Limitations of the KEYNOTE-775 Trial
Study Design and Population Constraints
The KEYNOTE-775 trial was an open-label study without blinding, which introduces potential bias in outcome assessment and patient/physician expectations, though progression-free survival was evaluated by blinded independent central review. 1
Key Methodological Limitations
Open-label design – Both patients and treating physicians knew treatment assignments, potentially influencing symptom reporting, supportive care decisions, and subjective outcome assessments, though objective radiographic endpoints were independently reviewed. 1
Selective patient population – The trial excluded patients who were candidates for curative surgery or radiation, limiting generalizability to the broader population of endometrial cancer patients who might have localized disease amenable to definitive local therapy. 2
Performance status restrictions – Real-world data demonstrate that patients with ECOG performance status 1-2 had significantly shorter PFS and OS compared to trial participants, suggesting the trial population was healthier than typical clinical practice populations. 3
Age and comorbidity selection – Patients aged >65 years required significantly earlier lenvatinib dose reductions (median 1.5 months) in real-world practice, indicating the trial may have underrepresented or selected healthier elderly patients. 3
Toxicity and Tolerability Issues
The trial reported that 88.9% of patients experienced grade ≥3 adverse events with pembrolizumab plus lenvatinib versus 72.7% with chemotherapy, but the open-label design may have influenced adverse event reporting and management strategies. 1
Treatment Modification Challenges
High dose modification rates – Approximately 65% of patients required lenvatinib dose reductions and 72% needed dose interruptions, raising questions about whether the starting dose of 20 mg daily is optimal for most patients. 4, 2
Real-world dosing discrepancy – In clinical practice, only 19.7% of patients actually started at the trial-mandated 20 mg dose, with 47.3% starting at 14 mg, suggesting the trial protocol may not reflect feasible real-world implementation. 5
Vulnerable patient exclusion – Patients non-eligible for KEYNOTE-775 participation (33% in one real-world cohort) had significantly shorter median PFS, indicating the trial excluded a substantial proportion of patients who would be considered for this therapy in practice. 3
Efficacy Endpoint Limitations
Modest absolute survival benefit – The median OS improvement was 5.4 months (17.4 vs 12.0 months) in the pMMR population, which while statistically significant, represents a relatively modest clinical benefit given the substantial toxicity burden. 4, 1
Short median PFS – The median PFS of 6.6 months in the combination arm, while superior to chemotherapy (3.8 months), indicates that most patients still progress within 6-7 months despite intensive dual-agent therapy. 4, 1
Lack of quality-of-life data – The primary publications do not report patient-reported outcomes or quality-of-life assessments, making it difficult to determine whether the survival benefit justifies the toxicity in terms of patient experience. 4, 1
Comparator Arm Concerns
The control arm used single-agent chemotherapy (doxorubicin or paclitaxel) rather than combination chemotherapy or other active regimens, potentially inflating the apparent benefit of the experimental arm. 1
Physician's choice chemotherapy – Allowing treating physicians to select between doxorubicin and paclitaxel introduces heterogeneity in the control arm and may not represent the most effective available chemotherapy option. 1
No comparison to hormonal therapy – For ER-positive endometrioid histology, letrozole plus palbociclib achieves median PFS of 8.3 months with likely better tolerability, but was not included as a comparator arm. 4
Biomarker and Subgroup Analysis Gaps
PD-L1 expression not predictive – The trial demonstrated that PD-L1 expression had limited predictive value for response, but this was an exploratory finding rather than a prospectively defined stratification factor. 2
Histologic subtype heterogeneity – Real-world data show the trial included diverse histologies (33% endometrioid, 41% serous, 10.1% carcinosarcoma), but subgroup analyses by histology were not adequately powered to detect differential effects. 5
dMMR subgroup underpowered – Only 130 of 827 patients (15.7%) had dMMR disease, limiting the ability to draw definitive conclusions about this molecular subgroup, particularly since single-agent immunotherapy is highly effective in dMMR tumors. 1
Generalizability and Implementation Challenges
Racial and ethnic diversity – Response rates, PFS, and OS were similar between White and Black patients in real-world analysis, but the trial's racial/ethnic composition and subgroup analyses were not comprehensively reported. 5
Healthcare system requirements – The regimen requires intensive monitoring (thyroid function every 4-6 weeks, liver enzymes, renal function, complete blood counts) and immediate access to supportive medications, which may not be feasible in all practice settings. 2
Cost and resource intensity – The combination of two expensive agents with high rates of dose modifications and supportive care needs creates substantial financial toxicity that was not assessed in the trial. 4, 1
Common Clinical Pitfalls Revealed by Trial Limitations
Starting dose too high – The trial's 20 mg lenvatinib starting dose proved intolerable for most patients in real-world practice, with 47.3% of clinicians empirically starting at 14 mg without apparent compromise in efficacy. 5
Inadequate patient selection – Patients with performance status 1-2, heavily pretreated patients, and those with organ dysfunction had significantly inferior outcomes, suggesting the trial results overestimate benefit in typical clinical populations. 3
Failure to identify optimal duration – The trial did not define optimal treatment duration or stopping rules for patients achieving complete response, leaving clinicians without guidance on when to discontinue therapy. 1