Participation in LIFE-Adult
The baseline participation in LIFE-Adult was substantially lower than in previous cohort and cross-sectional studies in Germany and worldwide [
1,
6‐
13,
25‐
28], with reported median participation of above 70% [
1]. This may be mainly due to the steady decline in participation in epidemiologic research over the past about four decades [
1,
5]. Less extensive recruitment procedures [
13] and certain characteristics of the target population, as presence of higher age groups and urbanity [
25,
27], may have contributed to the comparatively low response.
Reasons for nonparticipation
Our data on reasons for nonparticipation suggest that time and health constraints, as well as lack of interest contributed to the low participation. They are in line with other epidemiologic studies after which nonparticipation is predominantly justified with lack of time and/or interest [
7‐
9,
22‐
24,
28,
29]. Health reasons have been frequently given in some studies, too [
8,
9,
23,
29]. An accumulation of time reasons in younger persons and health reasons in older persons has also been reported [
28,
29]. We additionally found that higher educated persons more often cite time constraints, whereas lower educated persons more frequently mention lack of interest and health problems.
The length of the assessment is an important determinant of participation [
17]. Studies requiring a substantial amount of time have lower response rates than studies with lower participant burden [
17,
30,
31]. The extent of our baseline assessment may have been an obstacle to participation, especially for persons in the working age and those with a strong professional commitment.
The topic under investigation often influences response most [
17]. People are much more interested in participating in a study that explores an issue particularly salient to their lives [
30,
31]. The objectives of LIFE-Adult had been broadly formulated so that many of the invited persons might not have seen a personal significance. A diminishing enthusiasm for science in general could be of relevance, too [
31]. As our data suggest, a lack of interest in (this kind of) research is of particular concern in subgroups of the population.
Finally, participation in an epidemiologic study can be demanding in many respects [
31]. Our complex study design may have meant a great burden, particularly for the elderly who have limited physical resources – but regrettably also the diseases under study.
Selective participation in LIFE-Adult
It is widely recognised that not low participation itself but differences between participants and nonparticipants in relevant characteristics threaten the validity of a study [
4,
5,
17]. Our investigation suggests that those who participated in LIFE-Adult considerably differ from those not included in the study, particularly in terms of education and health status.
Our results are consistent with previous research that has predominantly shown that participants in epidemiologic studies are more likely to be married, highly educated, and employed in comparison with nonparticipants (e.g., [
6‐
10,
12‐
15,
28,
32]. The latter two characteristics are related to higher social status. On the one hand, persons with higher social status may be more time constrained. On the other hand, their overrepresentation in epidemiologic studies likely reflects greater health awareness and interest in science [
31].
Our findings are also in accord with the observation that nonparticipants in epidemiologic studies more often report poor subjective health [
6‐
8,
10,
12]. Our data further indicate that persons diagnosed with a common disease are less likely to participate in studies like ours. The impact of prevalent diseases on study participation has been investigated with conflicting results. Both no relation between disease status (including cardiovascular diseases, stroke, and diabetes) and response [
6,
14] and lower participation rates among diseased persons [
10,
15,
26,
28,
33], as well as higher participation associated with disease [
11,
23,
32] have been reported. The possible underrepresentation of ill persons in LIFE-Adult may be explained with several mechanisms, including lower health awareness, physical constraints hampering study participation, already high burden by frequent visits to the doctors, and satisfactory medical care (of course, representing a misunderstanding of the study’s aims).
Furthermore, our data are consistent with available evidence after which current smokers are underrepresented among study participants [
6‐
9,
14,
15,
28]. An unhealthy lifestyle is likely to be related to lower identification with the objectives of an epidemiologic study. Also, studies that are perceived to be concerned with socially undesired behaviour may have difficulties to recruit participants who practise such behaviour [
31].
Our observation that older people, in particular women, are less likely to be among the study participants is in line with some studies, too (e.g., [
8,
10,
13,
22,
23,
29]. Especially elderly women refused to participate because they had to take care for relatives, mostly their husbands [
8]. We found that also the response to the study’s questionnaires was lowest among elderly women, as observed in another study [
34]. This might partly reflect low familiarity with modern methods of data collection, as a preference for the paper to the computer versions of our questionnaires among elderly women indicates (data not shown).
Our results suggest that selection into the study population may be more pronounced in men than in women, whereas little difference seems to exist between age groups in the range from 40 to 80. Our findings are corroborated by few studies that also observed stronger relations of response to marital status, education, smoking status, and subjective health among males [
8,
9,
33], whereas age did not modify these associations [
8,
15]. Our observation supports the hypothesis that less health-conscious men are less willing to participate in surveys than their female counterparts [
35].
It is often argued that studies with a low response, typically below about 50%, are particularly prone to selection bias [
1,
17,
24]. However, studies with substantially higher response than LIFE-Adult, largely between 50 and 75%, mainly reported differences between participants and nonparticipants qualitatively similar to those found in our study as discussed above. The magnitude of these differences was also sizable in various studies (e.g., [
8,
10,
13,
28]. In line with these findings, a marked increase in response in a health survey from 37 to 60% brought about by multiple reminders did not eliminate existing differences between participants and nonparticipants [
13].
Impact of selective participation on study results
Selective participation in epidemiologic studies primarily affects the description of the health status of a population [
36‐
39]. For that purpose, study participants have to be representative of the target population with respect to the characteristics of interest. Therefore, as a consequence of overrepresentation of healthy and health-conscious persons in LIFE-Adult, frequencies of major risk factors and diseases in the Leipzig population will likely be underestimated. Weighting the study data to match the target population distribution for selected socio-demographic features is a common approach to correct for nonresponse in prevalence estimates [
2,
22]. The census and microcensus data inform us about the distribution of important socio-demographic characteristics in the Leipzig population, thus enabling us to calculate corresponding weighting factors. However, our regression models suggest that the differences between LIFE-Adult participants and nonparticipants in lifestyle and health variables may be attributed only to a small extent to differences in the distributions of age and education. Thus, weighting prevalence estimates of lifestyle and health characteristics for socio-demographic factors might insufficiently adjust for selection bias in LIFE-Adult.
The validity of analytic-epidemiologic studies is not necessarily impaired by selective participation [
36‐
39]. Estimates of exposure-outcome associations may be biased if selection into the study population depends on both the exposure and the outcome [
2,
5]. This situation, also termed differential selection, might particularly affect the internal validity of cross-sectional studies [
1]. Evidence for such bias comes from studies that could compare associations among study participants with those in the target or the total nonparticipant population. Among survey participants with low socio-economic status, subjective health was better compared to corresponding census participants [
12]. As a result of this differential selection, the survey underestimated the relation of socio-economic status to health. Furthermore, baseline associations between socio-demographic variables and health status partly differed in direction between participants in a cohort study and nonparticipants [
27]. We did not examine selection bias at estimates of cross-sectional relations due to the lack of relevant data on the target population and the likelihood of selective participation even in the short questioning. However, a differential selection related to sex as indicated by our findings may bias the effects of sex on health conditions [
35].
The validity of longitudinal studies is assumed to be primarily threatened by selective loss to follow-up, whereas selection at baseline is considered rather harmless [
30]. There are indications that participation in follow-up examinations follows similar selection patterns as participation at recruitment, particularly with regard to socio-demographic and lifestyle factors [
5,
25]. Yet, existing evidence suggests that effects on selected exposure-outcome associations are generally small as differential selection seems to be modest [
5,
40]. However, the actual impact of selective participation, both at baseline and at subsequent follow-ups, on the validity of prospective studies has to be further explored [
4,
5,
40].