Is there a Mediterranean migrants mortality paradox in Europe? Int

Full text

In her commentary
1
on our study of Pap smear screening and
cervical cancer in South Africa,
2
Dr Raffle makes three major
criticisms.
Dr Raffle cites the importance of validating screening histories
from records.
3
Given the lack of complete and available records
on Pap screening and therefore the need to rely on self-report,
the nurse interviewers in our study were specially trained to
conduct the interviews in a highly standardized manner
designed for the most complete ascertainment of information.
Dr Raffle states that the ‘unreliability of the data is betrayed by
the implausibly high 73% participation (in Pap screening)
amongst the controls’. The rate is not implausible: a recent
population-based survey by Fonn et al. reported a prevalence of
70.6% in the Western Cape.
4
In addition, as mentioned in our
article, until recently opportunistic screening was performed
when women attended family planning and antenatal clinics
in the public health sector. Attendance at these clinics is high in
the Western Cape: 73.7% of women of child-bearing age currently
use contraception and 91.7% attend antenatal clinics.
5
Dr Raffle states that ‘there is no description of how the date
of diagnosis was defined, of whether cases that became
advanced during the 6 months they were captured [sic], and of
whether fatal cases were included.’ We stated that incident
cases were enrolled no more than 6 months after diagnosis. In
fact, almost all of the cases were enrolled at the time of
diagnosis. There was no opportunity for the cases to progress
between the time of diagnosis and the time of enrolment. As we
indicated in our paper, we conducted an interview-based case-
control study; thus, fatal cases were not included.
Dr Raffle states that ‘healthy screenee bias’ cannot be ruled
out. We agree as indicated by the statement in our paper that
‘there may have been some distortion from uncontrolled
residual confounding’. Dr Raffle’s statement that we attempted
‘to remove potential bias by matching for decade of age, urban/
rural residence, race, education parity, age at first sexual activity,
use of contraceptives, and cigarette smoking’ is incorrect. We
used unconditional logistic regression to adjust for potential
confounding by those factors. Dr Raffle may be confused by the
fact that when we selected controls, we used series matching
(not individual matching) in order to obtain similar distribu-
tions in the comparison groups according to age, race, and area
of residence; these three factors were, in addition, allowed for
in the multivariate analysis.
Finally, a hypothesized explanation for why limited screening
may be effective against cervical cancer in South Africa is
currently in press.
6
References
1
Raffle AE. Commentary: Case-control studies of screening should carry
a health warning. Int J Epidemiol 2003;32:577–78.
2
Hoffman M, Cooper D, Carrara H et al. Limited Pap screening associated
with reduced risk of cervical cancer in South Africa. Int J Epidemiol
2003;32:573–77.
3
Weiss NS. Application of the case control method in the evaluation of
screening. Epidemiol Rev 1994;16:102–08.
4
Fonn S, Bloch B, Mabina M et al. Prevalence of precancerous lesions
and cervical cancer in South Africa-a multi-centre study. S Afr Med J
2002;92:148–56.
5
South Africa Demographic and Health Survey 1998: Burden of Disease
Research Unit—Reports. Medical Research Council of South Africa,
2001.
6
Shapiro S, Carrara H, Allan B et al. The act of taking a Papanicolaou
smear reduces the prevalence of human papilloma virus infection:
a potential impact on the risk of cervical cancer. Cancer Causes Control
2003 (in press).
DOI: 10.1093/ije/dyg329
Letters to the Editor
Case-control studies of screening should carry a health warning. Response
From MARGARET HOFFMAN,
1
LYNN ROSENBERG
2
and SAMUEL SHAPIRO
3
1
School of Public Health and Family Medicine, University of Cape Town,
South Africa.
2
Slone Epidemiology Center, Boston University, USA.
3
Slone Epidemiology Unit, Boston University, USA.
Correspondence: Prof. M Hoffman, Department of Public Health and Primary
Health Care, Faculty of Health Sciences, University of Cape Town, Anzio
Road, Observatory 7925, South Africa. E-mail: mh@cormack.uct.ac.za
1112
IJE vol.32 no.6 © International Epidemiological Association 2003; all rights reserved. International Journal of Epidemiology 2003;32:1112–1120
by guest on May 23, 2011ije.oxfordjournals.orgDownloaded from

Clinicians employ principles of epidemiology for decision-making,
consciously or subconsciously.
1
Though this is well known, teach-
ing and practice continue to be based on case studies, partially
endorsed or unendorsed reviews, and anecdotes. Clinicians fail
to appraise the evidence critically because of a poor understanding
of research methodology. Therapies backed by ample evidence
are underused because of a lack of knowledge or because
clinicians believe that results observed in clinical trials cannot be
translated into clinical practice.
2
Not surprisingly, only 15% of
medical interventions are based on solid scientific evidence.
1
Clinical medicine appears to consist of a few things we know, a
few things we think we know (but probably don’t), and lots of
things we don’t know at all.
3
In India, epidemiology is well established and mandatory in
undergraduate medical training as part of Social and Preventive
Medicine. Evidence-based medicine is a fashionable topic known
to medical personnel with international exposure but it is yet to
be routinely practised. Between November 2001 and January
2002, we performed a survey in three teaching and three non-
teaching hospitals in New Delhi to elicit perceptions about
epidemiology and its uses from 190 clinicians. We used a ques-
tionnaire consisting of 13 items, divided into five broad groups:
(1) perception of epidemiology, (2) enthusiasm for taking up
epidemiology as a career, (3) status of epidemiology, (4) attitude
to epidemiology, and (5) need for training. Each respondent was
asked to answer each item with his/her degree of agreement on
a four-point scale—strongly agree, tend to agree, no opinion, and
disagree. Public health professionals and clinicians already ex-
posed to epidemiological training were excluded from the survey.
Of 190 selected clinicians, 151 (79.5%) responded: 32 internists
(21.2%), 27 ophthalmologists (17.9%), 26 gynaecologists (17.2%),
10 gastroenterologists (6.6 %), and 24 from other specialties
(15.9%); 18 were non-specialists (11.9%), and no specialty was
reported by 14 (9.3%). Results of the survey are presented in
the Table. A surprisingly high proportion of the clinicians in this
survey agreed that epidemiology is a basic science for clinical
medicine and therefore necessary for a good clinician (86.1%).
The enthusiasm for taking up epidemiology as a career was
considerable (41.6%), in particular if better research and job
opportunities were provided and a modified, integrated
curriculum was introduced (60.9%). While 136 respondents
(90.1%) concurred that sound knowledge of epidemiology was
necessary for a good clinician, about a third felt that a trained
epidemiologist lacked clinical knowledge and was a waste of
time for clinicians. A high proportion of respondents felt that
epidemiology was a postgraduate subject choice for graduates of
lower rank (62.9%).
When comparing internists and allied subspecialties (n = 54)
with other respondents (anaesthesia, dermatology, gynaecology,
ophthalmology, orthopaedics, psychiatry, radiotherapy, and
surgery; n = 65), we found internists suggested more frequently
that lower ranking graduates took up epidemiology as a post-
graduate subject (77.4% versus 57.8%, difference 19.6%, 95%
CI: 3.0%, 36.1%), considered epidemiology more frequently to
LETTERS TO THE EDITOR 1113
Epidemiology through the eyes of clinicians
From C SHYAM
Table 1 Distribution of responses by clinicians overall to the 13 questions. Note that the categories ‘strongly agree’ and ‘tend to agree’ were
combined
Strongly
agree/ No
tend to opinion Disagree
Item No. agree (%) (%) (%)
Perceptions about epidemiology
Epidemiology is difficult to understand & retain 149 74 (49.7) 10 (6.7) 65 (43.6)
It is part of PSM
a
or Public Health & an MBBS
b
subject only 148 119 (80.4) 10 (6.8) 19 (12.8)
Matter of interest to field/research workers & not for clinicians or clinical researchers 149 35 (23.5) 12 (8.0) 102 (68.5)
It is a basic science for clinical medicine and is necessary for a good clinician 151 130 (86.1) 12 (7.9) 9 (6.0)
Enthusiasm for taking up epidemiology as a career
Would like to, if better research & job opportunities are provided 149 62 (41.6) 40 (26.8) 47 (31.5)
Would like to be an epidemiologist, if a modified, integrated curriculum
is introduced along with other clinical subjects/specialties 151 92 (60.9) 26 (17.2) 33 (21.9)
Status of epidemiology
Sound knowledge of epidemiology necessary for a good clinician 151 136 (90.1) 12 (7.9) 3 (2.0)
Epidemiology as a postgraduate subject is taken up by lower ranking graduates 148 93 (62.9) 28 (18.9) 27 (18.2)
A trained epidemiologist lacks clinical knowledge and so it is waste of time for clinicians 148 46 (31.1) 31 (20.9) 71 (48.0)
Within the medical field, epidemiology has a low status 146 80 (54.8) 22 (15.1) 44 (30.1)
Attitude to epidemiology
Teaching epidemiology or doing epidemiological work is as rewarding as treating a patient 151 73 (48.3) 30 (19.9) 48 (31.8)
Need for training courses in epidemiology
A posting in pre & para clinical departments including PSM/Public Health during
residency would help 151 121 (80.1) 18 (11.9) 12 (7.9)
A formal course in teaching for a medical teacher would be welcome 149 127 (85.2) 15 (10.1) 7 (4.7)
a
Preventive & Social Medicine.
b
Bachelor of Medicine and Bachelor of Surgery.
Dr RML Hospital, New Delhi 110001, India.
by guest on May 23, 2011ije.oxfordjournals.orgDownloaded from

Sirs—European directives and laws determine more and
more the conditions for epidemiological research within Europe.
The International Epidemiological Association (IEA) aimed to
investigate whether these directives facilitate or hinder epidemi-
ological research in Europe. It goes without saying that European
epidemiologists, with a large variety of different exposures related
to living conditions, occupations, diet, lifestyles, and health care
between EU countries, are in a good position to produce more
and better research if they get access to data from morbidity and
mortality registers.
In 1995 the European Union issued the EU Data Protection
Directive ‘on the protection of individuals with regard to the
processing of personal data and on the free movement of such
data’. (http://www.privacy.org/pi/intl_orgs/ec/final_EU_Data_
Protection.html). The aim of this Directive is to set minimum
standards for data protection in the various EU countries in
order to facilitate the free flow of data between these countries
(internal market motive) while at the same time offering a high
level of privacy protection. It contains certain exemptions to the
general principles in order to facilitate research with sensitive
data (recital 28 and article 13). Therefore the Directive should
also facilitate the sharing of data from morbidity and mortality
registries among epidemiologists in the EU. The directive states
that EU member states may set their own rules and regulations
with regard to availability of data from registries, but these
rules should not discriminate between member states. A report
in the British Medical Journal
1
showed that this EU directive led
to much confusion and delay in research in the UK.
We conducted a survey to compare access to data from reg-
istries in different European countries. Furthermore, we wanted
to know if data from one country can be used by epidemiologists
from other EU countries. A questionnaire was composed and
sent to all national epidemiological societies in 2000/2001, and
it was also available on the IEA website (www.iea.org).
Responses were received from Denmark (17), UK (4),
The Netherlands (2), and one response from Finland, Sweden,
Switzerland, France, Spain, Greece, Austria, Poland, and
Yugoslavia-Serbia. The answers of multiple responders within
countries varied to some extent. Where available, the answers
of the National Committees were used.
The EU Data Protection Directive was at the time of the survey
implemented in Denmark, the UK, The Netherlands, Finland,
Sweden, Austria, and Poland. The responders from other countries
were not sure. Being non-EU (candidate) member states, the
1114 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
have a low status within the medical field (75.5% versus 40.6%,
difference 34.9%, 95% CI: 18.1%, 51.5%), and refrained more
frequently from becoming an epidemiologists, even if a modified
integrated curriculum was introduced (29.6% versus 18.5%,
difference 11.1%, 95% CI: 4.2%, 26.6%).
The low opinion of epidemiology, particularly among internists,
is not entirely unexpected. Even David Sackett, one of the
founders of evidence-based medicine, realized the importance
of epidemiology and its relevance to clinical medicine rather
late.
4
Clinicians often refrain from formally applying the tools
of epidemiology during clinical decision-making, tend to be
categorical in expressing clinical outcomes, are uneasy about
uncertainty, and are reluctant to express this uncertainty using
probabilities.
5
Ever since its foundation in 1954, the International Epidemio-
logical Association has been concerned with education about
and promotion of the wider application and use of epidemiology.
However, no special emphasis for clinicians had been made.
Subsequently, under the auspices of the International Clinical
Epidemiology Network (INCLEN), faculties from six identified
institutions in India were trained abroad, so that they in turn
could develop local and regional capacity and expertise. How-
ever, it appears that further dissemination of this knowledge has
not taken place. Apparently the ‘trained’ could not become
‘trainers’ as envisaged.
This survey indicates a need to have new look at the issue
of epidemiological training for prospective clinicians in India.
Reorganizing the undergraduate and postgraduate medical
curricula may be necessary, particularly in the field of internal
medicine. Integrating epidemiology with clinical teaching, using
relevant examples to make it more interesting and easy to
comprehend and supplementing the formal classroom teaching
of epidemiological methods with teaching sessions during ward
rounds might be helpful.
References
1
Smith R. Where is the wisdom? BMJ 1991;303:798–99.
2
McMurray JJ. Failure to practice evidence-based medicine: why do
physicians not treat patients with heart failure with angiotensin-
converting enzyme inhibitors? Eur Heart J 1998;19(Suppl.L):
L15–21.
3
Naylor CD. Grey zones of clinical practice: some limits to evidence-
based medicine. Lancet 1995;345:840–42.
4
Sackett DL, Haynes RB, Guyatt GH, Tugwell P. Clinical Epidemiology:
a Basic Science for Clinical Medicine. Boston: Little Brown, 1985.
5
Fletcher RH, Fletcher SW, Wayner EH. Clinical Epidemiology—The
Essentials. Baltimore: Wilkins and Wilkins, 1989.
DOI: 10.1093/ije/dyg324
Access to data from European registries for epidemiological research: results from
a survey by the International Epidemiological Association European Federation
From HENRICA CW DE VET,
1
JACQUELINE M DEKKER,
1
EVERT BEN VAN VEEN
2
and JØRN OLSEN
3
1
Institute for Research in Extramural Medicine, VU University Medical
Center, Amsterdam, The Netherlands. E-mail: HCW.de_Vet.EMGO@med.
vu.nl
2
Medlaw Consult, Den Haag, The Netherlands.
3
The Danish Epidemiology Science Center, Denmark.
by guest on May 23, 2011ije.oxfordjournals.orgDownloaded from

Sirs—Compared with non-Hispanic Whites, Hispanics in the US
are poorer and less educated, and yet they enjoy a lower all-cause
mortality rate. This so-called ‘Hispanic Paradox’ has received
much attention over the past 20 years, both in the epidemiological
and demographic literature. Besides artefactual explanations (i.e.
possible under-reporting of Hispanic deaths on death certifi-
cates), competing theories fall into two categories: the ‘salmon
bias hypothesis’, according to which migrants are likely to
return to their country of origin after they retire or become
seriously ill, and, the ‘healthy migrant hypothesis’, according to
which those who migrate and remain in the host country are
the healthiest and strongest members of their population of
origin.
To date, two reviews have documented extensively the
wealth of literature on the Hispanic paradox. One, which is very
critical of the concept, focuses on low birthweight and infant
mortality;
1
the other, which is relatively supportive, covers
all the different health components involved in the paradox
(mortality, infant mortality, violence, AIDS, coronary heart
disease, stroke, cancer, and diabetes).
2
The first concludes that
the evidence supporting the paradox is fragile and highlights the
potential role of selective processes in explaining the migrants’
advantage, while the second puts forward the complexity of the
picture, with variations by age, gender, type of Hispanic group,
degree of acculturation, and specific disease or cause of death.
Recently, a paper—to be considered as a landmark in the
field—has very elegantly established that neither the salmon
bias, nor the selection of healthy Hispanic migrants into the US
LETTERS TO THE EDITOR 1115
Directive was not implemented in Switzerland and Yugoslavia-
Serbia.
Most Danish responders stated that the EU directive had
made epidemiological research less difficult. In the UK, Finland,
Austria, and Poland, the responders felt that research had become
more difficult. In The Netherlands and Sweden the situation
was considered unchanged.
In all countries data from mortality registries can be used
without individual consent in an entirely register-based research
project. Cause of death is difficult to obtain in The Netherlands,
in Poland, and in France.
All countries have a cancer registry and usually a number of
other morbidity registries. Scandinavian countries have a large
number of registries for several diseases. In most countries,
except Austria and Yugoslavia, data from these registries are
identifiable at an individual level. Data from these morbidity
registries can be used for epidemiological research based entirely
on registries without individual consent, except in Poland where
individual consent is necessary. In Austria and Yugoslavia these
data cannot be used at all for research.
Whether individual consent is needed for linkage of mor-
bidity registers to research data appears to depend on the type
of project in Denmark and the UK. In The Netherlands and
Poland individual consent is always needed, while in Sweden,
Switzerland, Spain, and Greece no consent is required.
We also asked if stored data with personal identifiers can be
used for another research purpose that requires
NO PERSONAL
CONTACT
to study members without renewed informed consent.
In Switzerland, Austria, and Poland these data cannot be used
for other research purposes. In The Netherlands and Yugoslavia
renewed informed consent is necessary. In other countries this
depends on the research question.
Obviously, there is variation between regulations in different
countries. However, the primary aim of this EU Directive was
not to make uniform regulations, but to make data also available
to other member states, without additional conditions. From
our survey it appeared that about half of the responders had
applied for use of data from other EU member states and they
had all received permission.
More detailed information on the study can be obtained from
www.iea-seminar.au.dk. The questionnaire can be found at
www.ieaweb.org (Click European Federation and then
‘Personal data’). The European Commission conducted its own
evaluation of the Directive.
2
The conclusion was that by now
all member states had implemented the EU directive, though
some were very late with it. As the background of the Directive
is driven mainly by economic issues, nothing is reported about
research data and health aspects. The accompanying ‘Analysis
and impact study’
3
reveals more clearly the different approaches
which have been taken by various countries when implement-
ing the Directive, e.g. in the definition of ‘personal data’. Some
countries consider coded data where the receiver of the data
does not have access to the code as personal data, while other
countries consider these as anonymous data (on the level of the
receiver). Obviously epidemiological research is better served
with the latter interpretation. The commission does not make a
choice between these two interpretations and considers this
subject open to further debate. Therefore it is important that
the IEA keeps surveying the consequences for epidemiological
research within Europe and will put possible negative aspects
on the European agenda.
References
1
Strobl J, Cave E, Walley T. Data protection legislation: interpretation
and barriers to research. BMJ 2000;321:890–92.
2
http://europa.eu.int/comm/internal_market/privacy/lawreport/data-
directive_en.htm
3
http://europa.eu.int/comm/internal_market/privacy/docs/lawreport/
consultation/technical-annex_en.pdf
DOI: 10.1093/ije/dyg314
Is there a Mediterranean migrants mortality paradox in Europe?
From M KHLAT
1
and N DARMON
2
1
Institut National d’Etudes Démographiques, 133 Bd. Davout, 75980 Paris
Cedex 20, France. E-mail: khlat@ined.fr
2
Institut National de la Santé et de la Recherche Médicale/Institut
Scientifique et Technique de la Nutrition et de l’Alimentation, 5 rue du Vert-
Bois, 75003 Paris, France.
by guest on May 23, 2011ije.oxfordjournals.orgDownloaded from

could explain the mortality advantage of the Hispanics, and
called for further research on cultural factors, especially those
involving favourable health behaviours, to shed more light on
this issue.
3
While much controversy surrounds the very exist-
ence and interpretations of the Hispanic paradox, it has been
argued that ‘investigation of this “paradox” may provide
additional insights into the ways that social factors affect the
health of the population at-large’,
2
and that ‘the Hispanic
paradox should be a motivator for further research ...’.
1
With this in mind, our purpose is to place the Hispanic
paradox in a wider geographical and cultural perspective, by
pinpointing the remarkable mortality advantage that some
Mediterranean migrant groups enjoy in Germany and France.
While the literature on migrants’ mortality is relatively scarce in
Europe, two methodologically sound studies have provided
quite an unexpected picture. In Germany, analysis of register
data has established that the age-adjusted mortality of the
2 million Turkish residents was consistently half that of the
Germans, and also less than half that of an urban population
in Turkey.
4
The former findings, which concerned males and
females equally, were confirmed by a cohort study (‘German
Socio-Economic Panel’) unlikely to be subject to the inaccuracies
of denominator figures.
5
Convergent features have emerged from
a study based on register and census data in France which high-
lighted the surprisingly low mortality of Moroccan immigrants,
whose number had swollen in the early 1980s to nearly
600 000.
6
Using an indirect demographic method originally
devised to estimate the completeness of vital statistics in developing
countries, the authors have found that the proportion of missing
deaths among Moroccans was about 23% for men, and negligible
for women. After having corrected the numbers of male deaths
accordingly, they found an adjusted life expectancy of 73.7 years
in the 1980s, as opposed to 71.3 in the French population, and
78.8 in females as opposed to 79.6 in the French population.
And yet, the socio-demographic profile of those migrants—they
are frequently illiterate, single, and employed largely as manual
workers (83%)—is expected to favour higher mortality.
Echoing the Hispanic mortality paradox in the US, the above
findings convincingly set the case for a Mediterranean migrants
mortality paradox in some Western European countries, and
have led to a questioning of the classical interpretations prevalent
in the epidemiological and demographic literature. Rather than
referring to a ‘salmon bias’, the authors of the European studies
have wondered about the existence of a ‘mobility bias’ resulting
from an inflation of the denominator base, due to migrants
often returning to their home country for short or long periods,
be it in relation to their health status or not. Indeed, the
Mediterranean countries of Southern Europe and North Africa
are geographically close to France and Germany, and migrants
often return to their home country during holidays or for family
reasons. Individual follow-up such as that implemented in
the German Socio-Economic Panel provides strong support for
the substantive nature of the mortality advantage, as does the
persistence of a substantial under-mortality in the Moroccans
study after correction for under-registration of deaths. Possible
explanations are then either the role of selective processes
(‘healthy migrant hypothesis’) or that of health-protective
behaviours.
The ‘healthy migrant hypothesis’ is particularly difficult to
investigate: first, mortality estimates in countries of origin are
not always available for comparative purposes; second, even
when they are, differences in health care between countries
render the comparisons meaningless, and; third, the regions of
origin of the migrants in their home countries are likely to be
different in their mortality profile from the national average. In
Abraido-Lanza et al.’s paper,
3
the mortality rates in Puerto Rico,
Cuba, and Mexico were found to be lower than those of the US,
and this was interpreted by the authors as consistent with a
cultural explanation of the Latino mortality paradox. Those
comparisons, however, were not very convincing, given that, as
pointed out by Landen,
7
they involved crude rather than age-
standardized mortality rates. In Europe, the French study is the
only one which has incorporated comparisons with the country
of origin of the migrants (Morocco). The picture which emerges
is that of a much higher life expectancy for the migrants than
that estimated for the population of Morocco: the gain in life
expectancy was 9.9 years for men, and 11.6 years for women.
This may reflect the healthy migrant hypothesis, but not neces-
sarily. Generally speaking, international comparisons are
extremely difficult to interpret, with countries differing in numer-
ous factors which influence all-cause mortality, among which
are social class, economic indicators, and effectiveness and
accessibility of health care and other services.
8
In addition, one may legitimately wonder to what extent the
selection of applicants for immigration on the basis of their
health is a plausible explanation of the mortality paradox.
Indeed, two questions are left unanswered: first, do the health
selection effects persist long enough to explain a mortality
advantage decades after it has taken place, and second, if the
migrants are healthier at entry, what are the factors underlying
their superior health? Concerning the first point, surprisingly
little has been published. Study of an industrial cohort in Great
Britain has shown that the ‘healthy worker effect’ was no
longer visible 15 years after entry in the cohort,
9
while con-
versely an analysis of the Assets and Health Dynamics of
the Oldest Old (AHEAD) in the US concludes that good health
of a population at young ages is maintained throughout the
lifespan.
10
As for Hispanics, the literature indicates that their
health and health behaviours deteriorate with accultur-
ation.
11
Clearly, more empirical and theoretical studies are
needed.
Supposing health selection is the key explanation to the
mortality advantage, then what are the factors underlying the
superior health of the migrants? Do they have lower mortality
rates just because the health checks have filtered out the
disabled and chronically ill? Or do they have more favourable
health behaviours? According to Uitenbroek and Verhoeff,
12
who have investigated mortality of the Mediterranean migrants
in Amsterdam, selection at entry is not a convincing explanation
for their remarkable life expectancy. Indeed, in their twenties and
early thirties ‘symptoms of the major causes of death, i.e. cancers
and cardiovascular disease are rarely present, and it is difficult
to imagine how these young people could have been selected
on their future susceptibility’ to those diseases. Those authors,
in accord with Razum et al.
4
are more supportive of the ‘unhealthy
re-migration hypothesis’; in fact a more elaborated version of
the ‘salmon bias hypothesis’ which assumes that migrants who
re-migrate are those who do not cope well socially and economically,
and that those migrants are more likely to experience higher
mortality in the future. This can be viewed as a kind of ‘indirect
1116 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
by guest on May 23, 2011ije.oxfordjournals.orgDownloaded from

selection’ on factors connected to both socio-occupational skills
and health capital, similar to that which has been conceptual-
ized regarding unemployment:
13
‘the mechanism would be that
both unemployment and health are related to a certain per-
sonality trait’. Could we imagine also that an indirect selection
is involved in the migration process, with factors connected to
both the will and capacity to migrate and to health? In relation
to the ‘salmon bias’ concept, Razum et al.
4
also question the
plausibility of re-migration of severely ill migrants, considering
that it is unlikely that Turkish residents return to their home
country when they suffer from ‘conditions such as cardiovascular
disease for which medical treatment in Germany is readily avail-
able and almost free.’ Economic considerations could therefore
deter sick migrants from going home, unless they are moribund,
but in this case do they have the strength to undertake a
journey back home?
Abraido-Lanza et al. consider that the role of cultural factors
involving favourable health behaviours is an attractive
hypothesis to be tested. If it is confirmed, then this would mean
that the migrants would be benefiting from the ‘best of both
worlds’:
14
the favourable habits of their country of origin and
the efficiency of the health care system of their host country. Of
the two studies in Europe, the one which has examined causes
of death and gathered data on health-related habits is the
French one, and the results are quite mixed: on one hand, the
death rates from cancers and cardiovascular diseases are much
lower among Moroccan males than in the French population,
on the other hand, their lifetime consumption of tobacco is
comparable, though there are indications that their nutritional
habits could be more favourable.
15
One important feature of
the Moroccan community in France is that they drink very little
alcohol, and this could play a major role in their mortality
advantage: of all countries in the European Community, France
is the one with the highest percentage of heavy alcohol
drinkers, and it is characterized by a high alcohol-related
premature mortality. There might be cohort effects involved in
the lower lung cancer mortality of the Moroccans in France,
with heavy smoking limited to the younger cohorts who have
not yet reached the age at which lung cancer rates start to
rise. Also, a role for alcohol consumption in lung cancer
aetiology has been suggested in some studies,
16
and one may
wonder whether Moroccans are protected from lung cancer in
part because they drink very little alcohol. Lastly, the potential
role of differential exposure to genetic factors of susceptibility
to lung cancer
17
is worthy of consideration. Greeks in
Australia are another Mediterranean migrant group which
was found to have an exceptionally high life expectancy in
spite of continuing high rates of cigarette smoking, and this
was attributed to the offsetting effects of the Mediterranean
diet.
14
One of the reviews of the Hispanic paradox concluded by
saying that there was a reasonable degree of certainty that
the paradox was real for some subgroups, among them older
Hispanics.
2
In France and Germany, the older cohorts of
migrants were precisely those which had the most difficult
and hazardous working conditions,
18,19
in the mines and the
automobile industry. They would, therefore, be expected to
have higher mortality. In fact, this was the case for migrants
from Eastern European countries but not for Mediterranean
migrants.
20
To date, the causes of the paradox are largely
unknown, be it in the US or in Europe, and, as pointed out by
Franzini et al.:
if the reasons are largely cultural, then the paradox will only
exist for as long as a large percentage of Hispanics remain
culturally distinct from the rest of the US ... a rare window of
opportunity now exists to learn more about how cultural
factors influence one’s health....
2
The migrants’ mortality paradox raises challenging questions
about the nature of the selective processes related to migration,
and those questions have a bearing on health-based selection in
general and on the potential role of indirect selection in
explaining part of the association between socio-demographic
factors and health.
The opening up of new research avenues is desirable to meet
the challenge, along the lines recently delineated by Schwartz,
Susser, and Susser.
21
More attention should be paid to the
historical context of migration and the past living conditions of
the different waves of migrants, before and after migration, and
to the legal and jurisdictional aspects of re-migration. Shifting
the emphasis from individual-based studies to studies of com-
munities as a whole and of the cultural factors that are related
to management of health and disease may provide explanatory
leads. As pointed out by Palloni,
1
studies of migrants in their
host countries should be complemented by:
studies of the sending populations, including those people
who have returned after being migrants, those who could
have been migrants but were not, and those who tried
unsuccessfully to be migrants.
Also, qualitative studies are potentially very informative with
respect to the acculturation process and its health-related
aspects, to the cultural representations of health, and the range
of motives for re-migration. Last but not least, an international
perspective on the subject would throw new light on the how’s
and why’s of this enduring epidemiological enigma.
References
1
Palloni A, Morenoff J. Interpreting the paradoxical in the Hispanic
paradox- Demographic and epidemiologic approach. In: Weinstein M,
Hermalin AI, Stoto MA (eds). Population Health and Aging—
Strengthening the Dialogue between Epidemiology and Demography. New
York: The New York Academy of Science, 2001, pp. 140–74.
2
Franzini L, Ribble JC, Keddie AM. Understanding the Hispanic
paradox. Ethn Dis 2001;11:496–518.
3
Abraido-Lanza AF, Dohrenwend BP, Ng-Mak DS, Turner JB. The
Latino mortality paradox: a test of the ‘salmon bias’ and healthy
migrant hypotheses. Am J Public Health 1999;89:1543–48.
4
Razum O, Zeeb H, Akgun HS, Yilmaz S. Low overall mortality of
Turkish residents in Germany persists and extends into a second
generation: merely a healthy migrant effect? Trop Med Int Health
1998;3:297–303.
5
Razum O, Zeeb H, Rohrmann S. The ‘healthy migrant effect’—not
merely a fallacy of inaccurate denominator figures. Int J Epidemiol
2000;29:191–92.
6
Courbage Y, Khlat M. Mortality and causes of death of Moroccans in
France, 1979–91. Population: an English selection 1996;8:59–94.
7
Landen M. Letter to the editor: Latino mortality rates. Am J Public
Health 2000;90:1798.
LETTERS TO THE EDITOR 1117
by guest on May 23, 2011ije.oxfordjournals.orgDownloaded from

In his commentary on our recent paper,
1
Dr Edward A Frongillo
2
criticizes our use of a well-described household food security
scale in Trinidad. There will always be some uncertainty con-
cerning the application of a given measure as there is no perfect
instrument to evaluate food security or dietary patterns in any
population. The household food security measure was used in
the US national Current Population Survey (which provided the
comparison data used in Dr Frongillo’s commentary) but the
application of the instrument to all groups in the multilingual,
culturally diverse US population ‘has not been examined suffici-
ently’ (ref. 3, p. 8). Questionnaire evaluation must be considered
when differences in literacy, language, dialect, or culture, as
well as socioeconomic status, may influence responses and
this consideration might suggest that an instrument should
be tailored to local requirements. It is advisable, however, to be
judicious in modifying such measures so as not to compromise
the validity or comparability of an instrument. Departures from
a previously tested template should only be undertaken to
guarantee enhanced performance of a measure. Dr Frongillo’s
comments appear to underestimate both the weight of evidence
required to justify an alteration to an established measure and
the limitations of local ‘validation’ studies. Before concluding
that a measure gives unsatisfactory results in a given local popu-
lation, or a particular group within a population, it is essential
to ensure that the findings cannot be ascribed to error or bias.
There is a relatively high risk that local questionnaire evaluation
studies, implemented within the short time scales suggested,
will lead to erroneous conclusions if sample sizes are too small
or if subjects are insufficiently representative.
While our data suggested an unexpected difference in the
frequency of food insecurity according to ethnicity, it would be
premature to conclude that the instrument had differential validity
in these groups. We had no prior hypothesis about ethnic differ-
ences in food insecurity. The study used cluster sampling with
the selection of a relatively small number of neighbourhoods.
Food insecurity, income, and ethnicity each showed evidence of
clustering within neighbourhoods. Imbalances in the character-
istics of different groups could arise through chance. A larger
study will be required to determine whether this finding will be
replicated. Dr Frongillo observes that the ranking of the preval-
ence of affirmative responses to the first two items differs in our
data as compared with the US data. This seems to overemphasize
the Guttman-like properties of the scale, since it is not clear that
an inability to afford balanced meals should always indicate a
greater severity of food insecurity than that for a person finding
that her food did not last and being unable to buy more. We
agree that the ‘balanced meal’ item may be unsatisfactory but
rather than concluding that this requires the adaptation of the
household food security scale in each local setting, special
consideration should be given to reviewing this item when the
instrument is revised.
A potential for misclassification of food insecurity status does
raise a concern that a possible true association between food
insecurity and obesity might be attenuated. In order to explore
this possibility, we repeated our previous analyses using the
same methods but with two modifications to the classification of
food insecurity. We first used a cut-point of three rather than
two to identify subjects who were food insecure. We then
omitted the ‘balanced meal’ item from the assessment of food
1118 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
8
Abraido-Lanza A, Dohrenwend B, Ng-Mak D, Turner JB. Letter to the
editor: Abraido-Lanza et al. respond. Am J Public Health 2000;
90:1798–99.
9
Fox AJ, Collier PF. Low mortality rates in industrial cohort studies due
to selection for work and survival in the industry. Br J Prev Soc Med
1976;9:225–30.
10
Swallen K. Do health selection effects last? A comparison of morbidity
rates for elderly adult immigrants and US-born elderly persons. J Cross
Cult Gerontol 1997;12:317–39.
11
Vega W, Amaro H. Latino outlook: Good health, uncertain prognosis.
Annu Rev Public Health 1994;15:39–67.
12
Uitenbroek DG, Verhoeff AP. Life expectancy and mortality
differences between migrant groups living in Amsterdam, The
Netherlands. Soc Sci Med 2002;54:1379–88.
13
Bartley M, Ferrie J, Montgomery SM. Living in a high unemployment
economy: understanding the health consequences. In: Marmot M,
Wilkinson RG (eds). Social Determinants of Health. Oxford: Oxford
University Press, 2000, pp. 81–104.
14
Powles J. The best of both worlds: attempting to explain the persisting
low mortality of Greek migrants to Australia. In: Caldwell J, Findley S,
Caldwell P, Santow G (eds). What we Know about Health Transition: the
Cultural, Social and Behavioural Determinants of Health. Canberra: Health
Transition Centre, 1990, pp. 584–94.
15
Darmon N, Khlat M. An overview of the health status of migrants in
France, in relation to their dietary practices. Public Health Nutr
2001;4:163–72.
16
Bandera EV, Freudenheim JL, Vena JE. Alcohol consumption and
lung cancer: a review of the epidemiologic evidence. Cancer Epidemiol
Biomarkers Prev 2001;10:813–21.
17
Garte S, Gaspari L, Alexandrie AK et al. Metabolic gene polymorphism
frequencies in control populations. Cancer Epidemiol Biomarkers Prev
2001;10:1239–48.
18
Schultze G. Premiere et deuxieme génération de migrants turcs en
RFA: mobilité professionnelle et son incidence sur le processus
d’intégration. In: Jund A, Dumont P, de Tapia S (eds). Enjeux de
l’immigration Turque en Europe. Paris: L’Harmattan, 1995, pp. 147–53.
19
Temime E. France, Terre d’immigration. Paris: Gallimard, 1999.
20
Brahimi M. La mortalité des étrangers en France. Population 1980;
3:603–22.
21
Schwartz S, Susser E, Susser M. A future for epidemiology? Annu Rev
Public Health 1999;20:15–33.
DOI: 10.1093/ije/dyg308
Food insecurity definitions and body mass index. Response
From MARTIN GULLIFORD, DEEPAK MAHABIR and BRIAN ROCKE
Department of Public Health Sciences, King’s College London, 42 Weston St,
London SE1 3QD, UK and Ministry of Health, Trinidad and Tobago. E-mail:
martin.gulliford@kcl.ac.uk
by guest on May 23, 2011ije.oxfordjournals.orgDownloaded from

insecurity, using a cut-point of two items out of five. The results
are shown in the Table. As expected, the estimated prevalence
of food insecurity was somewhat lower when the more restrictive
definitions were used. The associations of food insecurity with over-
weight (body mass index [BMI]  25 kg/m
2
) or obesity (BMI 
30 kg/m
2
) were similar using each of the three definitions of food
insecurity. Thus the test of our primary hypothesis concerning food
insecurity and obesity was robust to varying the definition of
food insecurity. The association of underweight (BMI  20 kg/m
2
)
with food insecurity was somewhat sensitive to the definition of
food insecurity, but this analysis was based on only 41 cases who
were underweight. This again points to the need for a larger study.
A fundamental problem with locally developed instruments is
that the ability to generate generalizable information may be
severely compromised. The application of robust standardized
questionnaire measures should generally be preferred to the
proliferation of locally developed or locally adapted measures.
Judgements on what is appropriate must be informed by know-
ledge of, and experience in, a given context. We doubt whether
experiences from Bangladesh or Burkina Faso will have much
relevance in the Caribbean region which has strong links with
Europe and North America through geography, history, trade,
culture, and language. Experience in the Caribbean shows
that the importation of health-related measurement scales is
feasible.
4
Our report describes one of the first applications of a
survey-based food security measure in the Caribbean. While our
report raises several issues which require further investigation,
the responses obtained using this instrument in Trinidad are, as
Dr Frongillo indicates,
2
‘similar to those of the 1993 New York
state sample’ (ref. 2, p. 516). The associations of food insecurity
with lower incomes, physical disability, and low consumption of
fruit and vegetables that we documented in our data, point to
the validity of the measure and are consistent with those of
other reports. Our survey has also provided the data required to
formally analyse the properties of the short form household
food security scale in the context of Trinidad and Tobago.
In spite of the uncertainties ‘the US food security measure is
understood to be broadly comparable across the US ... This
comparability will likely hold in many other countries as well,
but may not in some’ (ref. 3, p. 8).
References
1
Gulliford MC, Mahabir D, Rocke B. Food insecurity, food choices,
and body mass index in adults: nutrition transition in Trinidad and
Tobago. Int J Epidemiol 2003;32:508–16.
2
Frongillo EA. Commentary: Assessing food insecurity in Trinidad and
Tobago. Int J Epidemiol 2003;32:516–17.
3
Wolfe WS, Frongillo EA. Building Household Food Security Measurement
Tools from the Ground Up. Washington, DC: Food and Nutrition
Technical Assistance Project. Academy for Educational Development,
2000. http://www.fantaproject.org/downloads/pdfs/hfs_measure.pdf
Accessed 3 September 2003.
4
Gulliford MC, Mahabir D. Relationship of health-related quality of
life to symptom severity in diabetes mellitus; a study in Trinidad and
Tobago. J Clin Epidemiol 1999;52:773–80.
DOI: 10.1093/ije/dyg330
LETTERS TO THE EDITOR 1119
Table Frequency of food insecurity according to different definitions and associations with body mass index (BMI) category
Food insecurity definition
Excluding ‘balanced meals’
‘Standard’ 2/6 items ‘Conservative’ 3/6 items item (2/5 items)
BMI category (kg/m
2
)Total Freq.
a
(%) OR
b
(95% CI)
c
Freq. (%) OR (95% CI)
c
Freq. (%) OR (95% CI)
c
20.0 41 14 (34) 3.21 (1.17, 8.81)
d
10 (24) 2.72 (0.95, 7.85)
e
11 (27) 2.40 (0.86, 6.70)
f
20.0–24.9 149 38 (26) – 27 (18) – 32 (21) –
25.0
g
277 72 (26) 1.05 (0.61, 1.79) 46 (17) 0.87 (0.47, 1.60) 53 (19) 0.84 (0.47, 1.51)
30.0
g
120 35 (29) 1.08 (0.55, 2.12) 25 (21) 1.07 (0.51, 2.24) 27 (23) 0.90 (0.44, 1.84)
Not known 64 10 (16) 7 (11) 7 (11)
Total 531 134 (25) 90 (17) 103 (19)
a
Frequency.
b
Odds ratio.
c
Relative odds of BMI category contrasted with 20.0–24.9 kg/m
2
if food insecure, adjusted for age, sex, and ethnic group.
d
P = 0.024.
e
P = 0.064.
f
P = 0.094.
g
Note categories are not mutually exclusive.
by guest on May 23, 2011ije.oxfordjournals.orgDownloaded from

Sirs—As genetic studies of complex human diseases are relying
more and more on the epidemiological association paradigm,
1
it becomes a crucial issue to determine whether a population
is a homogeneous one or has hidden structures within it. If
the former is true, one can infer the genomic location(s) of the
putative susceptibility gene(s) for a particular disease by simply
comparing marker allele frequencies between ‘cases’ and ‘con-
trols’ recruited from the population. Whereas if the latter is true,
a naive case-control approach will produce an excess of false positive
results.
2
Here I propose a method to detect population stratification
(subdivision) using a panel of ‘single nucleotide polymorphisms’
(SNP).
3
The SNP are the most abundant type of human genetic
markers. Genotyping of SNP has the potential for automation
and the cost of doing it is expected to go down in the future.
Assume that a total of p SNP (indexed by i) have been geno-
typed in n subjects recruited from a population. For the ith SNP
marker, let b
i
represent the number of subjects with hetero-
zygous ‘Mm’ genotype, and let a
i
and c
i
represent the numbers
of subjects with homozygous ‘MM’ and ‘mm’ genotypes, respect-
ively (a
i
+ b
i
+ c
i
= n). Further define D
i
= 4a
i
c
i
– b
i
(b
i
– 1). The
expectation of D
i
can be written as the sum of two terms:
Under the null hypothesis of a Hardy-Weinberg population, the
first term becomes
(see ref. 4), and the second term, n · [2f
i
(1 – f
i
)], where f
i
is the
allele frequency of the ith SNP. Thus we see that E(D
i
) = 0 under
the null.
One can choose the p SNP to be unlinked or in linkage equili-
brium (widely spaced, say ~10 cM), such that the D
i
’s are
independent of one another. Therefore, we have
distributed asymptotically (for large p) as the standard normal
distribution under the null. Under the alternative hypothesis
that the subjects are recruited from a population with hidden
structures, we have E(D
i
)  0 due to the well-known Wahlund
phenomenon (an excess of homozygotes).
5
Thus, an upper
one-sided test based on Z, which combines the information of a
panel of SNPs, can be used to detect population stratification.
There are three important characteristics of such an
approach: (1) the method does not require knowing the allele
frequencies of a panel of SNP in advance; (2) the method is
contingent on the number of typed markers (p)being large,
whereas the number of subjects recruited (n) can be small (even
n =2 will do, provided p is large); and (3) the method involves
nothing more than simple arithmetic. The contingency table
χ
2
test
6
and the Hardy-Weinberg test
7
had previously been
proposed to test for population stratification. However, these
two methods will lead to false rejection or acceptance when the
number of subjects recruited is small or some cell frequencies
are small or zero. Another alternative would be to consider
the ‘structured association’ approach.
8–10
However, it demands
many computer-intensive modelling efforts, which are beyond
the scope of most epidemiologists.
If a global survey is to be conducted to detect possible hidden
structure in the human populations, due to cost and time
constraint, one probably would have to be content with just a
few subjects for each racial/ethnic group. The present approach
is a viable alternative in such a scenario.
References
1
Risch N, Merikangas K. The future of genetic studies of complex
human diseases. Science 1996;273:1516–17.
2
Ewens WJ, Spielman RS. The transmission/disequilibrium test:
history, subdivision and admixture. Am J Hum Genet 1995;57:455–64.
3
The International SNP Map Working Group. A map of human
genome sequence variation containing 1.42 million single nucleotide
polymorphisms. Nature 2001;409:928–33.
4
Hernández JL, Weir BS. A disequilibrium coefficient approach to
Hardy-Weinberg testing. Biometrics 1989;45:53–70.
5
Li CC. Population subdivision with respect to multiple alleles. Ann
Hum Genet 1969;33:23–29.
6
Pritchard JK, Rosenberg NA. Use of unlinked genetic markers to
detect population stratification in association studies. Am J Hum Genet
1999;65:220–28.
7
Deng HW, Chen WM, Recker RR. Population admixture: detection
by Hardy-Weinberg test and its quantitative effects on linkage-
disequilibrium methods for localizing genes underlying complex
traits. Genetics 2001;157:885–97.
8
Pritchard JK, Stephens M, Rosenberg NA, Donnelly P. Association
mapping in structured populations. Am J Hum Genet 2000;67:170–81.
9
Pritchard JK, Stephens M, Donnelly P. Inference of population
structure using multilocus genotype data. Genetics 2000;155:945–59.
10
Satten GA, Flanders WD, Yang Q. Account for unmeasured popu-
lation substructure in case-control studies of genetic association using
a novel latent-class model. Am J Hum Genet 2001;68:466–77.
DOI: 10.1093/ije/dyg301
Z
D
D
i
i=
p
i
i=
p
=
∑
∑










1
2
1
4
1
2
2
n
–f f
n
ii
⋅






(– )
EE E() .Dn
ab
n
bc
n
b
n
n
b
n
i
iii i i i
=⋅
+
⋅
+
−⋅






+⋅






4
2
2
2
2
1
2
2
1120 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
Detecting population stratification using a panel of single nucleotide polymorphisms
From WEN-CHUNG LEE
Graduate Institute of Epidemiology, College of Public Health, National Taiwan
University. No. 1, Jen-Ai Rd, 1st Sec, Taipei, Taiwan. E-mail: wenchung@ha.
mc.ntu.edu.tw
by guest on May 23, 2011ije.oxfordjournals.orgDownloaded from