The emerging obesity prevalence in developing countries is alarming( 1 – Reference Wang and Lobstein 4 ), especially among women of reproductive age (WRA), and recently this problem has been observed among WRA in the low socio-economic status (SES) population( Reference Ezzati, Vander Hoorn and Lawes 5 ). This increasing number is alarming because obesity increases the risk of non-communicable diseases( 1 , 6 ), and people with low SES have limited access to the treatment of non-communicable diseases that are the consequences of obesity. Moreover, WRA have the added risk of maternal and neonatal outcomes( Reference Rowlands, Graves and De Jersey 7 ).
Obesity is multifactorial, and each factor may lead to obesity as single factor or in combination with other factors( Reference Jebb 8 , Reference James 9 ). However, obesity mainly results from energy imbalance, which is directly associated with food intake and physical activity( 1 ).
The total energy intake is considered one of the dietary determinants that influences weight gain( Reference Du 10 ) and obesity( Reference Kant and Graubard 11 ). Recent studies have shown that high energy density is associated with high energy and fat intakes and low vegetable and fruit intakes( Reference Kant and Graubard 11 ). Energy density of food is influenced by the type and variety of food consumed( Reference Ello-Martin, Ledikwe and Rolls 12 ). Consumption of refined grains, added sugars and fats, snacks, beverages, fast foods and eating out of home contribute towards the intake of high energy-dense food( Reference Drewnowski 13 ). The food consumed by people determines their dietary pattern. Moreover, people from the low SES population consume high energy-dense food because it is more accessible and affordable( Reference Aguilar 14 ). However, evidence on the type of energy-dense food and dietary pattern leading to obesity in the low SES population is not yet convincing, especially in developing countries.
The present study used the Indonesian population data, because according to the 2010 Indonesian Basic Health Research Survey( 15 ) the prevalence of obesity among adult Indonesian men and women is 7·8 and 15·5 %, respectively, which increased to 19·7 and 32·9 %, respectively, in 2013( 16 ). Moreover, the prevalence of obesity was higher in urban areas than in rural areas. Khusun( Reference Khusun 17 ) demonstrated that the prevalence of overweight condition among low SES women in Indonesia increased from 15·6 % in 1993 and 21·8 % in 2000 to 30 % in 2007.
People should be informed on strategies for reducing energy intake in terms of the types of food and dietary patterns contributing to high energy intake, through simple informative messages on nutritional habits. Such public awareness measures are crucial considering the increasing prevalence of obesity among the low SES population. Studies exploring specific food consumption and intake patterns contributing to obesity among urban low SES women are limited. Therefore, we assessed the association of food intake and dietary pattern with obesity among women of the low SES population living in urban slum areas.
Methods
Study population and sampling procedure
This comparative cross-sectional study was conducted from February to April 2013, and a multi-stage random sampling method was used to enrol study subjects from three randomly selected slum areas in Central Jakarta – namely, Kebon Melati, Mangga Dua Selatan and Pegangsaan Village. In all, 600 apparently healthy women aged 19–49 years were screened for determining obese and normal-weight conditions according to BMI cut-off points for Asian people( 18 ). Pregnant and women with disabilities were excluded from this study. From the 600 subjects screened, 110 normal-weight and 110 obese subjects were randomly selected. The sample size was calculated by estimating differences in the mean energy intake among obese and normal-weight women with 80 % power, 95 % significance level and an expected mean difference of 2179·8 kJ (521 kcal)( Reference Alves, Falcão and Pinto 19 ). The final study subjects were 103 obese and 104 normal-weight women, because one subject refused to continue with the study, two subjects returned to their villages during data collection and the energy intake for four and six subjects were under-reported and over-reported, respectively.
Anthropometric measurement
Body weight and height were measured using standardised procedures( Reference Gibson 20 ). Body weight was measured using a SECA electronic weighing scale (Seca GmbH) to the nearest 0·1 kg. Height was measured using a Shorr Board (Shorr Productions LLC) to the nearest 0·1 cm. Weight and height measurements were recorded twice for each subject, and the average values of these two measurements were recorded for analysis. The BMI was calculated as weight (kg) divided by the square of height (m2). Subjects with BMI≥27·5 kg/m2 were classified as obese and those with BMI=18·5–22·99 kg/m2 were classified as normal-weight subjects( 18 ).
Structured interview
Structured interviews were conducted to collect information on the socio-demographic characteristics and physical activities of the study population. The International Physical Activity Questionnaire (IPAQ; short, 7-d recall) was used for assessing the physical activities of the study subjects. This questionnaire assesses walking and moderate- and vigorous-intensity physical activity profiles. Physical activity was expressed as the metabolic equivalent of task (MET)-min/week. According to the International Physical Activity Questionnaire( 21 ), the total MET-min/week score was categorised into three levels of physical activity: low, moderate and high.
Dietary assessment
Dietary habits of all study subjects were assessed using a single 24-h food recall and FFQ. The under-reporting of energy intake was defined using the Goldberg cut-off( Reference Goldberg, Black and Jebb 22 ) with a 99·7 % confidence limit. The over-reporting of energy intake was predicted on the basis of total energy expenditure values developed by McCrory et al. ( Reference McCrory, Hadjuk and Roberts 23 ).
Single 24-h food recall
Subjects were asked to recall their exact food intake, including beverages, during the previous 24-h period or the preceding day( Reference Gibson 20 ). Subjects were also asked to recall the portion size, processing method and composition of the food consumed. Sample food models were used to help the subjects assess their food portion sizes.
FFQ
The FFQ was developed after a group interview involving eleven normal-weight and twelve obese women from the study area. From the food items collected during the group interview, ninety-nine items including ninety foods and nine beverages were included in the FFQ. The study subjects were asked how often, on average, they consumed the foods over the previous 1 month( Reference Khani, Ye and Terry 24 ). The frequencies were categorised as never, 1–6 times/d, 1–6 times/week and 1–3 times/month. The frequency of consumption of each food item was calculated as the frequency of consumption per month.
The ninety-nine food items that were part of the FFQ were divided into twenty-two food groups (eighteen food and four beverage groups) for assessing the dietary patterns among the normal-weight and obese women. The food groups were based on similarity of their nutrient profiles, culinary usage among the foods or specific nutrients as study interests. Dietary patterns were explored through principle component analysis (PCA).
Data analyses
Data were analysed using SPSS version 16.00 (IBM). Normal distribution of data was analysed using the Kolmogorov–Smirnov test. Descriptive data are presented as mean values and standard deviations. χ 2 and t tests were used for assessing the association between variables and obesity status, with a significance level of 0·05. Energy and macronutrient intakes and energy density were calculated from the single 24-h food recall. Data from the single 24-h food recall were converted into energy and macronutrient intakes using NutriSurvey software for Windows 2004 and were calculated as the average of the mean intake.
The energy density of each subject was calculated by dividing the energy (kJ (kcal)) obtained from total food consumed, including beverages, within 24 h by the weight (g) of total consumed food, excluding the weight of all beverages( Reference Ledikwe, Blanck and Khan 25 , Reference Ledikwe, Blanck and Khan 26 ). The energy density for each group was determined by calculating the average energy density of all subjects in each group. The calculated energy density was classified into three groups according to the rankings of the energy density value of each subject from the tertile study samples. The first, second and third tertiles projected low, moderate and high energy densities, respectively.
Dietary patterns were identified from the FFQ. The factor analysis for the FFQ was performed through PCA for all twenty-two food groups included in the FFQ, and the factor solutions were expressed as frequency consumption per month. The factor analysis solutions were limited to the factors with an eigenvalue >1( Reference Ambrosini, Sullivan and de Klerk 27 ). Eigenvalue >1 denotes that the factor is a possible variable in the correlation( Reference Newby, Sc and Tucker 28 ). An orthogonal transformation using a varimax rotation was used for rotating the factors. This rotation was used for improving the interpretation and obtaining independent dietary patterns. Two factors that gave the highest variance explained were used for the analyses. The final factor solution after the rotations provided factor loadings for each factor or dietary patterns( Reference Ambrosini, Sullivan and de Klerk 27 ). The factor loadings represent the importance of a food group for determining the dietary patterns( Reference Sichieri 29 ). A positive loading means that the food group was positively associated with the investigated dietary pattern score, whereas a negative loading means that the food group is not associated with the dietary pattern score( 15 ). Moreover, a high factor loading suggests a large contribution from the food group towards the investigated dietary pattern( Reference Hu, Rimm and Smith-warner 30 ). Food groups with absolute factor loading >0·30 were considered significant contributors to dietary patterns( Reference Field 31 ). All food groups considered for the final factor solutions were used for calculating dietary pattern scores( Reference Ambrosini, Sullivan and de Klerk 27 ). The factor score for each subject was calculated by summing the intake of each food group weighted by the factor loadings, and the factor score for each dietary pattern was calculated by summing the factor scores of all subjects( Reference Khani, Ye and Terry 24 ).
Logistics regression analyses through the enter method were performed for exploring the association between a dietary pattern and the subject’s weight status. Two steps of modelling were used. In model 1, only dietary pattern score variables were included. In model 2, all possible confounders were included – namely, energy intake, energy density, age, marital status, smoking status, physical activity and occupation.
Results
The screening results (n 600) revealed that over-nutrition was substantially high in the study area; 68·1 % of the women screened were overweight or obese. The obesity prevalence was 37·5 %, evidencing the gravity of the already-high over-nutrition problem in urban slum areas. In total, data of 103 obese and 104 normal-weight women were analysed after screening and were compared for answering the study objective.
Table 1 shows the general characteristics of the study subjects. Age, ethnicity, marital status, occupation and education status of the obese and normal-weight women did not differ significantly. Most subjects were 30–49 years old, had Javanese ethnicity, married, unemployed and had completed junior and senior high school (29·8–33 % and 32–32·7 %, respectively) education. Furthermore, smoking status, use of hormonal contraceptives, parity and physical activity of the obese and normal-weight subjects also did not differ significantly. More than 70 % of the subjects were non-smokers, approximately 50 % used hormonal contraceptives and approximately 60 % had <3 children. Approximately 5, 47 and 48 % of the subjects showed low, moderate and high physical activity levels, respectively.
Table 1 General characteristics of the subjects
* χ 2 Test.
Energy and macronutrient intake and energy density
Table 2 lists the energy intake of the normal-weight and obese subjects. The energy, carbohydrate and fat intakes of the obese subjects were significantly higher (P<0·05) than those of the normal-weight subjects. The difference in the mean energy intake of the obese and normal-weight subjects was 929 kJ (222 kcal), whereas the mean energy differences for carbohydrate and fat intakes were 26 and 13 g/d, respectively. The protein intake and energy density did not differ significantly between the obese and normal-weight subjects; however, the mean energy density of the obese subjects was slightly higher than that of the normal-weight subjects.
Table 2 Energy and macronutrient intakes and energy density of normal-weight and obese subjects (Mean values and standard deviations)
* P<0·05.
† t Test.
Dietary patterns
Table 3 shows the separation of the food items consumed by the subjects according to their nutritive values into twenty-two food groups as follows: staple food, root and tuber, meat and poultry, seafood, organ meat, eggs, fried soyabean and soyabean products, non-fried soyabean and soyabean products, vegetables, mixed food, fruits, rujak (fruit salad in palm sugar dressing), fried snacks, non-fried snacks, biscuits and crackers, desserts, peanuts, oils and fats, sugar-sweetened beverages, milk and milk products, tea and coffee without sugar and tea and coffee with sugar and milk. Most food items in each food group were Indonesian traditional dishes.
Table 3 Classification of food items according to nutrient profiles
These twenty-two food groups were analysed through PCA, and the results are shown in Table 4. Two dietary patterns were revealed through PCA and subjectively named ‘more healthy’ and ‘less healthy’ based on their general nutritional values. The food groups with <0·30 factor loading – namely, sugar-sweetened beverages, tea and coffee without sugar and tea and coffee with sugar and milk – were excluded from the list. Moreover, four food groups that did not match the dietary pattern criteria based on their nutritional value were excluded from the list. The factor loading of desserts and organ meat were high in the ‘more healthy’ pattern, while the foods were actually considered unhealthy. On the other hand, the factor loadings of rujak and non-fried soyabean products were high in the ‘less healthy’ pattern, while the foods were actually considered healthy.
Table 4 Factor loadings obtained through principle component analysis for food groups that characterise the dietary patterns
* Mostly consumed as fried foods.
Food groups with >0·30 factor loading were considered as contributors to the dietary patterns. The more healthy pattern was comprised of food items with high factor loadings – namely, biscuits and crackers, fruits, non-fried snacks, seafood, peanuts, composite foods, staples, vegetables, milk and milk products and eggs; this patterns was considered more healthy because of the high factor loadings of fruits, vegetables and seafood. Moreover, the snacks included in this pattern were non-fried snacks. The less healthy pattern included food items with high factor loadings – namely, fried foods (fried snacks and soyabean and soyabean products) and oils and fats.
Association between the dietary patterns and BMI
Table 5 lists the logistic regression results, which indicate an association between dietary patterns and the nutritional status. Before adjustment for possible confounders (model 1), a high quintile score of the more healthy pattern was protective against obesity (OR 0·55; 95 % CI 0·22, 1·38), and a high quintile score of the less healthy pattern showed an increased risk of obesity (OR 1·60; 95 % CI 0·67, 4·22); however, the association was non-significant. After adjustment for possible confounding factors – namely, energy intake, energy density, age, marital status, smoking status, physical activity and occupation (model 2) – the association between the dietary pattern scores and nutritional status was stronger. The risk of obesity among subjects in the fourth to fifth quintile of the more healthy pattern score was three times lower than that among subjects in the first to third quintiles (OR 0·33; 95 % CI 0·12, 0·94). However, the obesity risk among subjects in the fourth to fifth quintile of the less healthy pattern score was 2·7 times higher than that among subjects in the first to third quintiles (OR 2·7; 95 % CI 0·95, 7·70). Energy intake was significantly associated with obesity. The obesity risk of subjects with an energy intake of >86 % of the reference daily intake (RDA) was 2·7 times higher than that among subjects with an intake of <86 % (OR 2·7; 95 % CI 1·38, 5·41). Married subjects were more obese compared with unmarried subjects (OR 2·16; 95 % CI 0·73, 6·36); marital status had no significant association with the risk of obesity.
Table 5 Crude and adjusted OR for obesity according to dietary patterns (Odds ratios and 95 % confidence intervals)
Ref., referent values.
* Binary logistic regression with the enter method.
Discussion
According to the 2010 Basic Health Research Survey, overweight and obesity levels in the community analysed in the present study were substantially higher than the overall overweight and obesity levels in Jakarta( 15 ). High overweight and obesity levels are concerning, especially considering the low SES of the urban slum community. The results of this study confirmed the hypotheses of the overweight problem in developing countries by showing evidence of problems related to high overweight and obesity levels, even in the slum community.
This study showed that energy intake was significantly higher among the obese than among the normal-weight subjects, even after adjustment for the dietary patterns and other potential confounding variables. This finding is consistent with the findings of similar studies from other countries( Reference Mokhtar, Elati and Chabir 32 , Reference Colditz, Willet and Stampfer 33 ) and confirms reports that high energy intake is the main determinant of obesity( 1 , Reference Du 10 ). The macronutrients that contributed to energy, carbohydrate and fat intakes were significantly higher among the obese than among the normal-weight subjects; however, no significant difference was observed for protein intake. Thus, obesity prevalence among women from the low SES population enrolled in this study was associated with high carbohydrate and fat intakes but low protein intake. The results clarified the contribution of carbohydrate intake to obesity, as reported in other studies( Reference Mokhtar, Elati and Chabir 32 , Reference Ma, Olendzki and Chiriboga 34 , Reference Mills, Perry and Reicks 35 ). Moreover, the contribution of fat intake towards obesity in this community was confirmed and was consistent with results of similar studies conducted in other countries( Reference Mills, Perry and Reicks 35 – Reference Ravussin and Tataranni 37 ).
No significant difference was observed in the energy density among obese and normal-weight subjects. This result was inconsistent with some studies that reported that energy density is associated with obesity( Reference James 9 , Reference Bes-Rastrollo, Van Dam and Martinez-Gonzalez 38 – Reference Mendoza, Drewnowski and Christakis 40 ); however, various other studies have reported no association( Reference Bes-Rastrollo, Van Dam and Martinez-Gonzalez 38 , Reference Stookey 41 ). This inconsistency was partly related to the non-uniformity in calculating energy density, particularly in the treatment of beverages; inclusion of all food items and beverages weakened the association between energy density with the outcome( Reference Ledikwe, Blanck and Khan 25 ) because the weight of beverages increased the day to day variations in the subject’s food consumption. In this study, the weight of beverages were excluded from the energy density calculations; however, this method also had a limitation because a subject who primarily drinks water would have high energy density values, and the effect of energy-containing beverages may be underestimated. However, the food consumed by the population under consideration in this study had relatively high energy density compared with that of the American population (6·7–8·45 kJ/g (1·6–2·02 kcal/g)) depending on age and calculated using similar method)( Reference Ledikwe, Blanck and Khan 25 ). Therefore, despite the lack of an association, high energy density food consumption should be considered a crucial obesity risk factor for this population.
Two dietary patterns were observed in this study and subjectively named more healthy and less healthy patterns. The food groups in the ‘more healthy’ pattern were considered healthier compared with those in the ‘less healthy’ pattern because the more healthy food groups included less or moderate oil and fats. In contrast to the ‘more healthy’ pattern, most food groups in the ‘less healthy’ pattern contained food items with oils and fats, mainly because of oil and food processing. Besides fried snacks and fried soyabean products, roots and tubers and meat and poultry products were the most consumed fried foods.
After adjustment for energy intake, energy density and other socio-demographic variables, the two dietary patterns showed an association with the BMI status. Subjects with a high quintile score in the ‘more healthy’ pattern had a lower risk of obesity than did those with a low score. Consistent with this result, subjects with a high quintile score in the ‘less healthy’ pattern had a high risk of obesity than did those with a low score. These results are consistent with those of similar studies from other countries, which have reported that intake patterns with a high intake of meat, fat, oil and sweets were associated with an increased risk of overweight and obesity, whereas a ‘healthy pattern’ with a high intake of vegetables, fruits, fish and whole grain was associated with a lower risk( Reference Okubo, Sasaki and Murakami 42 – Reference Dugee, Khor and Lye 44 ). However, in some other countries, considering the two dietary patterns, no association between the BMI and risk of obesity has been observed( Reference Song, Joung and Engelhardt 45 , Reference Shi, Hu and Yuan 46 ). These inconsistencies( Reference Togo, Osler and Sorensen 47 ) are likely attributable to the differences in the classification of foods and in the dietary assessment and analytical methods used.
The identified association became stronger after adjustment for energy intake, suggesting that the dietary pattern is a stronger influencer than is energy intake. Dietary patterns give a broader picture of food and nutrient consumption for studying the association between diet and a health outcome. Dietary patterns describe the whole diet, including the potential synergetic effects of foods and nutrients( Reference Hu 48 ).
Strengths and limitations of the study
This study was among the very few that have characterised food intake among obese and normal-weight women in urban slum areas in Indonesia. We used a comprehensive procedure to ensure clearer estimation of the food portion size and nutrient content, such as weighting duplicate food portions reported by the subjects and the cooking approach for foods with no available data on their nutrient contents in the food composition database. However, there were several weaknesses that may have influenced our study. The cross-sectional design of this study limited the ability of assessing the causal relationship between food intake and obesity status. Moreover, the physical activity assessment and 24-h food recall were prone to underestimation and overestimation; therefore, a study with specific settings is required to ensure validity and reliability of physical activity assessment. Alves et al.( Reference Alves, Falcão and Pinto 19 ), who assessed physical activity among obese women in urban slum areas in Brazil using the IPAQ 7-d short version, the same as that used in this study, showed that the assessment tools cannot distinguish leisure, occupation, household or transportation activities, which would had been useful for understanding behaviour patterns in the slum population( Reference Khani, Ye and Terry 24 ).
Conclusion and recommendation
The over-nutrition problem in our study population was severe, and the significance of food consumption in the risk of obesity was identified. Obese WRA had higher energy, carbohydrate and fat intakes than normal-weight WRA. Two dietary patterns were identified in this study: the ‘more healthy’ and ‘less healthy’ patterns. The ‘less healthy’ pattern was characterised by consumption of fried foods (snacks, soyabean, roots and tubers and meat and poultry products), whereas the ‘more healthy’ pattern was characterised by consumption of seafood, vegetables, eggs, milk and milk products and non-fried snacks. The dietary patterns were significantly associated with obesity. Furthermore, a high quintile score of the ‘more healthy’ pattern was associated with a low risk of obesity, whereas a high quintile score of the ‘less healthy’ pattern was associated with a high risk of obesity. Therefore, to overcome the obesity problem in this population, educating people on nutritional values is required for improving their diet and the type and portion of food consumed. Moreover, the population in urban slums have limited access to various food choices; therefore, investigating the association between food environment and dietary patterns is warranted. Consistent with the association of high energy intake with obesity, physical activity should be increased to compensate for the higher energy intake. Additional studies are required for validating the assessment of physical activity measurements among the study population.
Acknowledgements
The authors are grateful to the project team and thank all the study subjects for their participation in the study.
This study was financially supported by the Indonesian government through SEAMEO RECFON, University of Indonesia.
Y., H. K. and U. F. contributed towards developing the proposal, methods, guidelines and data analyses. Y. collected and analysed the data and was responsible for the first draft of the manuscript, and H. K. contributed substantially towards revising the manuscript. All authors contributed to revising the manuscript and approving the final version.
The authors declare that there are no conflicts of interest.
