In this report, we present an overview of the need for data on urban health for planning and allocating resources to address urban inequities. Such data needs to provide information on differences between urban and rural areas nationally, between and within urban communities. We discuss the limitations of data most commonly available to national and municipality level government, donor and NGO staff. We present two innovative approaches to improve the quality of the data. Finally, we assess how these approaches have the potential to improve responses to urban poverty. This information was originally presented at a special session at the International Conference on Urban Health, Dhaka, 2015.
Rapid and uncontrolled urbanisation is evident across the majority of low and middle-income (LMIC) countries. This growth is particularly evident in South Asia where urban populations are projected to rise from 45 to 62 % by 2050.1 Governments are struggling to respond to this scale of growth. The infrastructure—housing, sanitation, health care, education, fuel, electricity, roads—needed to support this expanding population is rarely available, particularly for the poorest. For one third of the world’s urban population, 828 million people, this means living in slum conditions1. These conditions are fuelling the deepening trend of inequities across a wide range of health and social outcomes.2
National level decision-makers, local governments, donors and communities in low income countries need data to understand and respond to these inequities, and particularly the needs of the poorest in urban and rural areas. We believe that current data frequently overlooks the urban poorest, defined in this paper as the homeless and those living in slum conditions1 whether in informal settlements or in rented permanent dwellings, sometimes dispersed among better-off households,3 making them invisible to planners and decision-makers.
Currently available data comes from several sources: Census data; routinely collected clinical data and cross-sectional household surveys. All these data sources have limitations. Our assessment highlights how the urban poorest are frequently absent from the data sources available to decision-makers at this macro level. Census data is commonly collected every 10 years; in the context for rapid urbanisation and highly transient urban poor populations, such data soon becomes out of date. Furthermore, censuses exclude the homeless and settlements that are seen as illegal.3 This situation is exacerbated by the length of time it can take for official lists of slums to be updated, leaving many slums unrecognized for years.4 Recent enumeration work in five Indian cities by the Urban Health Resource Centre (UHRC) found 40% of slums were unlisted and therefore unrecognized, this equates to 36% of all slum residents.5
Routinely collected clinical data aids understanding of the scale and trend of diseases and service use. However, clinical data excludes those managed by private medical practitioners, pharmacies, NGOs and traditional providers, thus underestimating prevalence and service use. Data is rarely disaggregated beyond male and female and provides no details on any other patient demographics such as age, level of poverty or home location. Publically available data is frequently aggregated to district or regional level and thus does not support small area planning. The move to electronic medical records has the potential to improve this data source considerably.6
Cross-sectional household surveys are a vital addition to the data available to decision-makers. Over the last twenty years or more, approaches and questionnaires have become standardized for many surveys allowing comparison across countries and over time. Over 200 Demographic and Health (DHS) and a similar number of Multiple Indicator Cluster Surveys (MICS) have been conducted since programmes began in 1984 and 1995, respectively.7
Whilst cross-sectional surveys have large sample sizes (between 5000 and 30,000 households), they do not collect samples of sufficient size to compare inter-urban or intra-urban disparities. In addition there are four methodological challenges which could lead to under-representation of the urban poorest and skew urban estimates in household surveys. Firstly, census data is commonly used to determine sampling frames, but this excludes ‘illegal’ settlements and the homeless.3
Secondly, inconsistent definitions of urban and rural may mean that peri-urban poor are miscategorized as rural; particularly when slums have burgeoned beyond the government defined urban boundaries.9 A third challenge is that household listing maps, produced during the second stage of sampling by survey implementers, often assume one dwelling is occupied by one household. Although questionnaires are designed to include non-typical residents including servants and extended family, they overlook whole households that share a dwelling. Multiple household dwellings may include households that split residence, for example with a dwelling in a rural village, households not listed on rental contracts, or households that view their residence as temporary as is common in poor neighborhoods.
A fourth challenge is the definition of a household. Many surveys follow the DHS definition of a household as “a person or group of related and unrelated persons who usually live together in the same dwelling unit(s) or in connected premises, who acknowledge one adult member as the head of the household, and who have common cooking and eating arrangements”.10 This definition can become problematic in urban areas: for example, in many of Dhaka’s slums several families share a cooking pot; whilst in Kathmandu, several individuals, often single men, share rooms with no cooking facilities, eating instead at street vendors. This multiple occupancy presents a challenge for survey enumerators who may not be aware of the poorest occupants within a dwelling.
Understanding inequities may also be constrained by the approach to assessing differences in wealth. DHS wealth quintiles are the most commonly used relative measure of wealth (see Tables 1 and 2). The measure is based on household ownership of physical assets such as water source type and cell phone ownership. They are calculated separately for urban and rural populations and then combined to account for the different value of the same asset in a rural versus urban context.11 Within urban areas, using physical assets to measure differences in wealth can prove misleading. Wealth includes income, saving, access to credit, and other financial assets beyond physical assets. For the poorest urban dwellers, high rents can keep a household in crippling poverty. Our work in Nepal highlights how those in some of Kathmandu’s informal settlements pay little or no rent and may be comparatively better-off than those living in better constructed formal dwellings paying high rents. These nuances are overlooked by a purely assets based categorisation.
To see whether the research community acknowledges these limitations when using cross-sectional data to understand rural–urban and intra-urban inequities we conducted a search of Global Health Ovid and Medline databases from 2000 to date in September 2015. We used the search terms ‘demographic health survey’ and ‘urban’. After the removal of duplicates, we identified 34 studies that had compared risk factors and outcomes between urban and rural populations. Overwhelmingly these papers find greater risks and worse health outcomes among rural populations when compared to urban populations. Only two of these papers4
12 recommend exploring differences by wealth categories within urban populations. If such surveys systematically under-represent the urban poorest, then the categorisation of ‘urban’ becomes a proxy for ‘wealthy urban’ rather than a representative reflection of the health risks and outcomes of the entire urban population. This bias provides an excessively rosy picture of the health of urban dwellers and masks the conditions and needs of the poorest. The value of disaggregation of urban DHS data as conducted in India by UHRC is highlighted by WHO and UN-Habitat1. If governments and donors do not have access to data which represents the needs of the urban poorest, it is unlikely that policies and resource allocation will address urban poverty.
To explore the extent to which household surveys can be used to understand urban health and identify health inequities, we assessed publically available reports and datasets of cross-sectional surveys in Bangladesh and Nepal. Six such datasets and survey reports were found (Tables 1 and 2). In order to make this assessment we used WHO’s Urban Health Equity Assessment Response Tool (HEART)13 which provides a comprehensive set of 12 core, 18 highly recommended and 6 optional indicators to explore different perspectives of urban health inequities (Fig. 1). The Urban HEART process recommends utilising existing data to assess these indicators. Following the assessment of indicators, the process advocates the involvement of a wide range of government, donor, and civil society stakeholders working together to prioritize and plan the responses to the identified health inequities.
Whilst the available surveys do provide a good coverage of most of the Urban HEART key indicators, particularly DHS, it is also clear that many of the highly recommended and optional indicators are missed. The main purpose of WHO’s Urban HEART process is to provide information to inform a response to inequities within urban areas. Current sampling methods may systematically miss the urban poorest, resulting in biased samples as described above. Further, the sample size of these surveys is insufficient to compare within and between urban inequities. For example, to compare infant mortality,14 a HEART indicator that is widely used as a gauge of health system impact on overall child health,15 a sample of 1000 to 1500 households of each the urban poorest and urban non-poor would be needed. In the 2011 Nepal10 and 2011 Bangladesh DHS,16 only 168 and 515 individuals respectively were sampled from the bottom wealth quintile, preventing estimation of HEART indicators in urban areas.