We read with interest Leas et al.’s analysis of the Current Population Survey data (1) relating the use of pharmacotherapy to outcome in attempts to quit smoking, particularly as a decade ago we published a similar analysis on some of the same data (2), not cited by Leas et al. We strongly agree with the editorial by Tindle and Greevy that a major issue in the analysis is serious confounding when treatment is assigned clinically or self-selected (3). This is a classic case of the well-known “confounding by indication” (4,5) (or by “severity”) problem, in which treatment is primarily used by those who already have the most severe conditions and are the most prone to poor outcomes. Leas et al. undertook heroic efforts to overcome confounding by propensity score matching, but this is insufficient to remove this bias. Those who are convinced they cannot succeed without treatment (and perhaps whose doctors are also so convinced) are most likely to use it. The limited data available in the survey cannot account for all the factors that may enter into smokers’ assessment of their own prognosis (eg, how motivated they are to quit, how difficult they found quitting last time they tried, what impediments to quitting they expect, etc.). When treatment is actively “assigned” by assessed of risk of failure, the resulting bias is near impossible to correct (5).

中文翻译：

---

RE：一项全国代表的美国吸烟者戒烟辅助药物的有效性

我们饶有兴趣地阅读了Leas等人对当前人口调查数据的分析（1），该分析将药物治疗的使用与尝试戒烟的结局联系起来，特别是在十年前，我们对某些相同的数据发表了类似的分析（ 2），未被Leas等人引用。我们非常同意Tindle和Greevy的社论，当临床分配或自行选择治疗方法时，分析中的一个主要问题是严重的混淆（3）。这是著名的“因指征混淆”（4,5）（或因“严重性”）问题的经典案例，在这种情况下，治疗主要是由那些已经患有最严重疾病且最容易患病的人使用的结果差。Leas等。进行了英勇的努力来克服倾向得分匹配带来的混淆，但这不足以消除这种偏见。那些相信如果没有治疗就无法成功的人（也许他们的医生也是如此相信）最有可能使用它。调查中可用的有限数据无法说明可能影响吸烟者自身预后评估的所有因素（例如，他们戒烟的动机，上次尝试戒烟的困难程度，预期戒烟的哪些障碍， ETC。）。当通过评估失败的风险积极地“分配”治疗时，所产生的偏差几乎是不可能纠正的（5）。他们上次尝试戒烟的困难程度，他们期望的戒烟障碍等）。当通过评估失败的风险积极地“分配”治疗时，所产生的偏差几乎是不可能纠正的（5）。他们上次尝试戒烟的困难程度，他们期望的戒烟障碍等）。当通过评估失败的风险积极地“分配”治疗时，所产生的偏差几乎是不可能纠正的（5）。