Rosner B, Colditz G A
Channing Laboratory, Brigham and Women's Hospital, Boston, MA 02115, USA.
J Natl Cancer Inst. 1996 Mar 20;88(6):359-64. doi: 10.1093/jnci/88.6.359.
In 1983, Pike et al. developed a mathematical model to quantify the effects of reproductive risk factors on the incidence of breast cancer. In 1994, we modified that model to correct some deficiencies in the original model, including a lack of terms for spacing of births and an inability to easily accommodate births after age 40 years. Our extended Pike model, while improving on the original, still has serious disadvantages, such as difficulty in translating model parameters into relative risks (RRs) and an incomplete fit to data that slightly overestimated incidence for premenopausal women with an early age at first birth and that underestimated incidence for post-menopausal women with a late age at first birth.
We undertook both the development of a new mathematical model to quantify the effects of reproductive risk factors on breast cancer incidence and validation of the model.
A new log-incidence model of breast cancer incidence was developed using nonlinear regression methods, and a study population consisting of 89,132 women in the Nurses' Health Study from which a total of 2249 incident cases of breast cancer were identified. Subjects were followed from the return of the 1976 Nurses' Health Study questionnaire until June 1, 1990, or until the last questionnaire was returned, until the development of any cancer, or until death, yielding 1,148,593 person-years of follow-up. The log-incidence models were fitted using iteratively reweighted least squares analysis.
The log-incidence model provided a better fit to that data than the extended Pike model, with parameter estimates interpretable in terms of RRs. This new model can be fitted using standard commercially available statistical software. In the model, younger parous women are generally at slightly higher risk than nulliparous women, which is true for both the observed and expected RRs, and older parous women, aged 55-64 years with an early age at first birth, are at lower risk than nulliparous women,while older women with a late age at first birth are at substantially higher risk than nulliparous women.
Log-incidence models, such as this one, provide an efficient framework for modeling the effect of lifestyle risk factors on breast cancer incidence that may be specifically targeted to certain time periods of a woman's reproductive life.
1983年，派克等人开发了一个数学模型来量化生殖风险因素对乳腺癌发病率的影响。1994年，我们对该模型进行了修改，以纠正原始模型中的一些缺陷，包括缺少生育间隔项以及难以轻松纳入40岁以后的生育情况。我们扩展后的派克模型虽然在原始模型基础上有所改进，但仍存在严重缺点，例如难以将模型参数转化为相对风险（RRs），并且与数据的拟合并不完全，对于初产年龄较小的绝经前女性，该模型略微高估了发病率，而对于初产年龄较大的绝经后女性，该模型则低估了发病率。
我们既开发了一个新的数学模型来量化生殖风险因素对乳腺癌发病率的影响，又对该模型进行了验证。
使用非线性回归方法开发了一个新的乳腺癌发病率对数发病率模型，研究人群包括护士健康研究中的89132名女性，从中确定了总共2249例乳腺癌发病病例。从1976年护士健康研究问卷返回开始对受试者进行随访，直至1990年6月1日，或直至最后一份问卷返回，直至发生任何癌症，或直至死亡，共产生了1148593人年的随访时间。使用迭代加权最小二乘法分析来拟合对数发病率模型。
对数发病率模型与扩展后的派克模型相比，对数据的拟合更好，其参数估计值可以用相对风险来解释。这个新模型可以使用标准的商业统计软件进行拟合。在该模型中，有生育史的年轻女性通常比未生育女性的风险略高，观察到的相对风险和预期的相对风险都是如此；而年龄在55 - 64岁、初产年龄较小的有生育史的老年女性，其风险低于未生育女性，而初产年龄较大的老年女性的风险则大大高于未生育女性。
像这样的对数发病率模型为模拟生活方式风险因素对乳腺癌发病率的影响提供了一个有效的框架，该框架可能专门针对女性生殖生活的特定时间段。
