MULTI-LEVEL SURVIVAL MODELS FOR DRUG ABUSE PREVENTION DATA

在基于理论的研究中， 物质使用预防必须随着事件持续时间而变化。 例如，一个特定的预防计划是否有效地延缓了 最初的酒精实验 在这个问题中， 酒精的初步实验。 再比如， 问题饮酒的发生与父母等因素的关系 饮酒和第一次饮酒的时间：这里的事件是 开始出现酗酒问题 出于各种有据可查的原因， 传统的线性回归技术不足以解决这些问题 问题;相反，经典的离散时间生存分析是 适当的统计方法。 然而，由于许多 经典的生存分析是为临床医疗环境开发的 而不是实地研究，机械地应用经典的 对预防问题的生存分析可能导致误导性结果。 本提案涉及生存分析方法的修改， 基于理论的预防/干预研究的独特情况。 特别是，经典的生存分析假设，在调整后， 风险因素个体是独立的，但这种假设往往是 由于其多级结构，例如， 学生被分到班级，班级被分到学校， 学校按学区划分。 这一多方面的结果- 学生不能被认为是独立的，因为 同一所学校的学生在本质上会更加相似 即使在考虑了风险之后， 因素 拟议的研究将开发生存分析方法， 认识并利用大多数药物的多层次结构 预防和干预虐待数据集。 这些方法将应用于 关于风险、保护因素、脆弱性和 青少年酒精预防试验数据的弹性。

Many fundamentally important questions in theory-based research on substance use prevention have to so with a duration of time ot an event. For example, is a particular prevention program effective at delaying initial experimentation with alcohol? In this question the event is initial experimentation with alcohol. For another example, what is the relationship between onset of problem drinking and such factors as parental drinking and time of first experience with alcohol: The event here is onset of problem drinking. For various well-documented reasons, traditional linear regression techniques are inadequate in addressing these kinds of questions; instead, classical discrete time survival analysis is the appropriate statistical methodology. However, because much of classical survival analysis was developed for a clinical medical setting rather than a field research setting, rote application of classical survival analysis to prevention questions can lead to misleading results. This proposal deals with the modification of survival analysis methods to the unique circumstances of theory-based prevention/intervention research. In particular, classical survival analysis assumes that after adjusting for risk factors individuals are independent, yet this assumption is often violated in prevention data because of it s multi-level structure, e.g., students are grouped into classes, classes are grouped into schools and schools are grouped into school districts. A consequence of this multi- level structure is that students cannot be considered independent, because students in the same school can be expected to be more similar in substance use than students in different schools even after accounting for risk factors. The proposed research will develop survival analysis methods that recognize and take advantage of the multi-level structure of most drug abuse prevention and intervention data sets. These methods will be applied to prevention questions about risk, protective factors, vulnerability, and resiliency in data from the Adolescent Alcohol Prevention Trial.