Combined effect of short stature and socioeconomic status on body mass index and weight gain during reproductive age in Brazilian women

Short stature, a marker for undernutrition early in life, has been associated with obesity in Brazilian women, but not in men. We tested the hypothesis that weight gain during the reproductive years could explain this gender difference. A national two-stage household survey of mothers with one or more children under five years of age was conducted in Brazil in 1996. The subjects were women aged 20 to 45 years (N = 2297), with last delivery seven months or more prior to the interview. The regions of the country were divided into rural, North/ Northeast (urban underdeveloped) and South/Southeast/Midwest (urban developed). The dependent variables were current body mass index (BMI) measured, BMI prior to childbearing (reported), and BMI change. Socioeconomic variables included mother’s years of education and family purchasing power score. A secondary analysis was restricted to primiparous women. The prevalence of current overweight and overweight prior to childbearing (BMI ≥25 kg/m2) was higher among shorter women (<1.50 m) compared to normal stature women only in the urban developed region (P < 0.05). After adjustment for socioeconomic variables, age, parity, BMI prior to childbearing, and age at first birth, current BMI was 2.39 units higher (P = 0.008) for short stature women living in the urban developed area compared with short stature women living in the urban underdeveloped area. For both multiparous and primiparous women, BMI gain compared to the value prior to childbearing was significantly higher among short stature women living in the urban developed region (P ≤ 0.04). These results provide clear evidence that short stature was associated with a higher BMI and with an increased risk of weight gain/retention with pregnancy in the developed areas of Brazil, but not in the underdeveloped ones. Correspondence


Introduction
Undernutrition early in life has been associated with obesity among children and adults in some (1)(2)(3)(4)(5)(6), but not all, studies (7)(8)(9).Short stature, a marker of early poor nutrition, was associated with obesity and abdominal fatness among women, but not among men in two population-based surveys in Brazil (6,10).Reproductive factors are a plausible explanation for this gender difference in the association between short stature and obesity.Parity has been positively associated with weight gain and with the onset of obesity in population-based studies in developed countries (11)(12)(13)(14)(15)(16), and also in Brazil, as shown by a previous analysis of the III Brazilian Demographic and Health Survey (DHS-96) (17).Events during the fetal and early postnatal growth period by causing permanent metabolic alterations through adult life may facilitate maternal fat deposition during pregnancy.Thus, low birth weight has been linked to adult glucose intolerance (18,19), and to activation of the cortisol axis during adult life (20,21).These metabolic changes due to insulin or cortisol could drive a greater fat deposition among short stature pregnant women.Therefore, in a sample of parous women we determined whether weight gain during the reproductive years increases the prevalence of obesity/overweight in women of short stature, and whether this association varies in different regions of Brazil.By design, this survey did not include nulliparous women.In order to better differentiate the effects of childbearing from underlying secular trends and aging on weight change a secondary analysis included only primiparous women.

Study population
The III DHS was conducted in 1996.A full description of the survey sampling frame can be found elsewhere (17,22).This was a nationwide, two-stage probability sampling survey that included only women with one or more children under five years of age.The response rate for households was 80.9%.The population eligible for the current analysis consisted of the 2368 parous women surveyed in the III DHS aged 20 to 45 years at the time of the interview after excluding pregnant women at the time of the survey (285 women), women who reported delivery seven months before or less at the time of the interview (621 women), first birth before the age of 15 (199 women; since they could have gained height since their first pregnancy interfering with the calculations of the prior body mass index, BMI).Also excluded were 41 women whose weight or stature values were missing and 30 with improbable values (stature <130 cm or weight <35 kg).The final study sample size was 2297 women.

Measurement of the variables
The dependent variable, current BMI (weight/height 2 , in kg/m 2 ), was calculated from weight and height measurements taken in the household by two trained interviewers.Weight before childbearing was recalled at the end of the 30-min interview (22).Weight was measured using Seca-UNICEF 890 microelectronic scales with lithium batteries, and height using Stanley anthropometric tapes.Anthropometrists were trained and standardized and quality control was carried out during fieldwork supervision.The technical error for stature was equal to 0.23 cm.Short stature was defined as the 5th percentile of the US population (150 cm) (23).
Purchasing power score as used by the Brazilian Association of Marketing Research is a combined score of highest level of education obtained by the family, with services and items available in the house such as housemaid, number of radios, VCRs, televisions, refrigerators, etc. (17,22).Regions of the country were divided into three categories: rural, urban area of the North/Northeast, which is the least developed region of the country, and the third region combining the urban areas of the South, Southeast, and Midwest, which are more developed regions.Throughout the text these three regions are called rural, urban underdeveloped and urban developed regions, respectively.Parity is the number of live births and stillbirths of each woman and education is the total number of completed formal years of education.

Analytical procedures
Means and percentages were weighted to reflect national estimates.In order to account for the variance associated with sampling, we used the "Proc surveyreg" procedure of the Statistical Analysis System (SAS).We regressed short stature on weight prior to childbearing, current weight, and weight change from prior to childbearing to current weight.Models included age, socioeconomic, geographic, and reproductive variables.When weight gain and current BMI were the dependent variables, BMI prior to childbearing was included in the model, since previous analysis of these data as well as data from other studies indicate that high BMI prior to childbearing is the most important predictor of weight gain associated with reproduction.In addition, since parity and short stature are highly associated with socioeconomic factors, we included in the models markers for socioeconomic level (purchasing power score and women's years of education).Plausible interactions between the co-variables and stature were tested jointly.Second-order terms were tested for age and parity (number of live births and stillbirths).The exclusion criterion for interaction and second-term variables was a P value >0.05.Both BMI and weight change were log transformed to normalize the data.To permit log transformation of weight change, the smallest negative weight difference was added to each weight change value.The analysis compared the risk of overweight and obesity by measuring prevalence ratios.

Results
The prevalence of current overweight and obesity (BMI ≥25 kg/m 2 ) was higher among short stature women living in the urban developed region of the country compared to short stature women living in rural or urban underdeveloped areas (Table 1).Among women living in the urban developed region the risk of overweight or obesity (BMI ≥25 kg/m 2 ) was 1.27 (46.0 vs 36.2%) when short stature women were compared to normal stature women.Prior to childbearing, overweight and obesity rates were higher among short stature women in all regions, although prevalences were much lower and differences according to stature were statistically significant only for the urban developed region (Table 1).
The prevalence of short stature was 23.9% in the rural area, 21.7% in the urban underdeveloped area, and 10.1% in the urban developed area.
Parity was significantly different in the three regions studied, with the highest parity level being observed in the rural area (Table 2).For most of the socioeconomic variables, statistically significant differences were found between the urban developed region and the urban underdeveloped region (Table 2).
After adjustment for age, reproductive factors and socioeconomic variables, current BMI was 2.39 units (0.86 +1.72 -0.19) higher among short stature women living in the urban developed area compared to normal stature women living in the urban underdeveloped region (Table 3).Yet the ß regression coefficient for the urban developed area (-0.19) was not statistically significant.Region was included in the model as two indicator variables with urban developed region as the reference group; therefore, for effect estimation both indicator variables and interaction terms were considered.
Thus, region of the country is a strong modifier of the association between short stature and BMI.In these multivariate models parity, previous BMI and age were important predictors of current BMI (P < 0.01) (data not shown).Table 3 also shows that, when analysis was restricted to primiparous women allowing a better control of cohort effect, the interaction term in the current BMI model was no longer significant (P = 0.07).However, overall models for BMI prior to childbearing and BMI change involved the same interpretation as the models including all women, with the same pattern and magnitude of association.

Discussion
In 1996, about 50% of Brazilian parous women living in the urban developed area and with stature shorter than 1.50 m were overweight or obese.Women with such stature in this region had a higher prevalence of overweight (RR = 1.27) than their counterparts with normal stature, but for the other two regions (rural and urban North/Northeast) short stature was not associated with BMI.This association was unchanged after adjustment for demographic and socioeconomic factors.This modification of region effect on the association between stature and BMI was not observed for BMI prior to childbearing (stature*rural and stature*developed terms had P values >0.60), after adjusting for socioeconomic factors.However, short stature and living in the urban region were predictors of BMI prior to childbearing (P < 0.05).Thus, short stature, a marker for undernutrition during childhood, appears to be related to higher BMI prior to childbearing in all regions and also to weight gain after pregnancy in the more developed region of the country.Analysis including only primiparous women showed the greatest BMI change with short stature in the urban developed region (+0.49+ 1.46 = 1.95 units of BMI).These findings support the hypothesis of preferential weight gain, or weight retention with pregnancy associated with short stature and may contribute to understanding the gender effect of short stature on BMI, as shown in two Brazilian studies which found a risk of greater BMI and waist-hip ratio among short stature women, but not among men (6)(7)(8)(9)(10).Also interesting is the lack of association between short stature and current BMI in the rural and the underdeveloped areas of the country.This important interaction, at least in this population, suggests that the health/ nutritional transition may modify the effect of early exposures, such as undernutrition, on chronic diseases.Our data may indicate that a minimum of food availability or the kind of food availability present in urban centers would be required for the expression of a metabolic programming at an early age.
It should be noted that the five regions of the country, North, Northeast, South, Southeast and Midwest, have special dietary and lifestyle habits.However, other anthropometric studies have shown that these regions, as clustered into rural, urban developed and urban underdeveloped regions in the present study, are important predictors of obesity (24).
In a previous study of Brazilian women we showed that the association between stature and overweight was not due to limitations of using BMI as an indicator of obesity, or to possible bias due to the persistence in adult life of socioeconomic conditions which produced the effects attributed to programming (10).In the present study we found an association of both current BMI and weight gain (BMI change) with short stature, but this association persisted after controlling for socioeconomic factors only in the more developed area of the country.Under the hypothesis of an incomplete control of confounding for socioeconomic condition operating in this population we would expect a greater association between stature and obesity in rural and underdeveloped urban areas, which had the lowest socioeconomic levels.However, after adjustment for socioeconomic factors there was no association in these areas.This finding of a statistical interaction between area of the country and stature may indicate that a western life style, such as that present in metropolitan areas of developing countries, may contribute to the manifestation of a metabolic programming early in life.
Although there was no confirmation of pre-childbearing weight other than the recall, as an attempt to validate the recalled pre-pregnancy weight we compared the recalled BMI distribution of the 21-to 29year-old women in our study with a measured BMI distribution of nulliparous women from a national survey.There was a good consistency in the percentile distributions as we presented in a previous paper (17).
In agreement with other studies we found that BMI prior to childbearing was the most important predictor of recent BMI (12,14,16,17).Population-based follow-up analysis of the United States population (NHANES I follow-up) showed that women with high BMI had an increased risk of greater weight gain with parity (21).
Adult short stature has been related to glycemia (18,19), insulin (13) and cortisol levels (21,22).Changes in insulin sensitivity during pregnancy increase fat deposition (25) and may explain how environmental conditions that allow marginal adequacy of energy intake may favor metabolic changes directed to fat deposition.These hormonal changes could explain how small increases in food availability are associated with a large increase in the prevalence of obesity.Energy intake and fat intake did not explain the association between short stature and obesity in a Brazilian population-based survey (10).

Table 2 .
Age, age-adjusted means and standard deviation (SD) of body mass index (BMI, kg/m 2 ), weight gain and reproductive and socioeconomic variables among parous women in three regions ofBrazil, 1996.
cData are reported as means ± SD. *Values with different letters are significantly different (P < 0.05, ANOVA).

Table 1 .
Prevalence of overweight and obesity by stature among parous women in three regions ofBrazil,  1996.
*P < 0.05 comparing short and normal stature women within each region (Student t-test).

Table 3 .
Regression coefficients (ß) of multivariate linear regression of short stature on body mass index (BMI) in three different models adjusted for the main confounding variables(Brazil, 1996).