In Reply We agree that observational comparison of outcomes across populations is dependent on the distribution of all attributes with the potential to affect the outcome of interest. For attributes that are captured by the database (eg, age, comorbidity, stage), a number of strategies were used to create balance (propensity matching) or adjust for imbalances (Cox proportional hazards). For attributes that are not captured by the database (eg, performance status [PS], smoking status), one must first hypothesize as to whether an imbalance is likely. First, we identified populations by a common event that tends to create balance in unmeasured attributes. All patients had undergone and survived lung cancer surgery. This would suggest that all patients had a reasonable PS (in the study by Kawaguchi et al,¹ 75% of patients who underwent surgery had a PS of 0). Whereas PS and smoking have not been examined in relationship to the timing of adjuvant chemotherapy in lung cancer per se, neither has been associated with delays for other tumor types.^2,3 Therefore, we believe that imbalances, should they exist for smoking and PS, would likely be subtle.

作为答复，我们同意，对人群结果的观察比较取决于所有可能影响感兴趣结果的属性的分布。对于数据库捕获的属性（例如，年龄、合并症、阶段），使用了许多策略来创建平衡（倾向匹配）或调整不平衡（Cox 比例风险）。对于数据库未捕获的属性（例如，性能状态 [PS]、吸烟状态），必须首先假设是否可能存在不平衡。首先，我们通过一个常见的事件来识别种群，该事件往往会在未测量的属性中创造平衡。所有患者均接受并幸免于肺癌手术。这表明所有患者都有合理的 PS（在 Kawaguchi 等人的研究中，¹75% 接受手术的患者 PS 为 0)。虽然 PS 和吸烟与肺癌本身的辅助化疗时间之间的关系尚未被研究过，但两者都与其他肿瘤类型的延迟有关。^{2 ,3}因此，我们认为，如果存在吸烟和 PS 的不平衡，可能会很微妙。