Materials and methods

Design of the study

The study villages. Four villages were included in the study; at baseline, two had populations of ~ 900 each and two were smaller, ~ 500 people each. Within each pair, assignment to Atole or Fresco supplementation was random (Martorell et al. 1995). The villages are located in the Department of El Progreso, a dry mountainous area northeast of Guatemala City. The temperature ranges from 14 to 38ºC, with the rainy season occurring from June to October.

Throughout the course of the studies, from 1969 to 1989, the primary source of income for most villagers was agriculture. The majority of the children's fathers made their living as wage laborers, tenant farmers or small land owners. No one in any of the villages reported being a large landholder. Few adult males reported being skilled tradesmen or merchants. Women infrequently reported having an occupation outside the home. Adult literacy rates continue to be low but have improved over time. Literacy among mothers increased from 30 to 60%, whereas, among fathers, rates increased from 46 to 67% from 1969 to 1989.

Subjects. At the time of the initial study in 1969 potential subjects were all children <7 y old living in the villages (i.e., all those born since 1962). In addition, all children who moved into the villages during the course of the project and all children born in the villages from January 1,1969 through February 28, 1977 were also potential study participants.

Subjects born between 1962 and 1965 were excluded from the behavioral assessments to lower costs and to focus on the children exposed to supplement at younger ages. The analysis reported here focuses on those subjects exposed to the nutrition treatment during the prenatal period and the first 2 y of life. This is a period of accelerated brain growth and is perhaps the time when development of the brain is most sensitive to the effects of nutrition. The critical period of exposure to supplementation is expanded to 3 y of age in analyses of physical growth data (Haas et al. 1995, Rivera et al. 1995) because of evidence of continued but declining effects of supplement intake on growth rates between 24 and 36 mo of age (Schroeder et al. 1995). All subjects selected in this paper were born between 1970 and 1974 and ranged in age from 13 to 19 y at the time of the follow-up study. The sample was comprised of 636 subjects, divided almost equally among Atole and Fresco villages.

Experimental intervention

The Atole supplement was a warm, thick, brown, sweet drink, similar to corn gruels given to children in rural Guatemala. It contained 11.5 g of protein and 163 kcal/682 kJ of energy per cup (180 mL). The Fresco supplement was a cool, clear, sweet drink like KoolAidä, also similar to common village drinks. It contained no protein but it had 59 kcal/247 kJ per cup, approximately one third the energy of the Atole. Both supplements were fortified with vitamins and minerals. At the time the longitudinal study began, it was assumed that the energy concentration of Fresco was insufficient to have a developmental effect. The supplements were available at a central feeding station 7 d/wk to every resident of the village twice daily (1000 and 1400 h). Ingestion was recorded only for target subjects, that is, pregnant and lactating women and children up to age seven. Because consumption was ad libitum, attendance at the feeding centers, after controlling for consumption, was used in the analyses to control for factors associated with supplement participation. Additional details are given in Martorell et al. |1995).

Socioeconomic indicators. Three socioeconomic variables are included in the analyses as potential confounders: house quality, maternal education and father's occupation. Data for each of these three variables were derived from census data obtained in 1987.

Because of the difficulty of obtaining direct measures of income in developing countries, a measure of house quality often is used as a proxy for social-environmental variables that affect cognitive growth and educational development (Johnston et al. 1987). Nine variables describing house quality were assessed: an overall rating of the type of house (1-4 scale), ownership of house (no = 0, yes = 1), number of rooms, type of floor (1-5), type of walls (1-7), type of roof (1-4), location of the kitchen (1 - 3), type of toilet (1 - 4) and number of household possessions (1-6). In all instances, higher scale scores reflected the higher quality of the dwelling. The measure used in these analyses was generated using factor analysis on the within-village standardized variables.

Maternal education consistently has been shown to be related positively to the cognitive development of the offspring (LeVine et al. 1991). In this case, informants reported both literacy (coded 0 = none, 1 = some) and the number of years of schooling completed successfully. The mean number of years of schooling for mothers was 2.1 y.

Occupational status is a carrier variable that may be associated with income, status in the community, availability of resources and family socialization practices. The indirect effects of parental occupation on the cognitive development of children are thought to occur through the earning capacity of the parent and the consequent resources for stimulation that it permits. Both mothers' and fathers' occupations were assessed; however, because only ~20% of women at follow-up reported having an occupation, mother's occupation was excluded from further analysis.

Data on 19 occupational categories were obtained and then subsequently collapsed into six categories for comparison with earlier census data from 1967; the original and recorded scales are highly correlated (r = 0.88). In preliminary analyses, the recorded scale demonstrated adequate linear properties and was used in all subsequent statistical calculations as an ordinal variable. A detailed description of the specific procedures used in the construction of these three variables and the related reliability and validity data have been reported elsewhere (Pollitt et al. 1993). The three variables were standardized within village and a composite score was constructed using the sum of the three scores.

Schooling variables. At the time of testing, all subjects had reached school age and most had received some schooling. Given the well-established relation between schooling and cognition (Ceci 1991), performance on psychoeducational tests was adjusted for schooling experience. This was particularly important for the assessment of treatment effects given that differences between Atole and Fresco villages on important schooling indicators (e.g., maternal and paternal education) favored Fresco villages before the intervention (Engle et al. 1992). Two schooling variables in the statistical analysis were the age the child started school and the maximum grade attained in primary school.

The psychological test battery

Two psychological test batteries were used in the follow-up study. One includes psychoeducational tests and the other includes information processing tasks. The psychoeducational test battery included tests of literacy, numerary, general knowledge, two standardized educational achievement tests and the Raven's Progressive Matrices. The achievement tests were part of the Interamerican Series used extensively in Guatemala by faculty from the Universidad del Valle in Guatemala City (for a detailed description of the tests see Pollitt et al. 1993). The purpose of this battery was to acquire a measure of general abilities, aptitudes and achievements that are influenced heavily by experience, education and cultural upbringing.

Information processing. Tests of simple, choice end memory reaction time (RT) (Sternberg 1966) comprised the computerized battery of tests to assess information processing. In addition, a paired associate test was administered as part of this battery. The intent of the battery was to assess the efficiency with which an individual processes information by focusing on speed of response in elementary cognitive tasks. In addition to measures of RT from three tests, two of the RT tests (i.e., choice and memory) also yielded a performance score (i.e., number of errors). In general, between-subject variability in RT tests is not accounted for by schooling and cultural background, yet test performance still maintains a low level correlation (-0.10 to -0.30) with g, a general ability factor. Theoreticians presently claim that RT is a sensitive indicator of differences in brain function (Eysenck 1986, Jensen 1991, Vernon 1987).

Procedure. Each of the four villages was visited twice by a research team, once during the dry season and once during the rainy season. The teams were rotated and each team visited each village during one round of testing. The presence of the team in the village varied from 3 to 9 wk depending on village size.

In each community, two staff members recruited subjects and made appointments for testing. All testing was done in community houses rented by the project and adapted accordingly. In addition to psychological data, subjects were given several examinations, including medical and anthropometric assessments. They also were interviewed extensively about sociodemographic characteristics.

Analytical strategy. The data were analyzed using a hierarchical regression model. This approach permits the estimation of variance accounted for by treatment alone as well as the identification of differential effects of treatment that may be related to particular characteristics of the population. By including interactive terms in the model, it is possible to identify the potential indirect pathways through which supplement could also have affected the outcomes of concern.

All independent variables were standardized. Individual characteristics (sex, age at testing and attendance at the feeding center with consumption par-tialled out) were entered first, followed by the socioeconomic status (SES) composite (sum of mother's education, father's occupation and house quality factor score), and then the two school indicators (age at school entry, maximum grade attained) and finally the treatment variable (entered as a categorical (1/-1)). In this way the percent of variance accounted for by the different predictors was estimated. In a subsequent step, two interaction terms were entered: treatment by SES and treatment by maximum grade. The results presented include the percent of variance accounted for at each step (R²), and F values and regression coefficients for each variable in the step in which it was entered, controlling for all other variables entered before this step.

A three-way interaction term (treatment by SES by grade attained) also was entered into the model but the results of these analyses were nonsignificant and will not be presented. Similarly, treatment by gender terms were entered in preliminary analyses but also were nonsignificant and dropped from the final model.