The Millennium Development Goals (MDG) are time-bound and quantified targets for addressing extreme poverty and hunger in its many dimensions across the low-income regions of the world, including income poverty, hunger, health, education and gender disparities(1, 2). While the proportion of undernourished people is decreasing slowly in many high-income countries of the world, in most of the low-income regions the proportion has risen over the past decade(2). According to the WHO(3), almost one-third of the total population in low-income regions of the world suffers from malnutrition. However, the figures vary across age groups. In Taiwan, the prevalence of malnutrition among elderly persons aged 60 to 80 years is reported to be between 2 % and 5 %(Reference Tsai, Chang, Lin, Chuang, Lin and Lin4). In Malaysia, 38 % of rural elderly people are found to be malnourished according to BMI, using a cut-off of <18·5 kg/m2 as the definition of undernutrition(Reference Suzana, Earland, Suriah and Warnes5). Notably, such findings are comparable with findings in elderly frail institutionalized individuals from high-income countries(Reference Ödlund, Koochek, Ljungqvist and Cederholm6, Reference Saletti, Johansson, Yifter-Lindgren, Wissing, Österberg and Cederholm7).
Elderly people are vulnerable to disease onset(Reference Joshi, Kumar and Avasthi8, Reference Prakash, Choudhary and Singh9), and diseases increase the risk of malnutrition(Reference Hickson10). Both the duration and the type of disease are independently associated with nutritional status among elderly people(Reference Cederholm, Jägren and Hellström11, Reference Cederholm, Jägren and Hellström12). Disease affects people’s food intake due to e.g. pain, anorexia, nausea and restricted mobility(Reference Morley13), with malnutrition as a subsequent occurrence. Moreover, malnutrition is found to be associated with several neuropsychological problems e.g. depressive symptoms(Reference Cabrera, Mesas, Garcia and de Andrade14) as well as impaired cognitive function(Reference Clarke, Birks, Nexo, Ueland, Schneede, Scott, Molloy and Evans15).
Poverty is a strong predictor of poor health(Reference Benzeval and Jodge16), and malnutrition is common among older persons who live in poverty(Reference Chen, Schilling and Lyder17). A recent study in urban slums of Bangladesh found significant associations between poor socio-economic status and poor nutritional status in adults(Reference Pryer and Rogers18). Positive associations between socio-economic status and nutritional status were also reported among a group of elderly people in Malaysia(Reference Suzana, Earland, Suriah and Warnes5).
Social factors, e.g. lack of social support, having a limited social network or being socially isolated, are also associated with nutritional status in elderly individuals(Reference Locher, Ritchie, Roth, Baker, Bodner and Allman19, Reference Pirlich, Schutz, Kemps, Luhman, Minko, Lubke, Rossnagel, Willich and Lochs20). Although research shows that malnutrition is related with disease(Reference Hickson10–Reference Cederholm, Jägren and Hellström12), socio-economic(Reference Suzana, Earland, Suriah and Warnes5) and social factors(Reference Locher, Ritchie, Roth, Baker, Bodner and Allman19, Reference Pirlich, Schutz, Kemps, Luhman, Minko, Lubke, Rossnagel, Willich and Lochs20), the relative contribution of the various types of factors to nutritional status of the elderly population in low-income countries is not well recognized.
Thus, the main aim of the present study was to investigate the associations of nutritional status with illness, socio-economic and social indicators, as well as their relative impact, on nutritional status among persons aged 60 years and over in rural Bangladesh.
Data were collected in a multidisciplinary, cross-sectional study of elderly people aged 60 years or more during August 2003 to January 2004. The study was conducted in Matlab, a rural area 55 km south-east of Dhaka city in Bangladesh where icddr,b: Knowledge for Global Lifesaving Solutions, Dhaka, Bangladesh (formerly known as the International Centre for Diarrhoeal Disease Research, Bangladesh) has been maintaining a Demographic Surveillance System (DSS) since 1966. The data collection methods have been described elsewhere(Reference Kabir, Ferdous, Cederholm, Khanam, Streatfield and Wahlin21).
A power calculation was performed to estimate the sample size according to effect variables such as functional ability before the study was conducted. Based on the calculation, a total of 850 elderly (≥60 years of age) individuals were randomly selected from the surveillance database. Among them, sixty-three died between sample selection and the start of data collection, thirty-eight refused to participate, eleven migrated, ninety-three could not be reached, eighteen were registered twice and, while cross-checking age, two persons were found to be below 60 years. A total of 625 subjects participated in the home interviews, 473 of whom underwent clinical examination at a medical centre. Information on complete nutritional status was available for 457 individuals. All participants were informed of the study objectives, and informed consent was obtained before home interviews and clinical examination. A total of 152 participants who were interviewed did not participate in the clinical examination. These drop-outs have been described previously(Reference Kabir, Ferdous, Cederholm, Khanam, Streatfield and Wahlin21); they were mainly women, older and had poor socio-economic status, but their health profile was not different from the respondents. The study was approved by ethics committees at icddr,b and Karolinska Institutet, Stockholm, Sweden.
Assessment of nutritional status
Nutritional status was assessed using the Mini Nutritional Assessment (MNA)(Reference Guigoz, Vellas and Garry22). The MNA comprises eighteen questions regarding anthropometry (e.g. BMI, kg/m2), global assessment (e.g. living independently, occurrence of acute disease or psychological stress, co-morbidity and neuropsychological problems), dietary assessments (e.g. number of whole meals consumed per day, consumption of dairy products, fruits and vegetables) and a subjective assessment regarding self-perception of nutritional and health status. The maximum score is 30. The MNA has been validated and mostly used for elderly people in high-income countries(Reference Ödlund, Koochek, Ljungqvist and Cederholm6, Reference Saletti, Johansson, Yifter-Lindgren, Wissing, Österberg and Cederholm7, Reference Pirlich, Schutz, Kemps, Luhman, Minko, Lubke, Rossnagel, Willich and Lochs20). As a result, some of the items in MNA, such as the cut-offs for BMI, are not relevant for elderly persons in a low-income setting. The present study used the BMI cut-off suggested by the WHO to classify adults as underweight with BMI < 18·5 kg/m2, a definition that has been used in other similar studies(Reference Kuzuya, Kanda, Koike, Suzuki, Shosuke and Lguchi23, Reference Visvanathan, Zaiton, Sherina and Muhamad24). In addition, due to lack of information on calf circumference in our study and the fact that nursing homes for ill elderly persons do not exist in Bangladesh, two MNA items were dropped. Considering this, a modified version of MNA(Reference Kabir, Ferdous, Cederholm, Khanam, Streatfield and Wahlin21), where the maximum score is 28, was used in the present study. A score of <15 indicates undernutrition, a score of 15 to 21·5 indicates risk of malnutrition and a score of ≥22 indicates that the person is well-nourished.
Medical diagnoses, assessment of depressive symptoms and cognitive function
Clinical examinations were conducted by physicians and included medical history, physical examination and collection of blood. Based on this information, medical diagnoses of each participant were decided by the physicians. A total of fifteen medical diagnoses were identified among the participants. In order to facilitate statistical analyses, medical diagnoses were assembled into the following categories: (i) acute infections, e.g. respiratory tract infection, symptoms of helminthiasis, leucorrhoea (i.e. vaginal discharge); (ii) chronic illnesses, e.g. arthritis, obstructive pulmonary symptoms, heart failure; and (iii) gastrointestinal disorders, e.g. gingivitis, inguinal hernia. Sensory impairments were defined as hearing impairment or impaired vision.
In order to assess depressive symptoms we used the Self-Reporting Questionnaire 20 items (SRQ20), which was originally developed by Harding et al.(Reference Harding, de Arango, Baltazar, Climent, Ibrahim, Ladrido-Ignacio, Murthy and Wig25) as a screening tool for common mental disorders in primary health care. The interviewers administered the twenty questions and the answers were coded as yes (1) or no (0). For the present purposes we calculated a symptoms summary score: 0–20 points, where higher scores indicate more depressive mood.
To assess cognitive function we used the Bangla Adaptation of Mini-Mental State Examination (BAMSE), originally developed by Kabir and Herlitz(Reference Kabir and Herlitz26). This instrument is able to assess cognitive function of elderly individuals irrespective of their literacy levels. The total score in BAMSE is 30 and higher scores indicate better cognitive ability.
Assessment of severity of disease
Low serum albumin concentration is a strong predictor of mortality and negative clinical outcomes(Reference Don and Kaysent27–Reference Sullivan, Roberson and Bopp29). In this sense and also suggested by Vellas et al.(Reference Vellas, Hunt, Romero, Koehler, Baumgartner and Garry30), serum albumin can be used as an indicator of disease severity. Cut-off values for serum albumin concentrations were based on the percentile distribution. The 25th percentile (35 g/l) was used as cut-off for normal level(Reference Tietz31), between the 10th and 25th percentile (33–34 g/l) was used as the indicator of low level and below the 10th percentile (<33 g/l) was defined as very low level of serum albumin. The levels of serum albumin were then coded as: normal = 1; low = 2; very low = 3.
A score of severity of disease was calculated for each participant as the product of serum albumin (1–3) and the number of medical diagnosis categories (1–3). Since the severity of depression and sensory impairments are not reflected by serum albumin levels, the severity of disease score included only acute infections, chronic illnesses and gastrointestinal disorder. At least one diagnosis in any of the three categories was considered to calculate severity score. Increasing product numbers from 0 to 9 indicate increasing burden of disease.
Demographic and socio-economic variables
Information on age, sex and literacy of each participant was collected from the DSS database. Information on monthly income e.g. from employment or pension fund, marital status and financial support from relatives was collected during the home interviews. Information on per capita daily household expenditure on food was calculated by using daily household expenditure on food as numerator and number of household members as the denominator.
The social network variable was created based on the number of children living in the same household and the number of children living in the same bari (a bari is a group of households, normally comprising members of the same family or close kin, sharing a common courtyard). It was categorized as follows: (i) very good social network = one or more children living in the same household and one or more children living in other households of the same bari (n 133); (ii) good social network = one or more children in the same household and none in other households of the same bari (n 187); (iii) poor social network = no children in the same household but one or more children in other households of the same bari (n 72); and (iv) very poor social network = no children living in either the same household or the same bari (n 65).
Descriptive analyses were performed to report the distribution of data and independent t tests were done to compare mean differences between men and women. In order to evaluate the relationship between potential predictor variables and MNA, hierarchical linear regression analyses were conducted. The following potential independent covariates were used in the regression analyses: categories of medical diagnoses, the severity of disease score, depressive symptom score, BAMSE score, income that was coded as having an income or no income, literacy coded into illiterate and literate. Marital status was coded as currently married or single. Financial support was coded as receiving regular support or not receiving any regular support. As previously described, the social network variable was labelled from very good social network to very poor social network (coded as very good = 3; good = 2; poor = 1; very poor = 0). In all analyses, age was used as a continuous variable. All statistical analyses were performed using the SPSS for Windows statistical software package version 15·0 (SPSS Inc., Chicago, IL, USA).
Demographic, socio-economic and social characteristics
As shown in Table 1, more than half of the respondents were women. Around one-third of the participants were literate and significantly more men than women were literate (P < 0·001). The majority of participants had no income, with significantly more women than men reporting lack of income (P < 0·001).
IQR, interquartile range; MNA, Mini Nutritional Assessment.
Significant difference compared with men: ***P < 0·001.
†Personal income from employment or pension fund.
‡Taka 69 = $US 1 during the study period.
§Support from relatives.
||Data missing for five individuals.
¶Data missing for four individuals.
A notably high proportion of the men were married, whereas a majority of the women were single (P < 0·001). Most of the participants received regular financial support and significantly more women than men reported receiving such support (P < 0·001).
Almost a third of the participants reported having either a poor or a very poor social network. More men than women reported having a good social network, i.e. one or more of their children lived in the same household. However, significantly more women than men reported having a very good social network (P < 0·001), meaning that one or more of their children lived in the same household as well as in the same bari.
Almost all participants had a mix of medical diagnoses (Table 2). The majority (71 %) of the participants had two to five medical diagnoses, and about 15 % had six or more diagnoses. More than three-quarters of the participants suffered from at least one acute infection (e.g. helminthiasis), 66 % were suffering from chronic illnesses, mainly arthritis; 81 % had gastrointestinal disorders dominated by upper alimentary tract disorder; and about a third displayed impairments in at least one sensory function. No significant sex differences were found in the distribution of acute infections or chronic illnesses. However, women were more likely than men to suffer from gastrointestinal disorders (P < 0·01), sensory impairments (P < 0·001) and depressive symptoms (P < 0·001).
SRQ20, Self-Reporting Questionnaire 20 items; BAMSE, The Bangla Adaptation of Mini-Mental State Examination; IQR, interquartile range.
Significant difference compared with men: **P < 0·01, ***P < 0·001.
†A symptoms summary score of the SRQ20 to assess depressive symptoms. SRQ20 is a screening tool for common mental disorders in primary health care. The tool comprises twenty questions; the answers to each question were coded as yes (1) or no (0), giving a total score of 20 with higher scores indicating high depressive symptoms.
‡Calculated as the number of medical diagnosis categories (1–3; i.e. acute infections, chronic illnesses or gastrointestinal disorders) multiplied by the level of serum albumin (1–3, where 1 = normal; 2 = low; 3 = very low).
§Data missing for two individuals.
||Data missing for fifteen individuals.
¶Data missing for fourteen individuals.
The mean serum albumin levels were significantly lower among participants who had acute infections (36·4 (sd 2·9) g/l v. 37·1 (sd 2·6) g/l; P < 0·05) or gastrointestinal disorders (36·4 (sd 2·9) g/l v. 37·1 (sd 2·5) g/l; P < 0·05) compared with participants who did not have these illnesses. However, serum albumin concentrations did not vary significantly across diagnoses within categories. Mean serum albumin level was significantly lower in women than in men (P < 0·001). The median score of severity of disease was 3.
Determinants of nutritional status
According to the MNA and as reported previously(Reference Kabir, Ferdous, Cederholm, Khanam, Streatfield and Wahlin21), a quarter of the elderly participants were classified as undernourished (Table 1). About two-thirds were found to be at risk of malnutrition, and 12 % of the participants were well-nourished.
Table 3 presents results of the hierarchical linear regression analyses which were performed to identify indicators independently associated with nutritional status. In these analyses, demographic information (sex and age) was entered in the first step. Next, disease-related indicators (i.e. categories of medical diagnoses) were entered in the second step, severity of disease score was entered in the third step, SRQ20 scores were entered in the fourth step and sensory impairments were entered in the fifth step. Information on socio-economic and social indicators (income, literacy, marital status, per capita daily household food expenditure, financial support and social network) were entered in step six. In addition, the analyses were repeated, reversing the entry order of the predictors, such that the socio-economic and social indicators were entered in the second step followed by disease-related indicators as before.
SRQ20, Self-Reporting Questionnaire 20 items.
†Calculated as number of medical diagnosis categories (1–3; i.e. acute infection, chronic illness and gastrointestinal disorder) multiplied by level of serum albumin (1–3, where 1 = normal; 2 = low; 3 = very low).
‡A symptoms summary score of the SRQ20 to assess depressive symptoms. SRQ20 is a screening tool for common mental disorders in primary health care. The tool comprises twenty questions; the answers to each question were coded as yes (1) or no (0), giving a total score of 20 with higher scores indicating high depressive symptoms.
Results indicate that, among the demographic indicators, being a woman was significantly associated with lower MNA score (P = 0·012). Among the disease-related indicators, acute infections (P < 0·001), gastrointestinal disorders (P = 0·007), higher disease severity score (P < 0·001) and increasing number of depressive symptoms as reflected by higher SRQ20 score (P < 0·001) were associated with lower MNA score. Among the socio-economic and social indicators, having an income, being literate and receiving regular financial support were significantly associated with higher MNA score (P = 0·009, P = 0·002 and P = 0·018, respectively). The resulting R 2 change for disease-related indicators combined and for socio-economic and social indicators combined was 0·205 and 0·038, respectively. Among the disease-related indicators, depressive symptoms accounted for 9 % of the variation in MNA scores. As a whole, the model explained 26·4 % of the variance. By reversing the entry order of predictors, the analyses resulted in almost identical results (data not shown).
Next, we wanted to evaluate the possible contribution of cognitive dysfunction to the MNA outcome. Since the cognitive variable in the MNA questionnaire was based on the BAMSE scores, this item, i.e. neuropsychological problem, was excluded from the MNA. The total MNA score was recalculated accordingly and the maximum score was now 26. Since literacy is a strong predictor of cognitive function(Reference Kabir and Herlitz26), BAMSE scores were entered as the last variable in the regression model after controlling for all other possible predictors including literacy. This procedure added another 2 % of explained variation in MNA scores, where better cognitive function as reflected by higher BAMSE score was significantly associated with higher MNA score (β = 0·16, P = 0·001; data not shown).
In the present study, the impact of disease, age, sex, socio-economic and social indicators on nutritional status, as measured by MNA, was examined. The main findings were that being a woman, suffering from acute infections, gastrointestinal disorders, high burden of disease, depressive symptoms and cognitive impairment, low income, illiteracy and absence of regular financial support were all associated with lower MNA score.
Almost all of the participants in our study suffered from at least one medical condition. A cross-sectional study in northern India reported that 89 % of the study participants (60+ years) were ill(Reference Joshi, Kumar and Avasthi8), a finding comparable with the findings in the current study. The range of illnesses found in the present study was wide, i.e. from acute infections to chronic illnesses as well as gastrointestinal disorders, which is in line with the findings of previous reports(Reference Prakash, Choudhary and Singh9, Reference Mostafa and Streatfield32). Mostafa and Streatfield(Reference Mostafa and Streatfield32) found that older persons in rural Bangladesh suffer from both chronic (76 %) and acute illnesses (51 %). They also observed that arthritis was the most prevalent chronic illness, whereas cough was the most prevalent acute illness(Reference Mostafa and Streatfield32). Disease has a negative impact on nutritional status(Reference Hickson10). As seen in the current study, malnutrition was linked to acute infections(Reference Wikby, Ek and Christensson33) as well as to gastrointestinal diseases(Reference Norman, Pichard, Lochs and Pirlich34).
In order to be able to grade severity of illness, we constructed a severity of disease score combining the disease prevalence and serum albumin level. After controlling for all categories of medical diagnoses, the severity of disease score showed significant associations with nutritional status in the regression analyses. This indicates that not only specific illnesses but also the combined burden of disease is independently associated with nutritional status among elderly individuals. Serum albumin has often been used as an indicator of nutritional status(Reference Omran and Morley35), but this interpretation of serum albumin has been questioned. Even in the most severe cases of anorexia nervosa, serum albumin levels remain normal(Reference Krantz, Lee, Donahoo and Mehler36). Hypoalbuminaemia may develop in association with advanced liver disease, severe congestive heart failure, nephritic syndrome and protein-losing enteropathies. Most often low serum albumin represents ongoing inflammation(Reference Sullivan37). Inflammation reduces hepatic albumin production in favour of acute-phase protein production(Reference Don and Kaysent27) and inflammation leads to capillary leakage of albumin. Hypoalbuminaemia is known to be an important predictor of morbidity(Reference Peralta, Rubery and Guzofski28). For these reasons, serum albumin was used in the present study to calculate a ‘severity of disease’ score rather than as a biochemical indicator of nutritional status. We believe that a factor that combines the occurrence of illness and the severity of illness better reflects health status than the diagnoses on their own.
Several studies have identified depression as one of the leading factors associated with weight loss(Reference Morley and Kraenzle38, Reference Chen39), and depressive symptoms have been suggested as one of the most important factors influencing nutritional status(Reference Chen39) in hospitalized(Reference German, Feldblum, Bilenko, Castel, Harman-Boehm and Shahar40) as well as community-dwelling elderly people(Reference Cabrera, Mesas, Garcia and de Andrade14). Considering the differences in the study populations, our findings confirm the observations made previously. Depressive symptoms as reflected by SRQ20 score were found to be a strong predictor even after controlling for other medical diagnoses. The gender differences in depressive symptoms were perhaps expected, as previous reports show similar findings(Reference Cabrera, Mesas, Garcia and de Andrade14). The findings might be explained by the poor socio-economic status, poor health status and lack of support observed among the female participants(Reference Chen, Wei, Hu, Qin, Copeland and Hemingway41).
Chen et al.(Reference Chen, Schilling and Lyder17) have pointed out that elderly persons who live in poverty are at greater risk of malnutrition. Supporting this well-known and obvious fact, nutritional status was positively associated with income and financial support in our study as well. Poverty, as reflected by lack of income and absence of regular financial support in the current study, may indeed lead to poor consumption because of inability to afford food rich in nutrients. As also seen in the current study, level of education influences the health and nutritional status of elderly people(Reference Pirlich, Schutz, Kemps, Luhman, Minko, Lubke, Rossnagel, Willich and Lochs20). Persons who are educated may be better informed about the nutritional quality of their diet and also about the impact of a nutritious diet on their health than less educated persons.
The gender aspect of malnutrition is not always recognized. In low-income countries, the nutritional deficiency that is observed among the general population may be even more evident in older women(Reference Bonita42). The current study lends support to such observations. While discussing the ageing population in India, Prakash(Reference Prakash43) reported low social status, discriminatory practices based on gender, early age in marriage, food taboos, multiple pregnancies and lack of attention to health to be responsible for the poor health of older women in India. This scenario may reflect the situation of older women in Bangladesh as well, where social and cultural practices are similar to those in parts of India. Poor health and low well-being in old age of rural women in many low-income countries is a result of high burden of work and insufficient food intake resulting in nutrition deficiency(Reference Bonita42). The current study found women to be a vulnerable group. Compared with men they had worse health status, negative socio-economic and social conditions as well as poor nutritional status. This scenario may not be unexpected, especially since older women in Bangladesh are more likely to be poor. A previous study in Bangladesh found that 80 % of rural elderly women reported being in unpaid work, which was mostly restricted to household work(Reference Kabir44). Receiving pension or social benefits in old age is not common in rural Bangladeshi society. Being involved in unpaid work or not receiving any social benefits leads these persons into economic insecurity, which may prevent ageing women from receiving better health care or from satisfying their hunger.
An interesting finding in the current study was the lack of association between social network and nutritional status. Although several studies have reported that socialization is associated with food intake, which leads to better nutritional status also for elderly people(Reference Pirlich, Schutz, Kemps, Luhman, Minko, Lubke, Rossnagel, Willich and Lochs20, Reference Martin, Kayser-Jones, Stotts, Porter and Froelicher45), our results do not support such findings. A possible reason for this may be that although the elderly persons have good social networks as defined here by children living in the same bari and/or household, they may share the same poor nutritional status as the younger household members. Another unexpected finding was that our results did not reveal any significant association between age and nutritional status.
Interestingly, although we found a significant and direct association between cognitive function and nutritional status, the BAMSE score added only 2 % of the variation in MNA scores. A possible explanation for this can be that in the context of Bangladesh older persons in general live with their children. If not always in the same household many live in the same bari as their children, thus having direct access to physical care from family members in contrast to the situation in Western societies. Consequently, the implications of impaired cognitive function in terms of inability to cook, shop, mix or eat food are taken care of by families or relatives in the Bangladeshi context. The small contribution of cognition to malnutrition may also be explained by the larger contribution by depressive symptoms. The dual association between depression and cognition is well known.
Some limitations of our study must be acknowledged. The construction of medical diagnoses categories is subjective and may not necessarily reflect the best categorization. Also, the severity of disease score that we used to indicate the burden of illnesses is not validated and to our knowledge has not been used in any other studies yet. Moreover, since it is a cross-sectional study, the results should be interpreted with caution.
Most public health research and interventions in low-income countries currently focus on the MDG(1), with the intention being to achieve these goals and meet the target by the year 2015. The present study clearly indicates that in low-income countries like Bangladesh, malnutrition is related to ill health as well as socio-economic indicators such as lack of income, illiteracy and absence of financial support. To focus only on the eradication of extreme poverty and hunger may not be enough to reduce the proportion of the undernourished population, which is the first of the eight goals mentioned in MDG. Indeed, socio-economic and social indicators, i.e. poverty and illiteracy, had a negative impact on the nutritional status of elderly individuals in the current study, but the impact was rather small compared with the disease indicators. Three out of six socio-economic indicators accounted for only 4 % of the variation in MNA scores compared with 20 % of the variation explained by a variety of health problems. Hence the disease situation also needs to be improved by better health-care facilities. The present study clearly shows that malnutrition is a combined result of medical illness, poverty, illiteracy and lack of financial support and that all these indicators make separate contributions to explain variation in nutritional status. A consequence of these findings is that the combat against malnutrition should likewise most probably gain by adapting a multi-factorial strategy.
In conclusion, the present study suggests that ill health and socio-economic and social indicators are independently associated with malnutrition and that these indicators combined are important for explaining nutritional status among elderly persons. However, both longitudinal data and further studies on the functional consequences of nutrition, ill health and socio-economic indicators are needed to provide more detailed information about the complex associations among malnutrition and the predictors studied here.
The study was supported by grants from the Department for International Development, UK to icddr,b: Knowledge for Global Lifesaving Solutions, and from the Swedish Medical Research Council and the Swedish International Development Agency to icddr,b and the Karolinska Institutet. None of the authors has any conflict of interest. All authors were engaged in planning of the study. T.F., T.C., Z.N.K. and Å.W. analysed the data. T.C., Z.N.K. and Å.W. contributed to the interpretation of the analyses. T.F. had the main responsibility of writing the manuscript and all authors provided direct input in the writing of the manuscript.