Department of Epidemiology, Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht, Utrecht University, Utrecht, the Netherlands.
Cancer Health Services Research, Centre for Cancer Research, Faculty of Medicine, Dentistry and Health Sciences, University of Melbourne, Melbourne, VIC, Australia.
JCO Clin Cancer Inform. 2024 Jul;8:e2400037. doi: 10.1200/CCI.24.00037.
Patient outcomes may differ from randomized trial averages. We aimed to predict benefit from FOLFOXIRI versus infusional fluorouracil, leucovorin, and oxaliplatin/fluorouracil, leucovorin, and irinotecan (FOLFOX/FOLFIRI), both plus bevacizumab, in patients with metastatic colorectal cancer (mCRC).
A Cox model with prespecified clinical, molecular, and laboratory variables was developed in 639 patients from the TRIBE2 trial for predicting 2-year mortality. Data from the CHARTA (n = 232), TRIBE1 (n = 504), and CAIRO5 (liver-only mCRC, n = 287) trials were used for external validation and heterogeneity of treatment effects (HTE) analysis. This involves categorizing patients into risk groups and assessing treatment effects across these groups. Performance was assessed by the C-index and calibration plots. The C-for-benefit was calculated to assess evidence for HTE. The c-for-benefit is specifically designed for HTE analysis. Like the commonly known c-statistic, it summarizes the discrimination of a model. Values over 0.5 indicate evidence for HTE.
In TRIBE2, the overoptimism-corrected C-index was 0.66 (95% CI, 0.63 to 0.69). At external validation, the C-index was 0.69 (95% CI, 0.64 to 0.75), 0.68 (95% CI, 0.64 to 0.72), and 0.65 (95% CI, 0.65 to 0.66), in CHARTA, TRIBE1, and CAIRO5, respectively. Calibration plots indicated slight underestimation of mortality. The c-for-benefit indicated evidence for HTE in CHARTA (0.56, 95% CI, 0.48 to 0.65), but not in TRIBE1 (0.49, 95% CI, 0.44 to 0.55) and CAIRO5 (0.40, 95% CI, 0.32 to 0.48).
Although 2-year mortality could be reasonably estimated, the HTE analysis showed that clinically available variables did not reliably identify which patients with mCRC benefit from FOLFOXIRI versus FOLFOX/FOLFIRI, both plus bevacizumab, across the three studies.
患者结局可能与随机试验平均值不同。我们旨在预测转移性结直肠癌（mCRC）患者接受 FOLFOXIRI 与氟尿嘧啶、亚叶酸钙和奥沙利铂/氟尿嘧啶、亚叶酸钙和伊立替康（FOLFOX/FOLFIRI）联合贝伐珠单抗，以及 FOLFOXIRI 与 FOLFOX/FOLFIRI 联合贝伐珠单抗治疗的获益情况。
在 TRIBE2 试验中，我们使用 Cox 模型对 639 例患者进行了预设的临床、分子和实验室变量分析，以预测 2 年死亡率。CHARTA（n=232）、TRIBE1（n=504）和 CAIRO5（仅肝脏 mCRC，n=287）试验的数据用于外部验证和治疗效果异质性（HTE）分析。这涉及将患者分为风险组，并评估这些组的治疗效果。性能通过 C 指数和校准图进行评估。C 获益指数用于评估 HTE 的证据。C 获益指数专门用于 HTE 分析。与常用的 C 统计量一样，它总结了模型的区分能力。大于 0.5 的值表示存在 HTE 的证据。
在 TRIBE2 中，经过过度拟合校正的 C 指数为 0.66（95%置信区间，0.63 至 0.69）。在外部验证中，C 指数分别为 CHARTA 中的 0.69（95%置信区间，0.64 至 0.75）、TRIBE1 中的 0.68（95%置信区间，0.64 至 0.72）和 CAIRO5 中的 0.65（95%置信区间，0.65 至 0.66）。校准图表明对死亡率的低估。C 获益指数表明 CHARTA 中存在 HTE 的证据（0.56，95%置信区间，0.48 至 0.65），但在 TRIBE1（0.49，95%置信区间，0.44 至 0.55）和 CAIRO5（0.40，95%置信区间，0.32 至 0.48）中不存在。
尽管可以合理估计 2 年死亡率，但 HTE 分析表明，在这三项研究中，临床可用的变量不能可靠地识别出哪些 mCRC 患者从 FOLFOXIRI 与 FOLFOX/FOLFIRI 联合贝伐珠单抗治疗中获益，而 FOLFOXIRI 与 FOLFOX/FOLFIRI 联合贝伐珠单抗治疗的获益情况。
