Yeomans Kinney A, Vernon S W, Shui W, Weber D V, Schell M, Vogel V G
The University of North Carolina Lineberger Comprehensive Cancer Center and Department of Epidemiology, The University of North Carolina School of Public Health, Chapel Hill 27599, USA.
Cancer Epidemiol Biomarkers Prev. 1998 Jul;7(7):591-5.
We evaluated the performance of a regression model in predicting enrollment status in a chemoprevention trial for breast cancer using a population independent of that from which the model was derived. In years 1 and 2 of recruitment, questionnaires were completed by eligible participants following attendance at informational meetings about the Breast Cancer Prevention Trial. The variables in the original model, based on women recruited in year 1, included not being able to take estrogen replacement therapy (ERT), concern about the side effects of tamoxifen, the possibility of getting a placebo, the out-of-pocket expenses associated with the trial, and disagreement with the statement "significant others would be reassured if the respondent was taking tamoxifen." These variables were used to predict enrollment status of women newly recruited to the trial in year 2. Among the 89 women in the study population who responded to the questionnaire, 66% did not enroll in the trial. By applying the original logistic regression model, enrollment status in the trial was correctly predicted for 72% of year 2 questionnaire respondents. Age and risk scores, as binary variables, were used in a derived logistic model to determine whether they provided additional predictive information on enrollment status. The resulting four-factor model, which predicted nonenrollment, included: age of > or = 50 years, not being able to take ERT, expressed concern that significant others would not be reassured if the respondent was taking tamoxifen, and concern about out-of-pocket expenses associated with the trial. This model correctly classified 76% of the respondents. The logistic regression models performed reasonably well in predicting enrollment status. Not being able to take ERT remained the strongest factor predicting nonenrollment. More research is needed to evaluate factors that motivate persons to seek participation in primary chemoprevention trials in culturally diverse populations.
我们使用一个与模型推导所基于的人群无关的群体，评估了一个回归模型在预测乳腺癌化学预防试验入组状态方面的性能。在招募的第1年和第2年，符合条件的参与者在参加关于乳腺癌预防试验的信息会议后填写问卷。基于第1年招募的女性的原始模型中的变量包括：不能接受雌激素替代疗法（ERT）、对他莫昔芬副作用的担忧、接受安慰剂的可能性、与试验相关的自付费用，以及不同意“如果受访者服用他莫昔芬，重要他人会放心”这一说法。这些变量被用于预测第2年新招募到该试验的女性的入组状态。在研究人群中回复问卷的89名女性中，66%没有参加该试验。通过应用原始逻辑回归模型，对于第2年问卷受访者，试验入组状态的预测准确率为72%。年龄和风险评分作为二元变量，被用于一个推导的逻辑模型中，以确定它们是否能提供关于入组状态的额外预测信息。由此得到的预测未入组的四因素模型包括：年龄大于或等于50岁、不能接受ERT、表示担心如果受访者服用他莫昔芬重要他人不会放心，以及担心与试验相关的自付费用。该模型正确分类了76%的受访者。逻辑回归模型在预测入组状态方面表现相当不错。不能接受ERT仍然是预测未入组的最强因素。需要更多研究来评估促使不同文化背景人群参与原发性化学预防试验的因素。
