Pembrolizumab-Based Neoadjuvant Therapy for Early Triple-Negative Breast Cancer: A Vietnamese Single- Center Target Trial Emulation

Technical Summary: Pembrolizumab-Based Neoadjuvant Therapy for Early Triple-Negative Breast Cancer in Vietnam

Problem Statement
Triple-negative breast cancer (TNBC) represents an aggressive subtype with limited treatment options beyond cytotoxic chemotherapy. While the phase III KEYNOTE-522 trial established pembrolizumab combined with neoadjuvant chemotherapy as a standard of care for stage II–III TNBC, real-world evidence (RWE) regarding its effectiveness and safety remains scarce, particularly in low- and middle-income countries (LMICs). In Vietnam, where TNBC prevalence is high and treatment access varies, there is a critical gap in data evaluating how pembrolizumab-based regimens perform in routine clinical practice compared to standard anthracycline-based chemotherapy. Traditional observational studies often suffer from confounding by indication, making causal inference difficult without the rigorous design of a randomized controlled trial (RCT).

Methodology
This study employed a Target Trial Emulation (TTE) framework to emulate a hypothetical randomized trial based on the KEYNOTE-522 design using retrospective real-world data from the Vietnam National Cancer Hospital (VNCH).

• Study Design & Population: A retrospective cohort study analyzed 170 women with newly diagnosed stage II–III TNBC who had not received prior systemic treatment, treated between October 2023 and December 2025.
• Treatment Arms:
  • Intervention: Pembrolizumab-based neoadjuvant therapy (KEYNOTE-522 regimen: paclitaxel/carboplatin + pembrolizumab followed by doxorubicin/cyclophosphamide + pembrolizumab).
  • Comparator: Standard anthracycline-taxane chemotherapy (4AC–4T: doxorubicin/cyclophosphamide followed by paclitaxel) without immunotherapy or platinum agents.
• Causal Inference: To address baseline confounding (e.g., age, menopausal status, tumor stage), the authors utilized stabilized inverse probability of treatment weighting (IPTW) combined with doubly robust g-computation. This approach estimated the Average Treatment Effect (ATE) on the primary outcome.
• Outcomes:
  • Primary: Pathological Complete Response (pCR), defined as ypT0/Tis ypN0.
  • Secondary: Safety outcomes, including treatment-related adverse events (AEs) and immune-related adverse events (irAEs), graded by CTCAE v6.0.
• Robustness Checks: The study included sensitivity analyses (e.g., excluding treatment discontinuations, Average Treatment Effect on the Treated [ATT] estimation) and quantitative bias analyses (calculating E-values and adjusting for unmeasured PD-L1 status) to assess the impact of unmeasured confounding.

Key Results

• Effectiveness: After adjustment, pembrolizumab-based therapy was associated with a substantial increase in the probability of achieving pCR compared to chemotherapy alone.
  • Risk Difference (RD): +40.4% (95% CI: 25.6–55.1%).
  • Risk Ratio (RR): 3.92 (95% CI: 2.41–7.09).
  • The crude pCR rates were 59.0% for the pembrolizumab group versus 14.7% for the chemotherapy group.
• Subgroup Analysis: The benefit was most pronounced in patients with highly proliferative tumors (Ki-67 ≥ 70%), showing an RD of 61.1% (95% CI: 40.7–79.3%). Benefits were consistent across other subgroups (age, stage, nodal status).
• Safety: Adverse event rates were comparable between groups. Grade ≥3 AEs occurred in 27.9% of the pembrolizumab group and 21.1% of the chemotherapy group. Immune-related AEs were rare (4.9%) and limited to the pembrolizumab arm.
• Bias Analysis: The E-value of 7.31 indicated that an unmeasured confounder would need a very strong association with both treatment and outcome to fully explain away the observed effect. Adjustments for potential PD-L1 imbalance did not alter the conclusion.

Key Contributions

1. First RWE from Vietnam: This study provides the first real-world evaluation of pembrolizumab-based neoadjuvant therapy for TNBC in Vietnam, a setting where such data was previously unavailable.
2. Methodological Demonstration: It successfully demonstrates the feasibility of applying Target Trial Emulation to oncology research in LMICs, utilizing causal inference methods to mitigate confounding in retrospective electronic medical record (EMR) data.
3. Clinical Insight: The findings suggest that the combination of pembrolizumab and platinum-based chemotherapy yields significantly higher pCR rates than standard non-platinum anthracycline-taxane regimens in routine practice; however, the study explicitly notes it cannot disentangle the independent contribution of the pembrolizumab component versus the platinum component to the observed benefit.

Significance and Claims
The paper claims that pembrolizumab-based neoadjuvant therapy is effective and safe for early-stage TNBC in a Vietnamese real-world setting, supporting its adoption in routine clinical practice. The authors emphasize that the observed magnitude of benefit (RD +40.4%) exceeds that of the pivotal KEYNOTE-522 trial, a difference they attribute to the lower pCR rates in the control arm of their study (which used a non-platinum regimen) compared to the platinum-containing control arm of the original trial. The authors clarify that the treatment contrast evaluated reflects the combined effect of adding both pembrolizumab and a platinum agent relative to standard non-platinum chemotherapy, rather than the isolated effect of pembrolizumab alone.

The study concludes that TTE is a viable and valuable complementary approach to randomized trials for generating evidence in resource-limited settings. However, the authors remain modest regarding the findings, acknowledging limitations such as the retrospective nature of the data, the single-center design, the small sample size of the pembrolizumab cohort (n=61), and the lack of long-term survival data. They note that while pCR is a strong surrogate, longer follow-up is required to confirm improvements in event-free and overall survival. Additionally, they highlight structural challenges in LMICs, such as fragmented health data systems and incomplete biomarker testing (e.g., PD-L1), which necessitate future infrastructure improvements for more comprehensive real-world studies.