HEALTH POLICY AND PLANNING; 13(4): 408^416 ß Oxford University Press 1998

Transcription

1 HEALTH POLICY AND PLANNING; 13(4): 408^416 ß Oxford University Press 1998 Research reports Ecological and geographic characteristics predict nutritional status of communities: rapid assessment for poor villages AGUSTIN KUSUMAYATI 1,2 AND RAINER GROSS 2,3 1 Faculty of Public Health, University of Indonesia, Jakarta, Indonesia, 2 SEAMEO-TROPMED Regional Center for Community Nutrition, University of Indonesia, Jakarta, Indonesia, and 3 Deutsche Gesellschaft fu«r Technische Zusammenarbeit (GTZ) GmbH, Eschborn, Germany The quality of poverty alleviation programmes relies heavily on appropriate targeting and priority setting. Major problems in assessing poverty include identification of the indicators of poverty and the methods used for its assessment. Nutritional status, expressed by anthropometric indices, has been proposed as a poverty indicator because of its validity, objectivity, reliability and feasibility. This study was conducted to explore the application of remote sensing to poverty mapping based on nutritional status at the community level. Relationships between the nutritional status within a community and the ecological characteristics of the community were investigated. Multiple linear regression tests were executed, and the resultant equations were tested for their validity in predicting communities with poor nutritional status. Among geographical and ecological indicators used, distance to the nearest market, main soil type, rice field area, and perennial cultivation area were found to be most useful predictors for the ranking of the communities by nutritional status. Among non-ecological determinants, food consumption, health service status and living conditions were also found as predictors. The highest correlation was found if total population was also taken into account in the regression model (R 2 = 0.69; p ). In the assessment of the sensitivity and specificity of the eight models studied, `undernutrition' was defined as a condition where a community belongs in the first quartile for nutritional status (highest prevalence of undernutrition), and the baseline nutritional survey was considered as a standard method for final diagnosis. Most models which included only ecological factors in the equations had lower sensitivity and specificity than models which included all determinant factors in the equations. All models which took into account the total population had higher sensitivity and specificity than those that did not take total population into account. The best model of those that took into account only the geographical and ecological characteristics of the community's living environment had similar sensitivity and specificity (80 % and 94.1 %, respectively) as the models that considered non-geographical and non-ecological variables in addition to geographical and ecological variables. In the case of West Sumatra, only four ecological and geographic characteristics were sufficient to predict poverty in villages. Since these characteristics could be surveyed by remote sensing, it may well be possible to use remote sensing for a rapid method for poverty mapping.

2 Introduction At the end of the 20th Century, poverty alleviation remains one of the most important worldwide goals of development. According to the World Bank (1992), more than 1.1 billion people live in poverty, with most of them in rural areas. This number is expected to increase despite many attempts at poverty alleviation. Due to the highly complex nature of the causes and effects of poverty, identification of effective strategies for poverty alleviation remains an extremely difficult and demanding task. Because resources for poverty alleviation programmes are always limited, priorities must be set, especially regarding the targeting of a programme area. For this reason, accurate socio-geographic identification of thepoorestareaisessentialforcost-effectiveprogramme implementation. However, two major problems in the process of poverty mapping are (1) selection of a valid, accurate and feasible indicator, and (2) selection of an adequate method to conduct the assessment. Definition of poverty Until the 1970s poverty was perceived in economic terms. A weak purchasing power was widely accepted as a major criterion for poverty. However, with increasing experience in poverty alleviation, it has been recognized that this way of understanding poverty is too narrow. Based on this experience and continuing worldwide discussion, the World Bank (1990) has changed the definition of poverty into a more comprehensive one: Poverty exists when individuals or group of individuals with inadequate access to resources are not able to properly satisfy their basic needs. Poverty indicators A suitable poverty indicator must (1) fit with this comprehensive definition, and (2) be valid, objective, reliable and feasible. Based on these criteria, the commonly used poverty indicator, income or expenditure, does not fulfill these requirements. Anthropometric indices of nutritional status have been recommended as suitable indicators of poverty (e.g. Bouis 1992; Cornia et al. 1987; and Pacey and Payne 1985). Anthropometric data are valid and relatively easily measured and are independent of statements made by impoverished individuals. Another important relationship that supports the appropriateness of using anthropometric indices as poverty indicators is the similarity between the determinants of nutritional status and the determinants of poverty. Research reports 409 Methods for poverty mapping If nutritional status can be accepted as an indicator of poverty, nutritional mapping should predict the geographical distribution of poverty. There are already methods for identifying areas with poor nutritional status, such as the baseline survey and rapid assessment procedures (RAP). However, both methods are time consuming and costly. Therefore, other methods should be developed for nutritional mapping, even if they are less accurate. During the last decade remote sensing has been applied to a broad range of fields. In remote sensing, satellite images are used to obtain a variety of types of geographical and ecological information. Several of these geographical and ecological factors may contribute to the nutritional situation of a community by influencing (1) the food production and food availability, (2) the disease pattern, and (3) the accessibility to various public services and resources. It was the objective of this study to investigate whether geographic and ecological characteristics which could be assessed by remote sensing are sufficient to predict the nutritional status of the population living in rural communities. The main questions were: (1) which geographical and ecological data obtainable by remote sensing can be used as predictors to rank the nutritional status of communities within a wide geographical area?; and (2) how well can the communities with the lowest nutritional status be identified? Subjects and methods In this study no primary data were collected; previously collected secondary data were used instead. Firstly, data on the nutritional status of several communities were compiled and served as the marker for the validation of the methodology. Nutritional and health-related data (see Table 1) and anthropometric data from a total of 1166 children under five years of age were collected from 43 villages in two crosssectional surveys (baseline nutritional surveys) that used the same survey methodology (Gross and Schoeneberger (1992): 1. Nutritional baseline survey in Pasaman, West Sumatra, Indonesia (carried out by the SEAMEO-TROPMED Regional Center for Community Nutrition, Jakarta, 1992), and

3 410 Research reports 2. Survey on the profile of a community's health/nutritional status and its determinants in some areas of OPHIR Palm Estate Project: Pasaman, Pangkalan, Sei Kunyit and Timpeh, West Sumatra, Indonesia (joint research by Akademi-akademi Kedinasan DEPKES RI Padang, 1992). These data were used to calculate the following indicators of nutritional status of each of the 43 villages: 1. underweight (= proportion of under-fives with weight-for-age Z-score ); 2. stunting (= proportion of under-fives with heightfor-age Z-score ); 3. wasting (= proportion of under-fives with weightfor-height Z-score ); and 4. undernutrition (= proportion of under-fives with height-for-age Z-score plus under-fives with weight-for-height Z-score ). Secondly, geographical and ecological data (see Table 2) for the 43 communities were gathered from two sources: 1. Data from the Landuse Survey 1991 in Kabupaten Pasaman (Brauns 1991) that was conducted to assess the development of landuse in Pasaman from 1971 to 1991, especially in relation to the implementation of the GTZ-Area Development Project West Pasaman (ADP) which started in 1970 and ended in Written documents from the Indonesian Central Statistical Bureau, Kabupaten Pasaman, West Sumatra (BPS 1991).

4 Research reports 411 Only geographical and ecological data obtainable by remote sensing were gathered and analyzed. None of the data were actually obtained by remote sensing, but were obtained by field survey. Statistical evaluation was carried out with SPSS for Windows Version (Chicago IL, USA). Several steps were taken during the data analysis, i.e: 1. Identifying parameters to be used as determinant variables in regression analysis. 2. Selecting the best indicator of a community's nutritional status to be used in regression analysis. 3. Standardizing the community's nutritional status for the under-fives' age distribution. 4. Developing models for multiple linear regression analysis and constructing the equation for each model. 5. Testing the sensitivity and specificity of the equations in identifying the poorest villages.

5 412 Research reports Non-geographical and non-ecological determinant variables such as food consumption, prevalence of diseases, level of education, health service status and living condition have multiple indicators. Therefore, composite indicators of these variables were developed to be used in multiple regression analysis. Table 1 presents components of composite indicators of non-geographical and non-ecological determinants. Geographical and ecological determinant variables had multiple indicators as well. However, since comparison of these indicators in predicting a community's nutritional status was the main interest of this study, all of these indicators were used in the multiple regression analysis. Table 2 presents variables and indicators of geographical and ecological determinants assessed in this study. The best indicator of the community's nutritional status was selected by comparing the R 2 and p values when the four indicators were regressed by the same set of determinants. While conducting this assessment, it was found that a community's nutritional status was strongly influenced by the age distribution of the under-five children. Since the age distribution of the surveyed children was not the same in each of the villages and anthropometric indices change with age, the community's nutritional status was standardized for the age distribution of the underfive children. Age standardization was done directly. One village was selected as a standard population; the children's age was classified into five categories, i.e. 0^11 months, 12^ 23 months, 24^35 months, 36^47 months and 48^59 months. For each village (except for the standard village) expected cases of undernutrition was calculated by multiplying each of its category-specific proportion of undernutrition by the category-specific population from the standard village. The directly standardized nutritional status of the village was determined by dividing the total number of expected cases by the total population of the standard village. There are several agricultural systems applied in Pasaman, i.e. rice field, perennial cultivation, permanent dry area, and shifting cultivation. For each system, we could identify the size of both the currently cultivated area and the fallow area. Therefore, two approaches were used in developing the multivariate models, i.e. models which include the total agricultural area (currently cultivated area + fallow area) in the equations, and models which include only the currently cultivated area in the equations. Because the total population in a village will influence the quantity of land available for producing foods for each individual, another two approaches were developed, i.e. models which take into account the total population to quantify the available land for producing foods, and models which don't include total population. In the assessment of the sensitivity and specificity of each of the models, `undernutrition' was defined as a condition where a community belongs to the first quartile of the ranking of the communities based on nutritional status. Baseline nutritional survey was considered as the standard method for `final diagnosis'. Sensitivity refers to the capability of the model in identifying an `undernourished' village, while specificity refers to the capability of the model in identifying a `well nourished' village. Results Table 3 presents the results of multiple regression tests for selecting the best indicator of the nutritional status of a community. All indicators showed at least two significant p-values (p50.05). The R 2 varied for different indicators as well as for different sets of determinants. Table 4 compiles all models and equations that resulted from the multiple regression analysis. All of the models showed very significant p-values (ranged from to ) and their R 2 varied from 0.38 to Table 5 presents the final results of this study, i.e. the sensitivity of the models in identifying an `undernourished' village, and the specificity of the models in identifying a `well nourished' village. The sensitivity ranged from 37.5 % to 80.0 %, whereas the specificity ranged from 80.0 % to 94.1 %. All models had better specificity than sensitivity. Discussion Indicators of a community's nutritional status Four possible indicators of a community's nutritional status were compared by assessing their R 2 and p values when they were regressed by the same set of determinants (Table 3). Six different sets of determinants were used. When the model only contained ecological variables (four out of the six sets), regression on the proportion of undernutrition gave the

6 Research reports 413 best R 2 indicating that the strongest association was between the proportion of the children with low anthropometric indices (either height and weight) and the studied poverty-related indicators at the community level. Therefore, if these poverty-related determinants are going to be used in predicting a community's nutritional status, the best indicator of the community's nutritional status is the proportion of undernutrition. The results of this assessment suggest that the underlying causes of undernutrition are beyond physiological problems. Its deeper roots may be dysfunctions in economic, demographic, cultural and ecological processes. The significant correlations also support the proposal of using nutritional status, expressed as undernutrition, as a poverty indicator, and highlight the basic concept of the proposed screening method for nutritional status mapping. The method appears especially appropriate for rural conditions, where food availability is mainly influenced by agricultural production. Location, climate, soil type and agricultural system When a community's nutritional status was regressed solely by a set of indicators of location, distance to the subdistrict capital and distance to the district capital had stronger influences than the other indicators of location (data not shown). In the study setting, and also in most of Indonesian rural areas, public services, sources of information, and the centre of area development are located in the subdistrict capital, or in the district capital. The more remote the rural area, the lower the accessibility to these resources. The lower the accessibility to public services and information resources, the worse the community's nutritional status. However, if indicators of location were put in the equation together with indicators of food availability/food production (i.e. climate, soil type, and agricultural system), distance to the market became more influential. This may be because the market is a place where economic activities occur. People can sell their agricultural products in the market and thereby obtain money that can be used to buy items to meet their needs for daily life, including foods which cannot be self-produced. In this study, altitude and rainfall did not have significant associations with the community's nutritional status due to the small variation of these variables. However, if the geographical area assessed by remote

7 414 Research reports sensing were larger, such as the whole of Indonesia, altitude and precipitation should be considered in the equation. The results of this study showed that soil type had an influence on the nutritional status of the community. The influence of soil type was less important than that of agricultural system, but more important than any other observed factor. Models which included currently cultivated area in the equations showed higher sensitivity and specificity than models which included total agricultural area in the equations (Table 5). This may be because food availability is influenced by the actual area which can produce foods or other agricultural products, and not merely the total available area which consists of both cultivated and fallow areas. If the total population was taken into account, the validity of the equations was

8 Research reports 415 variables such as food consumption, household welfare/caring capacity, health status, and living conditions (Table 5: models 2.1 to 2.4). In general, all models have better specificity than sensitivity, especially the models which take into account only geographical and ecological characteristics of the community's living environment. This means that they are better able to identify `well nourished' communities than to identify `undernourished' communities. better. Taking into account the total population, the community's nutritional status can be predicted by the size of the total agricultural area per person or the size of the currently cultivated area per person. Models which included only geographical and ecological determinants in the equations had lower validity than models which included both ecological and non-ecological determinants. This may be explained by the fact that a community's nutritional status is influenced by food consumption, prevalence of diseases, food availability, social welfare and caring capacity, health service status and living conditions, as well as by geographical and ecological characteristics of the community's living environment. Sensitivity and specificity of remote sensing `models' in identifying `undernourished' communities Eight models for predicting the rank order of the communities based on nutritional status were developed in this study and the validity of each model was tested in terms of its sensitivity and specificity in identifying `undernourished' communities. In general, the more variables taken into account in the model, the better the model's sensitivity and specificity in identifying `undernourished' communities. The first three models, which only take into account geographical and ecological variables (Table 5: models 1.1 to 1.3), have lower sensitivity and specificity than models which also take into account other If we compare the sensitivities of the eight models, three models have the best sensitivities (80 %), i.e. model 1.4, model 2.2, and model 2.4. This means that model 1.4, which only takes into account the geographical and ecological characteristics of the community's living environment, has similar sensitivity as models 2.2 and 2.4, which also take into account non-geographical and non-ecological variables. Thus remote sensing, which provides only geographical and ecological data, has the potential of providing insight into the socioeconomic situation. Possibility of using remote sensing technique for nutritional status mapping The study showed that, if not the nutritional status in absolute terms, at least the rank of the communities can be predicted solely by geographical and ecological factors. This finding leads to the question of whether remote sensing could be used as a rapid screening method for poverty mapping based on the community's nutritional status. However, the basic concept of this method should always be kept in mind. Geographical and ecological characteristics of a community's living environment influence the community's nutritional status mainly through food and agricultural production which determines food availability. This situation is obviously different from urban areas where food availability is mainly influenced by income. As a result it seems less likely that this method could be used in urban areas. Nutrition or poverty mapping can be determined by analyzing available secondary data such as records and reports on the health situation and health services/programmes in the area of interest. Nevertheless, in most developing countries these sources of information are rarely used, not even for health planning. A baseline survey is still commonly applied to gather data for planning, including planning for poverty alleviation programmes. However, remote sensing does have some advantages over

9 416 Research reports other methods used for nutrition or poverty mapping. Firstly, accessing and analyzing data obtained by remote sensing is cheaper and quicker than carrying out any type of survey; and secondly, it is possible to obtain geographical and ecological data from remote areas that are not covered by any services/programmes, and consequently about which no information has been gathered. It has to be borne in mind that this study was done on a very limited database of 43 villages and was not designed to be generalized to other rural agricultural areas. The results may not apply to other rural agricultural areas, even others in Indonesia. Therefore, similar studies need to be done to produce other appropriate models for identifying the poorest villages. By developing a model based on different and larger data sets, the sensitivity and specificity of the model will also be improved. It is expected that a more generally applicable model could be developed considering not only production area but also production systems and type of vegetation. Once such a model has been developed, remote sensing could be an interesting tool not only for the geographical and ecological, but also for the political, decision-making process regarding poverty alleviation. References ACC/SCN Some options for improving nutrition in the 1990s. United Nations Administrative Committee on Coordination 7Subcommittee on Nutrition, Geneva. Supplement to the SCN News No. 7. Bouis H The effect of income on the demand of food. Seminar on South-East Asian Agriculture in Transition: Implication for Food Policy. Jakarta, Indonesia. BPS, Biro Pusat Statistik Kabupaten Pasaman Kabupaten Pasaman Dalam Angka Brauns T Landuse survey 1991 in Kabupaten Pasaman. Area Development Project. Deutsche Gesellschaft fu«r Technische Zusammenarbeit (GTZ) GmbH, Padang, West Sumatra, Indonesia. Cornia GA, Jolly R and Stewart F Adjustment with a human face: protecting the vulnerable and promoting growth. Vol. 1. Oxford: Oxford University Press. FAO/WHO Major issues for nutrition strategies. International Conference on Nutrition, Food and Agriculture Organization of the United Nations, World Health Organization, Rome. Gross R, Altfelder S and Koch E To reduce poverty. Deutsche Gesellschaft fu«r Technische Zusammenarbeit (GTZ) GmbH, Jakarta, Indonesia. Gross R and Schoeneberger H Handbook for the assessment of the nutritional situation of a community. Experimental Edition. Deutsche Gesellschaft fu«r Technische Zusammenarbeit (GTZ) GmbH, Jakarta, Indonesia. Pacey A and Payne P Agriculture and development. Food and Agriculture Organization of the United Nations (FAO) and United Nations Children's Fund (UNICEF). London: Hutchinson Publishing Group. Scholz U Crop Geography for Agro-Ecological Characterization in Sumatra and Costa Rica. In: Agricultural environments ^ characterization, classification and mapping. Proceedings of the Rome Workshop on agro-ecological characterization, classification and mapping. Oxford: CAB International. UNICEF Strategy for improving nutrition of children and women in developing countries. New York: United Nations Children's Fund. World Bank World Development Report 1990: Poverty. Washington DC. World Bank World Development Report 1992: Environment and development. Washington DC. Acknowledgments The authors acknowledge the support of LUPAM/GTZ, Jakarta and SfPDM/GTZ, Jakarta. Hadimulyo and M Taufiq Dardjat are thanked for providing assistance in understanding the remote sensing concept; and Luc de Meester is thanked for providing landuse data of Pasaman. Biographies Agustin Kusumayati is an assistant lecturer of the Faculty of Public Health of the University of Indonesia in Jakarta, Indonesia. Rainer Gross is a senior advisor of the German agency of technical cooperation (GTZ) at the SEAMEO-TROPMED Regional Center for Community Nutrition at the University of Indonesia in Jakarta, Indonesia. Correspondence: Rainer Gross, SEAMEO-TROPMED Regional Center for Community Nutrition, PO Box 3852, Jakarta 10038, Indonesia. gtzseame@indo.net.id