Developing a cooperative multicenter study in Latin America: Lessons learned from the Latin American Study of Nutrition and Health Project

ABSTRACT This report examines the challenges of conducting a multicenter, cross-sectional study of countries with diverse cultures, and shares the lessons learned. The Latin American Study of Nutrition and Health (ELANS) was used as a feasibility study involving the most populous cities of eight countries in Latin America (Argentina, Brazil, Chile, Colombia, Costa Rica, Ecuador, Peru, and Venezuela) in 2014–2015, about 40% of the population of the Americas. The target sample included 9 000 individuals, 15–65 years of age, and was stratified by geographic location (only urban areas), gender, age, and socioeconomic status. Six principal challenges were identified: team structuring and site selections; developing a single protocol; obtaining ethic approvals; completing simultaneous fieldwork; ensuring data quality; and extracting data and maintaining consistency across databases. Lessons learned show that harmonization, pilot study, uniformity of procedures, high data quality control, and communication and collaboration across sites are imperative. Barriers included organizational complexity, recruitment of collaborators and research staff, institutional cooperation, development of infrastructure, and identification of resources. Consensus on uniform measures and outcomes and data collection methodology, as well as a plan for data management and analysis, communication, publication, and dissemination of study results should be in place prior to beginning fieldwork. While challenging, such studies offer great potential for building a scientific base for studies on nutrition, physical activity, and other health topics, while facilitating comparisons among countries.


Special report Fisberg et al. • Lessons from a multicenter study in Latin America
Growing interest in communitybased public health and policy interventions that reduce obesity and improve nutri tion has prompted multicenter studies to understand populationbased behaviors and outcomes. In contrast to multisite studies whose main purpose is to obtain a larger sample, multicenter studies are designed jointly by Principal Investiga tors (PIs) at all sites. This means that all are involved in planning the study proto col and procedures, are scientifically re sponsible for the study results, and participate actively in manuscripts and other dissemination activities. These studies have been increasingly valued due to large sample sizes, ability to ex plore differences across sites, and the in creased generalization of results (1). Multicenter collaborations also allow re cruitment of more diverse populations, within a much shorter time frame. While challenging, such studies offer great po tential for building a scientific base for the study of obesity and for planning health policies and intervention pro grams. However, such studies require heightened attention to detail, simultane ity, comprehensive planning, and collab oration with colleagues (2).
Several large, multicenter, observational studies have been conducted to investigate the nutritional and physical activity status of various populations (3 -6). The major ity of these studies have been performed separately by each country and unified later. In most cases, differing methodolo gies were used to assess food consumption and physical activity, or samples of each country were not representative of specific populations. In Latin America, few studies have been conducted that represent the re ality of each country and region.
The Latin American Study of Nutrition and Health (ELANS), however, is a crosssectional, multicenter study, that was conducted simultaneously in the ur ban populations of the most populous cities of eight countries in Latin America. This paper highlights the principle les sons learned from the ELANS, including the key decisions, challenges and barri ers, and logistical strengths and limita tions encountered during the design process, data collection, and data entry.

SURVEY METHODOLOGY
ELANS is a householdbased, multi national, crosssectional survey that aims to describe the nutritional status of peo ple in Latin America and to investigate food and nutrient intake, as well as phys ical activity levels among representative samples of urban populations. The total population of the eight study countries-Argentina, Brazil, Chile, Colombia, Costa Rica, Ecuador, Peru, and Venezuelarepresents about 40% of the population of the Americas. The target sample in cluded 9 000 individuals, 15 -65 years of age, and was stratified by geographic lo cation (only urban areas), gender, age, and socioeconomic status. The rationale and design of the study are reported in more detail elsewhere (7). In brief, the ELANS protocol was designed to collect data at the individual level using ques tionnaires (sociodemographic, dietary intake, and physical activity) and objec tive measurements (accelerometry and anthropometry).

CHALLENGE 1: STRUCTURING A TEAM AND SELECTING SITES
Selected ELANS countries differed in demographic and some socioeconomic indicators, reflecting the differences present in Latin America (Table 1).
Key factors in the ELANS project im plementation were the selection of team members and the inclusion of multidisci plinary professional, technical, and sci entific advisors with complementary expertise in epidemiology, energy bal ance, physical activity, and statistics. PIs were selected based on their ability to conduct epidemiological studies (in vestigator fieldwork experience, research background, and academic support structure for research) and diversity of geographic location in Latin America.
All of the academic/research teams were composed of researchers, under graduates, and postgraduate health sci ence students based in universities or in association with universities, with the exception of one based at a research institute ( Table 2). The organizational structure had three branches: academic team, technical support team (physical activity and food intake assessment, an thropometrics), and operational team ( Figure 1). The organizational structure of the ELANS was defined during the first design meetings among the re searchers, including the scope of the study, the complexity of communica tions, and what agreements would have to be reached. The study management and quality control strategies are de scribed elsewhere (7).
It is important to highlight that project coordination was centralized in an inter national team composed of two chair persons, a cochair, and two international project managers. This centralized ap proach ensured that any barriers that arose would be resolved rapidly, that communication among all teams would be effective, and that research uniformity would be maintained. By bringing to gether expert researchers in the field of physical activity and statistics, the

Lessons learned:
(a) A collaborative, multidisciplinary team of experts with extensive scientific experience and effective communication tools is essential.
(b) Supervision by an international team is crucial. The appointment of an experienced and proactive coordinating investigator to mentor and counsel other sites ensures consistency and support where it is needed.
(c) Designating/assigning an interna tional project manager who provides overarching communication and can ensure issues are quickly detected and measures are continuously improved.

CHALLENGE 2. DEVELOPING AND IMPLEMENTING A SINGLE STANDARDIZED PROTOCOL
At the start, when the decision to col lect common data across the sites was made, it became apparent that a single protocol and a uniform set of data    collection tools would be required. The process of creating a single protocol in volved 15 months of collaborative work with in person meetings and conference calls with the PIs. All decisions were guided by scientific evidence and field work logistics. The advice from experi enced external advisors with differing expertise was important.
Considering the large number of sites, eight in this case, a tools standardization process was developed to ensure equal standards for data collection (8) at every site. The choice of dietary method used in the ELANS was based on existing sur vey studies, such as NHANES (9) and HELENA (3), as well as the experience of some of the PIs. Repeated 24hour re calls (24hr) are considered useful tools in providing national and grouplevel esti mates of usual dietary intakes of individ uals, as well as in describing usual intake distributions of populations, using ap propriate statistical approaches and con trolling for intraindividual variability.
The standardization of procedures for 24hr administration and analysis was necessary to ensure equivalence of the dietary outcomes across participating centers. The Multiple Pass Method (10) was unanimously chosen. All PIs agreed to use the Nutrition Data System for Re search software (Minnesota University, Minneapolis, Minnesota, United States; NDSR) to allow ELANS intercountry comparisons. An extensive process of harmonization was initiated, as de scribed elsewhere (11).
The International Physical Activity Questionnaire has been validated in countries of Latin America; however, the Mexican (Spanish) version, (12) adapted by the International Study of Physical Ac tivity and Built Environments was se lected for use after cultural adaptations for wording and examples (13,14).
The objective measure of physical ac tivity was a crucial aspect of the ELANS that provided accurate estimates of phys ical activity and energy expenditure in this Latin American population. Due to logistic and financial matters, efforts were made to ensure that a range of 25% -40% of each sample would wear the accelerometer for 7 days.
A major concern was that all sites should use the same or equallyreliable equipment validated by previous studies.
It is noteworthy that the PIs faced many problems with the purchase of imported equipment (e.g., accelerometers, scales, stadiometers). Due to customs and mail delivery problems, all sites had to delay the start of the fieldwork.
The socioeconomic questions used in ELANS were designed based on ques tionnaires used by each national statis tics office, or otherwise used most frequently by each country (15 -20). Due to differing variables used across countries to determine Socioeconomic Levels (SEL), a rule was developed to equate the different classification sys tems of the eight countries (Table 3). Based on this, three levels of classifica tion were established for SEL and in cluded equivalent characteristics for all countries: high, medium, and low. The same procedure was performed to estab lish equalization across levels of educa tional in the eight countries, and also resulted in three classifications: primary to incomplete secondary schooling; complete secondary to incomplete higher/tertiary education (technical/ university); and complete higher educa tion (technical/university).  Lessons learned: (a) Input from local teams is essential during protocol development and plan ning of field work to ensure achievable goals.
(b) Developing a feasible and realistic protocol may be more difficult than first expected. Standardization and harmoni zation of foods in software require exten sive and innovative work. The use of accelerometers requires caution, as com pliance is difficult to achieve. Measure ment and definitions of socioeconomic status and educational level across coun tries require close attention to ensure comparability.
(c) Equipment availability and pro curement in different countries requires prior planning and ample time. Possible delivery delays and customs tariffs should be included in the schedule and budget.

CHALLENGE 3. OBTAINING ETHICS APPROVALS
The process for ethical reviews and timeframes for obtaining approvals vary greatly among countries. To ad dress this challenge and to reduce the burden of conducting multiple local ethics reviews, an external institutional review board (IRB) was selected to per form a centralized review of the proto col, the informed consent template, and other study documents. In this case, Western IRB (WIRB, Puyallup, Washing ton, United States) was chosen as an in dependent international organization. However, to obtain WIRB approval, each site also had to submit the protocol and informed consent through a local IRB. After approval of local IRBs, ELANS received WIRB approval.

Lesson learned:
(a) Obtaining ethics approvals for a multicenter study must be carefully planned and, preferably, centralized, with extra time allotted for reviews at multiple levels.

CHALLENGE 4. ACCOMPLISHING SIMULTAENOUS FIELDWORK
During the common protocol design, the complexity of the logistics-a large number of interviewers performing field data collection simultaneously at differ ing sites-it became apparent that a stan dardized and uniform application of questionnaires and measurements was needed. These led to the decision to use a Contract Research Organization (CRO) with offices in all eight countries in volved in ELANS. This CRO was respon sible for all participant sampling and recruitment and data collection and en try, except for dietary recalls, which were managed by each local institution. All procedures were continuously moni tored by the international team, and a weekly meeting of PIs was held through out the fieldwork.
Frequent contact with field supervi sors was necessary to ensure quality data collection. At the start, there were many mistakes due to the interviewers' lack of experience; later, there seemed to be less attention to detail. To accomplish the data collection on time, local interview ers were hired and trained to work in different parts of each country simulta neously. Interviewers with previous health research experience produced better quality interviews. Faster field work was experienced in some countries, such as Chile, where interviewers were health science students.
Standardized training and operating manuals were designed for use in all countries; however, revisions by the international team were made on an asneeded basis to ensure equivalent measures, described elsewhere (7). Three operating manuals provided wellstructured guidance for interview ers on anthropometric measurements, dietary intake recalls, and accelerometer use and preparation. A fourth operating manual was designed to ensure proper use of the NDSR software by researchers working with the dietary data. The inter viewers' work was continuously super vised during the data collection period.
To ensure functionality of data collec tion procedures, a pilot study was per formed before fieldwork began. It tested the tools and accompanying procedures at all the sites and identified performance differences across them (8). One of the main difficulties during the pilot study was a refusal to participate, due either to time constraints, lack of interest, or fear of strangers. To overcome this, sev eral strategies were used: leaflets on the ELANS project were distributed in the neighborhoods prior to selection of house holds; interviewers showed an official identification badge to prospective partic ipants and wore an identifying apron; and during the first visit, an informative letter, including the researchers' contact infor mation, was provided to each participant. Other issues identified during the pilot study were discussed and corrected, as needed.

Lessons learned:
(a) A pilot study is important to ensur ing feasibility, efficiency, and adherence to protocols and procedures. Although it delayed the study start, the pilot proved to be important in preventing measure ment errors and improving participant compliance.
(b) Thorough training is fundamental for interviewers and ample time should be allotted for this purpose. Interviewers with experience or training in health sci ences are preferable.
(c) Continuous supervision of inter viewers' tasks is critical for maintaining the data quality of the fieldwork.
(d) A good relationship and open communication between the academic/ research team and the CRO are key to conducting efficient supervision and monitoring procedures.

CHALLENGE 5. ENSURING DATA QUALITY
Data reliability and credibility are es sential to the success of ELANS. Threats to data quality were identified during the protocol design process, the pilot study, and the first weeks of the field work. Thus, the coordinating investiga tors/chairs proposed adopting procedures in all sites to ensure highquality data and reliable information. These included preparatory meetings; detailed operat ing manuals; site visits; technical visits to participating centers; interviewer train ing; close monitoring of data collection and data entry; retraining of interview ers when needed; concurrent query man agement and fieldwork supervision; partial database generation; inconsis tency checks of anthropometric, physical activity, and food intake data; and when possible, a return visit to the participant's household to correct unclear, incomplete, or questionable responses.
Monthly, each center sent informa tion on the status of interviews, entry of 24hr food records in NDSR, and the accelerometer files, plus reported any

Lessons learned:
(a) Maintaining regular, studywide communication among PIs, the CRO, and the accelerometer, nutritional, and statis tical centers is critical to data quality.

CHALLENGE 6. EXTRACTING DATA AND DATABASE CONSISTENCY
In all countries, except Ecuador and Venezuela, the CRO had computer tab lets available for data collection. When the electronic form was created, internal consistency checks were already pro grammed into the software to alert the interviewers to outliers and other issues during realtime data entry.
Throughout the fieldwork, data qual itycontrol was carried out periodically.
The types of inconsistencies that were found led the team to develop a second phase of consistencychecking post data collection. Each academic/research site team was responsible for this second phase. A specified procedure was used to detect possible errors of data consis tency before the generation of each final database ( Figure 2). All data collected were simultaneously reviewed by the CRO, the accelerometer center and its Physical Activity consultant team (PA), and each academic/research site team. The PA consultants and each academic/  research team performed participant identification (ID)matching between their databases (accelerometer database and food intake database, respectively) and the CRO database. Additionally, each academic/research team con ducted anthropometric and dietary intake consistency checks. Upon suc cessful completion, a general consis tency check was then performed by each academic/research team. Thereaf ter, final versions of the datasets were generated by the PA consultants, CRO, and academic/research team. If any step was not successfully concluded during this process, it was restarted, as shown in Figure 2 by connectors 1, 2, and 3. If any participant needed to be excluded from the sample, the academic team informed the CRO, and the case was replaced with a matched casesame gender, age group, SEL, and geo graphic area-from the oversample. Common sources of errors included simple typographical errors; consis tency errors, such as birth date and age did not concur; differing participant ID and/or demographic information be tween first and second visit; insufficient accelerometer wear time (hours/day and/or number of days); programming or accelerometer malfunctions; or ab sence of PA logs. After identifying the source of each error, most could be resolved by communication between the centers. If the entry was unclear, missing, or otherwise suspicious, field staff were contacted for correction or verification at participating house holds. Accelerometer data errors were corrected in the office when possible; when not, the individuals were moved to the group without accelerometer use (60% of the sample). When this group was full, the rest were assigned to the oversample group. As a considerable number of cases were found to be in valid due to insufficient wear time, pro gramming error, or device malfunction, the final percentage was reduced to 25% of the sample.
Each site was responsible for verify ing the quality of the data registered in its 24hr food records, and as needed, field staff were contacted for correction or verification at participants' house holds. Several steps were developed for analysis of dietary intake data consis tent with NDSR. This was a difficult phase that required validation from several researchers. Technical support and guidance were required by all the research teams. Special consideration was given to the variations in fortified local foods and processed food prod ucts, and to correcting micronutrient contents of foods according to local food composition tables. All of the re searchers provided input for this pro cess and approved the final version of the resulting document.

Lessons learned:
(a) Extensive expertise in dietary as sessment among the researchers is im portant because this sensitive part of the data collection required the greatest number of corrections to inconsistencies.
(b) Data cleaning demands close atten tion to detail and collaboration of multi ple stakeholders.
(c) Correct use of accelerometers relies on thorough interviewer training so that they can, in turn, effectively instruct par ticipants. Timely delivery of data to the PA team by the CRO is needed for correc tions to be made. A 70% accelerometer compliance rate should be taken into ac count, with fewer valid versus measured cases in the final sample.
(d) A data analysis and publication plan that ensures appropriate data management and dissemination must be developed with input from all team members and be followed meticulously through the end of the study.

FINAL CONSIDERATIONS
The development of a multicenter household crosssectional survey of the nutritional and physical activity status of adolescents and adults in eight countries of Latin America was an innovative and fundamental step toward better under standing behaviors and their relation ship to health in the region. This survey contributed to several lessons learned regarding the organization of a multi center study. The feasibility of perform ing such a study depends on many considerations, including standardiza tion of data collection, maintenance of high data quality, and collaboration across sites. Among the greatest barriers to conducting studies such as this one are the inherent organizational complexity, recruitment of collaborators and research staff, institutional cooperation, develop ment of infrastructure, and identifica tion of resources. By implementing the various strategies detailed, this project overcame various challenges to ensure the integrity of its crosssite data collection.
The entire research team at each par ticipating center and the coordinating center-the PIs, local investigators, con sultants, coordinators, research assis tants, data managers, system analysts, statisticians, network managers, and accountants-proved to be essential to study development and performance. Ideas and solutions, as well as planning for analysis and interpretation of re sults, arose naturally when coming from a team thinking and working to gether. The institutions involved all benefited from the exchange of knowl edge among researchers, the equipment acquisitions, the use of internationally recognized methodology, and wider study dissemination.
Multicenter cooperative studies while challenging, offer great potential for building a scientific base for studies on nutrition and health. Considering the methods and experience of other multi center studies while developing the study design of the ELANS was impor tant for surmounting possible obstacles and allowing faster progress. To this end, the lessons learned during ELANS pro vide new perspectives for better plan ning of financial, staff, and technological resources for similar future studies.