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 2 clinical specialists with expertise in the diagnosis and management of EC.

The clinical experts consulted by CADTH indicated that currently there are no standard, funded second-line treatment options for advanced or recurrent EC. The clinical experts agreed that patients who would most benefit from dostarlimab include those with identified dMMR or MSI-H recurrent or advanced EC. One of the clinical experts noted that dostarlimab could be used as monotherapy in first-line settings or later in the absence of effective treatments. The clinical experts noted that treatment with dostarlimab would not be suitable in patients with the following characteristics: very poor performance status; history of severe autoimmune disease; prior immunotherapy use; known uncontrolled central nervous system metastases and/or carcinomatous meningitis; poor medical risk due to a serious uncontrolled medical disorder; nonmalignant systemic disease or active infection requiring systemic therapy; or microsatellite stable (MSS) EC.

In the opinion of the clinical experts consulted by CADTH, treatment with dostarlimab should be discontinued in the case of disease progression, severe toxicity, or intolerability. The clinical experts indicated that the following outcomes would best assess response to treatment: overall survival (OS); response rate based on clinical and radiological investigation; progression-free survival (PFS); reduction of cancer burden and symptom improvement in activities of daily living; HRQoL; durability of response; and response to subsequent therapies.

In terms of clinically meaningful responses, the clinical experts recommended that in addition to clinical assessment of disease symptoms and duration of disease control, the use of standard immune-related Response Evaluation Criteria in Solid Tumours (irRECIST) for the assessment of response to immunotherapeutic treatments may be useful.

Clinician Group Input

A total of 7 clinician group inputs were submitted from the following groups: British Columbia Cancer Provincial Gynecological Oncology Tumour Group; McGill University Health Centre (MUHC), Division of Gynecologic Oncology; Ontario Health-Cancer Care Ontario (OH-CCO) Gynecological Drug Advisory Committee; Princess Margaret Cancer Centre (PMCC), Gynecologic Cancers Disease Site Group, Medical Oncology Group; Saskatchewan Cancer Agency (SCA); the Society of Gynecologic Oncology of Canada (GOC); and Sunnybrook Health Sciences Centre (SBHSC).

The views of the clinician groups were overall consistent with those of the clinical experts consulted by CADTH. The clinician groups indicated that the most important treatment goals are achieving disease control, delaying worsening of symptoms, prolonging survival, maintaining HRQoL, delaying disease progression, and an acceptable safety profile. All the clinician groups indicated that all patients with recurrent EC would benefit from an effective immunotherapy, but patients with dMMR or MSI-H subtypes would be the most likely to benefit from immune checkpoint inhibitor therapy. All groups recommended that patients diagnosed with metastatic EC should be offered platinum-based chemotherapy as first-line therapy. However, the British Columbia Cancer Provincial Gynecological Oncology Tumour Group did acknowledge that treatment with an immune checkpoint inhibitor may be an appropriate first-line therapy when chemotherapy is contraindicated or not desirable.

Description of Study

The GARNET trial (Study 4010-01-001)¹⁸ is an ongoing nonrandomized, noncomparative, multi-centre, open-label, phase I dose-escalation and cohort-expansion study in patients with recurrent or advanced solid tumours. The objective of part 2B of the GARNET trial was to evaluate the safety and antitumour activity of dostarlimab in patients with advanced solid tumours. Cohort A1 included patients with advanced or recurrent dMMR or MSI-H EC that had progressed on or after prior treatment with a platinum-containing regimen. Patients were enrolled from 123 sites in 8 countries (including 8 Canadian sites). Enrolment started on April 10, 2017, and is ongoing.

To be eligible, patients had to be at least 18 years of age, diagnosed with recurrent or advanced dMMR or MSI-H EC, and had to have progressed on or after no more than 2 lines of prior systemic therapy, with at least 1 of these being platinum-based doublet therapy. In addition, patients had to have adequate organ function and an Eastern Cooperative Oncology Group (ECOG) performance status of less than 1. The study consisted of 3 phases: the screening phase (up to 35 days before treatment), the treatment phase, and the follow-up phase. A total of 129 patients were enrolled in cohort A1. All patients received dostarlimab by IV injection (500 mg every 3 weeks for cycles 1 to 4, and 1,000 mg every 6 weeks from cycle 5 onward) for up to 2 years or until disease progression, treatment discontinuation, or withdrawal.

The co-primary outcomes of the GARNET trial were objective response rate (ORR) and duration of response (DOR). The secondary outcomes were OS, disease control rate (DCR), immune-related DCR (irDCR), PFS, immune-related PFS (irPFS), immune-related ORR (irORR), and immune-related DOR (irDOR). HRQoL was an exploratory outcome assessed by the European Organization for Research and Treatment of Cancer Quality of Life Questionnaire Core 30 (EORTC QLQ-C30) and the EQ-5D 5-Levels (EQ-5D-5L). The safety outcomes assessed included treatment-emergent adverse events (TEAEs), serious adverse events (SAEs), immune-related adverse events (irAEs), ECOG Performance Status, clinical laboratory measures, vital signs, electrocardiogram, physical examination, serum pregnancy testing, and concomitant medications.

The GARNET trial has 3 planned interim analyses that are contingent on a combined enrolment of 100, 200, and 300 patients, respectively, in cohort A1 and cohort F and 24 weeks of follow-up. Only data from the second interim analysis (IA-2) was reviewed by CADTH; it included a subset of patients from the first interim analysis (IA-1) (N = 72). Data for this subset of patients is available in Appendix 4. The data cut-off dates were July 8, 2019, March 1, 2020, and November 1, 2021, for IA-1, IA-2 (N = 105), and the third interim analysis (IA-3) (N = 143), respectively. The GARNET trial was sponsored by GlaxoSmithKline Inc.

Because only 2 (< 2%) enrolled patients had mismatch repair (MMR)-unknown but MSI-H tumours (MMR-unk/MSI-H), these patients were included with patients with dMMR tumours. Median age was 64 years (range = 39 to 80 years), median weight was 71 kg (range = 34.0 to 141.4), and median body mass index (BMI) was 27.97 kg/m² (range = 13.6 to 53.9 kg/m²). Most patients were White (> 75%). The most common histology type of EC was type I endometrioid carcinoma (67.6%), with grade 2 being the most common histology grade at diagnosis (39%). More than two-thirds (67.6%) of patients had stage IV EC. ECOG Performance Status of 1 was the most common (60%), followed by ECOG 0 (40%). All patients received prior anti-cancer treatment. Most patients (88.6%) had 2 or fewer lines of prior anti-cancer regimens, and a smaller proportion had 2 or more lines (11.5%). More than half (56.2%) of patients had received prior regimens for metastatic disease. The patient subgroups of interest, as identified in the CADTH systematic review protocol, included the following: International Federation of Gynecology and Obstetrics (FIGO) stage, histology of tumour type (e.g., type I, type II) and subtype (e.g., clear cell carcinoma), number and type of prior systemic therapies (e.g., chemotherapy, hormonal therapy), prior radiation, and progression-free interval after the most recent platinum-containing anti-cancer therapy.

Outcomes

The key outcomes from cohort A1 of the GARNET trial are summarized in Table 2. At the time of IA-2, the median duration of follow-up was 16.3 months, and the median duration of treatment (DoT) was 26 weeks.

The proportion of patients who achieved an ORR (complete response [CR] or partial response [PR]) was 44.8% (95% confidence interval [CI], 35.0 to 54.8). The Best overall response (BOR) was CR in 11 patients (10.5%), PR in 36 patients (34.3%), and stable disease in 13 patients (37.1%). The DCR was 57.1% (95% CI, 35.0 to 54.8). Of those who responded, 89.5% had an ongoing response. The median DOR was not reached, but 79% of patients who achieved an objective response had a DOR of at least 6 months.

At the time of IA-2, a large proportion of patients (66.7% and 45.7%, respectively) had no OS or PFS events. The median OS was not reached, but Kaplan-Meier (KM) estimates for the probability of survival at 6, 9, and 12 months were 80.9% (95% CI, 71.7 to 87.4), 75.1% (95% CI, 65.2 to 82.6), and 68.9% (95% CI, 58.3 to 77.4), respectively. The median PFS was 5.5 months (95% CI, 3.2 to not reached), with KM estimates of PFS by RECIST 1.1 of 48.6% (95% CI, 38.6 to 57.9) at month 6 and 47.5% (95% CI, 37.4 to 56.8) at month 9 and at month 12. In terms of HRQoL, the EQ-5D-5L visual analogue scale (EQ-VAS) and EORTC QLQ-C30 scores appeared stable over time. Summary data for the EQ-5D-5L descriptive system were not provided.

As part of the sponsor’s feedback on this CADTH reimbursement review report, the sponsor provided CADTH with a summary of the updated analysis (data cut-off November 1, 2021) for certain baseline characteristics, efficacy, and safety outcomes in the GARNET trial. The results of the updated analysis were, overall, consistent with those reported in the previous analyses performed as of March 1, 2020, data cut-off date. The additional results from the November 1, 2021, data cut-off date are available in Appendix 5.

Harms Results

A summary of the key harms reported in cohort A1 of GARNET are summarized in Table 2. Almost all patients (95.3%) experienced at least 1 TEAE. The most common serious TEAEs were abdominal pain, acute kidney injury, sepsis, pulmonary embolism, pyrexia, and urinary tract infection. Grade 3 or higher TEAEs occurred in 48.1% of patients, with the most common being anemia (14.7%), abdominal pain (5.4%), and hyponatremia (3.9%). No patient withdrew due to an adverse event (AE) as a primary reason. Study treatment discontinuation due to AEs occurred in 11.6% of patients, whereas AEs that led to study treatment interruption occurred in 24% of patients. The most common AEs leading to study interruption were anemia (3.1%) and diarrhea (2.3%).

One patient died due to a TEAE (aspiration) during the treatment period, and 4 patients died due to TEAEs (i.e., pleural effusion, pneumonia, sepsis, and shock) during the 90-day safety follow-up. None of the TEAEs leading to death were considered treatment-related, and no TEAEs were the primary cause of death during the long-term follow-up period.

The notable harms associated with dostarlimab included immune-related toxicity. The incidence of irAEs was 34.9% in cohort A1. The most frequently reported irAEs (≥ 5%) were diarrhea and hypothyroidism. A total of 7.9% of patients had a serious irAE, 12.7% had an irAE of grade 3 or higher, and 4.8% had an irAE that led to study treatment discontinuation. Most of the irAEs were considered related to the study treatment.

Table 2

Summary of Key Results From the Pivotal Study (IA-2).

Internal Validity

The main limitation of the GARNET trial is the single-arm design, which makes it challenging to interpret the data and determine whether the efficacy and safety events observed were attributable to dostarlimab. Formal hypothesis and statistical significance testing were not performed, limiting the ability to draw conclusions. Given that results were based on an interim analysis, some time-to-event outcomes, including median OS and DOR, were not reached due to data immaturity; therefore, the treatment effect observed with dostarlimab may be overestimated. The risk of overestimating HRQoL benefit and known subjective harms is also high, given the open-label trial design in which treatment was not blinded. To mitigate bias, the sponsor used a blind independent clinical review (BICR) to evaluate treatment response, with standardized criteria for certain efficacy outcomes (i.e., ORR, DOR, PFS, and DCR). Therefore, bias is less of a concern for these end points and OS, and more of a concern for subjective end points, including HRQoL and safety. It is also acknowledged that mature OS data will be confounded by the use of subsequent anti-cancer therapy received by some patients after progressive disease. No analyses were undertaken to account for the potential of confounding. Overall, the magnitude and direction of bias is unclear. The clinical experts agreed that in the absence of robust comparative data on PFS and OS, no firm conclusions could be drawn on how dostarlimab compares with other relevant treatment options, as causal inferences cannot be made from the results of a single-arm trial design.

HRQoL was identified as an important outcome by the patient and clinician groups providing input for this review. However, no conclusions could be drawn from the HRQoL data from the GARNET trial due to several limitations. Given the wide and overlapping CIs, the reduced number of patient responses over time, and the lack of statistical testing and a definition of what constituted a clinically meaningful response, it is not possible to draw conclusions with precision from the available data.

External Validity

Overall, the clinical experts consulted by CADTH agreed that the inclusion and exclusion criteria, baseline patient characteristics, concomitant medications, and prohibited medications present in cohort A1 of the GARNET trial were reflective of patients they see in clinical practice for the indication under review. There were no barriers to identifying patients who would most benefit from the treatment, given that testing for MMR and microsatellite instability (MSI) status is standard practice in Canada. The clinical experts indicated that no difference in treatment effect would be expected based on variation in disease-management practices across participating countries. In the opinion of the clinical experts, as long as patients have dMMR or MSI-H tumour status, dostarlimab would be appropriate to administer after any of the prior therapies received by patients in the trial. However, they noted that clinical benefit may be diminished in patients with more prior lines of systemic therapy. There were a limited number of patients included in the primary efficacy analysis dataset (n = 105) and very few patients from various ethnic backgrounds, which may reduce the generalizability of the results to a real-world practice setting. Furthermore, the subgroup analyses had no statistical comparisons and even smaller sample sizes, which limits generalizability to a broader population.

Description of Studies

The sponsor submitted 6 reports of indirect treatment comparisons (ITCs) — 3 reports of matching-adjusted indirect comparisons (MAICs), and 3 reports of inverse probability of treatment weighting (IPTW) analyses¹⁹ — which aimed to compare survival between dostarlimab from the phase I GARNET trial with the current treatment paradigm in advanced or recurrent EC.

Efficacy Results

The primary end point for all comparisons was OS. Other outcomes included PFS, ORR, DOR, time to treatment discontinuation (TTD), DoT, time to next treatment (TTNT), time to deterioration in HRQoL, and AEs; however, these were less frequently investigated, and outcomes specifically important to patients, including HRQoL, were not assessed. The results of the MAIC and IPTW analyses generally suggest that dostarlimab is favoured for OS over all the included comparators.

Harms Results

The sponsor-submitted MAIC and IPTW reports did not assess safety outcomes.

Critical Appraisal

Although the results of the MAIC and IPTW analyses generally suggest that dostarlimab is favoured for OS over all the included comparators, there was significant uncertainty in the results based on the clinical heterogeneity of the included populations, resulting in reduced sample sizes and wide CIs. There were important differences in the design of the comparator studies that limit the ability to draw strong conclusions about the effectiveness for dostarlimab compared with other treatments. An important limitation of all analyses was the fact that MMR and MSI-H status was unknown for all or most patients in the comparator trial, and it is therefore uncertain whether the comparator population in the ITC analyses would be eligible for treatment with dostarlimab, providing further uncertainty about the comparative effectiveness.

Unmet Needs

Currently, there is no standard second-line therapy for individuals with recurrent EC who have progressed on or after prior treatment with a platinum-containing regimen. This represents a critical unmet need in this patient population. Both clinical experts agreed that biomarker-driven treatments are needed to guide selection of the most effective and durable treatment option.

Place in Therapy

Currently, patients with advanced or recurrent EC receive carboplatin plus paclitaxel (another platinum chemotherapy doublet) or a single drug as first-line therapy. For a subset of patients with estrogen- or progesterone-receptor-positive indolent low-grade metastatic or recurrent EC, hormonal treatments, such as megestrol, medroxyprogesterone, letrozole, and tamoxifen, are considered. For those who have progressed on platinum chemotherapy and are considered to have platinum-resistant or refractory disease, second-line therapy with single-drug chemotherapy is considered, but it has a low expected response rate and a short DOR with no known survival benefit. Hormonal treatments may also be used for disease control but are not considered curative.

Dostarlimab will likely cause a shift in the current treatment paradigm for patients with dMMR or MSI-H metastatic or recurrent EC. The mechanism of action of dostarlimab would address the underlying disease-specific process and biomarkers for patients with dMMR or MSI-H EC. The clinical experts felt it would be preferable to initiate treatment with dostarlimab before other therapies.

Patient Population

The clinical experts agreed that patients with dMMR or MSI-H EC tumours would most benefit from an immune checkpoint inhibitor. One clinical expert added that patients with dMMR or MSI-H metastatic or recurrent EC of any histology would benefit from dostarlimab, regardless of symptoms or prior treatment; however, the other clinical expert noted that a smaller magnitude of benefit may be observed with increased previous lines of systemic therapy.

The clinical experts mentioned that the diagnosis of EC typically relies on biopsy, which may be conducted in clinic. If technical issues make it difficult to obtain samples from the endometrial lining during an office biopsy, the procedure can be performed in the operating room using dilatation and curettage. IHC testing for MMR status is relatively inexpensive and considered standard practice. If the MMR status of a sample is unknown, it may be further analyzed with genomic testing to determine MSI status. A valid test would involve NGS.

The clinical experts noted that treatment with dostarlimab would not be suitable for patients any of the following characteristics:

• very poor performance status
• history of severe autoimmune disease
• prior therapy with anti-PD-1, anti-PD-L1, or anti-PD-L2 drugs
• known uncontrolled nervous system metastases and/or carcinomatous meningitis
• poor medical risk due to serious uncontrolled medical disorder
• nonmalignant systemic disease or active infection requiring systemic therapy
• history of immunosuppression
• MSS EC.

Assessing Response to Treatment

According to the clinical experts, the most important goals of treatment for recurrent or advanced EC would be to improve OS and PFS; reduce symptoms of cancer; improve functional status (i.e., ability to perform activities of daily living); improve HRQoL; and reduce the burden of disease on patients and caregivers. One clinical expert noted that many patients with advanced or recurrent EC suffer from pelvic symptoms (e.g., unresectable disease in the pelvis causing bleeding or pain), lung symptoms (e.g., dyspnea from metastases), and neurologic symptoms (e.g., brain metastases) or bone symptoms (e.g., painful bony metastases), and that alleviation of these symptoms would be a benefit of treatment.

In terms of a clinically meaningful response, the clinical experts recommended that in addition to clinical assessment of disease symptoms and duration of disease control, the use of standard irRECIST³⁹ for the assessment of response to immunotherapeutic treatments is useful. One clinical expert noted that the Common Terminology Criteria for Adverse Events⁴⁰ tool for CT imaging can be used to assess response to treatment. The same clinical expert also suggested that cancer antigen 125 (nonspecific) may be used to assess treatment response in addition to other methods, although it is not commonly used in EC. In terms of the timing of assessments, the clinical experts recommend that response to treatment should be assessed radiologically every 3 months, with blood work every month and clinical assessments every 2 to 3 months.

Discontinuing Treatment

According to the clinical experts, treatment with dostarlimab should be discontinued when there are radiological and clinical signs and symptoms of disease progression, treatment toxicities (e.g., grade 3 or higher adverse reactions), or intolerability to treatment.

Prescribing Conditions

According to the clinical experts consulted by CADTH, the diagnosis, treatment, and monitoring of patients with EC should be undertaken by a specialist, namely a gynecologist oncologist, medical oncologist, and/or surgeon. Biomarker testing to identify dMMR or MSI-H status is also recommended.

Clinical experts recommend that dostarlimab be administered in a hospital clinic that has multidisciplinary medical supports to manage potential immune-related side effects. The clinical experts noted that treatment may also be administered in outpatient clinics.

Additional Considerations

The clinical experts noted that re-treatment with dostarlimab is possible if patients experience recurrence, as long as they had no signs of toxicity or intolerability while using the drug. If a patient completed their treatment and achieved a durable response for up to 2 years, and a significant time period elapsed before they progressed, then re-treatment with dostarlimab could be considered. The clinical experts noted that 50% of patients will have disease recurrence within 1 year of completing treatment for recurrent EC and will need to restart treatment.

Unmet Needs

The views of the clinician groups were overall consistent with the clinical experts consulted by CADTH, indicating that the most important treatment goals for advanced or recurrent dMMR or MSI-H EC are disease control, prolonged survival, delayed worsening of symptoms, maintenance of HRQoL, delayed disease progression, and an acceptable safety profile. Treatment for individuals diagnosed with recurrent EC is a critical unmet need. All clinician groups noted that although the prognosis for patients diagnosed with early-stage disease is good, for those with recurrent or metastatic EC, median OS is short. In fact, there is no effective second-line treatment for Canadians with EC. Response rates to second-line cytotoxic drugs are low (< 20%), with a median PFS of 3 to 4 months.

All clinician groups also noted that although Health Canada recently issued a NOC/c approval for pembrolizumab immunotherapy for dMMR or MSI-H EC, there is no funded access to this treatment. The lack of funding has created disparity in access to immunotherapy, as only those with insurance coverage or the capacity to self-pay can access pembrolizumab. Moreover, PMCC added that there is currently no patient-supported or compassionate-access program for immune checkpoint inhibitor therapies. The MUHC and SBHSC noted that the side-effect profile of combination therapy with pembrolizumab and lenvatinib was significant, with 66.9% of patients experiencing a grade 3 or 4 toxicity. The MUHC added that prolonging poor HRQoL with interventions associated with significant toxicity serves very little purpose.

Place in Therapy

All clinician groups indicated that all patients with advanced or recurrent EC would benefit from effective immunotherapy, but patients with MSI-H or dMMR subtypes would most benefit from treatment with an immune checkpoint inhibitor. All groups recommended that patients diagnosed with advanced or recurrent EC should be offered platinum-based chemotherapy as first-line therapy. However, the British Columbia Cancer Provincial Gynecological Oncology Tumour Group did acknowledge that treatment with immune checkpoint inhibitors may be an appropriate first-line therapy for patients who either refuse chemotherapy or for whom chemotherapy may be contraindicated or known to be poorly tolerated, provided dMMR or MSI-H status is confirmed. The MUHC further stated, “it is false economy to delay starting a highly effective treatment [dostarlimab] with minimal toxicity in this niche population.” The MUHC stated that trying second-line chemotherapy alone in this patient population is not advisable due to toxicity and lack of benefit.

MUHC indicated that when EC recurs in patients who initially responded well and dostarlimab is interrupted for any other reason, restarting treatment with dostarlimab would be appropriate. If patients progress on dostarlimab or other immunotherapies, palliative and supportive care options are considered.

Patient Population

Currently, the standard of care in Canada includes universal molecular characterization of all endometrial carcinomas for dMMR. Thus, identifying EC subtypes should not be a barrier to treatment. The British Columbia Cancer Provincial Gynecological Oncology Tumour Group and the GOC noted that patients whose EC is related to Lynch syndrome would benefit from treatment with dostarlimab, and PMCC added that those with polymerase-epsilon-mutation would benefit as well. The SCA noted that patients with EC suitable for treatment with dostarlimab should be PD-L1-naive and should not have undergone more than 2 prior treatments with a platinum-based therapy.

All clinician groups noted that patients who do not have a dMMR or MSI-H profile and/or have contraindications to immune checkpoint inhibitors are not suitable for treatment with dostarlimab. PMCC and the GOC also indicated that patients with MMR-proficient EC would not be suitable to receive dostarlimab as monotherapy treatment.

SBHSC noted that the EC patient groups less suitable for treatment with dostarlimab would be those that met all the following criteria:

• no previous treatment with platinum-based chemotherapy
• no dMMR seen on IHC
• poor medical risk due to a serious uncontrolled medical disorder
• poor performance status (ECOG Performance Status score of at least 3)⁴¹
• inadequate organ function
• a known immunodeficiency or current use of systemic steroids or other immunosuppressant medications.

Assessing Response to Treatment

The British Columbia Cancer Provincial Gynecological Oncology Tumour Group, COG, OH-CCO, and SBHSC noted that improvement in symptoms and physical findings related to advanced or recurrent EC (e.g., pain, bleeding, shortness of breath) would be considered a clinically meaningful treatment outcome. OH-CCO and PMCC stated that durable disease control with no adverse effects on HRQoL is an ideal treatment goal in this population. The SBHSC specified that PFS of at least 6 months would be a clinically meaningful outcome in this patient population. In addition, OH-CCO listed reduced caregiver burden as a meaningful treatment outcome.

The British Columbia Cancer Provincial Gynecological Oncology Tumour Group and the GOC also listed the following outcomes as clinically meaningful responses to treatment:

• maintenance or improvement of performance status and ability to perform activities of daily living
• evidence of disease regression from imaging studies
• disease stabilization (in patients with good baseline performance status and few disease-related symptoms).

OH-CCO and the GOC reported that standard clinical monitoring of therapy, including a physical examination, symptom review, and intermittent CT imaging, should be used to evaluate response to treatment. The MUHC recommended that after the first 3 cycles of dostarlimab, patients should be assessed for response to treatment with a CT scan. If there is no indication of progression and patients are well, the CT-monitoring interval may be spaced out to every 12 weeks. PMCC recommended that tumour assessment by CT scan or MRI be completed every 2 to 3 cycles (i.e., every 6 to 9 weeks). The British Columbia Cancer Provincial Gynecological Oncology Tumour Group and SBHSC stated that CT imaging should occur every 12 weeks, whereas the SCA suggested imaging every 3 to 6 months. The British Columbia Cancer Provincial Gynecological Oncology Tumour Group and SCA also added that laboratory values, as well as respiratory and pulmonary status, should be assessed to ensure that treatment response is occurring without toxicity.

Discontinuing Treatment

All clinician groups indicated that treatment with dostarlimab should be discontinued if patients experience disease progression or serious toxicities. SBHSC indicated that the following side effects warrant consideration of pausing or discontinuing dostarlimab:

• grade 2 anemia
• pneumonitis
• grade 3 colitis
• grade 2 asthenia
• grade 3 myalgia
• pemphigoid
• grade 3 increase in transaminases.

PMCC and SBHSC noted that due to the nature of immunotherapies and the possibility of pseudo-progression, patients with progression of disease on the first CT scan after initiation of treatment, but with no other symptoms, should continue treatment until further imaging demonstrates progression of disease. Similarly, the MUHC adds that in cases where disease progression occurs after 3 cycles with dostarlimab, patients should receive follow-up in 6 weeks to rule out pseudo-progression and discontinue treatment if progression is confirmed.

Prescribing Conditions

The British Columbia Cancer Provincial Gynecological Oncology Tumour Group, MUHC, OH-CCO, GOC, and SCA explained that dostarlimab is suitable to be delivered in the community and in outpatient and specialty clinics. Conversely, PMCC and SBHSC suggested that dostarlimab be delivered at cancer centres by gynecologic or medical oncologists. Both groups recommend that dostarlimab be administered in a chemotherapy suite with appropriate supervision by an oncologist familiar with gynecologic cancers and the management of immune-related adverse effects. The British Columbia Cancer Provincial Gynecological Oncology Tumour Group and the GOC noted that patients should be under the care of a treating physician or nursing staff with experience monitoring and evaluating patients for possible toxicities that may be caused by immune checkpoint inhibitor therapy. The GOC stated that this should not be a barrier to treatment because such therapies are now routinely used by most oncologists. The MUHC notes that for patients who do not live close to a treatment centre (e.g., patients in rural, remote, or Indigenous communities), it is appropriate to work closely with local physicians to share the responsibility of patient care.

Amendments and Protocol Deviations

The study protocol of the GARNET trial was amended 6 times, although versions 4.0 and 5.0 were not implemented. Amendments to the protocol did not substantially affect the results of the study.

Amendment 2 (October 31, 2016) included updates to the inclusion criteria, tumour assessment criteria, definitions for safety data, and sample size justification of cohort A1 and cohort A2.

Amendment 3 (October 9, 2017) included the addition of HRQoL assessments and interim analyses for patients with dMMR or MSI-H cancer in cohort A1 and cohort F combined.

Amendment 4 (July 3, 2018) included revisions to allow clinically stable patients without major safety issues to continue with dostarlimab treatment after confirmation of progressive disease. Changes were also made to blood sample collection and to increase the number of patients in cohort A2.

Amendment 5 (May 10, 2019) implemented changes to the definition of cohort A1 and cohort A2, based on feedback from a regulatory agency. A recommendation was made to select patients based on the MMR IHC result (local or central test) instead of NGS; hence, the central testing vendor was changed. The sample size for cohort A1 was increased and the assessment of HRQoL was revised from a secondary to an exploratory objective.

Amendment 6 (January 7, 2020) updates to the primary objective included evaluation of the safety and tolerability of dostarlimab, as well as the addition of 2 interim analyses for the combined enrolment of cohort A1 and cohort F to reach 200 and 300 patients, respectively, and for all patients to have measurable disease at baseline and at least 24 weeks of follow-up.

Figure 2

Study Schema.

Inclusion and Exclusion Criteria

The key inclusion and exclusion criteria used in cohort A1 of the GARNET trial are summarized in Table 4. Briefly, the trial enrolled adults aged 18 years and older who had been diagnosed with recurrent or advanced EC and who progressed on or after no more than 2 lines of prior systemic therapy (not including hormone therapy), with at least 1 of these being platinum-based doublet therapy. Patients had to have documentation of dMMR or MSI-H status and radiologically measurable disease that met Response Evaluation Criteria in Solid Tumors 1.1 (RECIST 1.1).⁴⁵ All EC histologies were permitted, except for endometrial sarcoma (including carcinosarcoma). At screening, patients had to have an ECOG Performance Status of 0 or 1, adequate organ function, a negative serum pregnancy test (if of childbearing potential), and 1 highly effective form of contraception throughout the study period. Prior therapy with an anti-PD-1, anti-PD-L1, or anti-PD-L2 drug was not permitted.

Baseline Characteristics

The baseline characteristics of patients who comprised the primary efficacy population of the GARNET trial are summarized in Table 7. At baseline, 105 patients in cohort A1 were identified as having dMMR or MSI-H EC (103 dMMR and 2 MMR-unk/MSI-H EC patients) for the primary efficacy analysis at IA-2. The demographic and baseline characteristics of patients with dMMR or MMR-unk/MSI-H tumours are presented as a total because of the small proportion of patients (n = 2) with MMR-unk/MSI-H tumours; the sponsor noted that the data remained similar when these patient groups were pooled. The median age was 64 years (range = 39 to 80 years), with 50.5% of patients between 39 and 65 years. Median weight was 71 kg (range = 34.0 to 141.4), and median BMI was 27.97 kg/m² (range = 13.6 to 53.9 kg/m²). Most patients were White (78.1%). The most common histology type was type I endometrioid carcinoma (67.6%); the most common histology grade at diagnosis was grade 2 (39.0%). At the time of study enrolment, more than two-thirds (67.6%) of patients had FIGO stage IV EC. An ECOG Performance Status of 1 was the most common (60%), followed by an ECOG Performance Status of 0 (40%). All patients received prior anti-cancer treatment that included surgery (90.5%), radiotherapy (70.5%), adjuvant or neoadjuvant treatment (53.3%), and/or bevacizumab use (4.8%). Most patients (88.6%) had previously received 2 or fewer lines of anti-cancer regimens; a small portion had previously received 2 or more lines (11.5%). More than half (56.2%) of patients had received prior regimens for metastatic disease.

Table 7

Summary of Baseline Characteristics of Patients in the GARNET Trial (IA-2, Primary Efficacy Analysis Dataset) .

Dose Modifications (Delay, Interruption)

Dosing delays were permitted in the case of medical or surgical events or for logistical reasons not related to study treatment (e.g., surgery, unrelated medical events, participant vacation). Continuation of the study treatment should have occurred within 28 days of the scheduled dostarlimab infusion. If a delay was more than 28 days, a patient may have been permitted to continue treatment after discussion with the sponsor.

Dosing interruptions were permitted in the event of toxicities. In general, treatment withheld for drug-related grade 3 toxicities could be resumed if toxicity resolved to grade 1 or lower. Similarly, for all irAEs, dostarlimab was held until the patient was considered clinically and metabolically stable and AEs resolved to grade 1 or lower. Dostarlimab was permanently discontinued after any drug-related grade 4 events and after some grade 3 immunologic-mediated AEs.

Dose reductions were not permitted.

Concomitant Medications

Concomitant medications were allowed to treat AEs and comorbidities. Concomitant medications included any medications other than study treatments taken on or after the initial study treatment dosing date. If systemic steroids were used as a part of irAE management, the total dose of daily steroids allowed was 10 mg or less of prednisone when dostarlimab treatment was resumed. Patients were also allowed to receive rescue medications and appropriate supportive care deemed necessary by the treating investigator.

To ensure accurate assessment of the safety and efficacy of dostarlimab, patients were prohibited from receiving the following therapies during the screening and treatment periods of the study:

• systemic anti-cancer or biological therapy
• immunotherapy not specified in the protocol
• chemotherapy not specified in the protocol
• investigational drugs other than dostarlimab
• radiation therapy within 3 weeks before study day 1 and during study treatment
  • note: palliative radiation therapy to a small field more than 1 week before day 1 of study treatment may have been allowed
• any surgery that involved tumour lesions
  • note: radiation therapy or surgery that involved tumour lesions was considered to be progressive disease at the time the procedure was performed.

Subsequent Therapies

Data were collected on whether patients received subsequent anti-cancer therapy after treatment discontinuation or progressive disease. However, these were not defined in the protocol or statistical analysis plan.

Overall Survival

OS was a secondary outcome of the GARNET trial and was defined as the time from the first dose of study treatment to death by any cause. Patients last known to be alive were censored at the date of the last known contact, as follows: OS in days = date of death or censoring – date of first dose + 1.

OS was assessed at EOT, at the safety follow-up visit, and every 3 months after treatment.

Disease Control Rate and Immune-Related Disease Control Rate

DCR and irDCR were secondary outcomes of the GARNET trial. DCR, per RECIST 1.1., was defined as the proportion of patients achieving BOR of confirmed CR, PR, or stable disease. irDCR, per irRECIST 1.1., was defined as the proportion of patients achieving irBOR or irCR. irPR or immune-related stable disease as assessed by the investigator.

BOR was programmatically derived, based on reported time point response assessments determined by a central reader at different evaluation time points, from the first dose until the first documented progressive disease. The irBOR, according to irRECIST, was programmatically derived, based on reported overall time point responses determined by investigators after assessment of radiological scans, at different evaluation time points from the first dose until documented disease progression.

BOR was determined according to the following rules:

• CR — at least 2 consecutive determinations of CR more than 4 weeks apart, with no other assessment between the 2 determinations other than not evaluable, CR, or missing.
• PR — at least 2 consecutive determinations of PR or better more than 4 weeks apart, with no other assessment between the 2 determinations other than not evaluable, CR, PR, or missing before disease progression (and not qualifying for a CR).
• Stable disease — at least 1 stable disease or non-CR or nonprogressive disease assessment (or better) at least 12 weeks to 10 days (i.e., ≥ 74 days) after baseline and before progressive disease (and not qualifying for a CR or PR). For an unconfirmed PR or CR to qualify, it must still meet the requirement of at least 12 weeks to 10 days.
• Progressive disease — disease progression after baseline. Note that a determination of CR followed at least 4 weeks later by stable disease resulted in a BOR of progressive disease.
• No disease — for central imaging data, when the independent radiologist cannot identify any disease at baseline, all subsequent assessments were documented as no disease if not declared progressive disease or not evaluable.

irBOR followed similar rules, except outcomes were immune-related (e.g., irCR, irPR). Clinical deterioration was not considered to be progressive disease. Only tumour assessments performed before the start of any new anti-cancer treatment were considered in the assessment of BOR and irBOR.

Progression-Free Survival and Immune-Related Progression-Free Survival

PFS and irPFS were secondary outcomes of the GARNET trial. PFS was defined as the time from the first dose to the earlier assessment of progressive disease or to death by any cause in the absence of progressive disease. irPFS time was defined, per irRECIST 1.1, as the time from the first dose to the earlier date of assessment of immune-related disease progression (irPD) event or to death by any cause in the absence of disease progression.

irPD events were considered when:

• 2 consecutive irPDs were observed; the date of the first was considered to be the date of the irPD event.
• the only or most recent tumour assessment before treatment discontinuation was irPD; the date was considered to be the date of the irPD event.

PFS and irPFS times were defined as follows: PFS or irPFS in days = date of PD or irPD event or death ÷ censoring – date of first dose + 1 (refer to Table 9 for details on censoring rules for PFS and irPFS).

Table 9

Censoring Rules for DOR, irDOR, PFS, irPFS.

Objective Response Rate and Immune-Related Objective Response Rate

ORR was the primary outcome of the GARNET trial and was defined, per RECIST 1.1 based on BICR, as the proportion of patients that achieved a BOR of CR or PR. irORR was a secondary outcome of the GARNET trial and was defined, per irRECIST 1.1, as the proportion of patients that achieved irBOR of irCR or irPR. Nonresponders were patients who did not have a post-baseline radiographic tumour assessment, who received post-baseline antitumour treatments (including surgery or radiation to tumour lesions) other than the study treatments before reaching a CR or irCR or a PR or irPR, or who died, progressed, or dropped out for any reason before reaching a CR or irCR or a PR or irPR.

Duration of Response and Immune-Related Duration of Response

DOR was a co-primary outcome of the GARNET trial and was defined as the time from first documentation of overall response leading to a confirmed CR or PR until the time of first documentation of disease progression or death. Censoring rules for DOR were similar to those for PFS (refer to Table 9 for details on censoring rules). irDOR was a secondary outcome of the GARNET trial and was defined, per irRECIST 1.1, as the time from first documentation of response leading to a confirmed irCR or irPR until the time of the irPD event or death. As with BOR and irBOR, clinical deterioration was not considered to be documented progressive disease.

Health-Related Quality of Life

The HRQoL outcomes measured in the trial included EORTC QLQ-C30 and EQ-5D-5L scores. Data from both instruments were exploratory outcomes in the GARNET trial and assessed at IA-2. HRQoL assessments were collected at scheduled visits (at each cycle) every 3 weeks (± 7 days) for the first 12 weeks, beginning on cycle 1, day 1, and every 6 weeks (± 7 days) thereafter while the patient was receiving study treatment. After treatment discontinuation, HRQoL assessments were collected from the remaining patients at the EOT visit, at the safety follow-up visit (90 days after EOT), and every 90 days (± 14 days) during the post-treatment follow-up period. A detailed discussion and critical appraisal of the HRQoL measures are provided in Appendix 3.

The EORTC QLQ-C30 is a questionnaire developed specifically to assess HRQoL in cancer patients. The questionnaire consists of 30 questions, 5 function scales (physical, role, cognitive, emotional, and social), 1 global health status/global quality-of-life scale, 3 symptom scales (fatigue, pain, and nausea and vomiting), and 6 single items that assess additional symptoms (dyspnea, appetite loss, sleep disturbance, constipation, and diarrhea) and financial impact.⁴⁸ Scales and single items range in score from 0 to 100, with higher scores on the functional and global health status/quality-of-life scales indicating higher levels of functioning and health status/quality of life, respectively. Higher scores on symptom scales or items represent a greater presence of symptoms.⁴⁸

The EQ-5D-5L version 2.0 is a generic HRQoL questionnaire for assessing a patient’s health status in terms of a single index value or utility score. There are 2 components of the questionnaire: a descriptive system that allows patients to rate their level of problems (none, slight, moderate, severe, extreme/unable) in 5 areas (mobility, self-care, usual activities, pain/discomfort, and anxiety/depression), and a visual analogue scale that allows patients to rate their overall health status from 0 (worst imaginable) to 100 (best imaginable).⁴⁹

Safety

Assessing the safety of dostarlimab was a primary objective of the GARNET trial. The primary safety variables included the following:

• TEAEs — all, SAEs, their relationship to the study treatment, and their intensity (based on National Cancer Institute Common Terminology Criteria for Adverse Events v4.03 severity grades)
• TEAEs leading to discontinuation
• TEAEs leading to death
• irAEs
• clinical laboratory measures (including hematology, chemistry, coagulation, thyroid function, and urinalysis) and vital signs
• ECG
• physical examination
• ECOG status
• serum pregnancy testing
• concomitant medications.

Sample Size Determination

For the final analysis of the co-primary end point (ORR) in cohort A1, the sample size of approximately 100 patients with dMMR or MSI-H EC (with potential enrolment of up to 165 patients) was planned after protocol amendment 5. The null hypothesis that the true response rate was 20% or less (expected ORR for conventional therapy) was tested against a 1-sided alternative hypothesis of a response rate of at least 40%. With 65 patients in cohort A1, there was 92% power to rule out an ORR of 20% or less when the true ORR was 40% at the 2.5% type I error rate (1-sided). The increase in sample size allowed the lower-limit boundary of the exact 95% CI to exclude a response rate of 25% or less, assuming the observed ORR was 35%.

Interim Analyses

The 3 planned interim analyses were based on the combined enrolment in cohort A1 and cohort F (nonendometrial cancer). IA-1, IA-2, and IA-3 were conducted when the combined enrolment reached 100, 200, and 300 patients, respectively. All enrolled patients in cohort A1 and cohort F had dMMR tumours with measurable disease at baseline and at least 24 weeks of follow-up. IA-1 had a data cut-off date of July 8, 2019; IA-2 had a data cut-off date of March 1, 2020; and IA-3 had a data cut-off date of November 1, 2021. IA-2 and IA-3 were added after amendment 6.

All interim analyses evaluated the outcomes listed in Table 8, with the exception of IA-1, which included irORR as a primary end point and patient-reported outcomes as a secondary end point.

Primary Outcomes

The co-primary outcomes in the GARNET trial were ORR and DOR. No formal hypothesis-testing or inferential analyses were performed in cohort A1, and no statistical comparisons were planned between cohorts. The analysis of ORR included summary statistics, including the number of patients and percentage for categorical variables and, for continuous variables, the number of patients, mean, standard deviation (SD), median, minimum, and maximum. Two-sided exact 95% CIs based on the Clopper-Pearson method were reported for the ORR.

DOR analyses were performed using KM methods and summarized by minimum, maximum, 25th, 50th (median), and 75th percentiles with associated 95% CIs, the number and percentage of events, the number and percentage of censored observations, and the median duration of follow-up. Censoring rules for DOR are presented in Table 7.

Secondary Outcomes

The analyses for OS were based on the safety analysis and primary efficacy analysis datasets. OS analyses were performed using KM methods and summarized by the 25th, 50th (median), and 75th percentiles with associated 95% CIs, as well as the number and percentage of events and censored observations. Censoring for OS was set at the last known date of contact.

The analyses of DCR and irDCR involved summary statistics, including the number of patients and percentage for categorical variables and, for continuous variables, the number of patients, mean, SD, median, minimum, and maximum. Two-sided exact 95% CIs based on the Clopper-Pearson method were provided for DCR and irDCR.

The analyses of PFS and irPFS were performed using KM methods and summarized by the 25th, 50th (median), and 75th percentiles with associated 95% CIs, as well as the number and percentage of events and censored observations. Censoring rules for PFS and irPFS are presented in Table 9.

The analyses of irORR and irDOR were similar to the those of the co-primary outcomes of ORR and DOR.

Exploratory Outcomes

No analysis plan, objectives, or minimally important difference (MID) for the EORTC QLQ-C30 or the EQ-5D-5L instruments were specified a priori in the statistical analysis plan; however, according to the study protocol, changes from baseline to each pre-specified study time point were to be measured and summarized using descriptive statistics. HRQoL assessments (EQ-5D-5L and EORTC QLQ-C30) were collected during scheduled visits for all patients in cohorts A1 and F enrolled under amendment 3 and subsequent amendments (i.e., every 3 weeks ± 7 days for the first 12 weeks, in alignment with study drug administration, and every 6 weeks ± 7 days thereafter, in alignment with tumour imaging assessments) while the patient was receiving study treatment. Once a patient discontinued treatment, patient-reported outcome assessments were performed during the EOT visit, the safety follow-up visit, and the post-treatment follow-up period every 90 days (± 14 days).

For the EORTC QLQ-C30, scores and changes in scores from baseline for each domain were summarized by the number of patients with values for the mean, SD, median, range, first and third quartiles, and 95% CI. Statistical graphs were produced for each of the 15 domains to show mean scores over time, with 95% CIs. Similar graphic displays were produced for the mean change from baseline over time. In terms of missing data, if at least half of the items from a particular scale were answered, it was assumed that the missing items had values equal to the average of those items that were present. With this method of imputation, none of the single-item measures could be imputed.

Safety Outcomes

Safety and tolerability were evaluated by monitoring the frequency, duration, and severity of AEs and irAEs. In general, all by-visit summaries of safety parameters were only summarized up to and including month 6 and at the treatment discontinuation visit. AEs were organized based on the Medical Dictionary for Regularity Activities version 23.0. Only TEAEs were analyzed, but all AEs occurring during the study were listed. TEAEs were tabulated by preferred term and system organ class. Patients with the same TEAE more than once had that event counted only once for each system organ class and once for each preferred term. By-patient listings of irAEs were presented, if appropriate.

Sensitivity Analyses

No sensitivity analyses were conducted for any of the outcomes outlined in the protocol of the GARNET trial.

Subgroup Analyses

Subgroup analyses were performed for ORR, DOR, and TEAEs. The following subgroup analyses for DOR and ORR were planned a priori in the statistical analysis plan:

• MSI status (MSI-H versus MSS versus unknown/missing)
• histology (patients with dMMR EC only)
• number of prior anti-cancer therapy regimens (1 versus ≥ 2)
• prior radiation therapy (yes or no)
• prior bevacizumab use (yes or no)
• BOR from last platinum-containing prior anti-cancer therapy (CR or PR, SD, progressive disease, or missing)
• progression-free interval from last platinum-containing prior anti-cancer therapy (< 6 months versus ≥ 6 months versus missing).

The following subgroups aligned with those pre-specified in the protocol for this CADTH review: histology, number of prior anti-cancer therapy regimens, prior radiation therapy, and progression-free interval from last platinum-containing prior anti-cancer therapy. Only the subgroups identified in the CADTH review protocol are reported in the Outcomes section.

Analysis Populations

The primary efficacy analysis dataset, by RECIST 1.1 per BICR, was defined as all patients in the safety analysis dataset with measurable disease at baseline (defined as the existence of at least 1 target lesion at baseline tumour assessment identified by BICR) who had the opportunity for at least 24 weeks of tumour assessment at the time of analysis (i.e., patients whose first dose of dostarlimab administration was on or before September 15, 2019).

The secondary efficacy analysis dataset, by irRECIST per investigators’ assessment, was defined as all patients in the safety analysis dataset with measurable disease at baseline (defined as the existence of at least 1 target lesion at baseline tumour assessment identified by investigator assessment) who had the opportunity for at least 24 weeks of tumour assessment at the time of analysis (i.e., patients whose first dose of dostarlimab administration was on or before September 15, 2019).

The safety analysis dataset included all patients who received any amount of study drug. In addition, all patient-reported outcome analyses were conducted on patients using the safety analysis dataset enrolled under amendment 3 or subsequent amendments.

Duration and Dose Intensity

Data on exposure to dostarlimab in cohort A1 at IA-2 are summarized in Table 12. Data were available for 129 patients with dMMR or MMR-unk/MSI-H. The median DoT with dostarlimab was 25.6 weeks. The maximum DoT with dostarlimab was 138.9 weeks. Approximately half of patients (53.5%) received dostarlimab treatment for up to 19 to 24 weeks. The median relative dose intensity was 100% throughout the study treatment period because dose reductions were not permitted. Twenty-eight patients continued study treatment despite initial progressive disease.

Treatment Compliance

Treatment compliance was based on number of infusions. At any time during the study, less than 6% of patients with dMMR EC had a missed infusion, and less than 20% had an infusion delay. From cycle 1 to cycle 4, there were 3 missed infusions (2.4%) and 11 infusion delays (8.7%) reported in patients with dMMR EC. From cycle 5 through the EOT, 7 missed infusions (5.6%) and 25 infusion delays (19.8%) were reported in patients with dMMR EC. One infusion interruption (0.8%) was reported between cycle 5 and EOT.

Table 12

Patients With dMMR or MSI-H EC on Treatment by Week Intervals in Cohort A1 of the GARNET Trial (IA-2, Safety Analysis Dataset).

Overall Survival

The results for OS from cohort A1 of the GARNET trial are summarized in Table 13. The median OS was not reached at the time of IA-2. The median follow-up time was 16.3 months. In total, there were 35 deaths (33.3%), and 70 patients (66.7%) were censored. The probability of patients in cohort A1 surviving to 6, 9, and 12 months was 80.9% (95% CI, 71.7% to 87.4%), 75.1% (95% CI, 65.2% to 86.6%), and 68.9% (95% CI, 58.3% to 77.4%), respectively.

Table 13

KM Analysis of OS for Patients With dMMR or MSI-H EC in Cohort A1 of the GARNET Trial (IA-2, Primary Efficacy Analysis Dataset).

Figure 3

KM Plot of OS for Patients With dMMR or MSI-H EC per RECIST 1.1 (IA-2, Secondary Efficacy Analysis Dataset).

Disease Control Rate and Immune-Related Disease Control Rate

The DCR results (based on investigator assessment) of the GARNET trial for cohort A1 are summarized in Table 14. The DCR was 57.1%, with 11 patients (10.5%) having a BOR of CR, 36 patients (34.3%) having a BOR of PR, and 13 patients (12.4%) having a BOR of stable disease. Forty-two (89.4%) patients at the time of IA-2 had an ongoing response.

irDCR by investigator assessment per irRECIST 1.1 (n = 113) was 63.7% for patients with dMMR or MSI-H EC, which was slightly higher than the DCR. Eight patients (7.1%) had an irBOR of irCR, 44 patients (38.9%) had an irBOR of irPR, and 20 patients (17.7%) had an irBOR of irSD. Forty-three (82.7%) patients at the time of IA-2 had an ongoing response. Detailed irDCR results are available in Appendix 4.

Table 14

Tumour Response in Patients With dMMR or MSI-H EC in Cohort A1 of the GARNET Trial per RECIST 1.1 Assessed by BICR (IA-2, Primary Efficacy Analysis Dataset).

Health-Related Quality of Life

EORTC QLQ-C30

EORTC QLQ-C30 data were available for 94 of the 129 patients in cohort A1 of the GARNET trial at the time of IA-2. Completion rates for the EORTC QLQ-C30 instrument declined over time. The completion rate at baseline was 100% and 58.5% by cycle 7 (n = 55). The mean scores and mean changes from baseline at each assessment point for EORTC QLQ-C30 (global health status or quality-of-life scale) are summarized in Table 15 and Figure 4. A definition for what constituted a clinically meaningful change from baseline in the study population was not provided. Overall, summary scores increased or remained stable from baseline to cycle 7 among patients who completed questionnaires.

Table 15

Summary of EORTC QLQ-C30 Results — Global Health Status and QoL Function From Baseline to Cycle 7 (IA-2, Safety Analysis Dataset).

A summary of the disease-related symptom subscales can be found in Figure 5. Patients who received dostarlimab and completed the questionnaire reported decreased symptoms or remained stable over time in key disease-related symptom subscales, including pain and fatigue.

Mean change in symptomatic AE response from baseline over time is illustrated in Figure 6. Most patients who experienced symptomatic AEs, including nausea, vomiting, constipation, diarrhea, or tiredness, remained stable or had improvement from baseline in these symptoms over the treatment course. Less than 25% of patients reported single-category worsening in these AE symptoms, and less than 6% reported 2- or 3-category worsening.

Figure 4

Mean Change in Global Health Status and QoL From Baseline (IA-2, Safety Analysis Dataset).

Figure 5

Pain, Fatigue, and Physical Functioning Mean Change From Baseline (IA-2, Safety Analysis Dataset).

Figure 6

Symptomatic AE Change in Response From Baseline (IA-2, Safety Analysis Dataset).

EQ-VAS

At baseline, the mean EQ-VAS score was 69.3 (SD = 19.2) (refer to Figure 7). The following changes were observed in the EQ-VAS scores over time:

• At week 12, the mean score was 77.1 (SD = 18.0) and mean change from baseline was 5.0 (SD = 12.6).
• At week 18, mean score was 77.4 (SD = 17.4) and mean change from baseline was 4.0 (SD = 15.2).
• At week 42, the change from baseline was 4.0 (SD = 16.2).

Greater change in scores was observed after EOT; however, the number of patients at each post-treatment visit was markedly low (< 5 patients).

Figure 7

Adjusted Mean Change From Baseline in EQ-VAS (IA-2, Safety Analysis Dataset).

Progression-Free Survival and Immune-Related Progression-Free Survival

The results for PFS in cohort A1 of the GARNET trial, based on BICR, are summarized in Table 16. At IA-2, 57 (54.3%) patients had a PFS event and median PFS was 5.5 (95% CI, 3.2 to NR) months. KM estimates of PFS by RECIST 1.1 were 48.6% (95% CI, 38.6% to 56.8%) at month 6 and 47.5% (95% CI, 37.4% to 56.8%) at month 9 and at month 12. Refer to Figure 8 for the PFS KM curve.

In terms of irPFS, 57.3% of patients had an irPFS event at the time of IA-2. The median irPFS time was 10.3 months (95% CI 5.2 to 18.0). Based on investigator assessment, the probability of having irPFS based on KM estimates were 54.7% at 6 months, 50.7% at 9 months, and 47% at 12 months for patients with dMMR or MSI-H EC (95% CI not reported). KM estimates for irPFS were higher than that of PFS by BICR. Refer to Figure 13 in Appendix 4 or the irPFS KM curve.

Table 16

KM Analysis of PFS in Patients With dMMR or MSI-H EC in Cohort A1 of the GARNET Trial per RECIST 1.1 Assessed by BICR (IA-2, Primary Efficacy Analysis Dataset).

Figure 8

KM Plot for PFS in Patients With dMMR or MSI-H EC per RECIST 1.1 Assessed by BICR (IA-2, Primary Efficacy Analysis Dataset).

Objective Response Rate and Immune-Related Objective Response Rate

The results for ORR (based on BICR) for cohort A1 of the GARNET trial are summarized in Table 12. The ORR results were generally consistent with the results for irORR.

At the time of IA-2, the ORR, per RECIST 1.1, was 44.8% (95% CI, 35.0% to 54.8%); 47 of 105 patients achieved either a CR (n = 11) or PR (n = 36) (Table 14). The median follow-up time was 16.3 months. The ORR results were generally consistent with the results of irORR; the irORR, per irRECIST 1.1, was 46% (95% CI, 36.6% to 55.6%), with 82.7% of responders having an ongoing immune-related response at the time of IA-2.

The ORR results for pre-specified patient subgroups of interest to the CADTH review are summarized in Table 17. The ORR was 49.2% in patients with 1 line of prior anti-cancer therapy, compared with 36.8% in patients who received 2 or more lines of prior anti-cancer therapy. The ORR was 46.6% in patients with prior radiation, compared with 40.0% in patients who had not received prior radiation. The ORR was 46.0% in patients with a progression-free interval of at least 6 months, compared with 39.5% in patients with a progression-free interval of less than 6 months. The ORR was 56.3% in patients with type II endometrioid carcinoma and 40.0% in patients with type I endometrioid carcinoma. Note that 95% CIs were not provided for any of the ORR subgroup analyses.

Table 17

Subgroup Analysis Results for ORR in Cohort A1 of the GARNET Trial (IA-2, Primary Efficacy Analysis Dataset).

The sponsor noted that there were too few patients of each endometrioid carcinoma type II subtype to make a meaningful comparison based on histologic subtype.

Duration of Response and Immune-Related Duration of Response

The DOR results (based on BICR) for cohort A1 of the GARNET trial are summarized in Table 18.

Of the 47 patients who achieved an objective response, 42 (89.4%) had an ongoing response at the time of the IA-2 data cut-off. The median duration of follow-up was 16.3 months; however, the median DOR was not reached among responders. The DOR in these patients ranged from 2.63 to 28.09 months as at the time of the IA-2data cut-off, with 37 patients (78.7% of responders) achieving a DOR of at least 6 months. Based on these results, the probability of responders maintaining a confirmed objective response, per RECIST 1.1, was estimated to be 97.9% (95% CI, 85.8% to 99.7%), 90.9% (95% CI, 73.7% to 97.1%), and 80.1% (95% CI, 56.8% to 91.7%) at months 6, 12, and 18, respectively.

The sponsor noted that the small sample size of responders hampered the ability to estimate the DOR of patient subgroups, so these data were not reported.

The results of investigator-assessed irDOR were generally consistent with DOR assessed by BICR (refer to Appendix 4 for further details). At the time of the IA-2, 43 patients had an ongoing immune-related response out of the 52 who achieved an irCR or irPR. The median irDOR was not reached among responders after a median follow-up of 16.5 months. Approximately 76.9% of responding patients achieved an irDOR of at least 6 months. Based on KM estimates, the probability of maintaining a response, per irRECIST, at 6 months was 96.1% (95% CI, 85.2 to 99.0) and at 12 months and at 18 months was 79.2% (95% CI, 62.1% to 89.2%).

Table 18

KM Analysis of DOR for Patients With dMMR or MSI-H EC in Cohort A1 of the GARNET Trial per RECIST 1.1 Assessed by BICR (IA-2, Primary Efficacy Analysis Dataset of Patients With Objective Response).

Adverse Events

The most frequently reported TEAEs (≥ 15%) in patients with dMMR or MSI-H EC were nausea, diarrhea, anemia, fatigue, asthenia, constipation, vomiting, abdominal pain, cough, arthralgia, urinary tract infection, and back pain. With the exception of anemia, these TEAEs were reported to be mild or moderate in severity in most patients.

Grade 3 or higher TEAEs occurred in 48.1% of patients in cohort A1. The most commonly reported grade 3 or higher TEAEs were anemia (14.7%), abdominal pain (5.4%), and hyponatremia (3.9%). Grade 4 TEAEs were reported for hyponatremia, pulmonary embolism, pneumonia, and sepsis. Sepsis was reported as a grade 4 TEAE in 3 patients with dMMR EC and 1 patient with MMR-unk/MSI-H. Other grade 3 or higher TEAEs included acute kidney injury (3.1%) and back pain (3.1%).

Serious Adverse Events

The percentage of patients that experienced a serious TEAE was 34.1% in cohort A1. The most common serious TEAEs were abdominal pain, acute kidney injury, and sepsis, each occurring in 3.1% of patients, and pulmonary embolism, pyrexia, and urinary tract infection, each occurring in 2.3% of patients.

Withdrawals Due to AEs

None of the patients in cohort A1 who withdrew from the GARNET trial had an AE as a primary reason. AEs that led to study treatment discontinuation occurred in 11.6% of patients, whereas AEs that led to study treatment interruption occurred in 24.0% of patients. The most common AEs leading to study treatment interruption were anemia (3.1%) and diarrhea (2.3%).

Mortality

TEAEs leading to death occurred relatively rarely in cohort A1 during the treatment period (n = 1; 0.8%) and included aspiration. TEAEs leading to death during the 90-day safety follow-up occurred in 4 patients (3.1%), and included pleural effusion, pneumonia, sepsis, and shock. None of the TEAEs leading to death were considered by the investigators to be treatment-related. No TEAEs were the primary cause of death during the long-term follow-up period.

Notable Harms

Notable harms as specified in the CADTH review protocol included immune-related toxicity.

The incidence of irAEs was 34.9% in cohort A1. The most frequently reported irAEs (≥ 5%) were diarrhea and hypothyroidism. A total of 7.9% of patients had a serious irAE, 12.7% had an irAE that was grade 3 or higher, and 4.8% of patients with dMMR EC had an irAE that led to study treatment discontinuation. Most of the irAEs were considered by the investigators to be related to study treatment.

Table 19

Summary of Harms in Cohort A1 of the GARNET Trial.

Internal Validity

The main limitation of the GARNET trial stems from the single-arm trial design and the lack of a comparator group. Such a design makes it challenging to interpret the efficacy and safety events attributable to dostarlimab, because all patients in cohort A1 received the same treatment. The lack of comparison with an active comparator or with standard of care or placebo precludes the ability to assess the relative therapeutic benefit or safety of dostarlimab. Formal statistical significance and hypothesis-testing were not performed, except for ORR. The risk of selection bias cannot be ruled out, given the lack of transparency regarding the selection and timing of the enrolment of patients in the trial. In the absence of such information, the risk of bias is unclear. GARNET is an open-label trial, and the study investigators and patients were aware of their treatment status, which increases the risk of detection and performance bias; this has the potential to influence results and outcomes in favour of dostarlimab if the assessor (investigator or patient) believes the study drug is likely to provide a benefit. Furthermore, if study personnel and patients knew that the treatment was dostarlimab (which is known to cause irAEs and other neurotoxicity), this also could have influenced the reporting of harms outcomes. To mitigate the impact of such bias, the investigators used a BICR to evaluate responses using standardized criteria for certain efficacy outcomes, such as ORR, DOR, PFS, and DCR. Bias is therefore less a concern for these efficacy end points and OS, and more of a concern for subjective end points such as HRQoL and safety. Overall, the magnitude and direction of this bias remain unclear.

Median OS was not reached at the time of IA-2, so the survival data from the trial were immature. Because results were based on an interim analysis, the treatment benefit may be overestimated and harms may be underestimated. In addition, the potential for confounding cannot be ruled out, given the lack of adjustment in the analysis for known confounders and treatment-effect modifiers that would typically be accounted for in an RCT. Interpretation of time-to-event end points such as OS or PFS is limited in single-arm studies. The clinical experts agreed that in the absence of robust comparative data on PFS and OS, no firm conclusions could be drawn on how dostarlimab compares with other relevant treatment options.

The open-label, single-arm design also limits the ability to interpret HRQoL data from the GARNET trial. Loss of patients completing HRQoL questionnaires over the course of the study led to small sample sizes at many assessment time points (in some cases < 5 patients) and resulted in imprecise measurement, with wide and overlapping CIs. Statistical analyses of changes in scale scores over time were not conducted, and a definition of what was considered a clinically meaningful response or minimally important difference was not reported. As well, summary data were not reported for the EQ-5D-5L descriptive system. For these reasons, no conclusions should be drawn from the HRQoL data. This is considered a significant limitation of the GARNET trial, because HRQoL was identified as an important outcome by the patient and clinician groups providing input for this review.

External Validity

Overall, the clinical experts consulted by CADTH agreed that the inclusion and exclusion criteria, baseline patient characteristics, concomitant medications, and prohibited medications in cohort A1 of the GARNET trial were reflective of patients they see in clinical practice for the indication under review. There were no barriers to the identification of patients who would most benefit from the treatment, given that testing for MMR and MSI status is standard practice in Canada. The population enrolled in the trial was consistent with the population expected to be treated in Canadian clinical practice, but the clinical experts noted that patients in the GARNET trial were slightly younger than those seen in clinical practice. The clinical experts also noted that patients with certain comorbidities (e.g., obesity) and of certain racial groups (e.g., Black) may have poorer prognosis; however, the findings of this study may not necessarily be generalizable to these subgroups, given the small sample size and lower level of representation of ethnic groups in the study (n = 2). Furthermore, the clinical experts indicated that no different treatment effect would be expected based on different disease-management practices in participating countries. In the opinion of the clinical experts, as long as patients have dMMR or MSI-H tumour status, dostarlimab would be appropriate to administer after any of the prior therapies received by patients in the trial. The majority of patients (88%) in cohort A1 of the GARNET trial had received 2 or fewer prior lines of systemic therapy before trial enrolment, and 11% of patients had received 3 or more lines of prior systemic therapy. However, the clinical experts noted that experience from clinical practice suggests that dostarlimab may have a lower magnitude of treatment benefit in patients with more prior lines of systemic therapy. This aligns with the results of the ORR subgroup analysis, in which the ORR was higher in patients with 1 line of prior anti-cancer therapy than in patients with 2 or more lines of prior anti-cancer therapy (49.2% versus 36.8% [95% CI, NR to NR]). However, these results are exploratory, and conclusions should be interpreted with caution. The clinical experts and clinician groups agreed that patients should not have previously been treated with immunotherapies. In cohort A1, 3 patients could not have their MMR tumour status determined, but their tumours were classified as MSI-H. The clinical experts noted that these 2 groups are generally thought to be synonymous, and data could be pooled together, especially given the small number of patients with MSI-H status. Concomitant medications received by patients in the trial appeared to reflect the medications that patients would receive in Canadian clinical practice, according to the clinical experts. It is also acknowledged that mature OS data will be confounded by the use of subsequent anti-cancer therapy received by some patients after progressive disease.

According to the clinical experts consulted by CADTH, OS, PFS, and HRQoL are clinically meaningful end points for patients with dMMR or MSI-H recurrent or advanced EC that has progressed on or after treatment with a platinum-containing regimen. Tumour response outcomes are also important in this patient population because of the accompanying delay in the worsening of symptoms and the slower decline in ECOG performance status.

There were a limited number of patients in the primary efficacy dataset (N = 105). The magnitude of the treatment-effect estimates observed in a small study sample may not be replicable in a larger study sample or generalizable to the target population in real-world clinical practice. Furthermore, the subgroup analyses had no statistical comparisons and even smaller sample sizes, which limits the generalizability to a broader population. These data should be considered exploratory in nature, and no conclusions should be drawn based on the results.

Description of MAIC Analyses

The sponsor submitted 3 MAIC reports to demonstrate the efficacy and safety of dostarlimab compared with relevant treatments for advanced or recurrent EC¹⁹:

• Two MAIC reports that compared individual patient data (IPD) from the phase I GARNET trial with a GARNET-like real-world evidence (RWE) cohort in the UK that received current treatment paradigms (hereafter called MAIC report 1), and a selection of 5 specific treatments for advanced or recurrent EC (hereafter called MAIC report 2) from the National Cancer Registration and Analysis Service (NCRAS) database.¹⁹
• One MAIC report that compared IPD from the phase I GARNET trial with relevant comparator data from the published literature (hereafter called MAIC report 3).¹⁹

Methods of MAIC Analyses

Study Selection Methods

The index trial in all cases was based on IPD from cohort A1 of part 2B of the phase I, single-arm GARNET trial (dMMR or MSI-H EC cohorts, safety population set; n = 129).¹⁹ Study selection methods to identify relevant comparators varied in the 3 sponsor-submitted MAIC reports.

For the 2 reports that used RWE as a comparator (MAIC reports 1 and 2), no systematic literature review (SLR) was conducted. Instead, the investigators created GARNET-like cohorts based on a subgroup of patients from the NCRAS in the UK, including a base-case analysis of current treatment paradigms, an analysis of 5 treatment-specific cohorts, and a sensitivity analysis of patients with an ECOG Performance Status of 0 or 1 at index. Notably, information on dMMR or MSI-H status (an inclusion criterion in GARNET and the indication under review) was not available in the RWE cohorts. In MAIC report 3, a SLR of clinical and observational studies was conducted in accordance with the highest standard for evidence synthesis (date last searched, April 5, 2020).⁵¹ Although there were no limitations to study eligibility by risk of bias, the authors noted that the included studies were generally at low risk of bias per the NICE tool,⁵² and bias was either high or unclear per the CASP tool.^19,53

The selection characteristics used to develop the GARNET-like cohorts and the population, intervention, comparison, outcomes, and study (PICOS) criteria used to identify published literature are summarized in Table 20.

Table 20

Selection Criteria for GARNET-Like Subgroups and for Relevant Comparator Studies for Sponsor-Submitted MAICs.

The outcomes evaluated in the MAICs are summarized in Table 21. OS was the pre-specified primary outcome in all MAICs. In MAIC report 3, PFS was also included as a pre-specified primary outcome. PFS and TTD were the pre-specified secondary outcomes of MAIC reports 1 and 2. These were summarized descriptively, as the measurement definitions for PFS and the time period evaluations associated with the RWE cohort were too dissimilar to those in GARNET. As PFS was not recorded in the NCRAS database, TTNT and TTD were used as proxies. Pre-specified secondary end points in MAIC report 3 included ORR, DOR, HRQoL, and SAEs; however, no definitions were provided for these outcomes in the report.¹⁹

Table 21

End Point Definitions in Included Studies.

MAIC Analysis Methods

Given that the GARNET trial was a single-arm trial, unanchored MAICs were necessary to conduct the ITCs because the single-arm nature precluded the use of anchored ITC methods. A summary of the analysis methods in each MAIC is provided in Table 23.

GARNET Versus NCRAS RWE Cohort MAIC Methods (MAIC Reports 1 and 2)

Analysis methods for the MAICs conducted using RWE data (MAIC reports 1 and 2) were identical, only differing in the RWE cohorts used. The GARNET trial was a phase I, open-label, dose-escalation study, whereas the RWE study was a descriptive, noninterventional study of patient-level data available through the NCRAS to describe characteristics, treatments, and outcomes of patients diagnosed with advanced or recurrent EC. The dataset for analysis consisted of IPD for all patients in the GARNET safety cohort (n = 129), and aggregated summary information from the initial RWE cohort of incident patients at second-line treatment (n = 999; n = 501 for patients with an ECOG Performance Status of 0 or 1 at index). Treatment-specific RWE cohorts included patients who received paclitaxel monotherapy (n = 116), carboplatin plus paclitaxel (n = 279), carboplatin plus liposomal doxorubicin (n = 141), liposomal doxorubicin monotherapy (n = 130), and carboplatin monotherapy (n = 93).¹⁹

Key differences were noted by the sponsor between the GARNET trial and the NCRAS RWE cohort’s inclusion criteria. One of the main differences between GARNET and the RWE cohort was the MMR or MSI status of patients. All patients in GARNET had dMMR or MSI-H EC, whereas this information was not available in the RWE cohort, so patients were included regardless of MMR or MSI status. Another key difference between GARNET and the RWE cohort’s inclusion criteria was that the RWE cohort used platinum doublet therapy as the key milestone regarding lines of therapy. The RWE cohort did not analyze the lines of therapy before the diagnosis of advanced or recurrent EC. The first-line therapy received for advanced or recurrent diagnosis was platinum doublet therapy for nearly all (99.3%) patients. Thus, the incident population at second-line therapy gave rise to the base case sample size (n = 999). Last, although all patients in GARNET had an ECOG Performance Status of 0 or 1, the inclusion criteria on ECOG Performance Status were partially met in the RWE cohort. In GARNET, ECOG Performance Status was ascertained at the index date of the trial (i.e., the start of second-line therapy), whereas in the RWE cohort, ECOG Performance Status was recorded at diagnosis. Also, there were many patients with ECOG Performance Status documented as “not recorded” in the RWE cohort. A sensitivity analysis conducted with the RWE cohort was restricted to patients with a known ECOG Performance Status of 0 or 1. This was referred to as the RWE cohort (ECOG ≤ 1).¹⁹

A total of 8 analyses (4 base-case and 4 scenario analyses) were conducted in the initial RWE cohort of MAIC report 1. Hazard ratios (HRs), 95% CIs, and corresponding P values were calculated only for the OS outcome (comparisons were descriptive for other outcomes), as follows¹⁹:

• OS in GARNET (N = 129) versus OS in RWE cohort (base case, n = 999; ECOG ≤ 1, n = 501)
• PFS in GARNET (N = 129) versus TTNT in RWE cohort (base case, n = 999; ECOG ≤ 1, n = 501)
• PFS in GARNET (N = 129) versus TTD in RWE cohort (base case, n = 999; ECOG ≤ 1, n = 501)
• TTD in GARNET (N = 129) versus TTD in RWE cohort (base case, n = 999; ECOG ≤ 1, n = 501).

A total of 4 analyses were performed for each treatment-specific RWE cohort in MAIC report 2. HRs, 95% CIs, and P values were calculated only for the OS outcome (comparisons were descriptive for other outcomes), as follows¹⁹:

• OS in GARNET (N = 129) versus OS in treatment-specific RWE cohorts
• PFS in GARNET (N = 129) versus TTNT in treatment-specific RWE cohorts
• PFS in GARNET (N = 129) versus TTD in treatment-specific RWE cohorts
• TTD in GARNET (N = 129) versus TTD in treatment-specific RWE cohorts.

A range of prognostic variables typically associated with survival in EC were identified in a targeted literature review (conducted May 2020), and validated through a panel of oncologists in Canada, Germany, and the UK, as follows¹⁹:

• race and/or ethnicity (Black, other, unknown versus White)
• age category (≥ 65 years versus < 65 years)
• ECOG Performance Status at treatment initiation (1 versus 0)
• histology at initial diagnosis (nonendometrioid, unknown versus endometrioid)
• FIGO stage (stages III and IV versus stages I and II)
• grade of disease at diagnosis (grades 3 and 4, unknown versus grades 1 and 2)
• number of prior platinum-based therapies (0 or 1 versus ≥ 2)
• prior surgery for study indication (yes versus no).

In MAIC report 1, assessment of the effect modification and prognostic value of each potential matching variable was done using a Cox proportional hazard model for each outcome (OS, PFS, and TTD). Cox regression models were fit separately for the GARNET data and the RWE data. Patient characteristics that exhibited an association at a level of significance of at least 0.1 in at least 1 of the 2 datasets were considered prognostic. P values evaluated in a multivariable model were obtained from a full model (including all variables listed above), and then backward stepwise selection, with a P ≤ 0.1 threshold, was applied. Effect-modifying status was assessed semi-empirically by comparing regression coefficients (or HRs) for the variable of interest from the 2 regression models fitted to the GARNET and the RWE dataset, where large differences between coefficients were considered to be potentially effect-modifying. The following variables were considered prognostic for each outcome¹⁹:

• OS: ECOG, race and/or ethnicity, grade, histology, stage at diagnosis, and prior surgery
• PFS: ECOG, and histology
• TTNT: Stage at diagnosis, grade, and prior surgery
• TTD: ECOG, race and/or ethnicity, grade, histology, and prior surgery.

The IPD from the GARNET trial were matched to the RWE cohort using the selected matching variables to estimate the above comparisons. A method of moments, as introduced by Signorovitch et al. (2010),⁵⁴ was used to allow a propensity score logistic regression model to be estimated without IPD for the RWE cohort.¹⁹ After matching, re-weighted outcome data from the GARNET trial and RWE cohort were compared. This included calculating KM survival estimates for OS, PFS, TTD, and TTNT at given time points after index treatment initiation, as well as creating KM curves. The primary end point analysis used a Cox proportional hazard model, with weights obtained using the MAIC method, to re-estimate HRs for dostarlimab in the GARNET trial versus the RWE cohort. In the treatment-specific RWE cohort, weighted Cox proportional hazard models were fitted, and OS HRs for dostarlimab versus the 5 comparator treatments of interest were re-estimated, together with 95% CIs and P values. For use in these regression models, OS data from the RWE cohort were digitized, and pseudo-IPD were constructed from KM curves using the method described by Guyot et al. (2012).⁵⁵ Log-cumulative hazards, as well as Schoenfeld residuals, were plotted for OS to examine the proportional hazards assumption (PHA) for the treatment group variable before and after matching.¹⁹

To assess the impact of each potential covariate on effective sample size (ESS) and estimated treatment effect, a “leave 1 out” approach was applied. This approach started with the calculation of ESS and a measure of treatment effect (HR for OS in the case of this study) for a full model, and then the calculation of the same 2 measures after removal of the covariate from the model; all remaining matching variables in the model were kept.¹⁹

The following matching scenarios were considered in both MAICs of the RWE cohorts¹⁹:

• Scenario 1 included all matching variables considered most relevant by expert opinion (grade, histology [1 patient from GARNET with unknown histology was removed], number of prior platinum-based therapies in the advanced or recurrent setting [patients from GARNET with 0 or > 2 therapies were removed]).
• Scenario 2 included all matching variables from scenario 1 except grade, because of poor reporting of grade in the RWE cohort (histology, number of prior platinum-based therapies in the advanced or recurrent setting). Scenario 2 was considered to be a sensitivity analysis of scenario 1 that was supposed to maintain a larger ESS.
• Scenario 3 included the matching variables that were considered prognostic, based on the empirical regression analyses, with the exception of grade. Moreover, ECOG Performance Status was only used as a matching variable in the sensitivity analysis for patients with an ECOG Performance Status of 0 or 1 from the RWE cohort.

GARNET Versus Published Literature MAIC Methods (MAIC Report 3)

Three studies were eligible for the MAIC (MAIC report 3) based on the SLR. GARNET was a phase I single-arm trial of dostarlimab. McMeekin et al. (2015)⁵⁶ was a global, open-label, phase III RCT comparing ixabepilone with paclitaxel or doxorubicin, Julius et al. (2013)⁵⁷ was a retrospective review of medical records in the US focusing on pegylated liposomal doxorubicin, and Mazgani et al. (2008)⁵⁸ was a retrospective analysis of carboplatin plus paclitaxel in patients with relapsed EC at 1 centre in Vancouver, British Columbia. The individual comparator studies were assessed for eligibility based on reporting of outcomes and appropriateness of the MAIC based on sample size and population (had to be adequately comparable to GARNET).¹⁹ Differences in eligibility criteria across studies were not discussed; however, MMR and MSI-H status was not accounted for in the comparator studies.

The specific (regression-type) modelling approach used in the analyses depended on the outcome data type. All P values presented were exploratory in nature and not conclusive. The tests performed were conducted post hoc, were not formally powered, and were not adjusted for multiplicity, so the reported P values are not confirmatory. When available, KM curves were digitized (using GetData Graph Digitizer 2.26) and the algorithm of Guyot et al. (2012)⁵⁵ was used to produce reconstructed IPD for the aggregate trial data. As the GARNET trial had less than 36 months of follow-up for OS, KM data from each digitized comparator study were restricted (censored) to 36 months or less to improve the comparability of results.¹⁹

For all studies, each KM curve was digitized separately and the reconstructed IPD were synthesized. Curves created from these reconstructed IPD were plotted against curves displaying the raw digitized survival probability time points, and visual comparisons were made against the published curves. These pseudo-IPD were considered alongside the IPD from the GARNET trial. A weighted Cox regression was then applied to the dataset, combining the reconstructed IPD with the GARNET data (no covariates other than treatment). To assess the assumption of proportional hazards in the Cox survival modelling between the treatment groups compared, a (MAIC-weighted) log-cumulative hazard plotted against log time was visually assessed for parallelism, and Schoenfeld residuals plots were produced. If the PHA was violated, an accelerated-failure-time model with a Weibull distribution was used.¹⁹

A method of moments analytical technique was used to produce individual patient-specific weights for patients from the GARNET trial to produce weighted prognostic means. Comparison of means between the aggregate trial data before and after adjustment of the GARNET trial, as well as the ESS were presented.¹⁹

In MAIC report 3, pre-specified prognostic factors were identified in the same manner as for the RWE cohort MAICs. Additionally, BMI, number of prior anti-cancer regimens, and MMR or MSI molecular profile were considered to be prognostic factors but were not included in weighting. Because of limitations of data availability, the GARNET data could only be matched on a small number of prognostic factors for McMeekin et al. (2015)⁵⁶ (age, ECOG, histology, race), Julius et al. (2013)⁵⁷ (age, race, number of prior chemotherapies), and Mazgani et al. (2008)⁵⁸ (histology).¹⁹ The distribution of baseline characteristics by treatment group were summarized before and after matching.

To minimize heterogeneity between separate study populations before statistical adjustments, the comparator study inclusion and exclusion criteria were applied to the GARNET IPD. The definition of outcomes in the comparator studies were also compared. Disease expertise was sought to identify the most likely criteria to have been used in each study.¹⁹ The inclusion criteria of GARNET that were modified to match the comparator studies and the resulting modified sample sizes are summarized in Table 22.

Table 22

Criteria and Sample Size Used in Each MAIC Published in the Literature (MAIC Report 3).

No sensitivity or subgroup analyses were conducted for MAIC report 3.

Table 23

Summary of MAIC Analysis Methods.

Results of MAIC Analyses

Results

GARNET Versus NCRAS Initial RWE Cohort (MAIC Report 1)

OS, PFS, and TTD in the GARNET ITT and initial RWE cohort are summarized in Table 25. For the initial RWE cohort, matching scenario 1, where studies were matched on histology, grade, and number of prior platinum-based therapies, resulted in the smallest ESS, with just 34 patients. The ESS was higher for scenario 2 and scenario 3, which did not include grade for 74 and 75 patients, respectively.¹⁹

Under all matching scenarios, dostarlimab was favoured over the current treatment paradigm for OS (scenario 1 HR = 0.52 [95% CI, 0.29 to 0.92]; scenario 2 HR = 0.35 [95% CI, 0.22 to 0.55]; scenario 3 HR = 0.31 [95% CI, 0.20 to 0.49]). Results for the sensitivity analysis of patients with an ECOG Performance Status of 0 or 1 were consistent with the base case. For PFS, 2 analyses were conducted: PFS from GARNET compared with TTNT from the RWE cohort; and PFS from GARNET compared with TTD from the RWE cohort. In general, the median PFS and PFS rate at 6, 12, and 18 months in all scenarios was higher than the comparator of current treatment paradigm, particularly for the analysis comparing GARNET PFS with RWE TTD. The median TTD and TTD rates at 6, 12, and 18 months were higher in all scenarios for dostarlimab than in the RWE cohort.¹⁹

The analysis of the leave-1-out approach to assess the impact of each potential covariate on ESS and estimated treatment effect is summarized for OS in Table 26. Omission of disease grade had the largest impact on ESS and estimated treatment effect, as a substantial proportion of patients in the RWE cohort had unknown grade.¹⁹

Table 24

Key Study and Baseline Characteristics in the GARNET and Comparator Studies.

Table 25

Survival Outcomes Before and After Matching — GARNET Versus Initial RWE Cohort (MAIC Report 1) .

Table 26

Summary of Leave-1-Out Approach — Initial RWE Cohort (Base-Case and Scenario Analyses, MAIC Report 1) .

GARNET Versus NCRAS Treatment-Specific Cohorts (MAIC Report 2)

For the treatment-specific MAICs, Table 27 summarizes the OS for dostarlimab compared with the 5 specific treatments in the RWE cohort. All results for scenario analyses using unweighted GARNET data were consistent with the primary analyses for OS. For paclitaxel monotherapy, the ESS in scenarios 1, 2, and 3 was 30, 72, and 63, respectively. Under all matching scenarios, the median OS (and HRs) favoured dostarlimab over paclitaxel monotherapy (scenario 1: HR = 0.36; 95% CI, 0.19 to 0.65; scenario 2: HR = 0.24; 95% CI, 0.15 to 0.40; scenario 3: HR = 0.18; 95% CI, 0.11 to 0.30), indicating that patients taking dostarlimab had a lower risk of death than patients receiving paclitaxel. The OS rates at 6, 12, and 18 months were greater for dostarlimab than for paclitaxel in the RWE cohort.¹⁹

For carboplatin plus paclitaxel, the ESS in scenarios 1, 2, and 3 was 36, 74, and 69, respectively. Dostarlimab was favoured over carboplatin plus paclitaxel only in scenarios 2 and 3 (scenario 2: HR = 0.48; 95% CI, 0.29 to 0.78; scenario 3: HR = 0.42; 95% CI, 0.25 to 0.68), but not scenario 1 (HR = 0.70; 95% CI, 0.39 to 1.28). The OS rates at 6, 12, and 18 months were greater for dostarlimab than for carboplatin plus paclitaxel in the RWE cohort.¹⁹

For carboplatin plus liposomal doxorubicin, the ESS in scenarios 1, 2, and 3 was 26, 69, and 74, respectively. Dostarlimab was favoured over carboplatin plus liposomal doxorubicin only in scenarios 2 and 3 (scenario 2: HR = 0.45; 95% CI, 0.27 to 0.76; scenario 3: HR = 0.40; 95% CI, 0.24 to 0.67), but not scenario 1 (HR = 0.74; 95% CI, 0.39 to 1.41). After matching, the 6-, 12-, and 18-month OS rates were greater for dostarlimab.¹⁹

For liposomal doxorubicin monotherapy, the ESS in scenarios 1, 2, and 3 was 37, 78, and 76, respectively. Under all matching scenarios, the median OS (and HRs) favoured dostarlimab over liposomal doxorubicin monotherapy (scenario 1: HR = 0.20; 95% CI, 0.09 to 0.44; scenario 2: HR = 0.17; 95% CI, 0.10 to 0.29; scenario 3: HR = 0.16; 95% CI, 0.09 to 0.27), indicating that patients taking dostarlimab had a lower risk of death than patients taking liposomal doxorubicin. Under all matching scenarios, the OS rates at 6 and 12 months were greater in the GARNET ITT cohort than in the RWE cohort.¹⁹

For carboplatin monotherapy, the ESS in scenarios 1, 2, and 3 was 23, 67, and 69, respectively. Dostarlimab was favoured over carboplatin in all matching scenarios (scenario 1: HR = 0.53; 95% CI, 0.29 to 0.98; scenario 2: HR = 0.32; 95% CI, 0.19 to 0.55; scenario 3: HR = 0.28; 95% CI, 0.16 to 0.48), indicating that patients taking dostarlimab had a lower risk of death than patients taking carboplatin monotherapy. Under all matching scenarios, the OS rates at 6, 12, and 18 months were greater in the GARNET ITT cohort compared with in the RWE cohort.¹⁹

Table 27

OS Before and After Matching — GARNET Versus Treatment-Specific RWE Cohort (MAIC Report 2).

Results for PFS for all comparators are summarized in Table 28. When compared with the RWE cohort TTNT and TTD, the median PFS was longer for dostarlimab than for all relevant comparators in all scenarios except scenario 1 of the TTNT comparison for paclitaxel monotherapy, carboplatin plus paclitaxel, carboplatin plus liposomal doxorubicin, and carboplatin monotherapy. For the TTD comparison, dostarlimab was favoured over all comparators in all scenarios except carboplatin monotherapy in scenario 1. The PFS rate for dostarlimab was greater in all scenarios and at all time points for the TTD comparison for all relevant comparators. In contrast with TTNT from the RWE cohort, the PFS rate for dostarlimab was greater than both paclitaxel and liposomal doxorubicin monotherapy at all time points and in all scenarios. The PFS rate was not greater at the 6-month assessment in any scenario for carboplatin plus paclitaxel, carboplatin plus liposomal doxorubicin, or carboplatin monotherapy. Otherwise, dostarlimab had a greater PFS rate at the 12- and 18-month time points in all scenarios.¹⁹

Table 28

PFS Before and After Matching — GARNET Versus Treatment-Specific RWE Cohort (MAIC Report 2).

For the outcome of TTD (Table 29), the median TTD and TTD rates were longer for dostarlimab than for all relevant comparators in the RWE cohort at all time points and across all 3 matching scenarios.¹⁹

Table 29

TTD Before and After Matching — GARNET Versus Treatment-Specific RWE Cohort (MAIC Report 2).

GARNET Versus Published Literature (MAIC Report 3)

Table 30 presents the ESS and weighted baseline characteristics for the MAICs versus published literature sources. A full list of baseline characteristics in GARNET after matching is summarized in Table 52. Seven patients were removed from GARNET who did not meet inclusion or exclusion criteria for McMeekin et al. (2015),⁵⁶ resulting in a MAIC base of 122 patients. These 122 patients from GARNET were then weighted for each prognostic variable available, resulting in an ESS of 87.274. Because of a lack of information on inclusion and exclusion criteria in Julius et al. (2013),⁵⁷ no patients were removed from the GARNET trial before weighting. After matching on available prognostic variables, the resulting ESS compared with Julius et al. (2013)⁵⁷ was 81.992. After the removal of 39 GARNET patients who did not meet the inclusion or exclusion criteria for Mazgani et al. (2008),⁵⁸ the resulting MAIC base was 90 patients, with an ESS after matching of 29.08.¹⁹

The covariate values used to derive the weights in the MAICs included age, ECOG, race, histology, and number of prior chemotherapies. Covariates not used in weighting varied across studies. Individual MAICs were conducted for each study identified in the literature.¹⁹

Table 30

Sample Size and Baseline Characteristics After Matching (MAIC Report 3).

The median OS and PFS for dostarlimab (GARNET) before and after matching compared with data from each individual study are summarized in Table 31. In all cases, the median OS was not reached before or after adjustment for the dostarlimab group. After adjustment, OS favoured dostarlimab over paclitaxel and over doxorubicin using data from the McMeekin et al. (2015)⁵⁶ study (HR = 0.407; 95% CI, 0.252 to 0.657) and over pegylated liposomal doxorubicin using data from the Julius et al. (2013)⁵⁷ study (HR = 0.287; 95% CI, 0.170 to 0.486). After adjustment, there was no difference in OS for dostarlimab compared with carboplatin plus paclitaxel using data from the Mazgani et al. (2008)⁵⁸ study (HR = 0.559; 95% CI, 0.256 to 1.220). Data for PFS were only available from the Mazgani et al. (2008)⁵⁸ study. After adjustment, there was no difference in PFS between dostarlimab and carboplatin plus paclitaxel.¹⁹

For OS, the test of the PHA did not show a violation for the comparisons with data from the McMeekin et al. (2015),⁵⁶ Julius et al. (2013),⁵⁷ or Mazgani et al. (2008)⁵⁸ studies (P = 0.28, 0.17, and 0.62, respectively). A Cox proportional hazards model with MAIC was fitted to the data. The HRs (0.407; 95% CI, 0.252 to 0.657 and 0.287; 95% CI, 0.170 to 0.486) indicated a difference between the 2 treatments for OS, with a 59.3%, and 71.3% lower risk of death for dostarlimab than for paclitaxel plus doxorubicin and for pegylated liposomal doxorubicin, respectively; median survival was not reported. For PFS, the PHA was violated (P = 0.014). As such, the accelerated-failure-time model with Weibull distribution was fitted to the data. Despite the accelerated-failure-time model, no difference was noted between dostarlimab and carboplatin plus paclitaxel in PFS.¹⁹

Table 31

Summary of OS and PFS Before and After Matching (MAIC Report 3).

The weighted ORR results for dostarlimab compared with paclitaxel or doxorubicin using the data from the McMeekin et al. (2015)⁵⁶ study and carboplatin plus paclitaxel using data from the Mazgani et al. (2008)⁵⁸ study are summarized in Table 32. For the comparison of dostarlimab with paclitaxel or doxorubicin in McMeekin et al. (2015),⁵⁶ dostarlimab was favoured over paclitaxel and over doxorubicin using data from the McMeekin et al. (2015)⁵⁶ study (OR = 0.202; 95% CI, 0.116 to 0.352). There was no difference between dostarlimab and paclitaxel plus carboplatin using data from the Mazgani et al. (2008)⁵⁸ study.¹⁹

Table 32

Summary of ORR Before and After Matching (MAIC Report 3).

SAEs were only compared for the GARNET and McMeekin et al. (2015)⁵⁶ studies and are summarized in Table 33. The results demonstrate no difference in SAEs between dostarlimab and paclitaxel or doxorubicin.

Table 33

Summary of SAEs Before and After Matching (MAIC Report 3).

Description of IPTW Analyses

In addition to the 3 reports of MAIC analyses, the sponsor submitted 3 IPTW and propensity score matching (PSM) analyses to compare clinical outcomes in patients who received dostarlimab in GARNET with patients who received other treatments for advanced or recurrent EC, as follows¹⁹:

• 1 analysis comparing IPD from GARNET with real-world patients receiving current treatment paradigms in a US real-world data cohort, using data from the Flatiron Health database (hereafter called IPTW report 1).
• 1 analysis comparing IPD from GARNET with a GARNET-like RWE cohort in the UK from the NCRAS database (the base-case RWE cohort described in the MAIC section, hereafter called IPTW report 2).
• 1 analysis comparing GARNET IPD with IPD from the doxorubicin arm of the ZoptEC trial, identified using the same SLR that was used to inform comparators for the previously described MAICs (hereafter called IPTW report 3).

Methods of ITPW Analyses

Study Selection Methods

Study selection methods varied in the 3 sponsor-submitted IPTW reports. The index trial in all cases was based on IPD from IA-2 for cohort A1 of part 2B of the phase I, single-arm GARNET trial (dMMR or MSI-H EC cohort; n = 129). The selection characteristics used to develop the GARNET-like cohorts for the IPTW analyses for IPTW reports 1 and 3 are summarized in Table 34. The selection criteria for IPTW report 2 (using data from the NCRAS database) are summarized in Table 20.¹⁹

In IPTW report 1, no SLR was conducted to identify relevant comparator studies. Instead, an external real-world control cohort was created, leveraging data from the Flatiron Health database, a US nationwide, longitudinal, demographically and geographically diverse database derived from electronic health record data from more than 280 cancer clinics. Patients with recurrent or advanced EC who had progressed on or after no more than 2 prior lines of systemic chemotherapy for advanced or recurrent disease, with at least 1 being platinum-based therapy, were identified from the Flatiron Health Endometrial Cancer Analytic Cohort. The inclusion and exclusion criteria for the real-world retrospective cohort study were divided into 2 categories: criteria that were used to create the Flatiron Health Endometrial Cancer Analytic Cohort (n = 4,264); and additional criteria (adult age; no more than 2 prior lines of therapy, including at least 1 line of platinum-based chemotherapy [prior hormone therapy was allowed but did not count toward line of therapy]; additional line of therapy after platinum therapy for advanced or recurrent EC, and an ECOG Performance Status of 0 or 1 at index) (n = 185 after applying all criteria).¹⁹

In IPTW report 2, no SLR was conducted to identify relevant comparator studies. Instead, data from NCRAS were used, as previously described, for the MAIC analyses (refer to Table 20). The initial RWE cohort base case (N = 999) was used for the IPTW analysis, with patients with an ECOG Performance Status of 0 or 1 included as a scenario analysis (N = 501). There were a few minor differences between patient-level data for the RWE cohort and aggregated data for the same cohort used previously to conduct the MAICs, including differences in age and race/ethnicity, although these changes would not likely affect the results.¹⁹

In IPTW report 3, in consultation with clinicians and payer advisors, doxorubicin was identified as 1 of the most used chemotherapies in EC patients who progressed on platinum-based treatments. A clinical trial investigating doxorubicin (ZoptEC) was identified from the same SLR that was used to identify relevant comparator studies for the MAIC comparing data from GARNET with that from published literature. ZoptEC was a phase III, open-label, randomized, active-controlled, international study evaluating the efficacy and safety of AEZS-108 and doxorubicin in advanced or recurrent EC patients who received 1 platinum plus taxane combination regimen as a first-line treatment and progressed. The trial was discontinued in 2017 as it did not achieve its primary or secondary end points. The transfer of the IPD for the doxorubicin arm (N = 255) of this trial was negotiated by the sponsor. Key differences in clinical criteria between the GARNET and ZoptEC trials included prior anti-cancer regimens (up to 2 prior lines in GARNET versus 1 in ZoptEC), biomarker status (dMMR or MSI-H in GARNET versus not evaluated in ZoptEC), and ECOG Performance Status (≤ 1 in GARNET versus ≤ 2 in ZoptEC). Although dMMR or MSI-H status was not known in ZoptEC, as previously noted, its influence as a prognostic or treatment-effect modifier in advanced or recurrent EC is unknown. Thus, it was assumed that the GARNET and ZoptEC patients had comparable molecular profiles.¹⁹

Table 34

Selection Criteria for GARNET-Like Subgroups and for Relevant Comparator Studies for Sponsor-Submitted IPTW Analyses.

The primary outcome of each analysis was OS, which was previously defined for the GARNET trial and the NCRAS RWE cohort (Table 21). In the Flatiron database cohort (IPTW report 1), OS was defined as the interval between the index date and the date of death, with the index date defined by the initiation date of therapy after post-platinum therapy (could consist of multiple lines of platinum therapy). OS was the only outcome for the IPTW analysis comparing GARNET with the NCRAS cohort (IPTW report 2). In the ZoptEC trial (IPTW report 3), OS was defined as time from date of the first dose of study treatment to the date of death by any cause.¹⁹

Other outcomes in the analyses using the comparison with the Flatiron database included TTD, previously defined for GARNET and icalculated as the time between the start date and end date of the index treatment, or censoring if the patient died, was lost to follow-up, or still on-therapy in the Flatiron database, as well as descriptive analyses of DoT and TTNT. In the comparison with the ZoptEC trial (IPTW report 3), additional outcomes included¹⁹:

• PFS, calculated as time from the first dose to the previous date of assessment of progression or death by any cause in the absence of progression based on the time of first documentation of disease progression
• ORR, defined as the proportion of patients that achieved BOR of CR or PR, per RECIST 1.1
• DOR, defined as time from the first documentation of overall response leading to a confirmed CR or PR, when confirmation is required by RECIST 1.1, until time of the first documentation of overall response that included disease progression or death
• TTD in HRQoL, defined as time from the first cycle to the earliest time with a decrease of at least 10 points (deterioration) from baseline in the global HRQoL; however, no rationale for this threshold was provided
• AEs, including TEAEs, SAEs.

IPTW Analysis Methods

The method of each IPTW analysis varied. A summary of the analysis methods in each MAIC is provided below.

GARNET Versus Flatiron Cohort Methods (IPTW Report 1)

Five patients in GARNET who received anti-PD-L1/2 therapies after dostarlimab were excluded because of differences in availability across countries. As such, the analysis set from the GARNET trial was the safety cohort (n = 124), as was the efficacy analysis dataset (n = 103). After the application of inclusion and exclusion criteria to the 4,564 patients from the Flatiron Health Endometrial Cancer Analytic Cohort, the final study population of the real-world retrospective cohort study was 185 patients.¹⁹

A feasibility assessment was conducted to determine whether the retrospective real-world cohort was suitable for use as an external control arm to the GARNET dMMR or MSI-H EC cohort. Feasibility was based on assessments of data quality and information availability, consideration of sample size, availability of key prognostic factors and outcomes, and comparability with the GARNET trial. Overall, the feasibility assessment revealed that the definitions of all prognostic variables were comparable between cohorts; however, there were noted differences in reporting of ECOG Performance Status, BMI, number of prior anti-cancer regimens, prior surgery, and MMR or MSI-H status. Notably, MMR or MSI status was not known for most patients, and there were concerns about the under-reporting of prior surgeries in the Flatiron cohort. Based on the results of the feasibility assessment, a decision was made to proceed with the analysis.¹⁹

Prognostic variables in EC were identified and validated in accordance with the methods described previously for the initial RWE cohort MAIC. The prognostic factors used as covariates in the models included race, age, ECOG Performance Status, histology, FIGO stage, BMI, disease grade, number of prior platinum-based regimens, and prior surgery.¹⁹

The KM method was used to describe the distribution of OS and TTD by cohort. Survival curves were presented graphically. Median OS (95% CI) and estimated survival probabilities at specific time points were reported by cohort. The HR for OS and its 95% CI for the cohort variable (GARNET versus real-world cohort) were calculated from a Cox proportional hazard model in which cohort was included as a single covariate. The corresponding P value was also reported. The PHA was checked graphically by log-cumulative hazard plots for the covariate and by Schoenfeld residual plots. To assess whether the PHA was met, an interaction between time and the study variable was added to the Cox models as a time-dependent covariate.¹⁹

Covariate-adjusted comparisons of OS and TTD between cohorts were performed using regression modelling (only for OS), IPTW, and PSM. Propensity scores for each patient in the real-world control cohort were estimated as a patient’s predicted probability of being assigned to the GARNET cohort, which was estimated from a logistic regression model. The propensity score model included the key prognostic variables as covariates.¹⁹

Propensity scores for each patient were estimated from a logistic regression model as a patient’s predicted probability of being assigned to the GARNET cohort. Variable selection in the propensity score model considered clinical relevance (determined by key opinion leaders and medical experts), frequency of reporting in the literature, and strength of association with OS. Two propensity score models were built¹⁹:

• A “lean” propensity score model that only included a small number of the most relevant of the variables; namely, histology, grade of disease at initial EC diagnosis, ECOG Performance Status, and number of prior platinum-based therapies in the advanced or recurrent setting.
• A propensity score model that included all variables identified except prior surgery for EC.

The overlap of the distribution of the propensity scores across the GARNET cohort and real-world control cohort was assessed with a histogram of the estimated propensity scores in the 2 cohorts. PSM formed pairs of subjects from the GARNET and the real-world cohorts, such that matched subjects had similar propensity score values. To account for the imbalance in sample size between the 2 cohorts, a 1:2 matching ratio was used, where 2 patients from the real-world cohort were matched to each patient from the GARNET cohort. The matching was based on the greedy nearest neighbourhood matching without replacement, using a caliper width of 0.2. After matching, OS and TTD were compared between cohorts in the matched sample. This included the calculation of KM curves and estimation of median survival and corresponding 95% CIs.¹⁹

Adjusted HRs were obtained from multivariable Cox regression models, and included the prognostic variables as covariates in addition to the cohort variable. The same 2 sets of covariates as in the propensity score models were considered in the regression models.¹⁹

IPTW was performed using weights from estimated propensity scores. Weights were calculated from propensity scores so that resulting estimates referred to the average treatment effect (ATE). To assess balance in baseline characteristics between cohorts before and after weighting and matching, absolute standardized differences were calculated for each covariate, in which the standardized difference was defined as the difference in means or proportions divided by the pooled SD. Using the IPTW method, OS and TTD were compared between the cohorts in the weighted sample. This included the creation of weighted KM curves and the estimation of weighted median survival and corresponding CIs. A weighted Cox regression model for OS was also fitted to estimate adjusted HRs for the cohort variable.¹⁹

GARNET Versus NCRAS RWE Cohort Methods (IPTW Report 2)

In the analyses comparing GARNET with the RWE cohort from NCRAS conducted to assess balance in baseline characteristics between cohorts before weighting, absolute standardized differences were calculated for each covariate, in which the standardized difference is defined as the difference in means or proportions divided by the pooled SD. Statistical differences in baseline characteristics between the 2 treatment cohorts were evaluated using t-tests for continuous variables and chi-square tests for categorical outcomes.¹⁹

As above, the KM method was used to describe median (95% CI) OS and estimated survival probabilities by cohort. The HR for OS and its 95% CI for the cohort variable (GARNET versus real-world cohort) was calculated from a Cox proportional hazard model that included cohort as a single covariate. The PHA was checked graphically with log-cumulative hazard plots for the covariate and plots of Schoenfeld residuals for the treatment variable over time. In addition, Schoenfeld global tests that evaluated independence between Schoenfeld residuals and time were also performed.¹⁹

Covariate-adjusted comparisons of OS between cohorts were performed using IPTW based on the propensity score. Propensity scores for each patient were estimated as a patient’s predicted probability of being assigned to the GARNET cohort, which was estimated from a logistic regression model. The propensity score model included the key prognostic variables, identified following the method previously described for the RWE cohort MAIC, as covariates.¹⁹

Three different propensity score models were fitted. The sets of covariates used in the 3 propensity score models were identical to the sets of matching variables that defined the 3 MAIC scenarios described previously. The overlap of the distribution of propensity scores across the GARNET trial cohort and the RWE cohort was assessed with a histogram distribution graph of propensity scores across cohorts.¹⁹

Adjustment for potential imbalances in relevant prognostic variables between cohorts was done with the IPTW method. In the principal analysis, weights were calculated from propensity scores so that resulting estimates referred to the ATE. As a secondary analysis, weights were calculated from propensity scores so that resulting estimates referred to the ATE for the controls. The use of the ATE for controls perspective was motivated by the fact that this approach was similar to the weighting approach previously used in the MAIC, where each patient in the control cohort was assigned a fixed weight of 1.¹⁹

Using the IPTW method, OS was compared between the cohorts in the weighted sample. This included the creation of weighted KM curves and the estimation of weighted median survival and corresponding CIs. In the weighted KM curves, numbers at risk represent the sums of individual weights of patients at risk, rather than the number of patients at risk. A weighted Cox regression model was also fitted to estimate adjusted HRs (and 95% CIs, P values) for the cohort variable.¹⁹

GARNET Versus ZoptEC Methods (IPTW Report 3)

To facilitate a comparison between the single-arm GARNET trial and control patients using doxorubicin in the ZoptEC RCT, IPD from GARNET and ZoptEC were merged to create a comparator control arm.¹⁹

To make the baseline characteristics between the 2 trials comparable before statistical analysis, exclusion criteria were applied to each trial. Patients in the ZoptEC comparator trial who would have been ineligible in GARNET were removed before analysis if such criteria were easily determined from the dataset. Patients were excluded from the ZoptEC trial if they had a follow-up longer than 36 months, and patients were excluded if they did not have an ECOG Performance Status score of 0 or 1. Patients were excluded from the GARNET trial if they had previously received more than 1 prior platinum-based therapy. The main analysis dataset included 92 patients from GARNET and 233 patients from ZoptEC, and was used for OS, PFS, and ORR.¹⁹ A summary of datasets in the analysis is provided in Table 35.

A sensitivity analysis that included all patients from the GARNET and ZoptEC trials was conducted to ensure that the removal of patients did not violate the positivity or exchangeability assumption for the IPTW method. The positivity assumption can be empirically verified by showing there is a positive probability of each treatment for each set of covariates, whereas the exchangeability assumption assesses the potential impact of unmeasured confounding on the results using the methods of Mortimer et al. (2005)⁵⁹ and Brumback et al. (2004),⁶⁰ respectively. Additional sensitivity analyses included 1 that used the safety analysis dataset to assess whether this would affect the OS results; another of TTD in HRQoL was conducted using a 15-point decrease as the threshold.¹⁹

Table 35

Summary of Included Patients in Each Analysis Dataset.

The main analysis dataset sample size — 233 for doxorubicin and 92 for dostarlimab — contained all available patients after the exclusion criteria were applied; thus, a sample size calculation was not applicable. However, a power analysis for OS was performed. If the HR for OS was greater than 0.65, the analysis would be underpowered; however, if the HR was less than or equal to 0.65, the analysis would be sufficiently powered. As with any statistical model using IPTW, power is decreased because IPTW inflates type I errors. Therefore, a stabilized IPTW was used, which is better suited in handling the inflation of type I errors and can also minimize any issues with statistical power.¹⁹

In the comparison between GARNET and the ZoptEC study, prognostic variables were identified and ranked with methods previously described for the RWE cohort published literature MAICs.¹⁹

A multivariable method with interaction terms for the assessment of effect modification was used for this analysis. For OS, Cox proportional hazards were used with covariates: treatment, candidate for effect modification (e.g., age or ECOG), and the interaction term (treatment × candidate of effect modification). If the P value for interaction was less than 0.05, the variable was considered an effect modifier. If effect modification was present, a subgroup analysis for each level of the effect modifier, in addition to the main analysis, was conducted.¹⁹

Because of the lack of randomization, there was a need to control for confounding bias at baseline, so an IPTW method was chosen. PSM was not chosen because there was a restricted number of patients in each arm coming from the clinical trial. A stabilized weight was estimated for each patient to reduce inflation of the type I error rate or to unintentionally increase statistical power. Once weighted, a Cox proportional hazards regression model with IPTW was used for the comparison of OS between dostarlimab and doxorubicin. KM analysis was used for PFS, DOR, and TTD in HRQoL. For binary outcomes such as an ORR, descriptive analyses were used. Specifically, the Clopper-Pearson method was used to calculate the 95% CI of the proportion. All assumptions, such as unmeasured confounding, positivity, and consistency, were checked in the validation of the IPTW method.¹⁹

Results of IPTW Analyses

Summary of Included Studies

Key study design characteristics and sources of heterogeneity of the studies included in the IPTW analyses are summarized in Table 36. There was considerable variation in the design of the 4 studies that were used in the IPTW analyses. The GARNET trial evaluated patients with dMMR or MSI-H EC, whereas the Flatiron, NCRAS, and ZoptEC studies included patients with advanced or recurrent EC, regardless of biomarker status. A total of 128 (69.2%) patients in the Flatiron database had unknown MMR or MSI-H status at index. There was also variation in the treatments included, with GARNET evaluating dostarlimab, the RWE cohorts evaluating the current treatment paradigms in advanced or recurrent EC, and the ZoptEC trial evaluating doxorubicin. Eligible patients typically had an ECOG Performance Status of 0 or 1; however, the ZoptEC trial enrolled patients with an ECOG Performance Status of 2, and ECOG status was unknown in most of the NCRAS RWE cohort. The line of therapy also varied across studies, with GARNET and the Flatiron RWE cohorts including patients with no more than 2 lines of therapy, 1 of which was a platinum doublet. The NCRAS real-world cohort used platinum doublet therapy as the key milestone regarding lines of therapy and did not analyze the lines of therapy before advanced or recurrent diagnosis. Moreover, the ZoptEC study only required 1 prior line of therapy.¹⁹ No consideration was noted regarding the differences in dosage, treatment or follow-up duration, or supportive care or cointerventions in the included studies.

Table 36

Key Study Characteristics in the GARNET and Comparator Studies.

GARNET Versus Flatiron Cohort (IPTW Report 1)

The baseline characteristics before weighting in the IPTW analysis of GARNET and the Flatiron RWE cohort are summarized in Table 37. In total, the GARNET safety population included 124 patients and the real-world cohort included 185 patients. There were 5 patients in GARNET (n = 129) not eligible for the comparison and excluded who received anti-PD-L1/2 after dostarlimab. In general, the populations of the GARNET trial and the Flatiron RWE cohort were relatively similar, except the GARNET trial had a higher proportion of White patients (75.0% versus 61.1%), endometrioid histology (66.1% versus 57.3%), FIGO stage I/II (43.5% versus 35.7%), and prior surgery (89.54% versus 56.2%). There was a substantial difference in disease grade between the GARNET and Flatiron cohorts because of the large amount of missing data for grade at diagnosis in the Flatiron cohort (38.4%). Comparability of the following variables between the 2 cohorts was limited because of discrepancies in variable definitions or data under-reporting of surgery and radiation in the Flatiron database: prior surgery for EC, history of radiation therapy before second-line therapy, and duration of the platinum- and progression-free interval at second line (because of the unavailability of response data in Flatiron, the platinum-free interval was measured differently in Flatiron and GARNET). The most frequent first-line regimen in the Flatiron cohort was carboplatin plus paclitaxel, received by the majority of patients (72.4%). During second-line treatment, the top 3 regimens were pegylated liposomal doxorubicin (17.8%), carboplatin plus paclitaxel (15.1%), and carboplatin plus docetaxel (9.7%). Only 105 patients received third-line therapy, with the most common being pegylated liposomal doxorubicin (15.2%), bevacizumab (9.5%), and carboplatin plus paclitaxel (7.6%).¹⁹

GARNET Versus NCRAS RWE Cohort (IPTW Report 2)

The baseline characteristics before weighting of the IPTW analysis of GARNET and the NCRAS RWE cohort are summarized in Table 37. In total, the GARNET cohort included 129 patients and the RWE base-case cohort included 999 patients. Patients in the GARNET trial were younger (51.2% versus 44.5% < 65 years). Overall, there were key differences in most characteristics of the patient populations; compared with the NCRAS cohort, GARNET had fewer White patients (76.0% versus 84.2%), and a greater proportion of patients with endometrioid histology (69.8% versus 42.4%), FIGO stage I and II disease (44.2% versus 22.1%), an ECOG Performance Status of 0 (42.6% versus 32.0%) or 1 (57.4% versus 18.1%), and a disease of grade 1 and 2 (67.4% versus 27.4%). The number of prior platinum-based therapies in the advanced or recurrent setting was 1 for all patients in the NCRAS RWE cohort, whereas it was 0 or 2 or more for 1.6% and 13.2% of patients in the GARNET cohort, respectively.¹⁹

GARNET Versus ZoptEC (IPTW Report 3)

The baseline characteristics before weighting of the IPTW analysis of GARNET and the ZoptEC trial are summarized in Table 37. The sponsor noted that the baseline characteristics were similar enough to warrant comparison. There were fewer White patients in GARNET than in the ZoptEC trial (76.0% versus 94.1%), and more patients with FIGO stage III and IV disease (85.3% versus 36.9%). The proportion of patients with endometrioid histology was similar. Comparability of some key variables, including BMI, disease grade, number of prior platinum therapies, and prior surgery, was limited because of a lack of reported data.¹⁹

Table 37

Baseline Characteristics of Patients by Cohort Before Weighting.

Results

GARNET Versus Flatiron Cohort (IPTW Report 1)

Propensity scores for each patient were estimated from a logistic regression model as a patient’s predicted probability of being assigned to the GARNET trial. Results from the lean and full propensity score models fitted on the GARNET trial compared with the Flatiron cohort are summarized in Table 53. The overlap of the distribution of the propensity scores was assessed with a histogram distribution graph of propensity scores across cohorts (refer to Figure 12 and Figure 13 in Appendix 4). The lean model demonstrated greater overlap than the full model, so only the lean model was used.¹⁹

The distributions of baseline and prognostic factors considered for ITC analyses for GARNET and the Flatiron cohort after lean model IPTW and PSM are summarized in Table 54. Before and after matching, baseline and prognostic variables of race, FIGO, and BMI were different in GARNET and the Flatiron cohort. The matched GARNET cohort had a sample size of 93, and the matched real-world cohort had a sample size of 103.¹⁹

Results for OS and TTD before and after IPTW and PSM are summarized in Table 38. Median OS after IPTW was not estimable in GARNET (95% CI, 15.4 to not estimable) and was 13.1 (95% CI, 8.3 to 15.9) in the Flatiron cohort. Median OS after PSM was not estimable in GARNET (95% CI, 17.1 to not estimable) and was 15.9 (95% CI, 10.4 to 33.2) in the Flatiron cohort. Survival rates were higher for dostarlimab than for the Flatiron current treatment paradigm cohort; after IPTW, 24-month OS rates were 0.529 (95% CI, 0.367 to 0.667) in the GARNET cohort and 0.338 (95% CI, 0.262 to 0.415) in the real-world cohort. After PSM, 24-month OS rates were 0.536 (95% CI, 0.387 to 0.664) in the GARNET cohort and 0.410 (95% CI, 0.305 to 0.512) in the real-world cohort. Although the point estimates differ, 95% CIs overlapped, indicating no difference in survival rates.¹⁹

Cox proportional hazards regression models were fitted for OS to examine the association between cohort (GARNET versus Flatiron) and the risk of death. The PHA was checked graphically with a log-cumulative hazard plot and with a plot of Schoenfeld residuals for the treatment variable over time. In addition, an interaction between time and study was added to the Cox model as a time-dependent covariate. If the interaction term was not statistically significant, it was reasonable to assume that the PHA was valid.¹⁹

Unadjusted Cox proportional hazards regression models for both GARNET and the Flatiron cohort suggested that GARNET patients had a lower risk.(HR = 0.447; 95% CI, 0.305 to 0.653). When interaction for study and time were included, there was no difference in risk of death between studies (HR = 0.989; 95% CI, 0.927 to 1.055). For the model fitted after IPTW, the interaction term was significant when only the study variable was included (HR = 0.559; 95% CI, 0.385 to 0.812) but was not statistically significant when study and time variables were included. After PSM, the interaction terms were not statistically significant (HR for study = 0.653; 95% CI, 0.413 to 1.033; HR for study and time = 1.030; 95% CI, 0.955 to 1.111) (Table 55), so it was reasonable to assume that the PHA holds.¹⁹

Results from the multivariable Cox proportional hazards model for OS where 2 sets of covariates were adjusted are summarized in Table 56. The interaction term of study and time was not statistically significant in either adjustment scenario, which indicated that the PHA was valid for study. The results suggested that the study variable favoured the GARNET trial for risk of death after adjustment for the covariates in both adjustment scenarios. Table 57 shows the Cox proportional hazards model results for OS adjusted for covariates and propensity scores. For the study variable, the HRs adjusted for propensity score and for propensity score and covariates were consistent.¹⁹

For the exploratory outcome of TTD before and after IPTW and PSM, the results were congruent with the primary analysis of OS, with higher TTD rates at 24 months in the GARNET trial than in the Flatiron database (Table 38). The TTD rate after IPTW was 0.480 (95% CI, 0.319 to 0.625) in GARNET and 0.149 (95% CI, 0.081 to 0.237) in the real-world cohort. After PSM, the TTD rate at 24 months was 0.397 (95% CI, 0.272 to 0.520) in GARNET and 0.173 (95% CI, 0.084 to 0.287) in the real-world cohort; however, the overlap in the 95% CIs indicated that there was no difference.¹⁹

Table 38

OS Before and After IPTW and PSM — GARNET Versus Flatiron Real-World Cohort (IPTW Report 1).

GARNET Versus NCRAS RWE Cohort (IPTW Report 2)

Baseline characteristics of the GARNET and NCRAS base-case cohorts after IPTW from 3 propensity score models are summarized in Table 58. Propensity score model 1 and 2 removed patients with unknown histology and patients with 0 or at least 2 prior platinum-based therapies from the GARNET cohort, which resulted in a sample size of 109 for the GARNET cohort; propensity score model 3 only removed patients with unknown histology from the GARNET cohort, which resulted in a sample size of 128 for the GARNET cohort. The standardized differences for variables used in the propensity score models were smaller after IPTW, compared with unweighted differences.¹⁹

Prior to IPTW, median OS could not be estimated for the GARNET cohort and was 10.3 months (95% CI, 9.2 to 11.1) for the RWE base-case cohort. The survival rates remained higher in the GARNET than in the RWE cohort during the study. At month 24, the survival rate was 0.578 (95% CI, 0.450 to 0.686) in the GARNET cohort and 0.215 (95% CI, 0.186 to 0.244) in the RWE base-case cohort. OS after IPTW that compared ATE weights for GARNET with the NCRAS RWE base-case cohort is summarized in Table 39. After IPTW, results for all propensity score models were consistent, with a greater median OS for dostarlimab than for the current treatment paradigm in the NCRAS RWE cohort (median OS was not evaluable at follow-up in GARNET and was 10.3 for all NCRAS models). Across all models after weighting, survival rates at 24 months ranged from 0.517 (95% CI, 0.326 to 0.678) to 0.606 (95% CI, 0.440 to 0.736) for GARNET and from 0.216 (95% CI, 0.187 to 0.246) to 0.218 (95% CI, 0.189 to 0.248) for the NCRAS RWE cohort. Using ATE for controls weights, results were consistent with ATE. Results of sensitivity analyses in patients with a ECOG Performance Status of 0 or 1 were consistent with the primary analysis.¹⁹

The HR for the cohort variable (GARNET versus real-world cohort) were calculated from a Cox proportional hazards model that included cohort as a single covariate. In all models, the HRs ranged from 0.310 (95% CI, 0.271 to 0.355) to 0.438 (95% CI, 0.386 to 0.498) (Table 39).¹⁹

The PHA was checked graphically with means of log-cumulative hazard plots for the covariate and plots of Schoenfeld residuals for the treatment variable over time. Both the graphs and the tests suggest that the PHA was valid for the unadjusted Cox models, as well as for the weighted Cox models based on weights calculated from propensity score model 3. The weighted Cox models based on weights calculated from propensity score models 1 and 2 do not seem to meet the PHA.¹⁹

Table 39

OS Before and After IPTW — GARNET Versus RWE Base-Case Cohort Propensity Score Models (ATE; IPTW Report 2).

GARNET Versus ZoptEC (IPTW Report 3)

The median OS of dostarlimab was not reached (95% CI, 17.150 to not reached), whereas the median OS for doxorubicin was 11.039 months (95% CI, 9.988 to 13.010). A naive HR was calculated using the lower-bound CI for dostarlimab, resulting in an unadjusted HR of 0.644. The main analysis was performed using a Cox proportional hazards model with stabilized IPTW to estimate the HR for OS (dostarlimab versus doxorubicin). A proportionality test was conducted to ensure that the PHA was not violated. As the P value was greater than 0.05, the PHA was supported, and the adjusted analysis used a Cox proportional hazards model with stabilized IPTW. KM curves for OS before and after adjustment are shown in Table 37. After adjusted stabilized IPTW, results of the Cox proportional hazards model suggested that dostarlimab was favoured over doxorubicin for OS (HR, 0.409 [95% CI, 0.277 to 0.605]). Survival time was significantly longer for dostarlimab, with median OS not reached (95% CI, 18.004 to not reached); for doxorubicin, median OS was 11.170 months (95% CI, 9.988 to 13.076).¹⁹

Figure 9

KM Curves for OS Without and With Adjusted Stabilized IPTW (IPTW Report 3).

A sensitivity analysis was conducted using the full safety analysis population from the GARNET (n = 129) and ZoptEC (n = 249) trials. Results were consistent with the main analysis, with dostarlimab favoured over doxorubicin for OS (HR, 0.403 [95% CI, 0.280 to 0.581]).¹⁹

ClinicalTrials.gov

Produced by the US National Library of Medicine. Targeted search used to capture registered clinical trials.

[Search terms – dostarlimab or Jemperli]

WHO ICTRP

International Clinical Trials Registry Platform, produced by WHO. Targeted search used to capture registered clinical trials.

[Search terms – dostarlimab or Jemperli]

Health Canada’s Clinical Trials Database

Produced by Health Canada. Targeted search used to capture registered clinical trials.

[Search terms – dostarlimab or Jemperli]

EU Clinical Trials Register

European Union Clinical Trials Register, produced by the European Union. Targeted search used to capture registered clinical trials.

[Search terms – dostarlimab or Jemperli]

Description

The EORTC QLQ-C30 is a self-reported instrument designed to measure the physical, psychological, and social functions of patients with cancers.⁴⁸ The EORTC QLQ-C30 consists of 30 items that are scored to create 5 multi-item functional scales, 3 multi-item symptom scales, 6 single-item symptom scales, and 2 global quality of life (QoL) scales (Table 43) Version 3.0 of the questionnaire is the most current version and has been in use since December of 1997.^75,76 It is intended for use in the adult population only.⁴⁸

Table 43

EORTC QLQ-C30 Scales.

Scoring

The EORTC QLQ-C30 uses a 1-week recall period to assess function and symptoms.⁷⁶ Twenty-eight questions are scored on a 4-point Likert scale (1: not at all; 2: a little; 3: quite a bit; 4: very much). The 2 questions that make up the global HRQoL scale are scored on a 7-point Likert scale with anchors at 1 (“very poor”) and 7 (“excellent”).

Raw scores for each scale are computed as the average of the items that contribute to a particular scale.⁷⁶ This scaling approach is based on the assumption that it is appropriate to provide equal weighting to each item that comprises a scale. There is also an assumption that, for each item, the interval between response options is equal (for example, the difference in score between “not at all” and “a little” is the same as “a little” and “quite a bit,” at a value of 1 unit). Each raw scale score is converted to a standardized score that ranges from 0 to 100 using a linear transformation, with a higher score reflecting better function on the function scales, higher symptoms on the symptom scales, and better HRQoL (i.e., higher scores simply reflect higher levels of response on that scale). Thus, a decline in score on the symptom scale would reflect an improvement, whereas an increase in score on the function and HRQoL scales would reflect an improvement. According to the EORTC QLQ-C30 scoring algorithm, if there are missing items for a scale (i.e., the patient did not provide a response), the score for the scale can still be computed if there are responses for at least one-half of the items. In calculating the scale score, the missing items are simply ignored — an approach that assumes that the missing items have values equal to the average of those items for what the respondent completed.

Psychometric Properties

The EORTC QLQ-C30 was originally validated in patients with lung cancer and head and neck cancer from various European and North American countries, as well as from Australia.^62,77,78 The scales of the EORTC QLQ-C30 have been found to assess distinct components of HRQoL; distinguishing between patients with different performance status and degrees of weight loss, and responsive to change over time.^62,77,78

A literature search was conducted to identify validation information of the EORTC QLQ-C30 in patients with EC and none were identified. Of note, there is a validated version of the EORTC QLQ that was developed specific for endometrial cancer: EORTC QLQ-EN24.⁷⁹ The EORTC QLQ-EN24 was designed to assess disease and treatment-specific aspects of the HRQoL of patients with EC. A validation study of the Mexican-Spanish version of the EORTC QLQ-EN24 in 189 patients with EC included a brief validation of the QLQ-C30.⁸⁰ The study confirmed the internal consistency and reliability of the QLQ-C30 and found that its convergent and discriminant validity (Cronbach alpha range = 0.77 to 0.89) was consistent with its original report.⁸⁰

In a further validation study of the EORTC QLQ-C30, patients with breast cancer (n = 121), ovarian cancer (n = 111) lung cancer (n = 160) and a heterogenous group of other cancers (n = 121) completed the questionnaire before and on day 8 of chemotherapy.⁶⁴ The item-domain correlations of the EORTC QLQ-C30 were not different across the primary tumour sites (i.e., breast, ovary or lung). The correlations for all items, except for item 5 (whether the responders needed help with eating, dressing, washing, or using the toilet) and the physical function domain (r = −0.3), were highly correlated within their own domain than with any other domains (r = −0.65 to 0.95). At day 8 of chemotherapy, the item-domain for item 5 and the physical function domain was 0.49 for the entire group and ranged from −0.38 for ovarian cancer to −0.55 for breast cancer. These higher values at day 8 suggest that item 5 was more relevant in the week after chemotherapy than before chemotherapy.⁶⁴ Similarly, items asking about vomiting showed a higher correlation with domains for nausea/vomiting on day 8 after chemotherapy (r = 0.89) than before chemotherapy (r = 0.74). The questionnaire also demonstrated good internal consistency for most domains at baseline and at day 8 (Cronbach alpha > 0.70). However, values were < 0.70 at baseline and day 8 for role function (0.66 and 0.53, respectively) and cognitive function (0.63 and 0.58, respectively). An examination of the discrimination of the domain scores according to primary cancer site found that the mean scores at baseline were not significantly different from each other in all groups for emotional function, cognitive function, or nausea/vomiting. However, the mean scores for each of the other domains differed between the group, with patients with breast cancer tending to have better physical function, role function, social function, less fatigue, and pain and better global HRQoL. Patients with lung and ovarian cancer reported lower scores for all these domains, with patients with ovarian cancer reporting the lowest scores for all domains. After chemotherapy, many of the differences seen at baseline between the groups were no longer evident. At day 8, patients with ovarian cancer has the smallest magnitude of change, with no change in role function, social function and global HRQoL, while being the only cancer group reporting a significant improvement in pain.⁶⁴

An analysis of data from a Canadian RCT of paclitaxel and cisplatin versus cyclophosphamide and cisplatin in the treatment of 153 patients with epithelial ovarian cancer found the EORTC QLQ-C30 adequately assessed the effect of expected toxicities on patients HRQoL during and after treatment.⁶⁵ At baseline, before the initiation of treatment, there was close agreement in the “mild or none” category between the symptoms recorded on case report forms (CRFs) and paired EORTC QLQ-C30 questions. The greatest degree of agreement ranged between 0.80 (95% CI, 0.75 to 0.86) to 0.98 (95% CI, 0.92 to 0.99). The pairing of lethargy with Question 18, and mood with Question 22 were slightly weaker in agreement compared to the other pairs at 0.72 and 0.73, respectively. The weakest pairs were constipation with Question 16, and lethargy with Question 18 at 0.44 and 0.44, respectively. During treatment and at the end of cycles 3 and 6, all but 1 symptom and HRQoL pairs demonstrated marked agreement ranging from 0.71 to 0.93. The 1 exception was the pair assessing symptom hair loss and Question 42 with a degree of agreement of 0.50 and 0.37 at cycles 3 and 6, respectively. A regression model predicting global HRQoL scores based on baseline grades of the most frequently observed toxicities and scores corresponding to HRQoL question found that the questions related to motor weakness (question 12), anorexia (question 13), mood (question 24), gastrointestinal pain (question 40) and vomiting (question 15) explained 60% of the variance in baseline global HRQoL on the EROTC QLQ-C30. When patients were off chemotherapy, 78% of symptoms and HRQoL pairs had high level of agreement (> 0.80).

Minimal Important Difference

A literature search was conducted to identify the MID of the EORTC QLQ-C30 in patients with EC and none were identified. Below is a summary of the MID of the EORTC QLQ-C30 in patients with cancer in general.

Change in the EORTC QLQ-C30 may be interpreted in terms of small, moderate or large changes in HRQoL.⁶⁷ A study of patients with breast cancer and small cell lung cancer estimated that a clinically relevant change in score on any of the EORTC QLQ-C30 scales to be 10 points.⁶⁷ Using an anchor-based approach to estimate the MID in which patients who reported “a little” change (for better or worse) on the subjective significance questionnaire (SSQ) had corresponding changes on a function or symptom scale of the EORTC QLQ-C30 of approximately 5 to 10 points. Patients who reported a “moderate” change had corresponding changes in the EORTC QLQ-C30 of about 10 to 20 points, and those who reported being “very much” changed had corresponding changes of more than 20 points.

A Canadian study estimated the MID for the EORTC QLQ-C30 among 369 patients with advanced cancer, the most common cancer being breast cancer, followed by lung, prostate, gastrointestinal, renal cell, and other cancers.⁶⁸ Patients completed the questionnaire at baseline and 1-month post-radiation. Using both an anchor- and distribution-based methods for improvement and deterioration, 2 anchors of overall health and overall HRQoL were used, both taken directly from the EORTC QLQ-C30 (questions 29 and 30) where patients rated their overall health and HRQoL themselves. Improvement and deterioration were categorized as an increase or decrease by 2 units to account for the natural fluctuation of patient scoring. With these 2 anchors, the estimated MIDs across all EORTC QLQ-C30 scales ranged from 9.1 units to 23.5 units for improvement, and from 7.2 units to 13.5 units for deterioration. Distribution-based estimates were closest to 0.5 SD.

Description

The EQ-5D-5L is a generic self-reported HRQoL outcome measure that may be applied to a variety of health conditions and treatments.^49,69 The first 2 components of the EQ-5D-5L assesses 5 domains: mobility, self-care, usual activities, pain/discomfort and anxiety/depression. Each domain has 5 levels: no problem; slight problems; moderate problems; severe problems; and extreme problems. A descriptive system that classifies respondents (aged ≥ 12 years) based on the following 5 dimensions: mobility, self-care, usual activities, pain/discomfort, and anxiety/depression. The EQ-5D-5L has 5 possible levels for each domain and respondents are asked to choose the level that reflects their health state for each of the 5 domains resulting in 3,125 possible health states.⁷⁰ The second component of the EQ-5D-5L part is a 20 cm visual analogue scale (EQ-VAS) that has end points labelled 0 and 100, with respective anchors of “worst imaginable health state” and “best imaginable health state.” Respondents are asked to rate their health by drawing a line from an anchor box to the point on the EQ-VAS which best represents their health on that day. Thus, the EQ-5D-5L produces 3 types of data for each respondent:

1. A profile indicating the extent of problems on each of the 5 dimensions represented by a 5-digit descriptor, e.g., 15121, 33211;
2. A population preference-weighted health index score based on the descriptive system; and
3. A self-reported assessment of health status based on the EQ-VAS.

Scoring

The EQ-5D-5L index score is generated by applying a multi-attribute utility function to the descriptive system.⁷¹ Different utility functions are available that reflect the preferences of specific populations (e.g., US or UK). Scores less than 0 represent health states that are valued by society as being worse than dead, while scores of 0 and 1.00 are assigned to the health states ‘dead’ and ‘perfect health,’ respectively.

Psychometric Properties

A literature search was conducted to identify validation information of the EQ-5D-5L in patients with EC and none were identified. Below is a summary of the psychometric property of the EQ-5D-5L in patients with gynecological cancers.

Among 530 patients with cervical cancer in Taiwan, the EQ-5D-5L was found to be both a reliable and valid assessment of HRQoL.⁷² The interclass correlation of the EQ-5D-5L was 0.8, with Cohen kappa values for the different dimension ranging from 0.54 to 0.73. Its convergent and discriminant validities were examined using the EORTC QLQ-30 and the clinical indicators of the functional performance assessment using the Karnofsky Performance Scale (KPF) and disease status. The EQ-5D was strongly correlated with all EORTC QLQ-C30 functioning scales, and its index and VAS scores were higher for patients with higher KPS score and disease-free status.⁷²

Among a group of 300 patients (mean age: 51.5 ± 11.5 years) with HPV-related cancer (i.e., head and neck cancer - 70%; cervical cancer - 13.4%; and nasopharyngeal cancer - 10%) in Indonesia, the EQ-5D-5L demonstrated good to excellent test-retest reliability across domains (ICCs: Mobility = 0.97; Self-care = 0.95; Usual activities = 0.79; Pain/discomfort = 0.84; Anxiety/depression = 0.82; EQ-5D visual analogue scale = 0.73) and good internal consistency (Cronbach alpha = 0.84).⁷³ Construct and convergent validity of the EQ-5D-5L in this population was assessed by mapping its subscales to those of the EORTC QLQ-C30. Significant correlation between almost all the dimension of the Indonesian version of the EQ-5D-5L with mapped subscale of the EORTC QLQ-C30, including physical function, role function, fatigue and pain was observed. Only the mobility dimension of the EQ-5D-5L was not correlated with the social function subscale of the EORTC QLQ-C30.⁷³

The responsiveness of the EQ-5D-5L was evaluated in a longitudinal study of 50 patients with cervical intraepithelial neoplasia in China. The EQ-5D-5L demonstrated only small to moderate responsiveness post-surgery. While scores for self-care and usual activities did not change, scores of mobility, pain/discomfort and anxiety/depression decreased by 0.003, 0.004 and 0.016, indicating improvement of these dimensions at follow-up.

Minimal Important Difference

A literature search was conducted to identify MID of the EQ-5D-5L in patients with EC and none were identified. Below is a summary of MID of the EQ-5D-5L in patients within the general population and in patients with other gynecological cancers.

A simulation-based approach using an instrumental-defined single-level transitions was used to estimate the MID of the EQ-5D-5L in the general population for each country-specific scoring algorithm. An estimated MID between 0.037 and 0.069 was determined for 6 countries (Canada, China, Spain, Japan, England, and Uruguay).⁸¹ The country-specific scoring algorithm were as follows Canada, 0.056 ± 0.011; China, 0.069 ± 0.007; Spain, 0.061 ± 0.008; Japan, 0.048 ± 0.004; England, 0.063 ± 0.013; and Uruguay, 0.063 ± 0.019. Differences in MID estimates reflect differences in population preferences, in valuation techniques used, as well as in modelling strategies. After excluding the maximum-valued scoring parameters, the MID estimates (mean ± SD) were as follows: Canada, 0.037 ± 0.001; China, 0.058 ± 0.005; Spain, 0.045 ± 0.009; Japan, 0.044 ± 0.004; England, 0.037 ± 0.008; and Uruguay, 0.040 ± 0.010.⁸¹

The MID of the of the EQ-5D-5L was determined in a longitudinal study of 50 patients with cervical intraepithelial neoplasia (CIN) in China using 3 methods: distribution-based, anchor-based, and instrument-defined methods.⁷⁴ The MIDs, by anchor-based, instrument-defined, and anchor-based methods were 0.041, 0064 and 3.12, respectively. The MIDs estimated in this study; however, only represents truly meaningful change of HRQoL scores at the group level, not the individual level.⁷⁴