IJE Advance Access published online on May 17, 2007
International Journal of Epidemiology, doi:10.1093/ije/dym091
Published by Oxford University Press on behalf of the International Epidemiological Association © The Author 2007; all rights reserved.
Cohort Profile: The Chennai prospective study of mortality among 500 000 adults in Tamil Nadu, South India
Vendhan Gajalakshmi1,2,*,
Richard Peto3,
Vendhan Chiruvai Kanimozhi1,
Gary Whitlock3 and
Dasarathi Veeramani1
1 Epidemiological Research Center, Chennai, Tamil Nadu, India.
2 International Agency for Research on Cancer, Lyon, France.
3 Clinical Trial Service Unit & Epidemiological Studies Unit, University of Oxford, UK.
* Corresponding author. E-mail: gajaerc{at}gmail.com
 |
How did the study come about?
|
|---|
A prospective study of half a million adults living in the city
of Chennai (formerly Madras) arose out of discussions at the
1994 International Cancer Congress in Delhi about how to assess
the effects of tobacco on health in different parts of India.
Chennai is the capital of the South Indian state of Tamil Nadu,
and it is India's fourth most populous city. Two large-scale
epidemiological studies of tobacco and other factors were established:
a case-control study
1 that could provide reasonably reliable
results quickly, and a prospective cohort study that could provide
more robust results over a longer period. (A parallel prospective
study of 100 000 adults, not included in this profile, is in
progress in the nearby rural area of Villupuram;
Figure 1.)
The case-control study, which involved 48 000 adult deaths during
1995–97 and 47 000 controls who had been living with a
case, indicated that smoking is a cause of, among other things,
about half of all tuberculosis (TB) deaths among men.
1 The prospective
cohort study, which recruited half a million participants between
1998 and 2001, is described here.
|
What does it cover?
|
|---|
The Chennai Prospective Study examines the associations of tobacco
smoking, quid chewing, alcohol drinking, obesity, blood pressure,
respiratory function and other factors with overall mortality,
and with cause-specific mortality.
Table 1 shows that in the
mid-1990s the age-standardized death rates for middle-aged men
and women in Chennai differed very substantially from those
in, for example, Japan, the United Kingdom and the United States.
All-cause mortality was twice as great in Chennai as in the
United Kingdom and the United States, and about three times
as great as in Japan. In both sexes, about half of the excess
mortality compared with these three countries was accounted
for by vascular diseases other than stroke—i.e. chiefly
by ischaemic heart disease (IHD). The main reasons for the high
IHD mortality rates in Chennai are not fully known. Furthermore,
there are still high TB death rates in Chennai, particularly
among men, but not in the other three countries. Within Chennai,
smoking accounts for most of the difference in TB mortality
between men and women, but even among women TB is still a major
cause of death. For these and other diseases, the study will
help quantify the evolving importance in this population of
some known causes of death such as tobacco, alcohol, high blood
pressure and obesity, and help identify and quantify some other
risk factors.
View this table:
[in this window]
[in a new window]
|
Table 1 Age-standardized mortality ratesa at ages 35 to 69 in Chennai (1995–97) and, for comparison, three high-income countries
|
|
|
Who is in the sample?
|
|---|
The study includes one-third of all adults aged 35 and over
in the city of Chennai. We sought to visit all houses in two
of the 10 administrative zones into which Chennai is divided
(
Figure 1), as well as all houses in either 30% or 50% of the
streets (selected randomly) in five of the other eight zones.
Participant information was collected through in-person interviews
of all aged 35 or over in each household visited. Only 5% (25
184) refused, and 500 816 took part. The mean age of participants
was 49 for women and 48 for men, and 97% were aged 35–74.
Eighty-five per cent were Hindu, and 7% each were Muslim and
Christian. Half (53%) were men. Seventeen per cent of the men
had been educated at university or college, compared with 6%
of women (
Table 2). Nearly one in three women, and 1 in 10 men,
had received no formal education. The most commonly reported
prevalent diseases at baseline were diabetes and hypertension;
few reported a history of IHD, stroke or TB (
Table 2).
How often are they being followed-up?
Follow-up, which can be especially challenging in lower-income
countries,
3 is being conducted not only by passive but also
by active methods. The passive methods involve record linkage
to routine mortality data collected from the Chennai Vital Statistics
Division for all adult (age 30 or more) deaths occurring in
Chennai since 1998. Participant and death records are matched
by manual and computerized (probabilistic
4) record linkage based
on name, age, sex, address and name of spouse or father. Even
though death registration is almost complete in Chennai, reliable
assessment of cause-specific mortality from the stated cause
on the death certificate is not yet possible because more than
half the medical causes of death are non-specific or inaccurate;
hence, all these deaths have to be supplemented by verbal autopsy
reports.
The verbal autopsy1,2 is a systematic retrospective inquiry, generally of family members, about the circumstances, events, symptoms and signs of illness and treatment details prior to a participant's death. It is an interactive process, with the respondent taking the lead in providing the narrative information, and the interviewer prompting where necessary for more details. The interviewers use an open narrative format for their report rather than taking the respondent through detailed questionnaires, as these would have to be impracticably long to cover a sufficiently wide range of possible causes for adult deaths. However, each interviewer has a short checklist of symptoms and signs that are used to help ensure the reports are appropriately informative. The verbal autopsy report, written in the local language by the interviewer, describes the chronological order of the appearance of various signs and symptoms and their progress, with details of treatment received, if any, the name(s) of the hospital(s) attended and any history of similar episodes. The interviewers are carefully trained non-medical graduates, and each report is independently reviewed centrally by two specifically trained medical doctors to arrive at a probable underlying cause of death. Strict supervision of the fieldwork and of the quality of the narratives is maintained by random checking of 5% of the interviews, partly to discover any inadvertent errors and partly to ensure that the original interviewer knows there is a 5% chance that the quality of the work will be checked just a few days later.
This verbal autopsy tool for adult deaths was originally developed and validated in 80 000 deaths in urban and rural areas of Tamil Nadu.1,5 It appeared generally satisfactory, so in recent years its use in nationwide surveys by the Registrar-General of India has been piloted.6 It is now being adapted for use in India's Sample Registration System, a large nationally representative survey of a million households in several thousand randomly chosen study areas, with prospective follow-up of all births and deaths.
The active component of follow-up in the Chennai study involves the field interviewers revisiting all households once during 2002–05, and then once in every two years (2006–07, etc.) indefinitely, supplemented where deaths have occurred by a verbal autopsy. In addition to repeat assessment of a few risk factors (tobacco smoking, quid chewing and alcohol drinking), surviving participants are asked about the occurrence (after the baseline interview) of any clinical events such as the onset of TB, diabetes, IHD, stroke, hypertension or cancer. If the study participant is not living at the address noted at the time of collecting the baseline data, then neighbours are asked when the person moved out and where he or she went.
A repeat survey of 20 000 individuals will be carried out in 2007–08 in randomly chosen streets in the study area to help correct for the ‘regression dilution’ bias7 that can result from random errors in baseline measurements. Participants will be re-assessed with repeat interviews and re-measurements as in the baseline survey. Further such surveys are planned at approximately five-year intervals thereafter.
|
What has been measured?
|
|---|
Among the items recorded at the baseline assessment were age,
sex, educational level, occupation, religion, mother tongue,
diet (vegetarian/not), tobacco smoking (including not only cigarettes
but also tobacco in the form of bidis, chuttas, cigars, pipes
and hookahs), quid chewing (with or without tobacco in the quid),
use of tobacco powder, alcohol drinking, family history of cancer,
and whether the participant had ever been diagnosed as having
TB, stroke, IHD, diabetes, hypertension or cancer (
Table 3).
In addition, height, weight, blood pressure, peak expiratory
flow rate and (in men only, for cultural reasons) waist circumference
were measured.
The questionnaires were checked centrally for consistency and
missing values by coding clerks, and then double-entered into
computers by 16 data-entry operators. As with the verbal autopsies,
a random 5% of the baseline assessments were checked one week
later by a fieldwork supervisor who re-interviewed and re-measured
participants in randomly selected households while blinded to
the original data. This random checking was done partly to ensure
that the fieldwork was reliably motivated (the original interviewer
knew that there was a 5% chance that the quality of their work
would be checked just a few days later).
|
What is attrition like?
|
|---|
Participants will be lost to follow-up if they move permanently
out of Chennai City, but only a very small proportion of participants
is likely to do so each year. It is hoped that those who move
address within the city can continue to be followed, but this
is not yet known. Hence, estimates of actual losses to follow-up
are not yet available.
|
What has it found?
|
|---|
Analyses of associations between possible risk factors and cause-specific
mortality will not start until 2008 when the second round of
follow-up is complete. By then, over 25 000 deaths will have
been recorded. On the basis of the verbal autopsy study of 48
000 deaths in 1998–99,
2 about 5% of the deaths that occur
at ages 35–69 in either sex will be due to stroke, about
40% will be due to IHD or other vascular diseases, and about
4% (women) or 8% (men) will be due to TB (
Table 1).
Table 4 shows the baseline distributions of various risk factors by age and sex. About one-third of the men aged 35–74 were current smokers of tobacco, and about one-third of these smokers used bidis. Bidis, which are widely smoked throughout South Asia, are smaller than Western-style cigarettes, and instead of the tobacco being wrapped in paper it is wrapped in the leaf of another plant (temburni). Smoking was rare among women: 99.9% reported never having smoked. In contrast, more women than men in every age group (and 10% vs 6% overall) reported quid chewing (either with or without tobacco in the quid). In both sexes, quid chewing was more prevalent at older ages. A quarter of men overall (and even more at younger ages) reported being current drinkers of alcohol, but almost all women (99.9%) reported never having been a regular drinker. One-third of the younger men who chewed tobacco also smoked it (whereas almost none of the women who chewed tobacco smoked it too). Figure 2 shows the distributions of the main smoking, chewing and drinking habits by educational level. Independently of age, bidi-smoking men tended to be less educated, as did beer- or brandy-drinking men, and also quid-chewing men and quid-chewing women (particularly, for both sexes, when the quid contained tobacco).
View larger version (17K):
[in this window]
[in a new window]
[Download PowerPoint slide]
|
Figure 2 Smoking, alcohol drinking and chewing habits in 264 848 men, and chewing habits in 235 968 women, by highest education level. Prevalences standardized for age using the age distribution of the study population. Boxes are point estimates, and vertical lines are 95% confidence intervals (not visible if smaller than box). ‘Univ.’ means university or college. The smoking graph omits about 0.3% of men who smoked things other than cigarettes or bidis, and the alcohol graph omits about 1% of men who drank any combination of the three categories of alcohol displayed
|
|
View this table:
[in this window]
[in a new window]
|
Table 4 Baseline distributions of selected disease risk factors in 500 816 adults aged 35 or more, by age and sex
|
|
In both sexes, the mean body mass index (BMI, kg/m
2) was higher
at younger ages, and at all ages the mean BMI was higher among
women than men. The mean BMI levels for men and older women
were much lower than in most Western countries, but the mean
for women aged 35–54 (24.0 kg/m
2) was similar to that
in some European countries (e.g. France, Sweden and Belgium
8).
Of the women aged 35–54, 24% had a BMI of 25–30
kg/m
2 and 10% had a BMI of 30 kg/m
2 or more. Unusually, in comparison
with Western countries, the mean systolic blood pressure above
age 55 was 1–2 mmHg higher among women than among men.
However, at ages 35–54 it was 1.5 mmHg lower among women
than among men, despite the mean BMI being higher among women.
|
What are the main strengths and weaknesses?
|
|---|
This study's main strength is the large number of participants
in a previously unstudied population, and the high background
mortality rate in that population, meaning that the study will
not need to wait a long time for a large number of deaths to
occur. It should, therefore, be able fairly quickly to assess
in detail many predictors of common causes of death in Chennai.
Cause-and-effect relationships will be generalizable not only
to Indian populations, but also more widely—for example,
the case-control finding that smoking is a major cause of death
from TB
1 has recently been confirmed by David Zaridze and his
colleagues in Siberia. Robust data collection methods—including
strict supervision, random checking and active follow-up—ensure
that all of the collected information is reasonably accurate.
The verbal autopsy tool is expected to overcome many of the
well-known shortcomings of cause-of-death information on adult
death certificates in most lower-income (and some higher-income
9)
countries. Correctly linking mortality data (from the Vital
Statistics Division) and baseline data for the same individual
is more challenging than in many settings because of the absence
of a unique identifier and the varying usage (including spelling)
of last names. However, modern computerized probabilistic matching
methods
4 can help with this problem, although some human judgement
(plus in some cases, extra fieldwork) is still necessary. Owing
to resource limitations, the study was not able to collect and
store biological specimens.
|
Where can I find out more?
|
|---|
Initial enquiries should be made to the principal investigator
(VG) in Chennai. An English-language version of the questionnaire,
and associated coding conventions, is available at:
www.ctsu.ox.ac.uk/chennai on the Clinical Trial Service Unit and Epidemiological Studies
Unit (CTSU) website.
|
Acknowledgements
|
|---|
We thank the Chennai Vital Statistics Division, The Cancer Institute
(WIA) in Chennai, and, most importantly, the study participants,
for their help in establishing the study. The study has been
funded by the Clinical Trial Service Unit and Epidemiological
Studies Unit, which receives core funding from the Medical Research
Council, the British Heart Foundation and Cancer Research UK,
with additional support for international studies of smoking
from the US NIH Fogarty International Center (grant no 5R01TW005993-02).
|
References
|
|---|
1 Gajalakshmi V, Peto R, Kanaka TS, Jha P. Smoking and mortality from tuberculosis and other diseases in India: retrospective study of 43 000 adult male deaths and 35 000 controls. Lancet (2003) 362:507–15.
[CrossRef][ISI][Medline]2 Gajalakshmi V, Peto R, Kanaka S, Balasubramanian S. Verbal autopsy of 48 000 adult deaths attributable to medical causes in Chennai (formerly Madras), India. BMC Public Health (2002) 2:7.[CrossRef][Medline]
3 Gajalakshmi CK, Shanta V. Methodology for long term follow-up of cancer cases in a developing environment. Indian J Cancer (1995) 32:160–68.[Medline]
4 Jaro MA. Probabilistic linkage of large public health data files. Stat Med (1995) 14:491–98.[ISI][Medline]
5 Gajalakshmi V, Peto R. Verbal autopsy of 80,000 adult deaths in Tamilnadu, South India. BMC Public Health (2004) 4:47.[CrossRef][Medline]
6 Jha P, Gajalakshmi V, Gupta PC, et al. Prospective study of one million deaths in India: rationale, design, and validation results. PLoS Medicine (2006) 3:e18.[CrossRef]
7 Clarke R, Shipley M, Lewington S, et al. Underestimation of risk associations due to regression dilution in long-term follow-up of prospective studies. Am J Epidemiol (1999) 150:341–53.[Abstract/Free Full Text]
8 WHO Global InfoBase team. The SuRF Report 2. Surveillance of chronic disease risk factors: country-level data and comparable estimates. (2005) Geneva: World Health Organization.
9 Mathers CD, Ma Fat D, Inoue M, Rao C, Lopez A. Counting the dead and what they died from: an assessment of the global status of cause of death data. Bull World Health Organ (2005) 83:171–77.[ISI][Medline]
10 WHO Statistical Information System. WHO mortality database. Accessed 27 March, 2007. Available at: http://www.who.int/whosis/mort/en/.
11 UN Population Division. World population prospects: the 2004 revision. In: ST/ESA/SER.A/244 (2005) New York: United Nations.
CiteULike 
Connotea 
Del.icio.us What's this?
Top
How did the study...
What does it cover?