Content uploaded by Nancy Hoeymans
Author content
All content in this area was uploaded by Nancy Hoeymans
Content may be subject to copyright.
Age and
Ageing
1998; 27: 35-40
Non-response bias in a study of
cardiovascular diseases, functional
status and self-rated health among
elderly men
NANCY
HOEYMANS12, EDITH J. M. FESKENS1, GEERTRUDIS A. M. VAN DEN BOS2, DAAN KROMHOUT1
'Department of Chronic Diseases and Environmental Epidemiology, National Institute of Public Health and the
Environment, PO Box 1, 3720 BA Bilthoven, The Netherlands
institute of Social Medicine, Academic Medical Centre, University of Amsterdam, Meibergdreef 15, I 105 AZ Amsterdam,
The Netherlands
Address correspondence to: N. Hoeymans. Fax: (+31) 30 2744450. E-mail: Nancy.Hoeymans@rivm.nl
Abstract
Objectives: to investigate to what extent differences in health status between respondents and drop-outs affected
the associations between cardiovascular diseases and functional status and self-rated health in a population-based
longitudinal health survey in elderly men.
Methods: during the 1993 survey of the Zutphen Elderly Study, a non-response survey was carried out. The
prevalence of myocardial infarction and stroke, disabilities in basic activities of daily living
(BADL)
and mobility, and
self-rated health were compared between non-respondents (n = 99) and respondents (n - 381). Associations
between myocardial infarction and stroke on the one hand and functional status and self-rated health on the other
were calculated for the total population and for the respondents to assess the amount of under- or overestimation of
these associations.
Results: the health of non-respondents was worse than that of respondents in terms of
stroke,
disabilities in
BADL
and mobility and self-rated health. Due to this selective non-response, the associations between cardiovascular
diseases and functional status and self-rated health were biased. Although most of the associations were slightly
overestimated, the most important bias was the underestimation by 57% of the association between stroke and
disabilities in
BADL
[total population: odds ratios (OR) =
6.1, 95%
confidence interval (CI) = 2.7-13-9; respondents
only: OR = 2.6, CI = 0.7-99].
Conclusion: selective non-response might lead to bias in the prevalence of disease, disabilities and self-rated
health as well as in the associations between disease and functional status and self-rated health. The direction
and magnitude of this bias varies according to type of disease and health outcome and is therefore difficult
to predict. The need to minimize non-response and to investigate its implications is recommended in every
study.
Keywords: functional status, non-respondents, myocardial infarction, stroke, study design
Introduction
In elderly people, cardiovascular diseases are an
important cause of diminished functional status and
well-being [
1
-
5].
However, studies on health outcomes
of cardiovascular diseases may be biased due to
selective non-response or drop-out. Many studies in
elderly populations have reported that the health status
of non-respondents is less than that of respondents
[6-10],
although examples of the opposite are also
found [11]. Furthermore, follow-up studies have
observed higher morbidity and mortality rates among
non-respondents than among respondents [12-14].
Non-response may lead to bias not only in prevalence
estimates of
diseases
and adverse health outcomes, but
also in the associations between diseases and health
outcomes [9, 13, 15, 16].
The Zutphen Elderly Study is a longitudinal study on
35
by guest on July 16, 2011ageing.oxfordjournals.orgDownloaded from
N. Hoeymans et al.
life-style, chronic diseases and health
and was
started in
1985.
A non-response survey was carried out during
the follow-up survey in 1993 to quantify possible bias
in measures of health status due to drop-out. The aim of
our study was to investigate to what extent differences
in health status between respondents and drop-outs
affected the association between myocardial infarction
and stroke on one hand and functional status and
self-
rated health on the other.
Methods
Study population
The Zutphen Elderly Study [17] was started in 1985 as a
continuation of the Zutphen Study [18], a longitudinal
population-based cardiovascular health study among
men born between 1900 and 1920 and living in the
town of Zutphen. In 1985, all survivors of the original
cohort and a random sample of all other men in the
same age range living in Zutphen were recruited. This
resulted in a target population of 1266 men, of whom
939 (74%) participated. This group formed the cohort
of the Zutphen Elderly Study. For the follow-up survey
of 1993, all 548 survivors of this cohort were
contacted. They received a letter in which the study
was explained and a response note in which they could
indicate whether they were willing to participate. A
reminder and a phone contact followed this initial
letter in order to reach as many men as possible. In
Spring 1993, the 390 men (71%) who indicated they
were willing to participate received a questionnaire by
mail and were visited 1 week later by one of our
research assistants to check the questionnaire for
inconsistencies or missing items and to carry out a
test for cognitive and one for physical function. The
questionnaire could also be completed by a relative or
caregiver.
In June 1993 those who indicated they did not want
to participate or who had not responded (n = 158)
received a very short questionnaire which they or a
relative or caregiver were asked to complete and send
back. In the accompanying letter it was explained that
it was important to have some information from non-
participants. When no reply was received after 2
weeks, non-respondents were interviewed by tele-
phone or visited at home.
Non-respondents who did not participate in the non-
response survey (n - 50) did not differ appreciably
from participants in the non-response survey (n
=
108)
as regards age, socio-economic status and baseline
health status. Regarding current health status, the
prevalence of myocardial infarction and stroke was
lower among these non-respondents than among the
participating non-respondents (myocardial infarction
14%,
stroke 4%). No information on functional status
and self-rated health was available for
this
group.
In this
report the term 'non-respondent' refers to participants
in the 1993 non-response survey. Complete data on
cardiovascular diseases, functional status and self-rated
health were available for
381
respondents and 99 non-
respondents.
Measurements
Questions on marital status, history of cardiovascular
diseases, functional status and self-rated health were
identical in the survey and in the non-response
questionnaire. Data on socio-economic status were
based on the 1985 survey. Socio-economic status was
recorded by life-long occupation in four levels: profes-
sionals, managers and teachers, small-business owners,
non-manual workers and manual workers. Marital
status was recorded in four categories: married, never
married, divorced and widowed.
Information on the prevalence of myocardial infarc-
tion and stroke was obtained from the (non-response)
questionnaire and verified with hospital discharge data
and written information from general practitioners. For
definite myocardial infarction the final diagnosis was
based on whether two of the following three criteria
were met: a specified medical history, i.e. severe chest
pain lasting for more than 20 min and not disappearing
in rest, characteristic electrocardiogram changes and
specific enzyme elevations. Stroke was denned as a
sudden onset of neurological paralysis lasting longer
than 24 h.
Functional status was measured as disabilities in
daily routine activities. The questionnaire we used
was adapted from the 11 countries study [19] and
described in detail in a previous publication [20]. In
short, the questionnaire consisted of 13 items each
mentioning one basic activity of daily living (BADL),
mobility or instrumental activity of daily living (IADL)
item. Participants who reported that they needed help
with at least one of the following activities: feeding
oneself,
getting in and out of bed, using the lavatory,
dressing and undressing, washing and bathing oneself
and walking between rooms were classified as
disabled in BADL. Respondents who stated that they
needed help with moving outdoors, using stairs,
walking at least 400 m or carrying a heavy object
for 100 m were classified as being disabled in mobility.
IADL disability was not taken into account in this
study.
Self-rated health was measured with a single-item
question: "We would like to know what you think
about your health. Please check what fits best in your
case.
Do you feel healthy, rather healthy, moderately
healthy or not healthy?" The value of this measure as a
predictor of mortality was shown in a previous study
[21].
For the logistic analyses, self-rated health was
dichotomized into good health (healthy or rather
healthy) and poor health (moderately healthy or not
healthy).
36
by guest on July 16, 2011ageing.oxfordjournals.orgDownloaded from
Non-response bias
Statistical analyses
Statistical analyses were carried out using
SAS,
version
6.11.
All tests were two-tailed and a /"-value of 0.05 or
lower was considered to be statistically significant.
Non-respondents were compared with respondents on
the demographic characteristics of age, marital status
and socio-economic status and on the health status
indicators prevalence of myocardial infarction and
stroke, disabilities in BADL and mobility and self-rated
health. Student
f-tests
for independent observations
were used for comparing means of continuous
variables and \2 tests for comparing prevalences.
Associations between myocardial infarction and
stroke on one hand and functional status and
self-
rated health on the other hand were analysed for the
total population and for the respondents only. The
prevalences of disabilities and poor self-rated health
among men with and without the disease were
assessed, adjusted for age. Age-adjusted odds ratios
(ORs) were calculated from logistic regression models.
We
assessed the under- or overestimation in the impact
of the diseases on functional status and self-rated health
that was due to non-response.
Results
Of the 158 non-respondents, 108 (68.4%) participated
in the non-response survey. The questionnaire was
returned by mail by 81 men, completed by telephone
by 10 and completed during a home visit by 17. Four
men were not approached, because they had stated
after the previous follow-up that they no longer wanted
to be contacted. The remaining non-respondents
refused to participate (n = 29) or could not be reached
in = 17).
Non-respondents and respondents did not differ
significantly in age and marital status (Table 1). The
socio-economic status of non-respondents was lower
than that of respondents. The health status was worse
among non-respondents than among respondents. This
was statistically significant for all health indicators,
except for
a
history of myocardial infarction. Adjustment
for
age,
marital status and socio-economic status did
not alter these differences.
Almost one-quarter (23.2%) of the non-response
questionnaires were completed by proxies, compared
with less than 5% in the response group (Table 2).
Table I. Demographic and health characteristics of respondents and non-respondents: Zutphen Elderly
Study,
1993
Mean age (SD)
Marital status (%)
Married
Never married
Divorced
Widowed
Socio-economic status
(%)"
Professionals, managers, teachers
Small business owners
Non-manual workers
Manual workers
Prevalence of chronic diseases (%)
Myocardial infarctionb
Stroke0
Functional status (% disabled)
Activities of daily living
Mobility
Self-rated health (%)
Healthy
Rather healthy
Moderately healthy
Not healthy
Respondents
in
=
381)
77.8 i4.4)
71.9
1.8
2.4
23.9
30.8
18.1
26.9
24.2
13.6
5.5
5.8
20.2
43.3
45.1
8.7
2.9
Non-respondents
in
=
99)
78.4 (5.2)
72.7
6.1
2.0
192
14.4
23.3
32.2
30.0
17.2
13.1
21.2
51.5
31.3
38.4
16.2
14.1
P-value
0.26
0.11
0.02
0.37
0.01
0.001
0.001
0.001
"Data on socio-economic status were available for only 364 respondents and 90 non-respondents.
"The
prevalence of myocardial infarction was
also
known on those non-respondents who did not
fill
out the non-response questionnaire (n
=
50).
The prevalence of myocardial infarction for the total group of non-respondents was 16.1.
cThe prevalence of stroke was also known on those non-respondents who did not fill out the non-response questionnaire (n = 50). The
prevalence of stroke for the total group of non-respondents was 10.1. This was still significantly different from the prevalence among
respondents
(P =
0.05).
37
by guest on July 16, 2011ageing.oxfordjournals.orgDownloaded from
N. Hoeymans et al.
Table 2. Self-rated health and functional status of respondents and non-respondents, classified as direct (non-)
respondents and proxies: Zutphen Elderly Study, 1993
Self-rated health
Healthy
Rather healthy
Moderately healthy
Not healthy
Functional status
Disabled in BADL
Disabled in mobility
No.
(and %) of subjects
Respondents
Direct
(n
=
364)
159 (43.7)
164(45.1)
31 (8.5)
10 (2.7)
19 (5.2)
66(18.1)
Proxies
(«= 17)
6 (35.3)
8 (47.1)
2(11.8)
1 (5.9)
3 (17.6)
11(64.7)
Non-respondents
Direct
(» = 76)
26 (34.2/
32(42.1)
9(11.8)
9(11.8)
8 (10.5)b
31 (40.8)a
Proxies
(n = 23)
5 (21.7)
6 (26.1)
7 (30.4)
5 (21.7)
13 (56.5)
20 (87.0)
BADL, basic activities of
daily
living.
"Significantly different from direct respondents, P
<
0.01.
""Borderline significantly different from direct respondents, P
=
0.08.
Proxies reported significantly worse health ratings and
functional status than the group of men who com-
pleted the questionnaire themselves. However, direct
non-respondents still differed significantly from direct
respondents in self-rated health and mobility and the
difference in proportion of men limited in
BADL was
of
borderline significance.
The impact of myocardial infarction on BADL
disabilities was not statistically significant, for respon-
dents or for the total population (Table 3). However,
men with a history of myocardial infarction had a
higher risk of disabilities in mobility [OR = 1.8; 95%
confidence interval (CD =
1.0-3.2]
and poor self-rated
health (OR
=
1.9;
95%
CI =
1.0-35).
Both effects were
slightly overestimated when only respondents were
considered. Stroke patients reported more disabilities
in
BADL
and mobility than men who had no history of
stroke
(BADL:
OR = 6.1; 95% CI = 2.7-139, mobility:
OR = 39; 95% CI =
1.9-8.3).
The non-responding
stroke patients, however, reported more disabilities
Table
3.
Age-adjusted associations [odds ratios
(OR)]
of a history of myocardial infarction and stroke with disabilities
and self-rated health, for respondents only and for the total
group,
and
%
over-
or under estimation in
OR
due to non-
response (OR error): Zutphen Elderly Study, 1993
Myocardial infarction
No.
of subjects
%
disabled/in poor health
Disabled in BADL
Disabled in mobility
Poor self-rated health
Stroke
No.
of subjects
%
disabled/in poor health
Disabled in BADL
Disabled in mobility
Poor self-rated health
Total
With
69
11.6
34.8
23.2
34
32.4
55.9
41.2
population
without
411
8.5
25.3
14.1
446
7.2
24.4
13.5
(n =
OR
1.6
1.8
1.9
6.1
3.9
4.5
480)
(95%
CI)
(0.7-3.6)
(1.0-3.2)
(1.0-3.5)
(2.7-13.9)
(1.9-8.3)
(2.1-9.3)
Respondents (n =
With
52
9.6
30.8
21.2
21
14.3
42.9
38.1
Without
329
5.2
18.5
10.0
360
5.3
18.9
10.0
381)
OR
2.1
2.2
2.4
2.6
3.0
5.5
(95%
CD
(0.7-6.2)
(11-4.3)
(11-5.2)
(0.7-9.9)
(1.2-7.8)
(2.1-14.3)
%
OR error*
31
22
26
-57
-23
27
BADL, basic activities of daily lMng.
*%
over- or under-estimation in OR due to non-response = (OR for response group-OR for total populationVOR for total population.
38
by guest on July 16, 2011ageing.oxfordjournals.orgDownloaded from
Non-response bias
than responding stroke patients, leading to a substan-
tial underestimation of the impact of stroke on
disabilities in BADL (57%) and to a lesser extent of
the impact on disabilities in mobility
(23%).
In contrast,
the impact of stroke on self-rated health (OR
=
4.5;
95%
CI = 2.1-9-3) was overestimated. All results were
adjusted for age. Additional adjustments for socio-
economic status and marital status did not change
these results.
Discussion
Non-respondents of the 1993 follow-up survey of the
Zutphen Elderly Study had a poorer health status than
respondents regarding the prevalence of stroke,
functional status and self-rated health. These differ-
ences led to a small overestimation of the risk of
disabilities and poor self-rated health for those with a
history of myocardial infarction and of the risk of poor
self-rated health for those who suffered a stroke. The
association between stroke and disabilities in activities
of daily living, however, was underestimated by 57%
because of non-response of the most severely disabled
stroke patients.
Differences in health status between respondents
and non-respondents, especially in self-rated health and
functional status, could have been because about one-
quarter of the non-response questionnaires were
completed by a proxy: spouse, child, caregiver etc.
Besides the probability that those who had proxies to
complete the questionnaire were really in a worse
health state, it has also been shown that health status
assessed by proxies gives a systematic underestimation
of the health status of the respondent [22-24].
Therefore, we also compared direct non-respondents
and respondents. Differences in self-rated health and
mobility between direct non-respondents and respon-
dents were smaller than between all non-respondents
and respondents, but remained statistically significant.
The finding that the difference in proportion of men
with BADL disabilities became borderline significant
could be due to the small number of men with
disabilities in BADL. Bias due to proxy measurements
cannot therefore fully explain the differences between
respondents and non-respondents.
Most of the non-response questionnaires were sent
back by mail
(75%).
The answers in the telephone and
face-to-race interviews might suffer from bias, due to
social desirability considerations in responding. How-
ever, additional analyses showed that health status did
not differ between these non-respondents and the non-
respondents who completed the questionnaire them-
selves. Based on these results we assume that our
results were not biased by this factor.
In our study, men with a history of myocardial
infarction did not report significantly more BADL
disabilities than men who did not suffer a myocardial
infarction, but they did report more disabilities in
mobility. Men with a history of stroke reported more
disabilities in BADL and mobility than men with no
such history. These results seem consistent with other
studies. Several studies have shown that stroke has a
strong negative impact on functional status [2-5, 25-
27].
Heart disease is found to have an impact on BADL
disabilities in some studies [25-28], but not in all [2].
Findings on the impact of cardiovascular diseases
on self-rated health are less consistent. Mulrow and
co-workers reported no significant effects of cardio-
vascular diseases on perceived health [2]. However, as
in our study, Stewart and colleagues observed signifi-
cantly worse health perceptions in those with a history
of myocardial infarction than in those without such a
history [1].
Non-response bias in these associations between
cardiovascular diseases and disabilities and self-rated
health was found. We expected that mainly the
healthiest myocardial infarction and stroke patients
would participate, leading to a systematic under-
estimation of the associations between these diseases
and health outcomes. However, this was only observed
for stroke and disabilities in BADL. The remaining
associations were biased in a different direction, albeit
to a lesser extent. The impact of myocardial infarction
on disabilities and poor self-rated health was slightly
overestimated,
as
was the impact of stroke on poor
self-
rated health. The finding that non-response bias
depended on type of disease was also reported by
Launer and co-workers [9], who found a biased OR of
stroke and diabetes on poor cognitive performance
due to non-response. Stroke patients with poor
cognitive function tended to participate more than
stroke patients with high performance, while diabetes
patients with poor cognitive performance tended to
refuse to participate.
Selective non-response might lead to bias in the
impact of cardiovascular diseases on functional
status and self-rated health. This bias can vary
according to type of disease and health outcome.
This study clearly shows the need for minimizing
non-response or drop-out and the need for research
on the implications of non-response in all studies.
Possible efforts that can be taken to minimize non-
response or drop-out are to emphasize the import-
ance of the participation of each individual in
the study, whether extremely healthy, extremely
unhealthy or anything between. It is also important
to underline the possibility that the questionnaire
can be completed by a proxy. Finally, a compensation
for participation in the study might lead to higher
response rates.
Acknowledgements
The Zutphen Elderly Study was financially supported
by the Netherlands Prevention Foundation, the Hague
39
by guest on July 16, 2011ageing.oxfordjournals.orgDownloaded from
N. Hoeymans et a\.
and the National Institute on Aging, Bethesda, MD,
USA. The authors would like to thank the fieldwork
team in Zutphen, especially
E.
B.
Bosschieter and
A.
H.
Thomassen-Gijsbers.
Key points
• Drop-outs from the Zutphen Elderly Study are less
healthy than those who remain in the study.
• Due to this selective drop-out, the associations of
myocardial infarction and stroke with functional
status and self-rated health are biased.
• The direction and magnitude of this bias varies with
type of disease and health outcome.
References
1.
Stewart
AL,
Greenfield S, Hays RD et
aL
Functional status
and well-being of patients with chronic conditions. Results
from the Medical Outcomes
Study.
JAMA
1989; 262: 907-13.
2.
Mulrow
CD,
Gerety
MB,
Cornell
JE,
Lawrence
VA,
Kanten
DN. The relationship between disease and function and
perceived health in very frail elders. J Am Geriatr Soc 1994;
42:
374-80.
3.
Bos GAM van den. The burden of chronic diseases in
terms of disability, use of health care and healthy life
expectancies. EurJ Public Health 1995; 5: 29-34.
4.
Ettinger WH, Fried LP, Harris T, Shemanski L, Schulz R,
Robbins
J.
Self-reported causes of physical disability in older
people: the cardiovascular healdi study. J Am Geriatr Soc
1994;
42: 1035-44.
5.
Jongbloed
L.
Prediction of function after stroke: a critical
review. Stroke 1986; 17: 765-76.
6. Rockwood K, Stolee P, Robertson D, Shillington ER.
Response bias in a health status survey of elderly people.
Age Ageing 1989; 18: 177-82.
7.
Heilbrun LK, Ross PD, Wasnich RD, Yano K, Vogel JM.
Characteristics of respondents and non-respondents in a pros-
pective study of osteoporosis. J din Epidemiol 1991;44:233-9.
8. Jay GM, Liang J, Iiu X, Sugisawa H. Patterns of non-
response in a national survey of elderty Japanese. J Gerontol
Soc Sci 1993; 48: S143-52.
9- Launer
LJ,
Wind
AW,
Deeg
DJH.
Nonresponse pattern and
bias in a community-based cross-sectional study of cognitive
functioning among die elderly. Am J Epidemiol 1994; 139:
803-12.
10.
Norton MC, Breitner JCS, Welsh KA, Wyse BW. Char-
acteristics of non-responders in a community survey of the
elderly. J Am Geriatr Soc 1994; 42: 1252-6.
11.
Tennant
A,
Badley
EM.
Investigating non-response bias in
a survey of disablement in the community: implications for
survey methodology. J Epidemiol Community Health 1991;
45:
247-50.
12.
Sheikh K. Predicting risk among non-respondents in
prospective studies. EurJ Epidemiol 1986; 2:
39-43-
13.
Bisgard KM, Folsom AR, Hong CP, Sellers TA. Mortality
and cancer rates in nonrespondents to a prospective study of
older women: 5 year follow-up.
Am
J Epidemiol 1994; 139:
990-1000.
14.
Heilbrun LK, Nomura
A,
Stemmermann GN. The effects
of non-response in a prospective study of cancer. 15 year
follow-up. Int J Epidemiol 1991; 20: 328-38.
15.
Austin
MA,
Criqui
MH,
Barrett-Connor
E,
Holdbrook MJ.
The effect of response bias on the odds ratio.
Am J
Epidemiol
1981;
114: 137-43-
16.
Greenland S. Response and follow-up bias in cohort
studies.
Am
J Epidemiol 1977; 106: 184-7.
17.
Feskens FJM, Bloemberg BPM, Pijls LTJ, Kromhout
D.
A
longitudinal study on elderiy men: the Zutphen study. In:
Schroots JJF
ed. Aging, Health and Competence. Amsterdam:
FJsevier Science Publishers, 1993.
18.
Keys
A.
Seven Countries: a multivariate analysis of death
and coronary heart disease. Cambridge, MA: Harvard
University Press, 1980.
19.
WHO. The Elderly in Eleven Countries. Copenhagen:
WHO,
1983-
20.
Hoeymans
N,
Feskens
EJM,
van den
Bos
GAM,
Kromhout
D.
Measuring functional status: cross-sectional and long-
itudinal associations between performance and self-report
(Zutphen Elderly Study 1990-1993). J Clin Epidemiol 1996;
49:
1103-1110.
21.
Pijls LTJ, Feskens EJM, Kromhout D. Self-rated healdi,
mortality and chronic diseases in elderly men. The Zutphen
Study 1985-1990.
Am
J Epidemiol 1993; 138: 840-8.
22.
Dorevitch
MI,
Cossar
RM,
Bailey
FJ,
et
aL
The accuracy
of
self and informant ratings of physical functional capacity in
the elderty. J Clin Epidemiol 1992; 45: 791-8.
23.
Epstein
AM,
Hall
JA,
Tognetti
J,
Son LH, Conant
L.
Using
proxies to evaluate quality of life. Can they provide valid
information about patients' health status and satisfaction
with medical care? Med Care 1989; 27: S91-8.
24.
Rothman ML, Hedrick SC, Bulcroft KA, Hickam DH,
Rubenstein LZ. The validity of proxy-generated scores
as measures of patient heakh status. Med Care 1991; 29:
115-24.
25.
Boult C, Kane RL, Louis TA, Boult L, McCaffrey D.
Chronic conditions that lead to functional limitations in the
elderly.
J Gerontol Med Sci 1994; 49: M28-36.
26.
Guccione
AA,
Felson
DT,
Anderson JJ et
aL
The effects of
specific medical conditions on die functional limitations of
elders in die Framingham
study.
Am J Public Healdi 1994; 84:
351-8.
27.
Guralnik
JM,
LaCroix AZ, Abbott RD et
aL
Maintaining
mobility in late life. I. Demographic characteristics and
chronic conditions.
Am
J Epidemiol 1993; 137: 845-57.
28.
O'Brien BJ, Buxton MJ, Patterson DL Relationship
between functional status and health-related quality-of-life
after myocardial infarction. Med Care 1993;
31:
950-5.
Received 22 January 1997
40
by guest on July 16, 2011ageing.oxfordjournals.orgDownloaded from