Bremer Troy M, Jacquemier Jocelyne, Charafe-Jauffret Emmanuelle, Viens Patrice, Birnbaum Daniel, Linke Steven P
Prediction Sciences, La Jolla, CA 92037-1354, USA.
Int J Cancer. 2009 Feb 15;124(4):896-904. doi: 10.1002/ijc.24001.
Single markers are insufficient to accurately assess risk of relapse for adjuvant therapy guidance in operable breast cancer patients. In addition, the accuracy and interpretability of current multi-marker tests is generally limited by their simply additive algorithms and their overlap with clinicopathologic risks. Here, we report the development and validation of a nonlinear algorithm that combines protein (ER, PGR, ERBB2, BCL2 and TP53) and genomic (MYC/8q24) markers with standard clinicopathologic features (tumor size, tumor grade and nodal status) into a global risk assessment profile. The algorithm was trained using statistical pattern recognition in 200 stage I-III hormone receptor-positive patients treated with hormone therapy. Continuous risk scores (0-10+) were then generated for 232 independent patients. In hormone therapy-treated patients, the profile achieved a hazard ratio of 6.2 (95% confidence interval [CI], 1.8-20) in high- vs. low-risk groups for time to distant metastasis with the low-risk group having a 10-year metastasis rate of just 4% (95% CI, 0-8%). Similar results were achieved in untreated patients and for disease-specific survival. In multivariate analyses with standard prognostic factors and clinical practice guidelines, the profile was the only significant variable. Furthermore, the profile reclassified as low risk over half of node-negative patients at elevated risk according to the guidelines, which could have spared such patients from unnecessary cytotoxic chemotherapy. It also accurately identified a group of high-risk patients within a guideline low-risk group. In summary, the profile intelligently combines biologically relevant marker pathways and established clinicopathologic risks to help guide breast cancer patients to the most appropriate level of adjuvant therapy.
单一标志物不足以准确评估可手术乳腺癌患者辅助治疗指导中的复发风险。此外，当前多标志物检测的准确性和可解释性通常受到其简单相加算法以及与临床病理风险重叠的限制。在此，我们报告了一种非线性算法的开发与验证，该算法将蛋白质（雌激素受体、孕激素受体、表皮生长因子受体2、B细胞淋巴瘤/白血病-2和肿瘤蛋白p53）和基因组（原癌基因c-Myc/8号染色体长臂24区）标志物与标准临床病理特征（肿瘤大小、肿瘤分级和淋巴结状态）整合为一个整体风险评估概况。该算法在200例接受激素治疗的Ⅰ-Ⅲ期激素受体阳性患者中使用统计模式识别进行训练。然后为232例独立患者生成连续风险评分（0-10+）。在接受激素治疗的患者中，该概况在远处转移时间的高风险组与低风险组中实现了6.2的风险比（95%置信区间[CI]，1.8-20），低风险组的10年转移率仅为4%（95%CI，0-8%）。在未治疗患者和疾病特异性生存方面也取得了类似结果。在与标准预后因素和临床实践指南的多变量分析中，该概况是唯一的显著变量。此外，根据指南，该概况将超过一半的高风险淋巴结阴性患者重新分类为低风险，这可能使这些患者免于不必要的细胞毒性化疗。它还准确地在指南低风险组中识别出一组高风险患者。总之，该概况智能地整合了生物学相关标志物途径和既定的临床病理风险，以帮助指导乳腺癌患者接受最合适水平的辅助治疗。
