Attrition in the Longitudinal Aging Study Amsterdam

Abstract

This study addresses the relation between attrition and characteristics of the study protocol, specifically contact frequency, and respondent burden. The study is based on data from a longitudinal study with side studies on various topics, so that respondents have differential exposure to these study characteristics. Attrition outcomes are refusal and ineligibility through frailty. The effect of side study contact frequency and respondent burden on these outcomes is examined in two analytical samples: (1) baseline participants surviving to the first follow-up after 10 months (sample I), and (2) first follow-up participants surviving to the second follow-up after 3 years (sample II). Attrition during the first study interval was higher than during the second study interval, 15.5 and 5.4%, respectively. In sample I, the request to participate in a side study on social network implied an increased risk of refusal to participate at first follow-up if subjects refused the request (RR 8.34). However, if subjects participated in the network study, their risk of refusal was decreased (RR 0.42). In sample II, requests to participate in one to four side study cycles increased the risk of refusal to participate at second follow-up if subjects participated in fewer cycles than requested (RR 9.21). If subjects participated in all side study cycles that they were approached for, even if the number of cycles was five or more, this had an opposite effect: it decreased the risk of refusal (RR 0.18). Ineligibility was not significantly associated with contact frequency or respondent burden. Furthermore, neither contact frequency nor respondent burden related refusal was selective with respect to socio-demographic characteristics and physical and mental health indicators. It is concluded that contact frequency is nonlinearly associated with attrition. The findings further suggest that designing a series of side studies within the “longitudinal paradigm” does not severely damage the study's validity in terms of selective attrition.

Introduction

With the growing number of longitudinal studies in the older population, there is a concomitant increase in the concern about loss to follow-up in participants. Prevention of loss to follow-up or attrition is of major importance for two reasons [1]. First, attrition implies that a diminishing number of respondents is available for analysis, leading to less precision of estimates. Second, attrition is likely to be nonrandom. If attrition is selective with respect to the study variables, the analyses may yield biased estimates.

Roughly, one might distinguish two categories of causes of attrition: causes that cannot be influenced by the researchers (unmodifiable causes) and causes that are amenable to change by the researchers (modifiable causes). The first category of unmodifiable causes includes characteristics of respondents. Much research has concentrated on this category, providing evidence on attrition-related characteristics such as being an older adult or living in a large city [1]. Although such evidence may guide researchers in allocating special effort to groups of respondents with attrition-prone characteristics [2], the researchers cannot change the characteristics themselves. The efforts of researchers, then, are to be categorized in the second category of modifiable causes. Research efforts, as well as aspects of the study design, are characteristics of the study protocol. The choice of elements in the study protocol is basically up to the researchers. Little is known about the actual association between specific study characteristics and attrition.

There are several study characteristics that may affect the attrition rate [1]. First, when the target population is defined based on unstable criteria (e.g., living in a specific geographic area), respondents who at baseline fulfilled the criteria may become ineligible at subsequent cycles. Second, attrition will be larger when respondents are no longer contacted after they have dropped out only once. Third, the subject of the study may affect attrition [3]. For example, people who are not politically interested are not likely to continue participation in a political science study. A fourth attrition-related study characteristic is the amount of personal attention that is paid to the respondents [4]. For example, data collection by mail or telephone is less personal than face-to-face interviews, so that respondents approached by mailed questionnaire are less persuaded to continue participation than respondents approached by an interviewer. A fifth factor that is likely to be related to attrition is the time investment required of the respondent. A short interview or a brief questionnaire of a few pages may yield better continuation rates than a long interview or a lengthy questionnaire. Finally, the contact frequency is a sixth factor that may affect attrition. Although a high frequency may facilitate the tracing of respondents, so that fewer respondents are lost to follow-up because they have moved with unknown destination, it also requires a greater time investment of the respondent. Whereas some studies have shown evidence of a positive association between number of contacts and rate of attrition 5, 6, other studies show no evidence of an association [7], and others even found a negative association 8, 9. There is also evidence of a nonlinear association, showing that attrition was highest after one or two contacts, and decreased when the number of contacts increased 10, 11, 12.

Several of these study characteristics refer to the burden experienced by the respondent at each contact (respondent burden). Clearly, time investment (fifth characteristic) can be expected to increase the respondent burden [13]. Likewise, respondent burden is related to the subject of the study (third characteristic). The burden experienced by the respondent can be expected to increase with decreasing interest in the content of the survey, and vice versa. Furthermore, respondent burden may increase with increasing mental effort required per contact. For example, an interview about the death of a beloved person may require more mental effort than an interview about leisure time spending.

Generally, the evidence on attrition-related study characteristics stems from comparison across studies. Because most studies differ in more than one design aspect, this evidence cannot be considered “hard.” Only when one specific study characteristic is varied within one study, can it be actually tested whether this characteristic increases attrition. Traditionally, a longitudinal study design includes a fixed-length interval between follow-up cycles, while in each cycle exactly the same questions are asked of the respondents implying the same respondent burden (the “Longitudinal paradigm”). More recently, subsamples of respondents involved in on-going longitudinal studies are often being used for side studies, thus allowing more researchers to make efficient use of the existing cohort [14]. The consequence of this practice is that respondents are exposed differentially to various study characteristics. Some respondents may be contacted only once in several years, whereas others may be contacted more frequently, even monthly or weekly. Moreover, the interest generated and the mental effort required may differ across side studies.

The present study is based on data from the NESTOR-study on Living Arrangements and Social Networks (LSN) which was continued in the Longitudinal Aging Study Amsterdam (LASA), an on-going longitudinal study with side studies on various topics. This article focuses on modifiable causes of attrition, and addresses the question whether differential exposure to study characteristics affects the attrition rate. In particular are addressed the frequency of contacts and the respondent burden per side study. Because there is a large variation in these study characteristics across respondents, while other aspects of the study protocol are the same for all respondents, this study enables an actual test of the relation of these characteristics with attrition. Furthermore, the selectivity of attrition is examined by comparing baseline characteristics of groups of respondents exposed to different study characteristics, in relation to their attrition rate.