Leigh J Paul, Schembri Michael
Center for Health Services Research in Primary Care, and Department of Epidemiology and Preventive Medicine, 2103 Stockton, Suite 2224, University of California, Davis, Sacramento, CA 95817, USA.
J Clin Epidemiol. 2004 Mar;57(3):284-93. doi: 10.1016/j.jclinepi.2003.08.006.
Debate surrounds the usefulness of the instrumental variables (IV) technique for medical research. The choice of an instrument for the technique has been contentious. This study estimated the effects of smoking on physical functional status. We chose an especially valid and strong instrument: cigarette price.
The data were a nationally representative cross-sectional sample of 34,288 persons aged 30 to 91 in 1996-1997. The sample was drawn from the Community Tracking Study. Number of cigarettes smoked per day was predicted by the average cigarette price for the state in which the subject resided. The outcome measure was physical functional status and was measured by the SF-12 physical functional index.
In multivariable models we found the following: cigarettes per day was strongly and negatively associated with the SF-12 index (P<.001); cigarette price was strongly and negatively associated with cigarettes per day (P=.002); the predicted cigarettes per day (the IV) was strongly and negatively associated with the SF-12 index in linear regression and tobit regression (P=.047 and P=.021).
Estimated coefficients from the IV method suggested that the effect of smoking on physical functional status was substantially larger than estimates that relied on conventional methods.
围绕工具变量（IV）技术在医学研究中的实用性存在争议。该技术中工具的选择一直存在争议。本研究估计了吸烟对身体功能状态的影响。我们选择了一个特别有效且有力的工具：香烟价格。
数据来自1996 - 1997年全国代表性的34288名年龄在30至91岁的横断面样本。样本取自社区追踪研究。每天吸烟的数量由受试者居住州的平均香烟价格预测。结果指标是身体功能状态，通过SF - 12身体功能指数进行测量。
在多变量模型中我们发现以下情况：每天吸烟量与SF - 12指数呈强烈负相关（P <.001）；香烟价格与每天吸烟量呈强烈负相关（P =.002）；预测的每天吸烟量（工具变量）在线性回归和托比特回归中与SF - 12指数呈强烈负相关（P =.047和P =.021）。
工具变量法估计的系数表明，吸烟对身体功能状态的影响远大于依赖传统方法的估计。
