Chappell R, Branch L G
Department of Statistics, University of Wisconsin, Madison 53706.
J Clin Epidemiol. 1993 Mar;46(3):309-12. doi: 10.1016/0895-4356(93)90079-g.
A dilemma appears in the design of any survey with followups intended to detect the occurrence of an event. Suppose that, at followup, the event of interest can only be determined to have occurred since the previous interview (such is often the case, for example, with onset of dependent behavior in the aged). Then a followup which takes place too soon runs the risk of observing no or very few events, while a followup which is conducted after too long a wait might observe events in nearly every subject, and so be equally uninformative. Thus we would have what one investigator compares to Goldilocks dilemma: is the interval before followup too short, too long, or just right? The problem is multiplied when more than one followup is intended. As a partial solution, a technique is described here which provides guidance on the spacing of followup waves in a multiwave study, preventing possibly serious inefficiencies in study design. Data from the Massachusetts Health Care Panel Study (MHCPS) on functional dependence in the elderly are used for a demonstration.
在任何旨在检测事件发生的带有随访的调查设计中，都会出现一个两难困境。假设在随访时，只能确定自上次访谈以来所关注的事件是否发生（例如，老年人出现依赖行为的情况通常就是如此）。那么，随访进行得太早，就有可能观察到没有事件或极少事件发生，而等待时间过长后进行的随访可能会在几乎每个受试者身上都观察到事件，从而同样无法提供有用信息。这样一来，我们就会遇到一位研究者所比喻的金发姑娘困境：随访前的间隔时间是太短、太长还是刚刚好呢？当计划进行不止一次随访时，问题就会变得更加复杂。作为一种部分解决方案，本文描述了一种技术，该技术可为多波研究中的随访波次间隔提供指导，避免研究设计中可能出现的严重低效情况。来自马萨诸塞州医疗保健小组研究（MHCPS）的关于老年人功能依赖的数据被用于演示。
