Attrition in a population-based cohort eight years after baseline interview: The Cornella Health Interview Survey Follow-up (CHIS.FU) Study
Introduction
The validity of longitudinal survey data can be threatened by several factors, including the dimensions of attrition: the response rate, the specific reasons for non-response, and the characteristics of non-respondents 1., 2.. The loss of subjects to follow-up can be a significant menace to the internal and external validity of longitudinal studies, when those lost differ from those found in the outcomes themselves or in the factors affecting outcomes 3., 4.. The study design depends upon the availability of subjects for repeated interviews. That is, the success of any longitudinal survey depends upon its subjects remaining in the study (5). If participants at the follow-up are not representative of all subjects from the baseline interview, intragroup changes cannot be generalized to the initial population.
Two categories of causes of attrition have been defined: causes that cannot be influenced by the researchers, including the characteristics of subjects; and causes that are modifiable by the researchers (6), such as research efforts and aspects of the study design. Of course, preventing loss to follow-up is the most desirable approach and is the only way to assure that selection bias will not occur (7). However, no matter how careful researchers are, there will always be such losses 8., 9., 10.. If the original sample was representative of a specific population, then survey analysis may provide misleading conclusions about changes in population characteristics over time if these individuals leave the sample in a non-random way (11).
Information on the type of attrition and on the possible determinants of attrition is important for a proper interpretation of the results of cohort studies (12). The most prominent types of attrition include those subjects who refuse further participation, those who cannot be located at follow-up, and those who have died during the follow-up period (13).
The best way to estimate non-response bias and its potential effect in epidemiological research is to know the characteristics of non-respondents (14). The comparison of the information available from the baseline allows an assessment of whether respondents differ from non-respondents (15). Therefore, the purpose of this study was to examine how response at follow-up varied from baseline sociodemographic data in a Spanish population-based cohort after 8 years of follow-up.
Section snippets
Methods
The Cornella Health Interview Survey Follow-up (CHIS.FU) Study is a population-based cohort study on lifestyles and their consequences on health. The cohort was set up with 2500 subjects (1263 women and 1237 men) randomly selected from the general population and interviewed in person in 1994 within the Cornella Health Interview Study (CHIS) 16., 17.. Cornella de Llobregat (http://www.cornellaweb.com) is a city with 85,000 inhabitants in the Metropolitan Area of Barcelona, Spain. We attempted to
Results
Table 1 shows the distribution of the study participants according to baseline variables and the result of the follow-up interview: subjects who responded to the general questionnaire, those who refused the interview, those found dead in the record linkage or after phone tracking, those found emigrated, and those finally non-traced. Table 2, Table 3, Table 4, Table 5 show the logistic regression analysis for each of these outcomes.
The variables associated with refusal in the bivariate analysis
Discussion
To detect a possible selection bias in follow-up we analyzed whether respondents differed from non-respondents. We have observed that the attrition was non-random, i.e., some characteristics were associated with specific causes of loss in the follow-up interview. Previous studies have reported response rates to telephone interviews and the presence of non-response bias related to lifestyle risk factors 20., 21., but few data have been published about health status differences among responders
References (35)
- et al.
Factors affecting attrition in a longitudinal smoking prevention
Prev Med
(1996) - et al.
Systematic bias in traumatic brain injury outcome studies because of loss to follow-up
Arch Phys Med Rehabil
(2003) - et al.
Is history of alcohol problems or of psychiatric disorder associated with attrition at 11-year follow-up?
Ann Epidemiol
(1996) - et al.
Attrition in longitudinal studies: How to deal with missing data
J Clin Epidemiol
(2002) - et al.
Características de no respondientes en un estudio de seguimiento de mayores de 65 años [Characteristics of non-responders in a follow-up study of those over 65 years of age.]
Gac Sanit
(1993) - et al.
Predictors of attrition in a longitudinal cognitive aging study: the Maastricht Aging Study (MAAS)
J Clin Epidemiol
(2002) Attrition in longitudinal population studies: Does it affect the generalizability of the findings? An introduction to the series
J Clin Epidemiol
(2002)Population-based cohort studies
Epidemiol Rev
(1998)- et al.
The relationship between prior functioning on cognitive and personality dimensions and subject attrition in longitudinal research
J Gerontol
(1988) - et al.
Advances in strategies for minimizing and adjusting for survey nonresponse
Epidemiol Rev
(1995)
Response and follow-up bias in cohort studies
Am J Epidemiol
Attrition in the Longitudinal Aging Study Amsterdam: The effect of differential inclusion in side studies
J Clin Epidemiol
Retaining and tracking cohort study members
Epidemiol Rev
Problems in designing and carrying out longitudinal research
Nord J Psychiatry
Attrition in prevention research
J Behav Med
Psychiatric and sociodemographic predictors of attrition in a longitudinal study: The Netherlands Mental Health Survey and Incidence Study (NEMESIS)
Am J Epidemiol
Characteristics of respondents and non-respondents to a mailed questionnaire
Am J Public Health
Cited by (0)
This study is partially funded by Fondo de Investigación Sanitaria (PI02/0261) and the Cornella de Llobregat City Council. M. Garcia received financial support from Instituto de Salud Carlos III (Network for Research in Epidemiology and Public Health, RCESP, C03/09).
- ∗
The CHIS.FU Study Group: Esteve Fernandez (principal investigator), Anna Schiaffino and Montse Garcia (study coordinators), and Mercè Martí, Esteve Saltó, Gloria Pérez, Mercè Peris, Carme Borrell, F. Javier Nieto, and Josep Maria Borràs (associate researchers).