Development of a proxy wealth index for women utilizing emergency obstetric care in Bangladesh
1Department of Health Sciences, University of Leicester, UK.
2Department of Public Health, University of Aberdeen, UK.
3Dugald Baird Centre for Research on Women's Health, Department of Obstetrics and Gynaecology, University of Aberdeen, UK.
4Department of Obstetrics and Gynaecology, University of Aberdeen, UK.
*Corresponding author. Department of Health Sciences, University of Leicester, 2nd Floor Adrian Building, University Road, Leicester, LE1 7RH, UK. Tel: +44 (0) 116 229 7261. Fax: +44 (0) 116 229 7250.E-mail: elp17{at}le.ac.uk
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionConclusionsReferences |
|---|
There are increasing concerns regarding inequities in access to health care, and hence calls for routine data collection to improve monitoring. For many developing countries, such as Bangladesh, increasing the availability and uptake of emergency obstetric care (EmOC) is vital in improving maternal health. It is crucial, however, that women of all socio-economic status benefit from this. This paper describes the development and validation of a proxy wealth index for assessing women's socio-economic status in Bangladesh as they are admitted to hospital. Existing poverty assessment tools are unsuitable for use in this context as they are too lengthy or need to be administered at household or community level. We sought to develop a tool with a limited number of indicators to allow quick administration and avoid interference with treatment. We also aimed to develop a pragmatic tool to be able to calculate a score in the field. The steps, involving selecting and weighting indicators, assigning a proxy wealth score and validating the score, are outlined. Indicators were selected from the Bangladeshi Demographic and Health Survey (DHS) data, which allowed comparison of socio-economic status between women using EmOC and those in the wider population. The tool proved quick and easy to use and was acceptable to women and their families. The validity of the tool was established by means of factor analysis. Our comparison with DHS data suggested that women using EmOC were significantly wealthier than women in the wider population. The implications of this, as well as the strengths and limitations of the proxy wealth index, are discussed. The proxy wealth index offers potential as a pragmatic and quick means of assessing poverty status in a busy hospital setting. Such a tool may enable monitoring of equity in access to treatment and identification of those least able to afford treatment, to enable any mechanisms in place to pay for care to be applied in a timely fashion, so avoiding delays in treating life-threatening complications.
Key Words: Wealth indices, poverty assessment, equity, access, maternal health
KEY MESSAGES
|
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsReferences |
|---|
One of the eight Millennium Development Goals (MDGs) is to reduce maternal deaths, from current levels of over half a million each year, by three-quarters by 2015 (United Nations, undated; WHO et al. 2004). Almost all maternal deaths occur in developing countries, and the difference between developed and developing countries is greater for maternal mortality than for any other health indicator (Commission on Macroeconomics and Health 2002
). Further, there is growing evidence of inequalities within countries as the poorest women face greatest risk of maternal death (Graham et al. 2004). Concern has been raised that making progress towards the MDGs will not necessarily reduce such inequalities (Moser et al. 2005) and therefore regular data monitoring is required (Victora et al. 2003; Nolen et al. 2005).
Inequalities in utilization of antenatal care, deliveries with a skilled attendant and place of delivery are evident in Bangladesh, as in many other countries (Peabody et al. 1999; Kunst and Houweling 2001). These are commonly measured through Demographic and Health Surveys (DHS) data; however, there are no comparable facility-level data on users of emergency obstetric care (EmOC). Ensuring availability and access to EmOC is considered vital in efforts to reduce maternal deaths (Fortney 2001). In Bangladesh, the vast majority of births take place at home and access to lifesaving treatment in the event of a complication is crucial. Efforts have been made to encourage routine data collection for process indicators such as proportion of all births in EmOC facilities, rates of caesarean section and met need (UNICEF et al. 1997; Islam et al. 2005). However, these do not include any measure of socio-economic status of women using EmOC. In this paper, we describe the development and validation of a tool for assessing women's poverty status in Bangladesh as they are admitted to hospital.
There is a vast literature on the definition and measurement of poverty, and more specifically on measuring inequalities in health and utilization of health care. Measuring poverty based on consumption or income is notoriously difficult in developing countries. In large, informal, predominantly agricultural-based economies, income and consumption expenditure can be hugely variable and measurement confounded by recall bias (McKenzie 2004). Rather than direct measures of living standards such as income and consumption, household characteristics and asset ownership are widely used as indicators of wealth (Lindelow 2006). Characteristics of housing and asset ownership are much less affected by seasonality and recall bias (McKenzie 2004). DHS (covering 70 countries worldwide) provide a valuable source of data on asset ownership and household characteristics as well as a large number of health, nutrition and health service utilization measures. These have been used, perhaps most notably, by Gwatkin et al. (2000) to construct asset indices and wealth quintiles through the use of principal components analysis in order to show socio-economic differences in key health, nutrition and population indicators. The choice of an asset-index was initially pragmatic rather than conceptual, given the data available in the DHS, but research has shown that the asset ownership can be taken as a reasonable proxy for consumption (Filmer and Pritchett 1998; Gwatkin et al. 2000).
With calls for increased monitoring and concern for equity in uptake of health services in obstetrics and other services, facility-based indicators of wealth need to be developed (Chowdhury et al. 2006; Pitchforth et al. 2006). The limitations of existing asset indicators are that they often involve many assets and are therefore impractical for administration in hospital where patients may be in life-threatening conditions and resources to administer and monitor socio-economic status are limited. Similarly, other existing quantitative and qualitative approaches for identifying the poor, such as the CASHPOR Housing Index and Participatory Wealth Ranking, require data collection at the household or community level (Mahbub et al. 1995; Simanowitz et al. 2000).
Our study sought to develop and validate a proxy wealth index in the context of EmOC in Bangladesh. A secondary aim was to compare the socio-economic status of women using EmOC with the wider population. This was part of a study that sought to compare the treatment and experience in hospital of women of different socio-economic status, and to identify financial support mechanisms for women least able to afford emergency obstetric care in a system resulting in high out-of-pocket expenses for users of care.
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsReferences |
|---|
Selection of proxy wealth indicators
At the outset of the project a number of criteria for the development of the poverty assessment tool were defined:
- The tool should be administrable as women are admitted to the hospital ward without interfering with medical care. The number of indicators should therefore be very limited.
- The indicators should be selected so that accompanying relatives can answer the questions in the instance of the woman herself being too unwell.
- The calculation of the index score should be possible in the field without the need for statistical analysis.
- The indicators used should allow some comparison with the wider population through available data sources such as DHS surveys.
1) Determination of relationship between individual variables and DHS poverty quintiles
Box 1 lists 13 DHS indicators that were considered possible indicators of poverty. To establish how well each might help differentiate between poverty groups, these variables were cross-tabulated against the poverty quintiles provided in the DHS data set.
Box 1 Key poverty indicators from Bangladesh DHS 1999/2000
|
Some of the indicators in Box 1 such as ‘has television’ showed little variation between quintiles 1 to 3, where ownership remained below 5%, then showed a marked difference to quintile 5 where ownership increased to over 70%. The four indicators that seemed to differentiate well across all five quintiles, increasing in proportion from quintile 1 to 5, were ‘type of toilet facility’, ‘main roof material’, ‘educational attainment’ and ‘literacy’. Therefore, we selected these four indicators for our proxy wealth index.
These were included as part of a 30-item questionnaire together with demographic details such as age, place of residence, marital status and parity, as well as hospital-related data, i.e. registration number and time of admission. Educational attainment was assessed by asking women first if they had ever attended school and, if so, the highest class they attended, following a pilot study that demonstrated that this was the most appropriate method. This allowed the interviewer to determine educational attainment (see Table 1).
|
2) Assigning ‘poverty score’
To allow categorization of women into poverty groups, we gave each of the responses a crude score (Table 2). However, as one indicator might be more significant than the others in differentiating between the poorest and not so poor, it was necessary to introduce some means of showing the relative importance of each indicator. The categorizations of the variables identified for inclusion in the creation of the score were rescaled to take values in the range of 0 to 1. Literacy (can read easily) showed the greatest difference between quintile 5 (5%) compared with quintile 1 (69%). This was the largest difference observed in the four indicators, and literacy was therefore assigned the greatest weight by multiplying the rescaled scores by 4 (Table 2). In a similar way, main roof, toilet type and educational attainment were assigned weightings through multiplying by factors of 3, 2 and 1, respectively.
|
To obtain a composite poverty score we then added each of the responses from the four individual indicators. Possible poverty scores thus ranged from 0 (poorest) to 10 (wealthiest). The median value was then used to categorize women into ‘poor’ and ‘non-poor’ groups.
3) Validating poverty scores
Factor analysis is commonly used in data reduction to identify a small number of factors that explain most of the variance observed in a larger number of variables (Kaplan 2004). Factor scores were generated for the study data and these were grouped into two poverty groups as above.
The DHS data were also used to assess the strength of our proxy wealth index. A factor analysis of all DHS assets, utilities and human capital variables (Gwatkin et al. 2000) was undertaken and used as a reference index (gold standard). The correlations between our three selected indicators and the reference index were then assessed based on our assigned weighting, no weighting and weighting by factor analysis. The level of agreement between the ‘gold standard’ and the three other indices was assessed using kappa analysis.
Comparison with wider population
Data collected from women during the study period were entered into the statistical package SPSS and cleaned. To assess how the study sample compared with women in the wider population, the sample was contrasted with DHS data for Bangladesh 1999/2000. Again, to make more directly comparable to our study population, only those women who were resident in Dhaka Division and who had had a previous delivery were included in the analysis. The complex sampling design and weighting applied to the DHS data meant that Stata, rather than SPSS, had to be used when making comparisons with DHS data (Kneipp and Yarandi 2002). In Stata, the Pearson 
2 statistic (test of independence) is corrected for the survey design and is converted into an F statistic (StatCorps 2003). The F value and associated statistic is reported for the analysis in this paper (Hogg and Tanis 1989).
Hospital selection
In the Dhaka Division of Bangladesh, where our study was based, less than a fifth of hospital-treated obstetric complications were treated in private facilities (ACPR and UNICEF 2001). Our study, therefore, focused on government provision. Government hospital services in Bangladesh include medical college hospitals, district hospitals, maternal and child welfare clinics, and upazila (sub-district) health facilities. The hospital selection criteria for our study were that it would: (1) provide comprehensive EmOC; (2) treat all obstetric complications; and (3) admit women of poor socio-economic status. A feasibility study identified a medical college hospital that met all of these criteria and this became the site for the study (Pitchforth 2004).
Data collection
Ethical approval for the study was granted by the Bangladesh Medical Research Council. As part of a mixed-methods study (Pitchforth 2004) a questionnaire was designed to collect demographic and poverty-related data in order to calculate a ‘poverty score’. During the data collection periods (from September 2002 to March 2003) all women admitted to the obstetric admissions ward in the selected hospital were invited to participate. The questionnaires were administered by EP and a local interpreter, usually at the women's bedside. If a woman was too unwell to complete the interview, but was willing to participate, an accompanying relative was asked to provide the information.
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsReferences |
|---|
During the period of the study, 774 women were admitted to the obstetrics ward. Of these, 638 (83%) patients and/or relatives completed the poverty assessment questionnaire. The questions appeared to be straightforward for the majority of women and could be completed swiftly, without interference with their treatment. Reasons for non-completion included patients leaving the hospital very soon after admission (n = 85), inappropriate admissions to the obstetrics department (n = 28), or patients being unable or unwilling to provide the information requested (n = 23). The characteristics of women in our sample are presented and compared with the wider population in Table 3.
|
The vast majority of women were married (98%), living in an urban location (69%) and Muslim (95%). These proportions were significantly greater than women in the wider population. The most notable difference is with respect to urban/rural residence. Less than a third of the study sample were rural residents, compared with almost three-quarters (72.6%, n = 2012) of women in the wider Dhaka Division population.
With respect to the individual proxy wealth indicators, over one-third (37.9%) of women in our sample reported having no education. The same proportion, although not all the same women, reported that they did not have any reading ability. All women reported having some type of toilet facility, the most common being a septic tank/modern toilet (41.1%). The most common type of roof material was a rudimentary/tin roof (50.9%).
By contrast, Table 3 also shows that women in the wider DHS Dhaka population were less likely to have any reading ability, and a greater proportion (50.9%) had no education. A far smaller population had a modern toilet (12.5%) and, although the most common type of roofing was also tin, a far smaller proportion of women in the wider population had a finished roof (cement/concrete) (10.9%) compared with the study sample (32.4%).
Poverty scores
The factor analysis suggested that only five women (0.78%) in our study had been wrongly classified into categories ‘poor’ and ‘non-poor’. Two women were classified in the ‘non-poor’ group when, according to the factor analysis, they should have been in the ‘poor’ group. The original scores for these women were checked. Both women had a higher education, but also a natural roof, which affected their poverty grouping. The three women who were misclassified as ‘non poor’ by factor analysis all had a septic tank as their toilet facility, two had a concrete roof, whilst the third woman had a natural roof but a high level of educational attainment. As only five women were misclassified, the factor analysis supported the poverty scores and grouping.
Figure 1 shows the distribution of poverty scores calculated by the same method for our study population and DHS data for women with previous deliveries in Dhaka Division. The distribution of poverty scores varies greatly between populations. The study sample had 10% of women with a poverty score of below 2, compared with 21.5% of the women in the Dhaka DHS sample. The distributional differences between the two groups, as shown in Figure 1, indicate that the study group was less disadvantaged. A comparison of means showed that the difference between the two populations was significant (study sample 5.6 versus DHS 4.18, p < 0.001). Women in our study sample had a significantly higher poverty score (indicating greater wealth) than mothers in the general Dhaka Division population, another indication that the study sample were ‘less poor’ than women with at least one delivery from the wider population.
|
The Pearson correlations between the factor scores and quintiles derived from the DHS data and the different forms of weighting were all above 0.77 and statistically significant (p < 0.001). As expected, the correlation between the two methods based on the woman's scores were highly correlated and significant (>0.9). The correlation between the factor scores derived using the four variables and the two scoring methods shown in Table 4 was also significant (p < 0.001).
|
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsReferences |
|---|
We aimed to develop a method of assessing the socio-economic status of women using EmOC that is appropriate for use as they are admitted to the obstetrics unit. This was in order to compare the socio-economic status of women using EmOC with women in the wider population, and to compare treatment and experience of ‘poor’ and ‘non poor’ women at the hospital. The four-indicator proxy wealth index we developed proved to be easy to use, even with the additional step of rescaling the categories, to be acceptable to women and their families, and valid in comparison with DHS data. The selection of indicators was pragmatic and straightforward and the use of publicly available data means that it will be easily replicable in different settings.
It may be argued that the initial selection of indicators across quintiles was subjective, by looking at the distribution across quintiles. Other measures such as the poor/rich ratios, using values from the wealthiest and least wealthy quintiles, could have been calculated and used as a means of selecting indicators. However, these are crude measures and effectively ignore the three middle quintiles (Gwatkin et al. 2000). Our approach deliberately sought to look at the distribution across all quintiles. There are likely to be advantages and disadvantages of any selection process but future work could explore the optimal way of selecting individual indicators. The advantage of using DHS indicators was that it allowed comparison of the poverty status of women using EmOC at the selected hospital with women in the wider Division population. The mean poverty score for women using EmOC was significantly higher for all three comparison scoring methods, indicating greater wealth, than for women in the wider population, providing some evidence that poorer women may not be using EmOC facilities.
We made a decision at the outset to include only four individual indicators in constructing our proxy wealth indicator. This may be considered a reductionist approach for the measurement of a highly complex concept. Indeed, as with other indices, our four-indicator proxy measure of wealth can only provide a partial view of the multi-dimensional concepts of poverty, inequality and inequity (Gwatkin et al. 2000). This said, the four indicators selected do have theoretical underpinning and are likely to show differences between women of different socio-economic status. Higher education levels of women are consistently shown to be associated with greater well-being of households and to be positively associated with health service use. The inclusion of educational attainment and literacy strengthens the validity of our four-indicator measure (Kauser et al. 1999; Chakraborty et al. 2003; Mukherjee and Benson 2003; Karim et al. 2006). Similarly, rudimentary housing and lack of sanitation are clearly associated with increased poverty (Islam 1996).
Whilst the four indicators selected from the DHS surveys appear theoretically robust and the factor analysis validated the composite scoring system, there are a number of problems with the approach used and further research is required to establish the validity of the poverty assessment. The questionnaire relied on self-reported levels of education, literacy and living conditions. It was not possible in our study to cross-validate the responses given. There was no benefit for women that would encourage deliberate misreporting of poverty-related questions but evidence from elsewhere suggests, for example, that self-reported literacy may be higher than actual reading and writing ability (LeVine et al. 2004). If women were assigned a poverty score that reflected a higher than actual socio-economic status, then this may go some way to explain the differences in poverty status between the study population and the DHS population. The differences were so marked, however, that it is unlikely that misreporting would account for all the difference. Future studies would benefit from cross-validation of responses.
Four indicators were selected for purely pragmatic reasons—it was important that the data could be collected as simply and as quickly as possible. The four indicators could all be fitted onto one side of paper. Further work is required to ascertain what the optimal number of indicators is. For example, the distribution of poverty scores across populations showed some evidence of clumping at various levels of the proxy wealth index. The use of more indicators may prevent such clumping (McKenzie 2004), but would add to the length and complexity of the index and the collection of the data itself.
To create our proxy wealth index we aggregated the individual responses. An equal weighting would have assumed an equal welfare value across all indicators but this may not be the case. Instead we assigned weighting after considering the difference across poverty quintiles for each indicator. The limitation of our approach was that the weighting of 1, 2, 3 and 4 times was somewhat arbitrary. However, using the DHS data, different weightings were compared and the high correlations suggest that the poverty grouping would have been similar whether no weighting, factor analysis or our assigned weighting had been used. The high correlations give some confidence to our results but further research could explore alternative methods of weighting.
Linked to the weighting and initial selection of poverty indicators is that the number of possible categories for each indicator may have influenced the proportions across quintiles. For example, literacy, with the fewest categories, showed the greatest difference between the wealthiest and poorest quintiles. Education had the greatest number of categories and was found to show the least difference. That said, the factorial analysis indicated that the poverty group scoring did differentiate well according to the DHS data. Finally, the decision to consider poverty as a dichotomized variable—‘poor’ and ‘non-poor’—fitted the aims of our study. However, doing so assumes that the groups are readily distinguished, and information may be lost in this way. Those women with a poverty score of less than 1 were considered in the same category as women with a poverty score of 5, although, in reality, there may have been considerable differences between these women in terms of literacy and education levels or housing conditions. Future application of the tool should consider the most appropriate grouping, if any, depending on the purpose of measurement (Karim et al. 2003).
Finally, another problem, in common with other indices, is that our proxy wealth index does not differentiate between rural and urban areas. It thus assumes that housing conditions, level of education and literacy are equivalent in all areas, whereas this may not be the case (Kauser et al. 1999; Falkingham and Namazie 2002). Future work should also consider the relevance of the tool to both rural and urban populations. The vast majority of women in our study had urban residences but this was in direct contrast to the wider population in Bangladesh. A greater proportion of women classified as ‘poor’ lived in rural areas (n = 118, 37.9%). The higher proportion of women in our study with modern toilets and finished roofs may reflect differences in urban buildings as well as differences in wealth.
A secondary aim of this paper was to compare the socio-economic status of women using EmOC with the wider population. The finding that users of EmOC in this facility were significantly wealthier than the wider population has a number of potential implications. First, it raises concerns regarding the maternal health of poorer women in Bangladesh as, in addition to lower use of antenatal care and skilled care at delivery, poorer women do appear less likely to utilize EmOC. The hospital in our study admitted all types of obstetric complications and women of all socio-economic status so it is not felt that poorer women could be accessing care elsewhere. Future work is needed to see if the differences in poverty profiles between users of EmOC and the wider population are accounted for by differences in clinical need. However, knowledge about the determinants of maternal mortality has long suggested that the poorest women are most likely to have greater need for EmOC (McCarthy and Maine 1992). Efforts to increase the uptake of EmOC in Bangladesh must therefore ensure that the poorest women are able to access care.
Secondly, it highlights the importance of adding a measure of socio-economic status to routinely gathered data and frequently used process indicators such as met need for EmOC. While such process indicators have broad application and, it can be argued, play an important role when the measurement of maternal mortality is so inherently difficult to measure (Hill et al. 2001), the broad nature of the indicators means it is not possible to look at inequalities or inequities in access to treatment between socio-economic groups. This is not a problem unique to developing countries and it is important that future monitoring of care should routinely include indicators of socio-economic status (Murray and Bacchus 2005). Our study provides one means of doing so that would not be over-burdensome for already overstretched facilities and can be carried out quickly so as not to interfere with treatment. Ensuring equity of access to care is important across all areas of health care, especially in countries such as Bangladesh where obtaining treatment relies on meeting out-of-pocket expenses. There is no reason to believe that the methods used in developing our proxy wealth index cannot be adopted in other areas of health care; it may be particularly useful for other areas handling emergency cases.
|
Conclusions |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsReferences |
|---|
Ensuring all women can access lifesaving treatment in the event of a complication in childbirth is vital in efforts to reduce maternal mortality. Despite suggestions that the poorest women are least likely to seek care in the event of a complication, little good quality evidence exists. Any efforts to increase uptake of EmOC must ensure that women of all socio-economic status are utilizing EmOC when required, and for this a poverty assessment tool is needed. The tool developed in our study proved to be straightforward, easy to use as women were admitted to hospital and acceptable for women and their families. Comparison with DHS data showed that the resulting proxy wealth index was also valid. The selection of indicators allowed comparison with the wider population and this reiterated the importance of incorporating poverty status questions into routine monitoring as the users of EmOC in the selected hospital were significantly wealthier than women in the wider population. Further research is required to assess the validity of the tool as it offers potential as a pragmatic and quick means of assessing poverty in a busy hospital setting.
|
Acknowledgements |
|---|
The authors thank the women and relatives and selected hospital for participating in the study. Mary Dixon-Woods and Sheila Bonas provided useful comments on earlier drafts of the paper. This study was part of a Ph.D. programme funded by the University of Aberdeen with additional funding from the Phil Strong Memorial Prize (British Sociological Association) and The Carnegie Trust Scotland.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsReferences |
|---|
ACPR, UNICEF. Review of availability and use of emergency obstetric care (EmOC) services in Bangladesh (2001) Dhaka: UNICEF.
Chakraborty N, Islam MA, Chowdhury RI, et al. Determinants of the use of maternal health services in rural Bangladesh. Health Promotion International (2003) 18:327–32.
Chowdhury ME, Ronsmans C, Killewo J, et al. Equity in use of home-based of facility-based skilled obstetric care in rural Bangladesh: an observational study. The Lancet (2006) 367:327–32.
Commission on Macroeconomics and Health. Improving health outcomes of the poor: report on working group 5 of the Commission on Macroeconomics and Health (2002) Geneva: World Health Organization.
Falkingham J, Namazie C. Measuring health and poverty: a review of approaches to identifying the poor (2002) London: Department for International Development.
Filmer D, Pritchett L. Estimating wealth effects without expenditure data or tears: an application to educational enrolments in states of India (1998) Washington, DC: World Bank.
Fortney JA. Emergency obstetric care: the keystone in the arch of safe motherhood. International Journal of Gynecology and Obstetrics (2001) 74:95–97.[CrossRef]
Graham WJ, Fitzmaurice AE, Bell JS, Cairns JA. The familial technique for linking maternal death with poverty. The Lancet (2004) 363:23–27.
Gwatkin DR, Rustein S, Johnson K, et al. Socioeconomic differences in health, nutrition and population in Bangladesh (2000) Washington, DC: World Bank.
Hill K, AbouZahr C, Wardlaw T. Estimates of maternal mortality for 1995. Bulletin of World Health Organization (2001) 79:182–93.
Hogg RV, Tanis EA. Probability and statistical inference (1989) New York: Macmillan Publishing Company.
Islam N. Sustainability issues in urban housing in a low-income country: Bangladesh. Habitat International (1996) 20:377–88.[CrossRef][Web of Science]
Islam MT, Hossain MM, Islam MA, Haque YA. Improvement of coverage and utilization of EmOC services in southwestern Bangladesh. International Journal of Gynaecology and Obstetrics (2005) 91:298–305.[CrossRef]
Kaplan D. The SAGE handbook of quantitative methodology for the social sciences (2004) London: SAGE Publications.
Karim F, Tripura A, Gani MS, Chowdhury AMR. Poverty status and health equity: evidence from rural Bangladesh. Public Health (2006) 120:193–205.[CrossRef][Web of Science][Medline]
Kauser F, Griffiths P, Matthews Z. Poverty and maternal health care utilisation in Maharashtra: associated influences on infant mortality and morbidity. (1999) Social Statistics Working Paper, 1999–04: Southampton: University of Southampton.
Kneipp MA, Yarandi HN. Complex sampling designs and statistical issues in secondary analysis. Western Journal of Nursing Research (2002) 24:552–66.
Kunst AE, Houweling T. A global picture of poor-rich differences in the utilisation of delivery care. Studies in Health Services Organisation & Policy (2001) 17:297–315.
Lindelow M. Sometimes more equal than others: how health inequalities depend on the choice of welfare indicator. Health Economics (2006) 15:263–79.[CrossRef][Web of Science][Medline]
LeVine RA, LeVine SE, Rowe ML, Schnell-Anzola B. Maternal literacy and health behaviour: a Nepalese case study. Social Science and Medicine (2004) 58:863–77.[CrossRef]
Mahbub A, Mayeed M, Das Roy R. A qualitative exploration of some socioeconomic issues in South Uddomdi, Matlab. (1995) Working Paper Number 10: Dhaka: BRAC-ICDDR,B.
McCarthy J, Maine D. A framework for analyzing the determinants of maternal mortality. Studies in Family Planning (1992) 23:23–33.[CrossRef][Web of Science][Medline]
McKenzie DJ. (2004) Measuring inequality with asset indicators. BREAD working paper No. 42: Cambridge, MA: Bureau for Research and Economic Analysis of Development.
accessed 23 August 2005. MEASURE DHS. Undated. Bangladesh: Standard DHS, 1999/2000. Online at: http://www.measuredhs.com/countries/metadata.cfm?ctry_id=1&surv_id=135.
Moser KA, Leon DA, Gwatkin DR. How does progress towards the child mortality millennium development goal affect inequalities between the poorest and least poor? Analysis of Demographic and Health Survey data. British Medical Journal (2005) 331:1180–2.
Mukherjee S, Benson T. The determinants of poverty in Malawi, 1998. World Development (2003) 31:339–58.[CrossRef][Web of Science]
Murray S, Bacchus L. Patient safety and adverse maternal health outcomes: the missing social inequalities ‘lens’. British Journal of Obstetrics and Gynaecology (2005) 112:1339–43.
Nolen LB, Braveman P, Dachs JN, et al. Strengthening health information systems to address health equity challenges. Bulletin of the World Health Organization (2005) 83:597–603.[Web of Science][Medline]
Peabody JW, Rahman MO, Gertler PJ, et al. Policy and health: implications for development in Asia (1999) Cambridge: Cambridge University Press.
Pitchforth E. Emergency obstetric care: needs of poor women in Bangladesh. In: PhD Thesis (2004) University of Aberdeen, UK. unpublished.
Pitchforth E, van Teijlingen E, Graham W, et al. Getting women to hospital is not enough: a qualitative study of access to emergency obstetric care in Bangladesh. Quality and Safety in Health Care (2006) 15:214–9.
Russell S. Illuminating cases: understanding the economic burden of illness through case study household research. Health Policy and Planning (2005) 20:277–89.
Simanowitz A, Nkuna B, Kasim S. Overcoming the obstacles of identifying the poorest families: using Participatory Wealth Ranking (PWR), the CASHPOR House Index (CHI), and other measurements to identify and encourage the participation of the poorest families, especially the women of those families (2000) Brighton: Institute of Development Studies.
StataCorps. Stata statistical software: release 8.0 ( 2003) College Station, TX: Stata Corporation.
Tabachnick BG, Fidell LS. Using multivariate statistics (1989) New York: HarperCollins Publishers.
accessed 15 September 2005. United Nations. Undated. UN Millenium Development Goals. Online at: http://www.un.org/millenniumgoals/.
Victora CG, Wagstaff A, Schellenberg JA, et al. Applying an equity lens to child health and mortality: more of the same is not enough. The Lancet (2003) 362:233–41.
UNICEF, WHO, UNFPA. Guidelines for monitoring the availability and use of obstetric services ( 1997) New York: UNICEF.
World Health Organization. Maternal Mortality in 2000: Estimates developed by WHO, UNICEF and UNFPA ( 2004) Geneva: World Health Organization.
Accepted for publication 1 May 2007.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||