Description
Health issue: substance abuse Describe your selected practice gap. practice gap selected (information bias)Explain how your treatment of this population/issue could be affected by having awareness of bias and confounding in epidemiologic literature.Explain two strategies researchers can use to minimize these types of bias in studies, either through study design or analysis considerations.Finally, explain the effects these biases could have on the interpretation of study results if not minimized.APA style with scholarly sources attached and include additional that are within the last 5 years.
Unformatted Attachment Preview
Enzenbach et al. BMC Medical Research Methodology
https://doi.org/10.1186/s12874-019-0779-8
(2019) 19:135
RESEARCH ARTICLE
Open Access
Evaluating selection bias in a populationbased cohort study with low baseline
participation: the LIFE-Adult-Study
Cornelia Enzenbach1,2* , Barbara Wicklein1, Kerstin Wirkner2 and Markus Loeffler1,2
Abstract
Background: Participation in epidemiologic studies is steadily declining, which may result in selection bias. It is
therefore an ongoing challenge to clarify the determinants of participation to judge possible selection effects and
to derive measures to minimise that bias. We evaluated the potential for selection bias in a recent populationbased cohort study with low baseline participation and investigated reasons for nonparticipation.
Methods: LIFE-Adult is a cohort study in the general population of the city of Leipzig (Germany) designed to gain
insights into the distribution and development of civilisation diseases. Nine thousand one hundred forty-five
participants aged 40–79 years were randomly sampled in 2011–2014. We compared LIFE-Adult participants with
both the Leipzig population and nonparticipants using official statistics and short questionnaire data. We applied
descriptive statistics and logistic regression analysis to evaluate the determinants of study participation.
Results: Thirty-one percent of the invited persons participated in the LIFE-Adult baseline examination. Study
participants were less often elderly women and more often married, highly educated, employed, and current
nonsmokers compared to both the Leipzig population and nonparticipants. They further reported better health
than nonparticipants. The observed differences were considerable in education and health variables. They were
generally stronger in men than in women. For example, in male study participants aged 50–69, the frequency of
high education was 1.5 times that of the general population, and the frequency of myocardial infarction was half
that of nonparticipants. Lack of time and interest, as well as health problems were the main reasons for nonparticipation.
Conclusions: Our investigation suggests that the low baseline participation in LIFE-Adult is associated with the typical
selection of study participants with higher social status and healthier lifestyle, and additionally less disease. Notably,
education and health status seem to be crucial selection factors. Consequently, frequencies of major health conditions in
the general population will likely be underestimated. A differential selection related to sex might also distort effect
estimates. The extent of the assessment, the interest in the research topic, and health problems of potential participants
should in future be considered in LIFE-Adult and in similar studies to raise participation and to minimise selection bias.
Keywords: Participation, Selection bias, Validity, Reasons for nonparticipation, Cohort study
* Correspondence: [email protected]
1
Institute for Medical Informatics, Statistics, and Epidemiology, University of
Leipzig, Haertelstrasse 16-18, 04107 Leipzig, Germany
2
LIFE – Leipzig Research Centre for Civilization Diseases, University of Leipzig,
Philipp-Rosenthal-Strasse 27, 04103 Leipzig, Germany
© The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
(2019) 19:135
Page 2 of 14
Background
Participation has declined over the past decades for all
types of epidemiologic studies [1]. The decreased willingness to participate in an epidemiologic study may
threaten the validity of the results. Those who volunteer
for study participation are often more likely to have
favourable exposure and health profiles compared to
those who do not. Consequently, estimates of prevalence, incidence, and exposure-disease associations may
be biased. This error is referred to as response bias or,
more broadly, selection bias [2]. Although being a potentially important precondition for the validity of an epidemiologic study, participation is often insufficiently
reported in the publication of the results [1, 3].
The presence of selection bias can usually not be inferred from the study data alone. We need to compare
study participants with nonparticipants or with the target population in terms of relevant characteristics to
judge possible selection effects on the study results [4,
5]. For such comparisons, we have to collect some core
information from nonparticipants as well, using short
questionnaires or secondary data. In addition, data on
the target population may be obtained from official statistics or representative surveys.
Using these methods, the potential for selection bias has
been investigated in epidemiologic studies in the general
population for many years (e.g., [6–15]. These studies have
predominantly shown that participants in baseline examinations of cohort studies and in cross-sectional studies are
more likely to be female and to have higher social status,
healthier lifestyles, and better subjective health than nonparticipants. Results are contradictory with respect to age
and prevalent diseases. These observations have been
made for participation rates of mainly above 50%.
The LIFE-Adult-Study is a recent population-based cohort study conducted in the city of Leipzig, Germany [16].
An extensive programme consisting of questionings, physical examinations, and biologic specimen collections was
established to better understand the distribution and the
development of civilisation diseases. With a response of
about 30%, the participation in LIFE-Adult was markedly
lower than in previous cohort and cross-sectional studies
that had examined selection bias. In light of this low participation and the claimed higher susceptibility of studies with
low levels of participation to selection bias [1, 17], we were
seeking for an in-depth understanding of the determinants
of response in our study.
Our primary objective was to evaluate the potential for selection bias in LIFE-Adult applying two independent
methods: (1) the comparison of LIFE-Adult participants with
the Leipzig population with regard to socio-demographic and
lifestyle characteristics using official statistics and (2) the comparison of LIFE-Adult participants with nonparticipants additionally considering health-related variables by means of
short questionnaire data. Furthermore, we investigated reasons for nonparticipation given in the short questionnaire by
describing their distribution and their relations to the individuals’ characteristics.
Enzenbach et al. BMC Medical Research Methodology
Methods
Study design and participants
LIFE-Adult-Study
LIFE-Adult is a cohort study designed (1) to estimate prevalences and incidences of common diseases and subclinical
phenotypes in the adult population of Leipzig and (2) to investigate the interplay of molecular-genetic and lifestyle factors in the development of these conditions.
Participants in LIFE-Adult are an age and gender
stratified random sample of the general population of
Leipzig mainly aged 40 to 79 years, which was drawn by
the registration offices. All selected residents were sent
an invitation letter with information on the study.
Persons who had not responded within four weeks received a reminder letter. Those who had not responded
within further two weeks were contacted by phone (see
reference [16] for more details on recruitment).
The baseline assessment took place between August 2011
and November 2014. All participants underwent a core assessment consisting of interviews and questionnaires, physical
examinations, and collection of blood and urine (average duration 5 to 6 h). Participants aged 60 to 79 years were invited
to additional assessments focusing on cognitive function and
depressive symptoms on two further days (average duration 3
to 4 h each).
The assessments were conducted in the LIFE-Adult study
centre, which is located in the city centre and easy to reach.
Participants received 20 Euro per visit to cover their travel expenses. They were also offered selected examination results in
written form. In addition, several public relation activities
were organised to raise participation.
Persons unwilling to participate in LIFE-Adult were
asked to fill in a short questionnaire, which was enclosed
in the first invitation and the reminder letter since
January 2012. The questionnaire comprised 17 questions
related to socio-demography, lifestyle, health status, and
reason for nonparticipation.
In the present investigation, we included participants
in LIFE-Adult who were in the study’s main age range
from 40 to 79. For the comparison with short questionnaire participants by means of regression analysis, we
further restricted the population to study participants
who had received the first invitation since January 2012.
Out of all short questionnaire participants, we considered those aged between 40 and 79.
Census and microcensus
We obtained data on the Leipzig population from the
census and the microcensus.
(2019) 19:135
Page 3 of 14
Data on the sex and age distribution within Leipzig
come from the national census, which is conducted
every ten years [18]. The data represent population updates by 30 June 2013 (based on census data from May
2011). At that time, about half of the LIFE-Adult population was recruited.
The microcensus is a representative statistics of the
population and labour market conducted annually in
Germany [19]. The sample comprises 1 % of all households. A fixed set of socio-demographic characteristics is
assessed each year using mainly computer-assisted interviews in the households. Respondents are obligated to
answer these questions, resulting in high response figures (e.g., unit-response 97.6% and item-response > 97%
in the year 2013 [19]). Additionally, variable topics are
addressed every four years on a voluntary basis. We used
public microcensus data of the year 2013 representing
the annual average. For each characteristic, extrapolated
numbers per sex and age strata were available. To prevent misinterpretation due to random error, numbers
less than 7000 for a given strata are generally not released and numbers below 10,000 should be interpreted
cautiously. We had to consider this when selecting and
handling the analysis variables.
illustration and explanation of the categories of individuals). We also calculated the recruitment efficacy
proportion by excluding from the denominator those
nonparticipant categories that cannot be influenced
much by the investigator [21], namely the persons
who could not be contacted and those willing to
participate.
We calculated participation in the short questioning
by relating the number of short questionnaire participants to all invited persons who did not participate,
namely refusals, nonresponders, persons who could not
be contacted, and persons willing to participate.
Enzenbach et al. BMC Medical Research Methodology
Variables
We selected major risk factors and health conditions as
variables for analysis. For the comparison of LIFE-Adult
participants with the Leipzig population, we considered
sex and age, as well as marital status, education, employment, and smoking status. For the comparison of LIFEAdult participants with short questionnaire participants,
we additionally chose physical condition and medically
diagnosed myocardial infarction, stroke, diabetes, and
cancer. We did not consider those items of the short
questionnaire for which corresponding data were not
available from study participants (e.g., sports activities)
or for which the assessment methods were not comparable between the two populations (weight status).
A detailed definition of each analysis variable in each
population is given in Additional file 1: Table S1.
Data analysis
Calculation of participation
We calculated participation in LIFE-Adult using two
different measures. The response proportion is the
percentage of persons that participated out of the
total number of persons who had been eligible for
study [20]. Our denominator comprised LIFE-Adult
participants, persons willing to participate, refusals,
nonresponders, and persons who could not be contacted, including persons with unknown address,
those who had died before contact could be made,
and persons with running invitations (see Fig. 1 for
Comparison of LIFE-Adult participants with the Leipzig
population and short questionnaire participants
We compared LIFE-Adult participants with the Leipzig
population and with short questionnaire participants
using descriptive statistics. We thereby investigated
whether there were sex or age differences in selective
participation. For this, we calculated relative frequencies
of study variable values according to sex and 10-year age
groups. We dichotomised variable values and chose
reference groups in a way that ensured reliable microcensus data. As only summary data were available from
official statistics, we could not indicate the precision of
the estimated frequencies at this stage of analysis.
We investigated the differences between LIFE-Adult
and short questionnaire participants in more detail by
means of logistic regression, taking into account the uncertainty of the estimates and explanatory factors. We
estimated odds ratios and 95% confidence limits. Participation in LIFE-Adult was the dependent variable. In a
first model series, we included each analysis variable
separately as independent variable. In a second model
series, we analysed the association of each variable with
study participation controlling for differences in the age
distribution between study and short questionnaire
participants. In a third model series, we examined to
what extent the observed associations may be attributed
to differences in social status by additionally including
school education as independent variable. We estimated
all associations separately for men and women according
to the observations in the descriptive analysis.
Calculation of completeness of the data
For all analysis variables, we calculated the completeness
of the data for LIFE-Adult and short questionnaire participants by sex and age. Completeness is defined as the
number of non-missing data divided by the total number
of the population. Missing data include questioning and
item nonresponse, the answer categories “I don’t know”
and “refusal of answer”, and erroneous data.
Enzenbach et al. BMC Medical Research Methodology
(2019) 19:135
Page 4 of 14
Fig. 1 Participation in LIFE-Adult and in the short questioning, age range 40 to 79 years. Sample 1 of LIFE-Adult participants was used for the comparison with
the Leipzig population and with short questionnaire participants using descriptive statistics (see Table 1). Sample 2 of LIFE-Adult participants was used for a
more detailed comparison with short questionnaire participants using logistic regression (see Table 2, as well as the “Methods” section for further explanation).
Invitation running refers to those invitees who had been sent an invitation few weeks before the end of the recruitment and who did not respond within that
time frame. Persons willing to participate are those invitees who had agreed to participate in LIFE-Adult but did not get an appointment because the targeted
total number of participants had been achieved. Refusals are those invitees who actively declined to participate by means of a response form enclosed in the
invitation letters or by phone. Nonresponders are those invitees who entirely ignored the invitation. Data available for analysis refers to the number of nonmissing data for each variable. Missing data include questioning and item nonresponse, the answer categories “I don’t know” and “refusal of answer”, and
erroneous data. ISCED 97 = International Standard Classification of Education 1997
Analysis of reasons for nonparticipation
The reason for nonparticipation had been asked in the
short questionnaire by the question “For which reasons
do you not want to participate in our study? Please state
the most important reason.” The answer categories comprised lack of time, job-related reasons, no interest,
doubts about the value of the study, health reasons,
moved, language reasons, no information on reasons,
other reason: which one.
Before the analysis, we combined non-exclusive categories, namely “lack of time” and “job-related reasons”,
“no interest” and “doubts about the value of the study”,
(2019) 19:135
Page 5 of 14
and “no information on reasons” and missing data. If
possible, we matched answers in the category “other reason” to more meaningful categories. However, we subsumed categories with very few cases (moved and
language reasons) in the category “other reason”. We
checked the “comment” field for nonparticipation reasons and replaced missing data if possible. We further
checked the fields “other reason” and “comment” to possibly identify the most important reason in case of multiple answers.
We calculated relative frequencies of the final reasons
for nonparticipation for all respondents and according
to sex, age (40 to 64 vs. 65 to 79 years), and school education as an indicator of social status.
We used SPSS (IBM SPSS Statistics), version 24, for
our calculations.
When comparing LIFE-Adult with short questionnaire participants, similar and additional differences
were observed (Table 1). LIFE-Adult participants were
more often married among those older than 50 years,
particularly in men. They had a higher school qualification and were more often current nonsmokers in
all ages with greater differences in men. LIFE-Adult
participants were more often employed in all age
groups and in both sexes. They were less often in
poor physical condition among men in all ages but
particularly at the age of 70 to 79. In women, this
difference was observed only in the oldest age group.
LIFE-Adult participants reported less often to have
been diagnosed with myocardial infarction and diabetes, irrespective of age and sex. With regard to
stroke, there was an analogous difference among
those older than 60 years. As to the frequency of diagnosed cancer, inconsistent and generally small differences between the two populations were found
across age and sex strata. The deviations of LIFEAdult from short questionnaire participants were particularly pronounced in education and health variables. For example, the frequency of high education
in male study participants was 1.3 times that of male
short questionnaire participants in the age range 50
to 69. For myocardial infarction, the corresponding
ratio was 0.5. Including in the analysis only those
LIFE-Adult participants invited since the beginning of
the short questioning did not affect the aforementioned differences (data not shown).
In the logistic regression analysis, in both sexes the
odds of being participant in LIFE-Adult was lower
among those aged 70 to 79, having low or no school
qualification, being in poor physical condition, and having been diagnosed with myocardial infarction, diabetes,
or stroke, whereas it was higher among those being
employed (Table 2, model 1). In addition, in men, the
odds of being LIFE-Adult participant was lower among
current smokers, whereas in women it was higher
among former smokers.
After adjustment for differences in the age distribution, physical condition remained associated with study
participation only in men (Table 2, model 2). In women,
the odds of study participation was also lower among
current smokers albeit weaker than in men. Additionally,
the odds of being LIFE-Adult participant was higher
among married persons in men. The associations of education, employment, and diagnosed diseases with study
participation remained directed as in the unadjusted
models, although slightly attenuated.
After further adjustment for school education, the
above mentioned associations between the individuals’
characteristics and study participation were still present
and only slightly weakened (Table 2, model 3).
Enzenbach et al. BMC Medical Research Methodology
Results
Participation in LIFE-Adult and in the short questioning
The numbers of individuals aged 40 to 79 at different
stages of the study are presented in Fig. 1. Nine thousand one hundred forty-five persons participated in
LIFE-Adult, resulting in a response proportion of 31%
and a recruitment efficacy proportion of 32.1%. Among
nonparticipants, 6475 persons filled in the short questionnaire, corresponding to a participation rate of 31.8%.
Participants in LIFE-Adult in comparison with the Leipzig
population and short questionnaire participants
In comparison with the Leipzig population, the percentage of women aged 75 to 79 was considerably lower in
LIFE-Adult (6.2% vs. 12.3%, Fig. 2). Compared to short
questionnaire participants, the percentage of both
women and men aged 75 to 79 was markedly lower in
LIFE-Adult (women: 6.2% vs. 12.9%, men: 7.8% vs.
12.3%).
LIFE-Adult participants differed from the Leipzig
population in all other selected characteristics (Table 1).
They were more often married at ages 40 to 69 years in
both sexes. They were higher educated in all age groups
with stronger differences observed in men. They were
more often employed in the considered age from 40 to
59 in both men and women. Finally, they were more
often current nonsmokers in men. The differences between LIFE-Adult participants and the Leipzig population were most pronounced in school education. For
example, the frequency of 1st stage tertiary education in
male study participants was 1.5 times the frequency in
the male Leipzig population in the age range 50 to 69
(see Table 1 for the corresponding frequencies). Regarding the other variables, the frequencies in LIFE-Adult
were less than 1.2 times the frequencies in the Leipzig
population.
Enzenbach et al. BMC Medical Research Methodology
(2019) 19:135
Page 6 of 14
a
b
Fig. 2 Age distribution in LIFE-Adult participants, the Leipzig population, and short questionnaire participants. a. Men b. Women
Completeness of the data
In LIFE-Adult, the completeness of the data was very
high (≥ 98.4%) for variables that had been assessed by
interview (see Table 3 for a selection of variables). For
these variables, the completeness was lower in short
questionnaire participants but above 95%, except for
school education. Among those older than 60 years, the
completeness was lower in LIFE-Adult than in short
questionnaire participants for variables that had been
assessed by questionnaires in LIFE-Adult, namely smoking and physical condition. The percentage of available
data was lowest among women aged 70 to 79 for questionnaire variables in LIFE-Adult (about 86%) and for all
characteristics in short
(mainly about 95%).
questionnaire
participants
Reasons for nonparticipation
In the raw data, reasons for nonparticipation were distributed as follows: lack of time 21.3%, job-related reasons
2.4%, no interest 12.6%, doubts about the value of the
study 3.9%, health reasons 11.7%, moved 0.7%, language
reasons 0.9%, other reason 5.7%, multiple answers 13.6%,
no information on reasons (including missing data) 27.2%.
After data preparation, six categories of nonparticipation
reasons remained. “Lack of time” was the most frequent
reason with 30.3%, followed by “no interest” with 19.0%
Enzenbach et al. BMC Medical Research Methodology
(2019) 19:135
Page 7 of 14
Table 1 Characteristics of LIFE-Adult participants, the Leipzig population, and short questionnaire participants by sex and age
Men
Age, years
Women
40 to 49
50 to 59
60 to 69
70 to 79
40 to 49
50 to 59
60 to 69
70 to 79
LIFE-Adult participants
48.8
65.3
80.7
85.3
52.7
65.8
68.1
55.5
Leipzig population
44.7
60.5
69.7
84.6
49.2
59.2
60.2
57.5
SQ participants
48.2
60.2
73.5
81.9
56.3
60.5
66.0
55.9
46.6
50.6
62.1
73.2
51.0
51.5
47.7
43.7
34.7
32.6
34.8
34.8
34.7
31.4
27.4
21.1
35.4
34.6
(40.2)
56.0
45.5
41.1
35.3
36.8
32.4
24.8
27.1
27.8
33.2
26.9
25.2
16.5
90.7
84.0
31.3
4.2
90.9
83.9
26.1
2.2
Married, %
Highly educated, %
LIFE-Adult participants
1st stage tertiary education
a
Hochschulreife
Leipzig population
1st stage tertiary education
SQ participants
Hochschulreifea
Employed, %
LIFE-Adult participants
Leipzig population
86.2
79.6
(38.9)
/
84.6
75.7
(24.9)
/
SQ participants
85.2
76.6
23.9
2.9
83.3
75.4
17.2
1.9
Nonsmokers of tobacco
66.1
68.7
82.2
93.0
71.1
71.4
87.0
94.6
Nonsmokers of cigarettes
67.4
70.0
83.9
94.4
71.4
71.7
87.2
94.6
57.7
63.1
73.6
92.1
70.0
74.7
85.0
97.0
59.8
62.3
75.8
87.5
67.3
67.1
86.0
92.9
LIFE-Adult participants
2.2
3.7
4.6
4.6
3.0
4.2
5.0
6.0
SQ participants
2.7
5.3
6.6
11.7
3.0
3.6
4.9
9.9
LIFE-Adult participants
0.5
2.5
5.2
9.2
0.1
0.9
1.8
2.6
SQ participants
0.7
4.9
10.0
13.1
0.7
1.1
2.5
4.4
LIFE-Adult participants
0.7
2.1
3.4
5.9
0.4
1.9
2.2
3.4
SQ participants
0.8
2.3
5.8
9.3
0.2
2.3
3.3
5.1
LIFE-Adult participants
3.4
9.7
20.5
23.0
1.8
6.3
13.4
19.8
SQ participants
6.0
13.2
30.1
34.7
3.5
12.7
21.0
28.1
Current nonsmoker, %
LIFE-Adult participants
Leipzig population
Nonsmokers of tobacco
SQ participants
Nonsmokers of cigarettes
Poor physical condition, %
Myocardial infarction, %
Stroke, %
Diabetes, %
Enzenbach et al. BMC Medical Research Methodology
(2019) 19:135
Page 8 of 14
Table 1 Characteristics of LIFE-Adult participants, the Leipzig population, and short questionnaire participants by sex and age
(Continued)
Men
Age, years
Women
40 to 49
50 to 59
60 to 69
70 to 79
40 to 49
50 to 59
60 to 69
70 to 79
LIFE-Adult participants
2.2
4.6
12.1
24.0
6.7
8.2
13.9
18.9
SQ participants
2.9
7.1
10.8
21.5
4.8
10.1
14.7
18.0
Cancer, %
Data for the Leipzig population: Percentages corresponding to less than 7000 cases are marked by “/”, percentages corresponding to less than 10,000 cases are
given in parenthesis. aHochschulreife = technical college or university entrance qualification, SQ = short questionnaire
and “health reasons” with 14.3%. The categories “other
reason” and multiple answers contained 6.0 and 4.2%, respectively. From 26.2% of the respondents, no reason for
nonparticipation was available. Within the “other reasons”,
“enough medical care” was mentioned particularly often
(in total 2.4%).
“Lack of time” was by far the most common reason
among the younger respondents (40 to 64 years) and was
reported much more frequently in this group (Fig. 3). In
contrast, the older respondents (65 to 79 years) gave
“health reasons” much more frequently, as well as “no
interest” and no reason for nonparticipation. Respondents
Table 2 Associations of individuals’ characteristics with study participation: LIFE-Adult participants versus short questionnaire
participants
Men
Model 1
Aged 40 to 44 y
Women
Model 2
Model 3
Reference
Model 1
Model 2
Model 3
Reference
Aged 45 to 49 y
1.11 (0.92–1.35)
1.14 (0.96–1.36)
Aged 50 to 54 y
0.99 (0.81–1.21)
1.16 (0.97–1.39)
Aged 55 to 59 y
0.81 (0.67–0.99)
1.01 (0.84–1.21)
Aged 60 to 64 y
0.80 (0.66–0.97)
1.07 (0.90–1.28)
Aged 65 to 69 y
0.85 (0.70–1.03)
0.80 (0.67–0.96)
Aged 70 to 74 y
0.69 (0.58–0.83)
0.68 (0.58–0.81)
Aged 75 to 79 y
0.53 (0.43–0.65)
0.42 (0.35–0.51)
POS/Realschulea
Reference
Reference
Reference
Reference
Hochschulreifea
1.13 (1.01–1.26)
1.16 (1.04–1.29)
1.06 (0.95–1.18)
1.07 (0.96–1.19)
Hauptschulea
0.40 (0.35–0.47)
0.45 (0.38–0.53)
0.45 (0.39–0.52)
0.56 (0.48–0.66)
Other/no qualification
0.56 (0.42–0.76)
0.58 (0.42–0.78)
0.33 (0.24–0.45)
0.35 (0.25–0.48)
Married
1.08 (0.98–1.20)
1.25 (1.12–1.39)
1.20 (1.08–1.34)
1.04 (0.95–1.14)
1.02 (0.93–1.12)
1.00 (0.91–1.10)
Employed
1.61 (1.47–1.78)
1.56 (1.36–1.78)
1.42 (1.24–1.64)
1.79 (1.64–1.96)
1.63 (1.43–1.86)
1.53 (1.34–1.75)
Never smokerb
Reference
Reference
Reference
Reference
Reference
Reference
Former smokerb
0.90 (0.80–1.00)
0.94 (0.84–1.05)
0.98 (0.87–1.10)
1.33 (1.18–1.50)
1.19 (1.05–1.34)
1.19 (1.05–1.34)
Current smokerb
0.72 (0.63–0.81)
0.62 (0.54–0.70)
0.69 (0.60–0.79)
1.07 (0.95–1.20)
0.86 (0.76–0.98)
0.88 (0.77–1.00)
Poor physical condition
0.50 (0.40–0.62)
0.55 (0.44–0.69)
0.60 (0.48–0.75)
0.79 (0.64–0.97)
0.91 (0.73–1.12)
0.99 (0.79–1.22)
Myocardial infarction
0.50 (0.40–0.61)
0.56 (0.46–0.70)
0.57 (0.46–0.70)
0.50 (0.36–0.72)
0.61 (0.43–0.87)
0.65 (0.46–0.94)
Stroke
0.58 (0.46–0.75)
0.67 (0.52–0.86)
0.68 (0.53–0.88)
0.61 (0.45–0.83)
0.71 (0.52–0.96)
0.79 (0.58–1.09)
Diabetes
0.55 (0.48–0.62)
0.60 (0.52–0.68)
0.62 (0.54–0.71)
0.50 (0.44–0.58)
0.58 (0.51–0.67)
0.62 (0.53–0.71)
Cancer
0.87 (0.75–1.01)
1.03 (0.88–1.20)
1.01 (0.86–1.19)
0.91 (0.79–1.04)
1.02 (0.88–1.17)
1.02 (0.88–1.18)
Association measures are odds ratios (95% confidence limits). The dependent variable is participation in LIFE-Adult vs. participation in the short questioning.
Model 1: crude association of each analysis variable with study participation, model 2: adjustment for age (40 to 44, 45 to 54, 55 to 64, 65 to 69, 70 to 74, 75 to
79 years), model 3: adjustment for age and school education (Hauptschule, POS/Realschule, Hochschulreife, other/no qualification). For dichotomous variables, the
reference category is not shown. Example of interpretation: In male persons with the diagnosis of a myocardial infarction, the odds of being LIFE-Adult participant
is 0.50 times as big as the odds of those without a diagnosis of myocardial infarction (model 1)
a
School qualification: Hauptschule = certificate of primary education, POS/Realschule = certificate of polytechnic secondary school/secondary education,
Hochschulreife = technical college or university entrance qualification. bSmoking status refers to cigarette smoking. y = years
Enzenbach et al. BMC Medical Research Methodology
(2019) 19:135
Page 9 of 14
Table 3 Completeness (%) of selected variables in LIFE-Adult participants and short questionnaire participants by sex and age
Men
Age, years
40 to 49
Women
50 to 59
60 to 69
70 to 79
40 to 49
50 to 59
60 to 69
70 to 79
School education
LIFE-Adult participants
100
100
99.9
99.7
100
99.9
99.9
100
SQ participants
93.8
93.5
93.9
93.9
96.4
95.7
95.1
91.9
LIFE-Adult participants
100
99.9
99.6
99.9
100
99.9
99.7
99.9
SQ participants
95.2
95.5
96.5
97.1
96.9
96.9
97
95.5
Employment
Cigarette smoking
LIFE-Adult participants
97.4
95.3
94.0
89.9
97.8
97.2
95.1
86.8
SQ participants
95.2
94.7
95.8
96.8
96.4
96.8
94.6
94.0
LIFE-Adult participants
97.9
96.4
94.4
87.6
98.5
96.6
91.7
85.5
SQ participants
97.2
96.6
96.8
97.7
97.9
97.5
97.0
95.0
Physical condition
Myocardial infarction
LIFE-Adult participants
98.7
99.4
98.8
98.6
99.7
99.1
99.6
98.9
SQ participants
95.8
97.5
96.3
96.5
97.7
97.3
96.9
94.8
Completeness is defined as the number of non-missing data divided by the total number of the sample. Sample 2 of LIFE-Adult participants (see Fig. 1) was used.
SQ = short questionnaire
with high school education stated time reasons much
more frequently and had less missing information (in the
younger age group only). In contrast, lower educated persons more often answered with “no interest” and “health
reasons”. There was also a tendency of men giving more
often “no interest” as the reason for nonparticipation compared to women.
Discussion
Key results
LIFE-Adult is a cohort study aimed at providing prevalence and incidence estimates for the Leipzig population,
as well as insights into the development of common
diseases.
In the study’s main age range from 40 to 79, 31% of
the invited persons participated in the baseline
examination.
We compared these study participants with both the
target population and short questionnaire participants to
evaluate the potential for biased study results due to selective participation. Both approaches suggest that participants in LIFE-Adult are less often elderly women and
more often married, highly educated, employed, and
current nonsmokers. In addition, the data of the short
questioning point to LIFE-Adult participants being less
often in poor health. The differences between LIFEAdult participants and the comparison populations were
particularly pronounced in education and health variables. Besides, they were partly stronger in men than in
women.
Nonparticipation in LIFE-Adult was most often justified with lack of time, lack of interest, and health problems. The reason for nonparticipation strongly depended
on age and education of the respondent.
Strengths and limitations
In contrast to some other countries [5
