|Home | About | Journals | Submit | Contact Us | Français|
This paper describes the main methodological aspects of a cohort study, with emphasis on its recent phases, which may be relevant to investigators planning to carry out similar studies. In 1993, a population based study was launched in Pelotas, Southern Brazil. All 5,249 newborns delivered in the city’s hospitals were enrolled, and sub-samples were visited at the ages of one, three and six months and of one and four years. In 2004-5 it was possible to trace 87.5% of the cohort at the age of 10-12 years. Sub-studies are addressing issues related to oral health, psychological development and mental health, body composition, and ethnography. Birth cohort studies are essential for investigating the early determinants of adult disease and nutritional status, yet few such studies are available from low and middle-income countries where these determinants may differ from those documented in more developed settings.
Birth cohort studies are receiving growing attention. Early life characteristics such as birthweight, infant and childhood growth patterns, and breastfeeding practices have been linked to morbidity in childhood, adolescence and adulthood. In the 1980’s, Barker proposed the “fetal origins of adult diseases” hypothesis, based on studies associating birthweight and weight at one year with cardiovascular disease risk latter in life.1 A modified version of this hypothesis was proposed by Lucas8 in the 1990s, which he referred to as the “programming” hypothesis. More recently, Singhal & Lucas,11 based on experimental and observational data, proposed the “growth acceleration hypothesis”, shifting the focus from intrauterine growth restriction to the velocity of growth in early life. The debate on these issues is still rife.
The vast majority of birth cohort studies are from high-income countries. However, newborns from low and middle-income countries may differ substantially from those. For example, intrauterine growth retardation is much more frequent,10 as are anthropometric deficits, slow growth in childhood and infectious diseases.15 This stresses the need to carry out studies in less-developed settings.14
Cross-sectional, case-control and retrospective cohort designs have been used to investigate these issues, but the quality of data on early exposures is often poor and may lead to bias or misclassification. Therefore, prospective studies are warranted, particularly those that follow a group of people since birth to adulthood. The first national prospective birth cohort study was started in 1946 in the United Kingdom16 and was followed by many similar studies in high-income countries. The high costs of such studies, and the complex logistics involved in following up a large number of individuals over long time periods, has limited the number of such studies in low and middle-income countries.6
In Pelotas, Southern Brazil, a population-based birth cohort study was started in 1982, and subjects have been followed up at several points in time during infancy, childhood, adolescence and young adulthood. The Pelotas 1982 Birth Cohort Study became the largest and longest running birth cohort study in a developing country.6 The success of this study encouraged us to plan a new birth cohort one decade later. Due to funding issues, it was not possible to start the new cohort in 1992, but only in 1993.
Pelotas currently has 320,000 urban inhabitants, being located at the extreme South of Brazil, near the Uruguayan border. Its main economic activities are rice production, commerce and education. At the time of the study, the infant mortality rate was 21 deaths per thousand births.
The main objectives of the 1993 cohort were: (a) to evaluate time trends in health indicators throughout the lifespan, through a comparison with results of the 1982 study; (b) to test specific hypotheses which were not possible to address in the 1982 cohort; (c) to improve data quality with the lessons learned from the 1982 study.
The early phases of the Pelotas 1993 Birth Cohort Study were described in previous publications.3,13 In the present paper, we provide a summary of the methods of the early visits (1993-94) as well as a detailed description of the visits and sub-studies that took place since 1997.
The 1982 study was originally designed as a perinatal health survey, rather than a birth cohort. As a consequence, detailed information necessary for long-term follow-up were not initially collected. In contrast, the initial interview for the 1993 cohort included information on the address and telephone number of the cohort member’s family; address and telephone number of at least two relatives or close family friends; and place of employment of the parents. The research team also asked families if they were planning to move, and if so, where to. In addition, the full names of the child and of both parents were collected – in 1982 only the mother’s name was recorded; because many children still did not have a name on the occasion of the hospital interview, the child’s name was often missing. Names, addresses and phone numbers were confirmed in all follow-up visits. It is important to note that all these contact data were kept in a confidential database with restricted access.
The 1982 cohort showed that most infant deaths and hospital admissions were concentrated in the first few months of life, and also that critical events related to feeding patterns took place at these early ages. By the early 1990s, the international literature was starting to highlight the importance of early “critical windows” in which exposures would affect long-term health.7 To allow such analyses, the 1993 cohort included visits at the ages of one, three, six and 12 months, in contrast to the 1982 cohort for which the first follow-up visit took place at 12 months. Also, the high burden of disease faced by low-birthweight children in the 1982 cohort led us to oversample these children in the 1993 follow-up visits.
The 1993 study also allowed to correct some of the pitfalls of the earlier cohort. In 1982, birth length was not collected because it was deemed to be a complex measurement requiring special equipment and training. Birthweight was obtained in 1982 from the regular hospital scales (that were periodically calibrated by the research team). In 1993, birth length was measured using standard equipment and protocols, and each hospital was provided with a new electronic scale. In 1982, gestational age was based on the reported date of the last menstrual period, but this information was missing for 20% of the cohort. In 1993, we also collected this variable for comparison purposes, but in addition the interviewers were trained on the Dubowitz scale based on the physical examination of the newborn.4 This allowed us to use the first variable in reports comparing both cohorts, and the second indicator when analyzing the 1993 cohort alone.
Table 1 shows the main follow-up visits of the 1993 cohort study, including year of data collection, target ages, sampling strategies, as well as the number of eligible individuals and follow-up rates.
All five maternity hospitals were visited daily during 1993. Mothers who lived in the urban area answered a questionnaire containing demographic, socioeconomic, reproductive, behavioral, care seeking and morbidity variables (Table 2). Weight (precision of 10 g), length (precision of 1 mm) and head circumference (precision of 1 mm) of all newborns were measured. There were 16 refusals, and information was obtained on 5,249 live births and 55 fetal deaths. The latter were included in the mortality statistics, but are not computed in the birth cohort study. This phase of the study, as well as the home visits at six and 12 months, were financed by the European Economic Commission.
For the one and three-month visits, a systematic sample of 655 cohort members was selected. The one-month visit started in February 1993 (when children born in January completed one month) and ended in January 1994 (when children born in December 1993 completed one month). The three-month visits were carried out between April 1993 and March 1994. Follow-up rates were 99.1% and 98.3%, for one and three months, respectively. Standardized questionnaires were applied to the mother, with emphasis on morbidity, feeding patterns, health services and medicines utilization. Children were weighted and their length, head and abdominal circumferences were measured (Table 2). These visits were funded by the World Health Organization for assessing breastfeeding patterns, early growth and related factors. Because of funding limitations, it was only possible to follow up a smaller number of children at these ages than in the other visits.
A different sampling approach was used in the six and 12-month visits, for which larger funds were available. To better assess the consequences of low birthweight, all 510 children with this condition were visited, as well as a 20% systematic sample of the remaining children. The 655 children eligible for the one and three-month visits were included in the sample, so that these children were examined five times – at birth and at one, three, six and 12 months. The different sampling fractions for low birthweight and other children required weighted analyses of these datasets. The questionnaires used in the previous follow-ups were adapted and information on the same variables was collected (Table 2). In addition, the Denver II motor development scale5 was applied to all children. Response rates were 96.8% at six and 93.4% at 12 months. There were very small differences in follow-up rates according to family income and to birthweight;13 at least 92.2% of the children in any category of these variables were traced.
In 1997-8, we attempted to locate the 1,363 children examined at 12 months; 1,273 (93.4%) were identified, representing a cumulative response rate of 87.2%. Out of the 90 children who were not interviewed, 61 had moved out of the city. Other reasons for non-response were incorrect addresses (18 children) and refusals (five children).
During the home visit, mothers were interviewed about socioeconomic and demographic variables, environmental characteristics; health services utilization and morbidity; nutrition and psychological development; and child care. The children were weighed with portable digital scales and measured with locally made stadiometers (Table 2). Standardization sessions for anthropometric measurements were carried out prior to and during data collection. In addition, quality control included a short questionnaire repeated for 5% of the interviews. This phase of the study was funded by the Brazilian Program for Centers of Excellence in Research (PRONEX).
The experience gained with the 1982 cohort suggested the whole cohort, rather than subsamples, should be visited. Use of subsamples meant that full data were only available for a very small subgroup of the whole cohort, those seen in every visit. Also, for prospective studies of rare events such as morbidity or mortality, it is essential to have information on the whole cohort at key points in time. Therefore, unlike all previous visits to the cohort, the 2004-5 visit was aimed at locating all children born in 1993.
Our previous experience, particularly the difficulty of tracing subjects based on old addresses (due to high mobility), suggested a multi-pronged strategy for locating the adolescents:
As in the previous phases of the study, field supervisors repeated 10% of the interviews for quality control purposes. Half of the families who had a telephone (72% of those interviewed) were contacted to confirm that the interview had taken place, and any criticisms on the questionnaire or on the interviewer were recorded.
The 2004-5 questionnaire was divided into four blocks. The first collected detailed contact information, including at least three different addresses of the family and relatives, phone numbers and place of work. In the second block, mothers were asked about family characteristics, maternal and paternal variables, and morbidity history of the adolescent. The third block was an interviewer-applied questionnaire to the adolescent, including diet, physical activity and other variables. Confidential information – including smoking, alcohol and drug use – was included in a fourth questionnaire that was self-applied and identified only by the adolescent’s serial number. Table 2 provides a list of the main topics of the questionnaire.*
In this visit, physical examination comprised measurements of weight, height, subscapular and triceps skinfolds, blood pressure and pulse (Table 2). Interviewers were standardized prior to the start of the field work, and standardization sessions were repeated every two months. The measurement equipment was calibrated weekly.
Several sub-studies including the cohort members were also carried out. Table 3 shows these studies, with respective years, target ages, sampling strategies, main objectives, and number of individuals included. As described in a previous publication, infant mortality was actively monitored by weekly visits to all hospitals, civil registration offices and the local health authorities.9 Deaths continued to be monitored after the first year, except for the hospital visits.
Four sub-studies (not shown in Table 3) are currently underway or will be started shortly. These include:
The complete 1993 cohort database now includes over 2,500 variables. A data dictionary is available in two versions. In the first, variables are organized according to when they were collected, that is, by visit. In the second, they are grouped by categories (e.g. socioeconomic, anthropometric, and others). The initial letter of each variable name indicates the follow-up visit, e.g. A for the perinatal survey, B for the 1-month visit, C for the 3-month visit) and so on. As far as possible, each indicator has the same variable name in the different follow-ups, differing only regarding the first letter. The main database is available both in Stata and SPSS formats. All questionnaires and interviewer guides used in previous follow-up visits are also available in electronic and paper formats.
Of the 5,249 live born children, 141 (2.7%) are known to have died; other children may also have died after migrating out of the city, but no information on these deaths is available. In 2004-5, 4,482 adolescents were contacted. Among these, there were 30 refusals (0.7%) and 4,452 adolescents were interviewed. By adding those who are known to have died to those interviewed, the follow-up rate was 87.5%.
An unexpected finding was that the door-to-door visits to all households in the city, leading to the identification of 3,254 cohort subjects, failed to locate 27.4% of the 4,482 adolescents who were eventually traced.
Table 4 shows follow up rates in the 2004-5 visit according to key baseline characteristics. No differences were observed according to sex or birthweight. Follow-up rates tended to be higher for adolescents belonging to low and middle-income families, and those born to mothers with intermediate levels of education. At least 79.9% of all children in each subgroup were traced.
Prospective birth cohort studies provide evidence on life-course determinants that are difficult, if not impossible, to investigate through other study designs. In the oldest available prospective birth cohort, the 1946 British Births Study, subjects are now approaching their seventh decade of life.16 Several new birth cohorts have been launched around the start of the new millennium, including one in Britain12 and another in the USA.* Three birth cohorts are currently being followed up in Pelotas: the 1982 and 1993 cohorts, and a new one launched in 2004.2 In addition to providing evidence on the life-course approach, the existence of three cohorts allows the investigation of secular trends by comparing children of the same age in the three cohorts.
These cohorts have addressed not only the study of biological determinants of adult disease, but also emphasized the study of social and cultural factors. The broad social inequalities that characterize our country allow the demonstration of the long term effect of early-life poverty.
Several lessons were learned in the recent phases of the study. The overall follow up rate was quite high and almost nine out of 10 cohort members were traced. However, the fact that the door-to-door visits failed to identify about one quarter of the adolescents who were eventually located suggests that this strategy alone is insufficient for the future phases of the study. In the past, we had successfully used this approach to locate a high proportion of the 1982 cohort. It is possible that increased concerns with security are responsible for many families failing to provide information to the interviewers.
A multidisciplinary study team is essential for birth cohort studies. Epidemiologists, clinicians, nutritionists, anthropologists, psychologists, biologists, physical educators, dentists and statisticians have played essential roles in the study. In the future, further inputs will be sought from economists, sociologists and educators.
A critical issue affecting long-term cohort studies is “funder fatigue”, namely the fact that no single funding agency is likely to provide long-term support to a single study, regardless of how successful it may be. Over the last 12 years, five different agencies provided financial support.
The next visit to the cohort is planned for 2008, when the adolescents will be 15 years old. In addition to collecting information on variables measured in the 2004-5 visit, special emphasis will be given to sexual and reproductive factors. Because the 1982 cohort members were visited at the age of 15 years, this visit will allow a direct comparison between the two cohorts at the same age. DNA samples will be collected in that visit, to create a bank for future studies of chronic diseases. We look forward to continuing this study for many years to come.
The authors are particularly thankful to the colleagues who participated in earlier phases of the cohort study, including Ricardo Halpern of Universidade Luterana do Brasil; Elisabeth Weiderpass of Karolinska Institute/Stokolm; Bernardo Horta, Silvia Fonseca and Moema Chatkin of Universidade Federal de Pelotas/Brazil; Juraci Cesar of Fundação Universidade de Rio Grande/Brazil; Elaine Albernaz, Elisa Hallal and Rafael Carmo of Universidade Católica de Pelotas/Brazil; Marco Aurélio Peres of Universidade Federal de Santa Catarina/Brazil; Karen Peres of Universidade do Oeste de Santa Catarina/Brazil; Jonathan Wells of Insitute of Child Health/England.
Supported by the Wellcome Trust initiative entitled Major Awards for Latin America on Health Consequences of Population Change, European Union (Grant n. GR072403MF), the National Program for Centers of Excellence - PRONEX (Brazil), Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq - Grant n. 66.1279/1996-0), and the Ministry of Health (Brazil) (MS-FNS - Grant n. 2799/2003).
*Full version in Portuguese is available from http://www.epidemio-ufpel.org.br/_projetos_de_pesquisas/coorte1993/ [2005 out 31]
*National Center for Education Statistics (NCES). Early Chidhood Longitudinal Study, Birth Cohort (ECLS-B). Available from http://nces.ed.gov/pubs2005/children/ [31/10/2005]