This is the old United Nations University website. Visit the new site at http://unu.edu

**Secular trends in adult and
child anthropometry in four Guatemalan villages**

**Marie T. Ruel, Juan Rivera, Hilda Castro,
Jean-Pierre Habicht, and Reynaldo Martorell**

*Secular trends in the stature and head circumference of
adults born between 1905 and 1959 and in the length of
three-year-old children born between 1965 and 1985 from four
villages of eastern Guatemala are analysed. Data were collected
before (1968), during (1969-1977), and after (1988) a
longitudinal protein-energy supplementation trial conducted in
the four villages. No secular trends are observed in the
age-adjusted height or the head circumference of the adults
studied; similar results are obtained whether a longitudinal or a
cross-sectional method is used to correct height for the effects
of ageing. A positive and significant linear trend is observed in
the length of the three-year-olds. The estimated increase of
2.5-3.3 cm seen over this 20-year period represents only
approximately 27% of the total deficit in length seen in these
children in 1968 (11 cm).*

A secular trend of increasing stature in adults has been observed in developed countries over the last century [1, 2]. This increase in height has been attributed to general improvements in socio-economic conditions, environmental sanitation, and health care associated with industrialization which have led to improved nutrition and decreased morbidity. Very little is known, however, about trends in the stature of adults in the developing world, where poverty and insalubrious conditions remain highly prevalent. Studies in poor populations of Colombia [3], Mexico [4, 5], and Brazil [6] have documented the absence of a secular trend in adults' height. As noted by the authors, this result could be expected since these populations have experienced little change in sanitary facilities, dietary patterns, and the availability of health care over the last century [4].

The present study examines whether there is a secular trend in the height and head circumference of adults from four rural villages of Guatemala born between 1905 and 1959. The availability of anthropometric data on the same individuals measured on two occasions, with an interval of approximately 14 years, allowed us to estimate the height loss associated with ageing in this population and to correct for it in the analysis of secular trends. Various methods and correcting factors have been proposed to adjust for this well-documented phenomenon [3, 7-10], but the general consensus is that population-specific correction factors should be used when possible because of large variability between populations both in the importance of the shrinkage and in its timing. In this study, therefore, we used a longitudinal method to partition the effects of age-associated height reduction from those of secular trends. We also compared this method with a commonly used cross-sectional method based on the association between stature and subischial height [3]. Finally, we assessed the existence of a secular trend in the length of three-year-old children born between 1965 and 1985.

The data were collected by the Institute of Nutrition of Central America and Panama (INCAP) before, during, and after a longitudinal supplementation trial conducted in the four villages. During the intervention, two villages were randomly assigned to receive a high-energy, high-protein drink (atole) and the other two were assigned to a low-energy, non-protein drink (fresco). The supplements were made available at a central location for all pregnant and lactating women and all children younger than seven years. Health services were provided in all four villages throughout the supplementation period (1969-1977). In 1988-1989 INCAP carried out a follow-up study of the participants.

Two cross-sectional surveys were also conducted in the same villages. One took place in 1968, before the intervention trial, and the second in 1988-1989 (referred to, for simplicity, as the 1988 survey), 11 years after the intervention was terminated. Data from the two surveys and from the longitudinal trial were used for the present analysis. Only relevant information about these data sets and about the analytical methodology used are presented here. The reader is referred to other publications for a detailed description of the design of the original study and data-collection procedures [11, 12].

**Secular trends in adults **

**Height**

Two methods were used to assess the existence of a secular trend in the height of adults (specifically, parents of participants of the longitudinal study, 19691977) born between 1905 and 1959, controlling for the effect of ageing (age-associated shrinkage). The first method took advantage of the availability of two measurements at approximately a 14-year interval on 498 adults, the first of which was taken during the supplementation trial (in 1974 1 year) and the second during the 1988 survey. The second method, commonly used in cross-sectional studies, is based on the high correlation between subischial height (the difference between stature and sitting height) and stature [3]. The rationale behind this method is that the length of certain long bones is highly correlated with stature and is influenced by secular trends but not by age-associated shrinkage. For instance, stature and sitting height are assumed to reflect the effects of both ageing and secular trend, whereas subischial height reflects only the effects of secular trends [13]. This second method was used to compare the results obtained using longitudinal information with those derived from a cross-sectional data set. The availability of longitudinal information on the sitting height of women also allowed us to test whether the assumption of stability in subischial height throughout ageing was verified in this sample.

The longitudinal method involved four steps. First, the difference in height per decade was calculated for each subject using the following formula:

(1)

Second, the mean change in height (cm/decade) was computed for each 10-year age group, based on the age of the subjects in 1988. Third, the mean changes per decade were used cumulatively to adjust each adult's height for the effects of ageing. For instance, if an adult was 51 years of age in 1988, the height was adjusted for the measured mean change per decade observed in this sample for adults between 30 and 40 years plus the mean change per decade observed between 40 and 50 years. (It was assumed that height reduction did not occur before age 30 [3].) Finally, age-adjusted heights were regressed on the year of birth, to assess the existence of a secular trend. All analyses were done separately for each sex. (Data for subjects born before 1930 are presented for men but not for women, because only parents of preschool children were measured in 19691977 and few mothers were born before 1930 whereas the age range for fathers was much broader.)

The 1988 data were used for the cross-sectional method, which consists of computing the partial regression coefficient of stature on age, controlling for subischial length [3]. The equation is as follows:

height = b

_{0}+ b_{1}(age) + b_{2}(subischial length) (2)

The coefficient for age (b1) is interpreted as being the rate of shrinkage per year (ageing effect), and is used in the following step to compute the age-adjusted height for each individual, using the following equation:

age-adjusted height = observed height + b

_{1}(age - 30) (3)

where b_{1} is the
coefficient for age obtained in equation 2. Since shrinkage is
assumed not to occur before the age of 30 [3], the age used in
the equation is the difference between the actual age and 30
years. As in the case of the previous method, the age-adjusted
height was regressed against the year of birth to test for the
secular trend effect.

**Head circumference**

Measurements of head circumference from the 1988 survey were regressed on the year of birth, by sex, to assess the existence of a secular trend in adults born between 1905 and 1959.

**Secular trend in the length of three-year-olds**

Information from the three data sets was used for this analysis. The two cross-sectional surveys were used to estimate length at three years of age in 1969 and 1988 respectively, using regression analysis. Longitudinal data collected throughout the supplementation period (1969-1977) were used to compute the mean length of three-year-olds (7 days) for each two-year period (from 1969 to 1976) and for 1977.

The method used to estimate length at three years of age from the cross-sectional data consisted of regressing the length of all children 9-60 months old against age and age squared, and using the regression coefficients obtained to estimate length at 36 months. Separate regressions were done for each year (1968 and 1988) and treatment group (atole and fresco). The interaction between sex and treatment was not statistically significant and therefore the analyses were done with both sexes combined.

The analysis of a secular trend in the children was complicated by two aspects related to the nature of the data available. First, because of the use of both estimated and observed length values, the assumption of equal variance could not be made. Second, because of unequal spacing between measurements, coefficients commonly used to estimate linear and quadratic trends could not be used [14].

The first problem was addressed by calculating separate estimates of variance for the longitudinal and cross-sectional data. For the longitudinal data, analysis of variance was used to estimate the variance of a model that included treatment (fresco, atole) and cohort (data collection years), and the two-way interaction term between these variables. The variance estimate obtained was 15.50. For the cross-sectional data, separate regressions of length on age were done for each treatment and data-collection period (1968, 1988). The variance estimates obtained ranged between 13.97 (fresco, 1968) and 19.78 (atole, 1988).

The problem of unequal spacing between measurements was addressed by using the method proposed by Robson [15] to construct special polynomials for unequally spaced data. The formulae used for the estimation of contrasts (L) and their variance (V) are presented in the appendix (p. 252), which also shows an example of the methodology used. Two sets of confidence intervals were calculated for all contrasts, one using the smallest variance obtained (13.97) and one using the largest one (19.78). Since the results were similar in terms of the statistical significance of the various contrasts, only those using the largest variance estimate are presented. All analyses were done using the micro-computer version of SAS, release 6.03. Probability values smaller than or equal to .05 were considered statistically significant.