MULTIPLE INFORMANT DATA IN PSYCHIATRY

使用多个告密者(家长、教师、儿童、临床医生、 同行)通常被认为是获取信息的最佳途径 关于儿童的心理健康、社交功能和服务使用，以及 结果。在组合数据时遇到的一些困难 多个告密者包括告密者之间的系统差异 关于一些结果，相对较低水平的线人 协议，以及一些告密者的信息缺失。一批 目前，研究人员在多信息者中使用了策略 研究，但没有一种单一的方法被接受为标准， 尤其是在评级是绝对的情况下。 这项提议寻求资金，以进行最近的一系列研究。 由我们自己发起，旨在为分析开发方法学 重复的(纵向)分类结果从多个 告密者。对于这项工作，我们提出了一种基于似然的方法 多元Logistic回归及其在分类中的推广 数据)，这在概念上类似于重复的多变量分析 用于持续结果的措施。在统一的框架下，这 方法将使我们能够：1)测试风险因素的影响是否 精神病理学或服务使用的预测因素因告密者和 获得所有告密者的综合估计，如果他们不这样做的话)“边缘” 分析)；2)评估线人协议(“协议”分析)；3) 评估时间对精神病理学或服务使用的影响(纵向 分析)；和4)充分利用所有数据，即使一些孩子 从告密者处或在某个时间点丢失数据。 我们计划开发软件，将这种方法应用于现有的 数据(例如，康涅狄格州使用父母对儿童心理健康的调查 和教师评级以及波士顿一项关于精神疾病的纵向研究 儿童和青少年)，到目前为止还没有得到充分的分析。这个 所提出的方法也将直接适用于未来的分析 从最近的几项NIMH倡议中获得的数据，包括 多地点联合国履约协助方案项目和其他基于社区的协作 学习。

The use of multiple informants (parents, teachers, children, clinicians, peers) is generally regarded as the best approach to obtain information about children's mental health, social functioning, and service use and outcomes. Some of the difficulties encountered in combining data from multiple informants include systematic differences between informants with regard to some outcomes, a relatively low level of informant agreement, and missing information for some informants. A number of strategies are currently employed by researchers in multiple-informant studies, but no single approach has been embraced as a standard, especially for the case where ratings are categorical. This proposal seeks funding to pursue a line of research recently initiated by ourselves and aims to develop methodology for the analysis of repeated (longitudinal) categorical outcomes obtained from multiple informants. For this work, we propose a likelihood-based method (multivariate logistic regression and its extensions for categorical data), which is similar in concept to multivariate analysis of repeated measures used for continuous outcomes. Under a unified framework, this approach will allow us to: 1) test if the effects of risk factors of psychopathology or of predictors of service use vary by informant and obtain a combined estimate across informants if they do not )"marginal" analyses); 2) assess informant agreement ("agreement" analyses); 3) assess the effect of time on psychopathology or service use (longitudinal analyses); and 4) fully utilize all the data, even if some children are missing data from an informant or at a time point. We plan to develop the software to apply this methodology to existing data (e.g., a Connecticut survey of children's mental health using parent and teacher ratings and a Boston longitudinal study of mental illness in children and adolescents) that have not been fully analyzed so far. The proposed methods will also be directly applicable to the future analysis of data obtained from several recent NIMH initiatives, including the multi-site UNOCCAP project and other community-based collaborative studies.