A study of the geographical clustering of districts in Uttar Pradesh using nutritional anthropometric data of preschool children.

BACKGROUND &
OBJECTIVES: Worldwide variations in human growth and its genetic and environmental factors have been described. In this study, an attempt was made to assess the morphological differences and similarities among under 5 year children of rural areas of Uttar Pradesh State in India, and to determine differences or similarities of body size among children living in diverse regions.
METHODS: For this purpose, a cross-sectional district nutrition profile study conducted during 2002-2003 was used. The data on 10,096 children drawn from 1080 villages in 54 districts were part of the district level Diet and Nutrition Assessment survey. The mean values for height and weight for 54 districts were taken as the input data for subsequent analysis. The data were first normalized by means of principal component analysis (PCA) and then K-means clustering was performed.
RESULTS: The PCA and cluster analysis yielded four distinguishable clusters or patterns in the anthropometric data of children. These clusters were ordered according to the average body size (weight and height) of children. The mean stature and body weight of these children in cluster I were 3.2 cm and 1.4 kg higher than those of cluster IV indicating differences between clusters. Also, the variations between clusters in their social, demographic, health and nutrition parameters were compared. INTERPRETATION &
CONCLUSIONS: The use of PCA and cluster analysis methods and their merits in studying the Uttar Pradesh preschool children growth variations are discussed. These results helped in identifying the districts with higher prevalence of undernutrition and the contributing factors.

Growth and development of children in a community are largely influenced by the environment they live in, which includes a host of factors related to socio-economic, socio-cultural and agro-climatic conditions. Worldwide variations in human growth patterns have been described12. In this study, an attempt was made to extract the nutritional patterns of under-five children using anthropometric data collected at district level in the State of Uttar Pradesh in India. This study was undertaken particularly in Uttar Pradesh because it is the most populous State in India, having about 200 million population3. Therefore, geographic clusterization of districts in large areas such as a State or region of a country based on nutritional status of the children may help in identifying factors that have significant influence on the growth and development of the communities and to design and implement appropriate Region/State specific strategies for prevention and control of undernutrition.

The cluster analysis technique is an easily replicable way of constructing classifications, which has attracted widespread popularity across diverse scientific disciplines4. Mahalanobis et al5 employed this method to make an anthropometric survey of the united province in 1949. Vasulu and Pal6 studied the relationship between anthropometric differentiation and cultural diversity in the Yanadi tribe in different regions of India. This method has been successfully used in China on anthropometric measurements to classify growth profiles of children7 and in India to carry out social marketing strategies for control of vitamin A deficiency89 as well as to identify the patterns in nutritional data of children. Monda and Popkin10 used cluster analysis to create patterns of overall activity and inactivity in a diverse sample of Chinese youth and to evaluate its use in predicting overweight status. Dietary patterns of different populations in US have been reported using principal component analysis (PCA)-cluster analysis technique11 as also the obesity patterns of Hungarian children12.

District nutrition profile (DNP) survey carried out in the State of Uttar Pradesh, India jointly by National Institute of Nutrition (NIN), Hyderabad, and Institute of Applied Statistics and Development Studies, Lucknow, India during 2002-200313 and reported during 2003-2004. The study collected information included collection of data on household socio-economic, socio-cultural and demographic particulars, nutritional status of individuals in terms of anthropometry and clinical examination, average food and nutrient intakes at the household level and breast feeding and child rearing practices prevalent in the community with the following specific objectives: (i) to assess the food and nutrient intake of different segments of the population of the State; (ii) to assess the nutritional status of representative segments of population in terms of anthropometry and clinical status; and (iii) to assess the knowledge and practices of mothers on breastfeeding, child rearing and socio-cultural aspects of food consumption in relation to health and disease.

In the present study, the data on anthropometry and socio-economic variables were utilized from the DNP survey with the following objectives: (i) to form geographical clusters in the State of Uttar Pradesh, based on the anthropometric data of weight and height of preschool children; (ii) to identify differences or similarities in the nutritional status of children living in different clusters; and (iii) to study the household demographic, socio-economic differentials of the children between the clusters.

Material & Methods

Sample design: A cross-sectional design was used for carrying out DNP survey. In each district, village formed the Primary Sampling Unit (PSU) and the Households (HHs), the Secondary Sampling Unit (SSU). Thus, a total of 400 HHs were covered from 20 villages by covering 20 randomly selected households from each village. Considering the large variations in the district, due representation was given to all the blocks in the district while selecting the villages, by adopting stratified random sampling procedure coupled with probability proportion to size (PPS).

Subjects: The anthropometric data, viz. heights and weights available on 10,096 preschool children (1 to 5 yr of age) from a total of 87,491 individuals of different ages of both the sexes from 54 districts of Uttar Pradesh was considered for analysis. The geographic distribution of the districts is shown in Fig. 1. The mean values for height and weight for 54 districts were computed which consists of 10,096 preschool children was taken as the input data for the purpose of analysis.

Fig. 1

Geographical distribution of UP State (54 districts).

Variables: Two anthropometric measurements viz. heights and weights were collected using standard equipment and procedures. The investigators were trained and standardized in the survey methodologies by the scientists of NIN, before initiating actual data collection in the field. Two anthropometric measured were used viz; (i) height (cm); and (ii) body weight (kg).

Statistical methods: Descriptive statistics and one way analysis of variance with post hoc comparisons were computed. PCA was used as it captures most of the variation and co-variation in multivariate data through a few combinations of the original variables having key objectives of data reduction and interpretation. The principal components were extracted so that the first few principal components denoted as PC(1), PC(2)....PC(k), accounted for a large fraction of the total variation in the data (the components themselves being mutually uncorrelated).

The data were first normalized by means of PCA and then clustering was performed using SPSS 19.0 statistical software14, using the following procedure: (i) The means of each variable for 54 districts were computed. (ii) An inter-variable correlation coefficient matrix was derived. (iii) The PCA was extracted by adopting the following criteria: (a) criterion for accuracy of selecting principle component was 0.005, (b) the minimum variance for extracting each component was 0.5, (c) the value of all measures was transformed into principle-component scores.

The cases were clustered by k-means cluster method15 using Euclidian distance which was calculated by the formula given below.

Wherein dij is the distance between any two cases (I and j) in a group, Xik and Xjk are the principal component scores of the kth principal component (k=1, 2, 3…m; here m=1).

The procedure for clustering was done by MacQueen method16 as follows:

Step 1: Partition the items into k initial clusters

Step 2: Proceed through the list of items, assigning item to the cluster whose centroid (mean) is nearest. Re-calculate the centroid for the cluster receiving the new item and the one in which the item is removed

Step 3: Repeat step 2 until no reassignments take place.

Assessment of nutritional status: Nutrition and health factors: The extent of undernutrition among pre-school children was assessed by Standard Deviation (SD) classification by using World Health Organization(WHO) growth standards17, in terms of stunting (height for age,

Results

The mean heights and weights by district are listed in Table I. Only one principle component could be extracted from the data. The ‘Eigen’ value resulting from this component was ≅ 1.4 and could explain 70 per cent of the variation (Table II).

Table I

District wise mean and standard deviation of height and weight of Uttar Pradesh pre-school (1-5 yr) children

Table II

Principal components extracted from anthropometry

The data were first transformed into PC score for the 54 districts which formed the input for performing cluster analysis using K-means cluster analysis method. Four different clusters or patterns were observed in the data. A visual representation cluster analysis represented as dendogram (Fig. 2) shows the clusters being combined and the values of the distance coefficients at each step. Looking at the dendogram, it appears that the four cluster method described may be appropriate, since the clusters are easily interpretable and occurs before the distance at which cluster becomes very large. The dendrogram rescales the actual distances to numbers between 0 and 25, preserving the ratio of the distances between steps.

Fig. 2

Dendogram of cluster analysis.

These clusters were ordered according to the average body size (weight and height) of children. The body sizes which formed into different small clusters are listed in Table III. The mean stature and body weight of these children in cluster I was 3.2 cm and 1.4 kg higher than those of cluster IV indicating difference between clusters.

Table III

Comparison of weight and height for different clusters

Geographic factors: The cluster analysis which resulted in the formation of four distinguishable clusters is presented in Fig. 3. The clustered groups as depicted on the map by different legends, showed a gradual decline in the nutritional status of the children. The cluster I on the top of the map showed prosperous growth as compared to others. The map showed the geographic dissimilarities in the body size of children. Many of the districts for which the data were clustered, were geographically adjacent. Most of the districts under cluster I were from the western part of the State, such as Ghaziabad, Farrukhabad, Etawah, which is relatively prosperous region of the State. The majority districts viz. Bijnor, Saharanpur, Meerut, Aligarh, Mathura, Maharajgunj, etc. grouped into cluster II ranked as second best segregation, are from Western and Eastern regions of the State considered to be developed regions. In contrast, in cluster IV having lowest body size of children, most of the districts viz. Sitapur, Rae bareli, Jalaun, Lalitpur, Hamirpur belonged to Central and Budelkhand regions which are considered to be underdeveloped regions of the State. The above segregation of districts was in conformity with the data of District Level Household Survey on Reproductive and Child Health (DLHS-RCH)18.

Fig. 3

Growth pattern (Clusters) of children of UP State (54 districts).

Socio-economic factors: The clusters were compared in relation to their socio-economic parameters such as extent of land holdings, type of house, community, per-capita monthly income, density of population to see whether it was an artifact or if any relationships existed. It was found that the districts in cluster I were relatively more developed than in the other clusters (Table IV). The differences observed between clusters were both in terms of ‘population density’ and per capita income. It was also observed that the proportion of underprivileged communities such as Schedule Caste and Schedule Tribe population was much lower in the cluster I, compared to clusters III and IV.

Table IV

Comparison of social and economic indicators

Demographic factors: Demographic factors like sex ratio of population, birth order, children covered for nutrition assessment, literacy status, per cent married below 18 yr of age were compared among the clusters319. It was observed that the districts in cluster I were better off when compared to clusters II, III and IV, with respect to all the above variables (Table V). The sex ratio (females for 1000 males), a good indicator of demographic change was 921 in cluster I, as against 881 in cluster IV. The literacy status was 65 per cent in cluster I, compared to 52 per cent in cluster IV. These findings indicated that the development of the areas in cluster I was much higher than in the other three.

Table V

Comparison of demographic variables of the clusters

To assess the proportion of children with underweight (low weight for age), stunting (low height for age) and wasting (low weight for height) according to WHO growth standards using the above three criteria were least in the districts of cluster I, as compared to cluster IV. The extent of underweight was 33 per cent in cluster I, 45 per cent in cluster II, 52 per cent in cluster III and 58 per cent in cluster IV, indicating the extent of undernutrition was higher in clusters II, III and IV when compared with cluster I, though the differences were also higher between cluster II and cluster IV. Similar pattern was observed for stunting and wasting (Table VI). The health parameters like per cent women undergoing antenatal check-ups in different clustered districts ranged from a high (50%) in cluster I to 47 per cent in cluster IV. In the present study the institutional deliveries were relatively more in cluster I (20%), compared to cluster IV (14%), indicating better health care utilization in cluster I districts (Table VI).

Table VI

Comparison of health indicators of clusters

Discussion

In the present study four distinguishable clusters were found using PCA and clustering methods. Both the body weight and height parameters showed significant mean difference among the clusters. The clustered data mostly belonged to geographically adjacent districts of UP. A marked variation in clusters I and II was seen which signifies the development based on socio-economic and demographic variables in comparison to cluster III and IV. This analysis also helped in identifying the districts with relatively higher prevalence of undernutrition and the factors contributing to the same.

There are number of methods available for clustering4, but the methods of PCA and cluster analysis were used in this study because with PCA method the values of anthropometric variables in each case are transformed into principal component scores, which reflect body size of children more comprehensively than any single variable, and the cluster analysis performed by calculating the distances as well as considering the magnitude of difference between variables, helped in avoiding the drawbacks of other methods which use correlation coefficients as the similarity measure and tend to be sensitive to shape at the expense of magnitude6.

The studies in India and elsewhere demonstrated that socio-economic and demographic variables influence the nutritional status of children affecting their weight and height20–22.

Our results showed that the body size of children was different between different clusters, i.e. between developed and underdeveloped areas. For example, most of the districts in clusters I and II, which are considered as good clusters in terms of their better nutritional, health, social and demographic indicators, are located in the western and eastern part of the Uttar Pradesh State, which are considered to be prosperous regions.

This study had the limitation that other factors such as the ecological conditions, lifestyle, which might influence the nutritional status of the preschool children, were not considered. However, this study has identified possible areas of intervention for improvement in the nutritional status of children. The results of cluster analysis, are not only of interest, in terms of geographical, biological, ecological and anthropometric similarities but may also facilitate the planners and policy makers to conceive and implement appropriate action programmes for improvement in the nutritional status of the community in general and preschool children, in particular.