Authors' response to: cohort effects explain the increase in autism diagnosis: an identifiability problem of the age-period-cohort model.
Authors' response to: cohort effects explain the increase in autism diagnosis: an identifiability problem of the age-period-cohort model.
复制标题
作者的回应：队列效应解释了自闭症诊断的增加：年龄-周期-队列模型的可识别性问题。
DOI：
10.1093/ije/dyu214
复制
发表时间：
2014
影响因子：
7.7
通讯作者：
Bearman,PeterS
中科院分区：
文献类型：
--
作者：
Keyes,KatherineM;Susser,Ezra;Cheslack-Postava,Keely;Fountain,Christine;Liu,Kayuet;Bearman,PeterS
汉森和帕纳1得出结论，当应用年龄-时期-队列分析时，我们对模型约束的假设可能对所获得的结果产生重大影响。他们使用我们对加州1992年至2003年出生的儿童进行的自闭症诊断研究的数据来证明这一点。我们同意;所有年龄段队列模型都有识别问题，因此模型本身不能指导我们。我们使用了一个经典的基于约束的方法，但即使是现代的方法，如内在的estimator 3和层次年龄-时期-队列4的方法已受到抨击的潜在偏见，由于无法验证的假设有关的数据结构。[5，6]因此，尽管汉森和帕纳是正确的，人们可以以改变结果解释的方式来约束模型，但这肯定不是我们的数据或分析方法所独有的问题。汉森和帕纳表明，人们可以生成一个模型，其中自闭症诊断趋势由周期效应而不是队列效应来解释，如果我们假设在8岁到12岁之间有一个线性的负斜率。然而，在这个发展时期，自闭症诊断不太可能与年龄呈线性负相关，因此使用这种限制的有效性是值得怀疑的。此外，自闭症诊断的周期效应意味着所有儿童，无论年龄大小，随着时间的推移，自闭症诊断的相对增加都是相同的。自闭症的诊断最常发生在3至6岁之间，8然而，很少有诊断是在9岁之后进行的。虽然周期效应不会改变不同年龄段诊断相对差异的潜在分布（3-6岁的人比其他年龄组更容易被诊断），但我们应该看到所有年龄组的诊断都有所增加。考虑到这种增加的证据有限，即使在存在模型的情况下，解释自闭症诊断增加的周期效应的可解释性也很弱。
Hansen & Parner1 conclude that when applying age-period-cohort analyses, the assumptions that we make about model constraints can have a pivotaleffect on the results obtained. They demonstrate this using data from our study2 of autism diagnoses in California from children born from 1992 to 2003. We concur; all age-periodcohort models have identification problems, and thus the model alone cannot guide us. We used a classical constraint-based method, but even modern methods such as the intrinsic estimator3 and hierarchical age-period-cohort4 approaches have been under fire for potential bias due to unverifiable assumptions about underlying data structure. 5, 6 Thus, whereas Hansen & Parner are correct that one could constrain the model in such a way as to change the interpretation of the results, this is certainly not an issue unique to our data or our analytical approach.Hansen & Parner show that one could generate a model in which autism diagnoses trends are explained by a period-effect rather than a cohort-effect, if one assumes a linear negative slope between ages 8 and 12 years. However, it is unlikely that autism diagnoses have a linear negative relationship with age over this developmental time period7 and thus the validity of using such a constraint is dubious. Further, a period-effect in autism diagnoses would imply that all children, regardless of age, experience the same relative increase in autism diagnoses across time. Autism diagnoses most frequently occur between the ages of 3 and 6 years, 8 however, and few diagnoses are made after age 9 years. Although a period-effect would not change the underlying distribution of relative differences in diagnoses across age (those aged 3–6 yearsmore likely to be diagnosed than other groups), we should see an increase in diagnoses across all age groups. Given that there is limited evidence for such an increase, the plausibility of a period-effect explaining the increase in autism diagnoses is weak, even in the presence of a model that suggests it.