Response to Pisano, Gastonis, Sparano, et al.
Response to Pisano, Gastonis, Sparano, et al.
复制标题
对 Pisano、Gastonis、Sparano 等人的回应。
DOI：
10.1093/jnci/djab056
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
Miglioretti,DianaL
中科院分区：
文献类型：
--
作者：
Kerlikowske,Karla;Bissell,MichaelCS;Sprague,BrianL;Buist,DianaSM;Henderson,LouiseM;Lee,JanieM;Miglioretti,DianaL
我们感谢Pisano等人。(1)感谢他们对我们的手稿感兴趣，并有机会澄清我们研究的概念方法和结果。我们的研究是专门设计来确定乳腺肿瘤分类方法，最准确地确定妇女在最高风险的5年乳腺癌死亡率用于筛查有效性的研究（1）。我们计算了敏感性、特异性和阳性预测值（PPV），以评估每种分类方法，使用晚期与非晚期癌症的二元结果。在所有癌症中，美国癌症联合委员会（AJCC）预后病理学II期或更高阶段总体上是预测5年乳腺癌死亡最准确的（灵敏度<$76.7%，特异性<$81.6%，PPV <$21.0%），而乳腺X射线断层摄影成像筛查试验（TMIST）方法具有高灵敏度特异性（41.1%）和阳性预测值（9.1%）较低。我们还使用每种分类方法的多个类别计算了受试者工作特征曲线下的时间依赖性面积，以预测5年乳腺癌死亡率（2）。对于AJCC分期系统，我们使用8个分期类别，IA至IV。对于TMIST肿瘤分类，我们构建了一个6类变量：非晚期加上我们研究中观察到的5个TMIST晚期类别，从最差到最佳生存率排序。我们发现，当比较不同肿瘤分类的相同妇女时，AJCC预后病理分期在预测5年乳腺癌死亡方面的区分度比AJCC解剖分期和TMIST肿瘤分类有统计学意义。我们还通过检测模式评估了肿瘤分类方法。使用二元和受试者工作特征曲线下的面积结果表明，预后病理分期最好地预测5年死亡率的整体和屏幕检测，间隔，和临床检测到的癌症。值得注意的是，我们的结果在预测10年乳腺癌死亡率时以及在不同种族和民族中是一致的。我们赞赏TMIST研究者选择终点的原因与我们的研究不同-识别通常考虑化疗治疗的癌症，假设接受数字乳腺断层合成摄影的女性的TMIST晚期癌症比例低于数字乳腺X射线摄影组，因此较少接受数字乳腺断层合成摄影的女性可能需要化疗。除了降低乳腺癌死亡率外，避免化疗也是一个重要的结果。TMIST结果是否支持这一假设将是有趣的，因为我们惊讶地发现5个TMIST晚期癌症亚类中有3个具有非常高的5年生存率（95%），这可能是因为女性接受了化疗。如果TMIST研究者使用II期或更高的预后病理分期来确定入组样本量，那么TMIST将需要比计划的165000例入组样本量更大的样本量，因为我们的研究筛选人群中II期或更高的预后病理分期的患病率为18%，而TMIST晚期癌症的患病率为56%。AJCC预后病理学II期或更高是评估筛查方案的有临床意义的中间结果，因为它是5年乳腺癌死亡率的准确预测因子。我们建议，当主要目标是评估降低乳腺癌死亡率的能力时，AJCC预后II期或更高阶段应该是筛查计划有效性研究的主要结局（3）。
We thank Pisano et al.(1) for their interest in our manuscript and for the opportunity to clarify our study’s conceptual approach and outcomes. Our study was specifically designed to ascertain the breast tumor classification method that most accurately identifies women at highest risk of 5-year breast cancer mortality for use in studies of screening effectiveness (1). We calculated sensitivity, specificity, and positive predictive value (PPV) to evaluate each classification method using a binary outcome of advanced vs nonadvanced cancer. Among all cancers, American Joint Committee on Cancer (AJCC) prognostic pathology stage II or higher was overall the most accurate for predicting 5-year breast cancer death (sensitivity ¼ 76.7%, specificity ¼ 81.6%, PPV ¼ 21.0%) while the Tomosynthesis Mammographic Imaging Screening Trial (TMIST) method had high sensitivity (96.1%) but low specificity (41.1%) and PPV (9.1%). We also calculated the time-dependent area under the receiver operating characteristic curve using multiple categories for each classification method to predict 5-year breast cancer mortality (2). For AJCC staging systems, we used the 8 staging categories, IA through IV. For TMIST tumor classification, we constructed a 6-category variable: nonadvanced plus the 5 TMIST advanced categories ordered from worst to best survival observed in our study. We found AJCC prognostic pathologic stage had statistically significantly better discrimination for predicting 5-year breast cancer death than AJCC anatomic stage and the TMIST tumor categories when comparing the same women across tumor classifications. We also evaluated tumor classification methods by mode of detection. Using both the binary and area under the receiver operating characteristic curve outcomes shows that prognostic pathologic stage best predicts 5-year mortality overall and for screen-detected, interval, and clinically detected cancer. Notably, our results were consistent when predicting 10-year breast cancer death and across racial and ethnic groups. We appreciate that TMIST investigators chose an endpoint for a different reason than our study—to identify cancers generally considered for chemotherapy treatment with the hypothesis that women undergoing digital breast tomosynthesis would have a lower proportion of TMIST advanced cancers than the digital mammography arm, and thus fewer women undergoing digital breast tomosynthesis may require chemotherapy. Avoiding chemotherapy is an important outcome in addition to reducing breast cancer mortality. It will be of interest to see if TMIST results support this hypothesis because we were surprised to find 3 of 5 TMIST advanced cancer subcategories have very high 5-year survival (95%), possibly because women received chemotherapy. Had the TMIST investigators used prognostic pathologic stage II or higher to determine enrollment sample size, TMIST would have required a larger sample size than the 165000 planned enrollment given the prevalence of prognostic pathologic stage II or higher is 18% of our study screening population compared with the TMIST advanced cancer prevalence of 56%. AJCC prognostic pathology stage II or higher is a clinically meaningful intermediate outcome to evaluate screening programs given it is an accurate predictor of 5-year breast cancer mortality. We suggest AJCC prognostic stage II or higher should be the primary outcome for studies of screening program effectiveness when the primary goal is to assess the ability to reduce breast cancer mortality (3).