This is the authors' abstract. We don't add key points for this paper.
To the Editor: Cohort studies of older people face particular problems of attrition due to death, as well as other types of dropout that occur over the course of the study.1-4 Few studies report on multiple types of attrition,5 and most use baseline data to examine predictors of attrition.1, 3, 4 This study aims to examine whether the same factors consistently predict attrition over multiple waves of a longitudinal study of older women. Data from 12,432 women born from 1921 to 1926 who participated in the Australian Longitudinal Study on Women's Health (ALSWH; http://www.alswh.org.au/) who were surveyed over six waves (1996, 1999, 2002, 2005, 2008, 2011) were used.6 Vital status was determined by linkage to the National Death Index. Participants were categorized at each wave as respondents, nonrespondents (women known to be alive who did not complete that survey but had not formally withdrawn), withdrawn (women known to be alive who asked not to receive any more surveys), or dead. The predictor variables (Table 1) were based on factors identified in previous studies1, 4, 5 and geographical location to account for the oversampling of women from rural and remote areas. Data were collected at each wave with the exception of smoking status (Waves 1, 2, 6 only) and country of birth and education (Wave 1). For variables with missing data from more than 400 respondents (arbitrarily chosen based on the data) at each wave, a separate “missing data” category was used. Multinomial logistic regression, with respondent status as the outcome, was used to model the associations with the predictor variables at the previous wave (or when most recently collected). Wave (centered at Wave 3) was included in the model to account for temporal effects. Interaction terms between wave and each individual variable were included in another model to assess whether any effect changed over time. Multilevel models were used with robust variance estimates to take account of repeated measures on individuals. Analyses were conducted in Stata version 13 (Stata Corp., College Station, TX). By the end of Wave 6, 4,055 women had responded, 979 had not responded, 2,430 had withdrawn, and 4,968 had died; of those who died, 1,925 had previously withdrawn. More than 70% of eligible women (those alive and who had not withdrawn) responded at each wave. Different variables were associated with each form of attrition and item nonresponse was related to future attrition (Table 1). Women born in non-English-speaking countries had a higher likelihood of nonresponse or withdrawal than those born in Australia, although significant interactions with wave suggest that this risk decreased over time: nonresponse, relative risk ratio (RRR) = 0.84, 95% confidence interval (CI) = 0.77–0.92; withdrawal, RRR = 0.89, 95% CI = 0.80–0.98. No other significant interactions were observed (all P > .30). This is the first study to report on patterns of multiple types of attrition using time-varying factors in a longitudinal study of older people, and the findings are consistent with those of previous studies that used data collected at baseline only (e.g., lower education, less-healthy behaviors, poorer health1, 3-5 were associated with attrition). Foreign-born participants in the U.S. Health and Retirement Study were more likely not to respond at some waves and withdraw, although the timing of attrition was not investigated.7 Findings from the ALSWH suggest that attrition of women born in non-English-speaking countries occurs across the course of the study but is most pronounced in earlier waves. The influence on attrition of all other factors remained remarkably constant over the 15 years of the ALSWH. Although questions about race or ethnicity were not asked in ALSWH, 73% of participants in this cohort were born in Australia, so the findings may not generalize to other populations or to men. Death occurred not only in the cohort being studied, but also in the general population.8 If factors related to death are the same as in the general population,1 losses due to death do not affect generalizability from the study to the general population, but for factors that are consistent predictors of attrition across all waves, any biases would accrue consistently. In conclusion, the same factors that predicted attrition due to death, withdrawal, and nonresponse at first follow-up continued to do so over multiple waves in a longitudinal cohort of older women. Groups of interdisciplinary researchers at the Universities of Newcastle and Queensland conceived and developed the ALSWH. We thank all participants for their valuable contribution to the study. This work was supported by the Australian Government Department of Health and the Australian National Health and Medical Research Council (Centre of Research Excellence APP1000986).
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Gardiner et al. (2015) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: