Determinants of weight gain in pregnant women attending a public prenatal care facility in Rio de Janeiro , Brazil : a prospective study , 2005-2007

The objective of this study was to evaluate the determinants of weight gain during pregnancy. The study adopted a prospective cohort design with four follow-up waves and included a sample of 255 pregnant women that received prenatal care at a public health care facility in Rio de Janeiro, Brazil. A mixed-effects linear longitudinal regression model was used, having as the dependent variable the weight assessed in four follow-up waves, and as independent variables: demographic, socioeconomic, reproductive, behavioral, and nutritional data. Mean weight gain was 0.413kg per gestational week, consistent with recommendations by the Institute of Medicine. Per capita family income and smoking were associated with total weight gain during gestation. According to the longitudinal multiple linear regression model, age (β = 0.6315), menarche (β = -2.3861), triglycerides (β = 0.0437), blood glucose (β = 0.1544), and adequacy of energy consumption (β = -0.0642) were associated with gestational weight gain. Special attention should be given to these subgroups, due to increased risk of excessive weight gain. Nutritional Epidemiology; Weight Gain; Pregnancy Introduction Weight gain during pregnancy includes three components: the product of conception (fetus, placenta, extracellular and amniotic fluid), maternal tissue expansion (uterus, breasts, and blood volume), and maternal fat reserve 1,2. Normal mean total gestational weight gain is estimated at 12.5kg 1,2,3,4. Weight gain measured at various moments during pregnancy is the most common way of assessing the pregnant woman’s nutritional status, besides indirectly assessing fetal growth 4,5, since it is more sensitive (as compared to other anthropometric indicators) to acute intra-gestational stress 3. It is thus an important indicator for observing normal evolution of pregnancy and is widely used in prenatal care 4. Pre-gestational maternal nutritional status and gestational weight gain have been studied systematically, due not only to the growing prevalence of deviations from their normal values, but especially to their determinant role in gestational outcomes 2 and the lifelong consequences for the mother and child 6,7. Low birth weight, intrauterine growth retardation, and prematurity are related to insufficient weight gain during pregnancy 8,9,10,11. Meanwhile, excessive maternal weight gain is associated with macrosomy, increased cesarean rate, diabetes mellitus, pregnancy-induced hypertension, and postpartum weight retention 2,12,13,14,15,16,17,18. ARTIGO ARTICLE S272 Cad. Saúde Pública, Rio de Janeiro, 24 Sup 2:S272-S284, 2008 DETERMINANTS OF WEIGHT GAIN IN PREGNANT WOMEN S273 Cad. Saúde Pública, Rio de Janeiro, 24 Sup 2:S272-S284, 2008 Complex interactions between maternal biological and metabolic factors influence the weight gain pattern during pregnancy, besides numerous social factors that can act as determinants. The latter include pre-gestational nutritional status, schooling, socio-demographic situation, skin color, work, physical activity, calorie intake, smoking, alcohol consumption, reproductive history, and prenatal care 2,4,19,20. Although the amount of gestational weight gain and obstetric outcomes are well documented in the scientific literature, the determinants of gestational weight change have received relatively little attention. The objective of the current study was thus to investigate demographic, socioeconomic, reproductive, behavioral, and nutritional determinants of weight change during pregnancy.


Introduction
Weight gain during pregnancy includes three components: the product of conception (fetus, placenta, extracellular and amniotic fluid), maternal tissue expansion (uterus, breasts, and blood volume), and maternal fat reserve 1,2 .Normal mean total gestational weight gain is estimated at 12.5kg 1,2,3,4 .
Weight gain measured at various moments during pregnancy is the most common way of assessing the pregnant woman's nutritional status, besides indirectly assessing fetal growth 4,5 , since it is more sensitive (as compared to other anthropometric indicators) to acute intra-gestational stress 3 .It is thus an important indicator for observing normal evolution of pregnancy and is widely used in prenatal care 4 .
Pre-gestational maternal nutritional status and gestational weight gain have been studied systematically, due not only to the growing prevalence of deviations from their normal values, but especially to their determinant role in gestational outcomes 2 and the lifelong consequences for the mother and child 6,7 .
weight as measured at the fourth and first followup waves.
The covariables used in the current analysis were collected at baseline (8 th -13 th gestational week) and separated in five blocks: demographic and socioeconomic variables: age (18-24, 25-40  years), self-reported skin color (white, brown, black), marital status (married, living with partner, single), schooling (≤ 8, > 8 years), per capita family income (≤ 0.5; 0.5-1.0;> 1.0 times the minimum wage); and work outside the home (yes, no); reproductive variables: age at menarche (< 12; ≥ 12 years), parity (primiparous, multiparous); inter-pregnancy spacing; behavioral variables: smoking (non-smoker, former smoker, smoker), alcohol consumption (non-consumer, former consumer, consumer), intensity of physical activities (sedentary, light activity).Information on smoking also included questions on age when the individual began smoking, number of cigarettes per day, and time during which the woman smoked.However, none of these variables showed an effect on weight change.The alcohol consumption variable was based on a questionnaire on frequency of alcohol consumption that included information on the type of alcoholic beverage consumed, frequency, and amount, in addition to age when the individual began drinking; nutritional variables: pre-gestational nutritional status (underweight, normal weight, overweight, obese), adequacy of calorie intake (below adequate, adequate, more than adequate), daily protein, lipid, and carbohydrate energy intake; biochemical variables: hematocrit (%), hemoglobin (g/dL), blood glucose (mg/dL), total cholesterol total (mg/dL), low density cholesterol (LDL-cholesterol, mg/dL), high density cholesterol (HDL-cholesterol, mg/dL), and triglycerides (mg/dL).
The women were weighed on a digital scale with a minimum capacity of 2.5kg and maximum of 150kg, accurate to 0.1kg (Filizzola PL 150, Filizzola Ltda., São Paulo, Brazil).Stature was measured in duplicate with a Harpeden portable stadiometer (Harpenden Inc., UK) with a minimum capacity of 70.1cm and maximum of 208.5cm, with a maximum allowable variation of 0.5cm between the two measurements.All the anthropometric measurements were taken by trained interviewers and standardized according to recommended guidelines 21 .
Pre-gestational body mass index [BMI = weight (kg)/stature (m 2 )] was obtained during the first follow-up wave, the maximum period for establishing a pre-gestational nutritional diagnosis based on the weight measurement.The BMI cutoff points were those proposed by the World Health Organization (BMI < 18.5kg/m 2
The pregnant women's level of physical activity was investigated by means of a scale measuring the intensity of physical activities 22 .First, the mean weekly frequency with which each of the activities listed on the questionnaire was investigated.Subjects were then asked about the mean duration of these activities, and this value was multiplied by the weekly frequency of the respective activity, thus obtaining the mean weekly time for each activity.The mean weekly time for all the activities was then summed.The mean weekly time for each activity was multiplied by an intensity factor associated with the activity (MET -metabolic equivalent test) 23 and a new sum total was performed.The pregnant women's level of physical activity was then estimated by dividing the second total by the first.Finally, the pregnant women were categorized according to level of physical activity and classified according to the cutoff points proposed by Chasan-Taber et al. 24 : sedentary (METgest ≤ 1.5) and light (METgest > 1.5).
Food consumption was measured by means of a semi-quantitative food frequency questionnaire (Food Frequency Questionnaire -FFQ) 25 , previously validated for adults, containing a food list with 81 items and eight possible answers for frequency of current consumption (> 3 times/day, 2-3 times/day, once/day; 5-6 times/ week, 2-4 times/week, once/week, 1-3 times/ month, and never/rarely).The overall frequencies thus obtained were transformed into daily frequencies and multiplied by the amounts of foods consumed, thereby obtaining the daily consumption of energy, carbohydrates, lipids, and proteins.The databanks adopted for this purpose were the Brazilian Food Composition Table (TACO) 26 , the table of the Brazilian Institute of Geography and Statistics (IBGE) 27 , and the USDA (United States Department of Agriculture) table 28 .The analyses excluded 33 women with energy consumption less than 600 or greater than 6,000kcal, who were considered outliers.The food intake energy adequacy was calculated according to the recommended energy for age, weight 29 , and estimated level of physical activity, considering adequate gesta-tional weight gain 1,30 .Energy consumption was considered adequate from 90 to 110%, low when it was less than 90%, and high when it was greater than 110%.
The blood tests of the pregnant women participating in the study were done on 5mL blood samples, drawn by a designated professional from the health unit.Subjects had been fasting for at least 12 hours before drawing the blood sample.The samples were stored in the laboratory under controlled conditions and standard temperature (6ºC), and were analyzed by a properly trained lab technician.Blood samples were centrifuged (3,000rpm) for plasma separation and extraction, after which they were immediately frozen at -20ºC for subsequent analysis.Hemoglobin and hematocrit tests were performed using an automatic counter.The enzymatic method was used to obtain total cholesterol and triglycerides, while the colorimetric method was used to obtain HDL-cholesterol.LDL-cholesterol, was measured by the formula proposed by Friedwald et al. 31  Data on bodyweight and gestational age were also obtained from medical charts of the pregnant women who began follow-up but who missed any of the subsequent interviews.Weight measurements were also taken at the time of prenatal visits, by trained health professionals, using only the measurements that fit the period proposed for each of the study's follow-up waves.
Data were entered using Epi Info version 6.02 (Centers for Disease Control and Prevention, Atlanta, USA).The consistency of the resulting databank was confirmed using SPSS 13.0 (SPSS Inc., Chicago, USA) and S-Plus 2000 (Mathsoft Inc., Seattle, USA).
The statistical analysis initially involved the description of the pregnant women participating in the cohort, according to established procedures like means and 95% confidence intervals (95%CI) for the continuous variables, in addition to the mean weight measured at each follow-up and the total gestational weight gain according to the selected covariables.The Student t test and analysis of variance (ANOVA) were used to test the equality of the mean total gestational weight gain.
Mixed-effects linear longitudinal regression modeling was used as the statistical procedure, with weight measured at each follow-up as the dependent variable and the selected covariables as independent variables, based on a minimum model, controlling for gestational age in weeks.The categorical variables were analyzed in the factor format, allowing estimates for each of the strata in relation to the reference category.
The mixed-effects procedure generates regression coefficients and their respective standard errors.The explanatory variables displaying a significant association (p-value < 0.20) with the outcome were candidates for the multiple model.The mixed-effects linear model was adopted based on its great statistical efficiency 32 .
The pattern of losses was evaluated based on the final follow-up rate (number of pregnant women who began/concluded follow-up).The results were analyzed using the chi-squared test for proportions.
The study was approved by the Institutional Review Board of the Martagão Gesteira Institute of Childcare and Pediatrics at the Federal University in Rio de Janeiro and by the Rio de Janeiro Municipal Health Secretariat, and complies with Ruling n o .196/1996 of the Brazilian National Health Council 33 .All of the participants signed a free and informed term of consent, having received all the necessary clarifications.

Results
The final rate of loss to follow-up was 32.1%.However, the pattern of losses was random in relation to the different variables listed in Table 1.
The data analysis included 255 pregnant women in the first follow-up wave (baseline).Of these, 197 appeared for the second interview, 186 completed the third, and 173 concluded the entire follow-up with the fourth interview, as displayed in Figure 1.For the sample of pregnant women as a whole, 11.7% (n = 23), 17.9% (n = 33), and 45.1% (n = 78) had their weight data retrieved from their medical charts at the second, third, and fourth follow-up waves, respectively.No significant differences were found between the mean weights as measured directly and retrieved from medical charts, after controlling for gestational age.
Table 2 shows the baseline means for several selected variables for the 255 pregnant women  1).Mean total weight gain was 11.7kg (95%CI: 11.0-12.4).In addition, 33.4% of the women began the pregnancy with excess weight (12.2% with obesity) and 5.5% with pre-gestational underweight, according to the WHO cutoff points, while the values according to the Institute of Medicine guidelines were 13.7%, 15.3%, and 18% for overweight, obesity, and underweight, respectively (results not shown).
Tables 3 and 4 show the mean weight measurements at each follow-up wave and the mean total weight gain according to the selected variables.Mean weight was higher in pregnant women with per capita family income greater than one minimum wage, as was total weight gain (12.5 versus 10.8kg).Former smokers showed higher total gestational weight gain (13.9kg), as well as higher mean bodyweight throughout pregnancy.
Pre-selection of variables for the mixed-effects linear model (Table 5) showed a positive association between weight gain and the following covariables: gestational age, age, marital status (married/living with partner), parity, smoking (non or former smoker), alcohol consumption (none/former), pre-gestational BMI, stature, blood glucose, and triglycerides.Meanwhile, age at menarche and adequacy of energy consumption were inversely associated with weight gain.The variables inter-pregnancy spacing, alcohol consumption (none/former), and HDL-cholesterol showed statistical significance < 0.20 and were also candidates for the multiple model.

Discussion
The study's findings demonstrate that mean gestational weight gain was 0.413kg per week of pregnancy, consistent with the weight gain recommended by the Institute of Medicine for women with pre-gestational BMI within the normal range 1,34 .Analyzing the mean weight gain per trimester of pregnancy, our results were similar to those of Nucci et al. 35 based on data collected by the Brazilian Study on Gestational Diabetes from 1991 to 1995 with more than three thousand pregnant women in six Brazilian State capitals.In the current study, mean weight gain was 0.41kg and 0.43kg per week in the second and third trimesters, respectively (data not shown in tables).In Nucci et al. 35 , the figures were 0.43kg and 0.40kg per week for the second and third trimesters, respectively.
The prevalence rates for women who began pregnancy with excess weight and underweight were 33.4% and 5.5%, respectively, based on the WHO cutoff points 4 .These figures disagree with those found in health services in the States of São Paulo (24% and 21%) 19 , Pernambuco (26.3% # Variable with losses of 5 to 10% due to absence or inconsistency of collected information. and 25.4%) 36 , and Paraíba (27% and 23%) 11 .A lower prevalence rate for excess pre-gestational weight was described by Kac et al. 13,14 , who found 17.5%, based on the results of a cohort of women monitored for nine months postpartum in Rio de Janeiro from 1999 to 2001.A multicenter study by Nucci et al. 16 showed 29% with excess weight and 6% with underweight, similar to our findings.Importantly, the use of different diagnostic criteria for evaluating pre-gestational nutritional status can explain the differences in various studies for low pre-gestational weight, since only Nucci et al. 16 and the current study used the WHO cutoff points 4 , while Kac et al. 13,14 used the Institute of Medicine criteria 1,34 and other studies used the cutoff points proposed by Atalah et al. 37 , which differ from the WHO guidelines as to the diagnosis of underweight.
According to the linear multiple regression model, maternal age, serum triglycerides and blood glucose, age at menarche, and adequacy of energy consumption remained associated with gestational weight gain.
For each increase of one year in the woman's age, there was an increase of 0.631kg in weight gain.In other words, comparing 20 and 30-year-old pregnant women, the difference in total weight gain could reach nearly 6kg.Age is a non-modifiable factor and has proven to be an important determinant of anthropometric measurements 38 .For pregnant women, the literature shows lower weight gain in younger women 39 , which is consistent with the current study's findings.
We observed an inverse association between weight gain and age at menarche.In other words, the earlier the menarche, the higher the gestational weight gain, on average -2.4kg per year.In the epidemiological context, age at menarche has been inversely associated with excess bodyweight in childbearing-age women, and early age at menarche has been reported as an important risk factor for future obesity 40,41 .Some authors suggest that this association can be attributed to excess weight in childhood 42,43 .Other studies 44,45 confirm that menarche before 12 years of age can represent predisposition to fat accumulation, as well as to the tendency to deposit adipose tissue during pregnancy in women without excess pre-gestational weight.Early menarche also leads to early initiation of the reproductive cycle, which can influence hormonal activity and body fat deposition due to physiological immaturity.Gunderson et al. 44 reported that women with age at menarche ≤ 12 years showed 2.57 times the odds of presenting postpartum overweight.A study of childbearing-age women in Belo Horizonte by Kac et al. 45 showed a 3.2 odds ratio of obesity in women with age at menarche less than 12 years.Helm et al. 41 found that early menarche was associated with excess weight (OR = 5.0; 95%CI: 2.4-10.6).
Higher triglycerides in early pregnancy were directly associated with weight gain, although the coefficient was relatively small (0.043kg per 1mg/dL).Changes in lipid metabolism during pregnancy lead to an accumulation of maternal fat in early and mid-pregnancy.During early pregnancy, estrogen, progesterone, and insulin levels are increased, favoring lipid deposition and inhibiting lipolysis.Butte 46 demonstrated that lipoprotein lipase activity is increased in adipose tissue in the femoral region from the 8 th to the 11 th week of pregnancy.This hypothesis provides a probable explanation for the association between triglyceride levels in early pregnancy and greater subsequent weight gain.
Serum glucose concentrations were positively associated with weight gain, estimated at 0.154kg per 1mg/dL, meaning, for example, that a woman who begins pregnancy with a blood glucose of 110mg/dL would experience a 4.6kg greater weight gain as compared to women with blood glucose of 80mg/dL.Saldanha et al. 47 , studying the relationship between gestational weight gain and glucose tolerance in women, observed that weight gain was significantly higher in women with gestational diabetes than in those with normal blood glucose.Studying a cohort of 700 women to evaluate the relationship between BMI, glucose tolerance, and gestational outcomes, Bo et al. 48observed that hyperglycemia in pregnancy was a risk factor for excess gestational weight gain (OR = 1.06; 95%CI: 1.02-1.10).The results of the current study corroborate the findings from the studies quoted above.
An inverse association was observed between adequacy of energy consumption in the first trimester and weight gain during pregnancy.For each percentage point drop in adequacy, there was a weight gain of approximately 0.064kg.Importantly, bodyweight for the pregnant women was measure up to the 13 th week to calculate ad-equacy of energy consumption 1,34 .A possible explanation is that some women may display a significant weight drop in early pregnancy, since they are more susceptible to nausea and vomiting 1,3 .This can overestimate the percentage adequacy for those who lost weight and underestimate it for those who gained weight.Another plausible explanation is the underreporting of food consumption (as measured by the FFQ) by pregnant women with greater bodyweight.In the study that validated the FFQ used in the present study, Sichieri & Everhart 25 observed that energy consumption by women decreased with the increase in energy expenditure, suggesting that women with excess weight underreported their calorie consumption.Stulbach et al. 19 evaluated the impact of socio-demographic factors, gestational history, and pre-gestational nutritional status on excess weight gain in pregnancy, and found a high frequency, particularly in women with the following characteristics: more schooling (RR = 1.9; 95%CI: 1.22-2.97),unmarried/without a partner (RR = 1.7; 95%CI: 1.06-2.59),primiparous (RR = 2.1; 95%CI: 1.20-3.85),and pre-gestational overweight/obesity (RR = 2.0; 95%CI: 1.04-3.92).A recently published study with a similar design and using multilevel analysis identified the effect of schooling, marital status, parity, and baseline BMI on weight gain in pregnant women enrolled in a public prenatal clinic in the city of São Paulo, Brazil 49 .
The results of the linear longitudinal regression model in the current study showed an effect on weight gain, particularly from biological and food consumption variables, which may only indirectly reflect the pregnant woman's socioeconomic status, thus partially differing from the findings of Stulbach et al. 19 .Meanwhile, the results shown in Tables 3 and 4 showed other associations between total weight gain, income, and smoking.
It is suggested that low schooling and low socioeconomic status are associated with increased risk of insufficient gestational weight gain 2 .In the current analysis, lower socioeconomic status of pregnant women, as represented by per capita family income, showed lower total gestational weight gain (10.8kg).
Studies focusing on the association between bodyweight and smoking have produced inconclusive evidence 50 .Smoking and bodyweight show an inverse association, and smokers, especially women, frequently gain weight when they quit the habit.Smoking is known to induce an acute reaction in the metabolic rate that can lead to a reduction in relative food consumption as compared to non-smokers 51 .For pregnant women, it has been documented that smokers gain less weight during pregnancy as compared to non-smokers and former smokers 1 .Mongoven et al. 52 observed that women who quit smoking gained more weight than smokers, regardless of pre-gestational nutritional status.In addition, smokers had a higher risk of low weight gain as compared to those who quit smoking (28% versus 13%) 52 .Oslon & Strawderman 53 found an association between smoking and weight gain in pregnancy, whereby women who smoked 1.5 packs a day gained significantly less weight.Hellerstedt et al. 54 found that smokers showed lower mean weight gain as compared to non-smokers (9.3 and 9.7kg, respectively), when the women were obese at the beginning of pregnancy; how- ever, women with normal baseline weight gained less weight when they were non-smokers (13.4 versus 15.0kg).
In the current study, smoking was associated with total gestational weight gain.Pregnant women who were former smokers showed the highest total weight gain (13.9kg), followed by smokers (11.7kg) and non-smokers (10.9kg).Our results differ from those of part of the literature, since the current study showed that non-smokers gained less weight by the end of pregnancy.One possible explanation for the discordant findings is the way the smoking habit was measured, which should have included information on quitting, since some women may only have stopped smoking shortly before the first interview.
Losses to follow-up are an important issue for studies with a prospective cohort design, and can thus be considered a potential limitation in the current study.Although care was taken during the study to minimize this problem, losses by the end of the fourth follow-up interview amounted to 31.8%, consistent with similar studies 19 .Meanwhile, no selective losses were observed according to socioeconomic charac-teristics and pre-gestational nutritional status of women who began follow-up as compared to those who concluded the study.A particularly important aspect of the current study was its design and analytical approach.A prospective design, for example, allows calculating important indicators like mean gestational weight gain, rarely used due to the difficulty in calculating it.In addition, the use of mixed-effects models is relevant due to the greater precision and the power to detect significant differences as compared to cross-sectional studies or even the results from analyses of variance for repeated measurements 32 .
The results of the current study can be expected to be applicable to pregnant women with the same characteristics as this sample.Thus, special attention should be given during prenatal nutritional follow-up for older women, those with early menarche, high serum triglycerides and glucose, lower-than-recommended energy consumption for satisfactory weight gain, and specific population sub-groups like former smokers and pregnant women with lower socioeconomic status.
Epidemiologia Nutricional; Ganho de Peso; Gravidez Contributors P. L. Rodrigues participated in the data collection, elaboration, statistical analysis, and interpretation and preparation of the final paper.E. M. A. Lacerda collaborated in the analysis and revision of the final draft.M. M. Schlüssel contributed to the data interpretation and elaboration and revision of the final draft.M. H. C. Spyrides participated in the statistical analysis and critical revision of the final draft.G. Kac participated in all stages of the study, from the planning to the final revision.

Figure 1 Flow
Figure 1 Flow chart of recruitment, loss-to-follow-up rate, mean gestational age, and mean weight measured at each follow-up interview.Rio de Janeiro, Brazil, 2005-2007.

Table 1
Distribution of selected variables comparing losses to complete follow-up and fi nal follow-up rate in pregnant women attending a prenatal clinic in Rio deJaneiro, Brazil, 2005-2007.
* p-value refers to chi-squared test of proportions.
* Variables with losses < 2% due to absence or inconsistency of collected information; ** "n" refers to 128 non-primiparous pregnant women; *** Variable with losses of 10 to 20% due to absence or inconsistency of collected information;

Table 5
Pre-selection of variables for the weight change mixed-effects model in pregnant women attending a prenatal clinic in Rio de Janeiro, Brazil, 2005-2007.Categorical variables for which the first category is the reference and the second is the risk. *

Table 6
Final model mixed-effects linear longitudinal regression analysis for weight gain in pregnant women attending a prenatal clinic in Rio deJaneiro, Brazil, 2005-2007.
β: linear longitudinal regression coeffi cient; fi xed effect: refl ects the mean of the overall criteria; random effect: refl ects how specifi c criteria deviate from the overall mean; residual: unexplained variation.