C30 scoring manual. Brussels (BE): EORTC; 2001: https://www.eortc.org/app/uploads/sites/2/2018/02/SCmanual.pdf. Accessed 2021 Mar 8.

47.Davda J, Kibet H, Achieng E, Atundo L, Komen T. Assessing the acceptability, reliability, and validity of the EORTC Quality of Life Questionnaire (QLQ-C30) in Kenyan cancer patients: a cross-sectional study. J Patient Rep Outcomes. 2021;5(1):4. PubMed

48.Luo N, Fones CS, Lim SE, Xie F, Thumboo J, Li SC. The European Organization for Research and Treatment of Cancer Quality of Life Questionnaire (EORTC QLQ-c30): validation of English version in Singapore. Qual Life Res. 2005;14(4):1181-1186. PubMed

49.Osoba D, Rodrigues G, Myles J, Zee B, Pater J. Interpreting the significance of changes in health-related quality-of-life scores. J Clin Oncol. 1998;16(1):139-144. PubMed

50.Snyder CF, Blackford AL, Sussman J, et al. Identifying changes in scores on the EORTC-QLQ-C30 representing a change in patients' supportive care needs. Qual Life Res. 2015;24(5):1207-1216. PubMed

51.Bedard G, Zeng L, Zhang L, et al. Minimal important differences in the EORTC QLQ-C30 in patients with advanced cancer. Asia Pac J Clin Oncol. 2014;10(2):109-117. PubMed

52.Wong CK, Chen J, Yu CL, Sham M, Lam CL. Systematic review recommends the European Organization for Research and Treatment of Cancer colorectal cancer-specific module for measuring quality of life in colorectal cancer patients. J Clin Epidemiol. 2015;68(3):266-278. PubMed

53.Ganesh V, Agarwal A, Popovic M, et al. Comparison of the FACT-C, EORTC QLQ-CR38, and QLQ-CR29 quality of life questionnaires for patients with colorectal cancer: a literature review. Support Care Cancer. 2016;24(8):3661-3668. PubMed

54.Gujral S, Conroy T, Fleissner C, et al. Assessing quality of life in patients with colorectal cancer: an update of the EORTC quality of life questionnaire. Eur J Cancer. 2007;43(10):1564-1573. PubMed

55.van der Hout A, Neijenhuijs KI, Jansen F, et al. Measuring health-related quality of life in colorectal cancer patients: systematic review of measurement properties of the EORTC QLQ-CR29. Support Care Cancer. 2019;27(7):2395-2412. PubMed