Contents - Previous - Next
This is the old United Nations University website. Visit the new site at http://unu.edu
Patterns and predictors of participation and consumption of supplement in an intervention study in rural Guatemala
Dirk G. Schroeder, Haley Kaplowitz, and Reynaldo Martorell
Children's participation and supplement intake patterns in an eight-year, community-based intervention trial in Guatemala are presented. The supplements were either high-energy, high-protein atole or low-energy, no-protein fresco.
Participation rates were between 65% and 85%, with few differences by village (N = 4) or child age (0-7 years). The percentage of days of attendance at the supplementation centre and the volume of supplement consumed, however, were significantly higher in villages that were given atole for children under four years old. After the age of four, children who received fresco consumed more volume. Supplement energy was significantly greater for children receiving atole at all ages. Proximity to the supplementation centre and larger family size were significant predictors of attendance for both supplements. Low socio-economic status was highly associated with increased attendance for children consuming atole, but not fresco.
Analyses using these data should consider the possible biases introduced by differential self-selection between supplement types, particularly those that were potentially confounded by attendance or socioeconomic status.
Adequate participation and consumption of the supplement are necessary if a nutrition intervention is to have impact . As opposed to controlled clinical trials in which consenting individuals are randomly assigned to treatment or control, nutrition intervention trials are often community-based, with participation voluntary. The participants are thus self-selected and often differ systematically from non-participants . Factors that influence participation include distance from the distribution centre, socio-economic status, composition of the supplement, sex of the target individuals, and family size, among others .
Different self-selection between treatment types may confound the association between intervention and impact . For example, consider a village-level comparison of two supplements, A and B. Suppose that the high consumers of supplement A tend to be individuals of low socio-economic status (SES), whereas those of supplement B are of high SES. Because SES may be strongly related to many outcomes of interest (e.g., growth), a simple group comparison would underestimate the potential effects of supplement A related to B. More appropriate analyses would examine the results stratified by SES.
Designs in which participants are compared with non-participants, or in which the amount of supplement is related to the outcomes, are particularly vulnerable to self-selection bias.
We describe and characterize children's participation and supplement-consumption patterns in a longitudinal intervention trial conducted in rural Guatemala. Specifically, descriptive analyses provide a general overview of patterns by supplement type and age, as well as temporal changes as the study progressed.
Original design and intervention
A prospective, community-based supplementation trial was conducted between January 1969 and August 1977 to test the hypothesis that improved nutrition in early childhood results in better growth and development. Randomization was at the level of the village and stratified by village size. Two villages (large, Conacaste; small, San Juan) received a high energy, high-protein gruel (atole) and two control villages (large, Santo Domingo; small, Espíritu Santo) received a low-energy, no-protein drink (fresco). The atole provided 163 kcal and 11.5 g of protein per cup (i.e., 180 ml) and the fresco 59 kcal per cup.
The supplements were distributed daily, free of charge, at supplementation centres established in each of the four villages. The subjects were served one cup of supplement on arrival, and were offered additional cups if they desired them. The number of cups provided each subject was recorded, and leftovers were measured carefully to the nearest 10 ml. This allowed for unusually precise measurement of individual supplement intake. Attendance was voluntary and open to all members of the community; an individual could visit the centre to receive the supplement in mid-morning, mid-afternoon, or both. Attendance was summarized in the data tapes as daily attendance (morning and/or afternoon). Data on attendance at the centre, supplement intakes, growth, health, diet, socio-economic status, and mental development were routinely collected on pregnant women and children under 7 years of age.
Measurement intervals and sample
Attendance and supplement consumption were summarized on the master tape for three-month intervals from birth to 24 months of age, six-month intervals from 24 to 48 months, and twelve-month intervals from 48 to 84 months. For multivariate analyses, data were further summarized into three age categories: birth-1 year, 1-3 years, and 3-7 years.
For the descriptive analyses, only children who met the age criteria and were permanent residents in one of the four villages over the entire measurement interval were included. An eligible non-participant was a child who was of the correct age and a resident in a village but who did not attend the supplementation centre during the interval. For the multivariate analyses, children who were eligible for at least 12 consecutive months during the interval were included.
Three indicators of programme participation and intake patterns were analysed: attendance at the supplementation centre, volume of supplement consumed, and energy from the supplement.
Attendance patterns are presented as both the percentage of children within each age interval who were non-participants and the percentage of days within a given measurement interval that participants attended. Participation was defined as attending the centre at least once during the interval. Because the measurement intervals varied in length for different ages, presenting attendance as percentages (rather than in number of days) facilitates comparison between intervals.
Volume is presented as the average millilitres of supplement consumed per day during the interval. Energy intakes are presented as mean kilocalories derived from the supplement per day. Even if non-participants are eliminated, distribution of these two variables is non-normal (right-tailed): the majority of children took in moderate amounts of the supplement, whereas a small percentage consumed significantly more than their peers. Descriptive analyses were also done using transformed (square root) variables to normalize their distributions; however, statistically significant differences and interpretation of the results were identical for the non-transformed measure.
To investigate possible temporal changes over the duration of the study, analyses were also done by year. Patterns for children 1-3 years old are presented as representative examples.
Descriptive analyses were done stratifying by sex, but no significant differences were identified. Sex as a predictor of attendance and supplement intake is addressed in the multivariate analyses.
Multivariate analyses were executed with the objective of modelling the predictors of attendance at the centre and energy derived from the supplement. Because it was presumed that the factors responsible for attendance and consumption levels would be very different between the two supplements, separate models were generated by supplement type. In addition, because attendance and consumption levels were measured repeatedly on the same children at various ages, it was necessary to summarize data over the three age categories. These categories were selected on the basis of theoretical considerations (e.g., that mothers would accompany infants, children of 1-3 years would be coming with mothers or siblings, and those of 3-7 years would come to the centre on their own) as well as patterns observed in the descriptive analyses. Separate multivariable models were run for each category.
Six possible predictors of attendance and intake of supplement were considered in the multivariate models (table 1). The distance to the centre from the child's home, originally recorded as 1 to 5 based on the walking time, was recoded 1 (closest) to 3 (farthest). The SES score was derived by principal components analysis on the basis of housing-quality and possession variables; it has been standardized (e.g., mean = O. SD = 1). The year of birth (coded 1962 = 1, 1963 = 2, etc.) was included to control for possible temporal changes over the study's duration. The variables for sex, village, and family size are self-explanatory.
TABLE 1. Definitions of the variables used in the multivariate analyses
|Attendance||Percentage of days child attended supplementation centre during each age interval, birth-1 yr, 1-3 yrs, 3-7 yrs|
|Energy from supplement||Average daily calorie intake from supplement during each age interval (to normalize distribution, the square root of this value was used)|
|Distance to centre||1 = closest, 2 = medium, 3 = farthest|
|Family size||Number of people in family (range 1-12)|
|SES||Composite score for socio-economic status based on characteristics of home and possessions variables|
|Sex||1 = male, 2 = female|
0 = Conacaste, 1 = San Juan
Fresco: 0 = Santo Domingo, 1 = Espíritu Santo
|Birth year||1962 = 1, 1963 = 2, etc. (range 1-15)|
Tobit methods were used to model the predictors of attendance. This approach is appropriate for distributions that contain a mass of discrete observations as well as a continuous range of values [7, 8]. In this case, we wanted to model the predictors of non-attendance (also referred to as "left-censored" values) together with the normally distributed percentage of days attended. (The Tobit analyses were done using the SAS LIFEREG procedure and NORMAL distribution option [SAS, 1990].) An important assumption with this method is that factors that caused children to be eligible non-participants are the same as those that caused children to attend the centre infrequently. An alternative approach was to subdivide these analyses into two parts: attenders versus non-attenders, using probit or logistic regression methods, and levels of attendance, using parametric methods after eliminating non-attenders . A comparison of these approaches provided very similar results; the Tobit models are presented here, as they have the advantage of summarizing the information more concisely.
To model the predictors of energy derived from the supplement, the dependent variable was transformed (square root) to normalize its distribution. In addition, only participants are included in the sample. Attendance at the centre was entered as a covariate to identify predictors of consumption controlling for attendance patterns. In other words, once a child got to the clinic, what were the factors that led him or her to consume more or less supplement?
Interactions were tested in the multivariate analyses. Probability values of less than .05 are referred to as statistically significant. (All analyses were conducted using the PC-SAS statistical package, version 6.04.)
Contents - Previous - Next