STATISTICAL ISSUES IN MENTAL HEALTH EPIDEMIOLOGY

生物统计学研究提出了解决几种方法 心理健康流行病学家需要获得 可靠和有效的数据，精神障碍的比率， 一般社区。 问题来自所有三个阶段， 精神病学的流行病学研究：建立诊断 可靠性，流行率调查的设计，以及 多元数据分析。 具体目标是：1.发展 估计kappa及其标准误差的方法，以及构建 使用有效的重测设计的置信区间，其中 是否包括在重新测试阶段取决于第一次测试的结果。 test. 2.制定选择诊断有效性病例的规则 研究基于两个或两个以上的独立诊断时，一个单一的 诊断的预测能力很差。 与以前制定的规则不同， 这些将不会假设诊断灵敏度在所有的 病人的类型。3.为了评估各种设计的效率， 罕见精神疾病的两阶段调查 将特别注意 用于采用多元筛选规则的设计，特别是 那些试图筛查不同精神疾病的人。 4.制定流行率标准误的估计方法 从两阶段调查中得出的估计数，并用于建立信心 在这些流行率估计值附近。5.批判性地审视 通过包括错误指定逻辑回归模型的效应 独立的变量，结构上不相关的几率， 给出结果。 如果这种影响是不平凡的， 在逻辑模型规范中指导数据分析师的规则。 最后， 6.研究排列检验评估 当多变量结果不是多变量时的多变量假设 正常 将开发一个FORTRAN计算机程序来实现 多元排列测试，该方法的结果将是 相对于其他稳健方法进行评价。 不会有新的数据 在拟议的研究中收集。 心理健康调查数据 智力迟钝和儿童残疾将被用来说明 目标3和4中开发的方法，以及来自以下病例对照研究的数据： 抑郁症的心理社会风险因素将被用来说明 目标5和6中制定的方法。

Biostatistical research is proposed to address several methodological problems that arise from mental health epidemiologists' need to obtain reliable and efficient data about the rates of mental disorders in the general community. The problems come from all three phases of epidemiologic research in psychiatry: the establishment of diagnostic reliability, the design of prevalence surveys, and the statistical analysis of multivariate data. The specific aims are:, 1. To develop methods for estimating kappa and its standard error, and for constructing confidence bounds using an efficient test-retest design in which inclusion in the retest phases is dependent on the outcome of the first test. 2. To develop rules for selecting cases for diagnostic validity studies based on two or more independent diagnoses when a single diagnosis has poor predictive power. Unlike previously developed rules, these will not assume that diagnostic sensitivity is constant across all types of patients. 3. To evaluate the efficiency of various designs for two-phase surveys of rare mental disorders. Special attention will be given to designs that employ multivariate screening rules, especially ones that attempt to screen persons for different psychiatric disorders. 4. To develop methods for estimating the standard error of prevalence estimates derived from two-phase surveys, and for constructing confidence bounds around these prevalence estimates. 5. To examine critically the effects mis-specifying a logistic regression model by including independent variables that are not structurally related to the odds of a given outcome. If such effects are nontrivial, to develop practical rules to guide data analysts in logistic model specification. Finally, 6. To investigate the usefulness of permutation tests to evaluate multivariate hypotheses when the multivariate outcome is not multivariate normal. A FORTRAN computer program will be developed to implement multivariate permutation tests, and the results of this method will be evaluated relative to other robust methods. No new data will be collected in the proposed research. Data from surveys of mental retardation and childhood disability will be used to illustrate the methods developed in aims 3 and 4, and data from a case-control study of psychosocial risk factors for depression will be used to illustrate methods developed in aims 5 and 6.