Abstract
We aimed to measure the contribution of national factors, particularly health system characteristics, to the individual likelihood of professionally attended delivery (“safe delivery”) for women in low- and middle-income countries. Using Demographic and Health Survey data for 165 774 women in 31 countries, we estimated multilevel logistic regression models to measure the contribution of national economic and health system characteristics to likelihood of attended delivery. More health workers, higher national income, urbanization, and lower income inequality were associated with higher odds of attended delivery. Macrosocial factors increase utilization of attended delivery and may be more efficient in reducing maternal mortality than are interventions aimed at individual women.
As of 2010 it appears that Millennium Development Goal (MDG) 5, to reduce by three quarters the maternal mortality ratio between 1990 and 2015, is unlikely to be achieved in many low-income countries.1,2 Despite the acknowledged importance of professionally attended delivery in reducing maternal mortality, attended delivery rates are low in many parts of the world.3,4 In the 68 countries that accounted for 97% of child and maternal deaths globally, only half of all deliveries were attended by a health professional, and the rate of increase in professional attendance was among the slowest of all MDG health interventions.1
Much of the research on low utilization of attended delivery focuses on individual predisposing and enabling factors. However women make delivery decisions within a community and national context, and little is known about the interplay between systemic factors and individuals’ delivery choices. In particular, at a time of rising interest in health system strengthening, to what extent do stronger health systems contribute to greater take-up of this essential service for women with low preexisting likelihood of attended delivery? Multilevel analysis is a method well suited to identifying and quantifying the relative influence of nested determinants on individual outcomes. Only 1 study has used this approach, but that study focused on the effect of maternal age rather than national factors.5
Building on this work, we aimed to measure the contribution of national factors, particularly health system characteristics, to the individual likelihood of professionally attended delivery (“safe delivery”) for women in low- and middle-income countries. To this end we explored how the national-level determinants mitigate individual-level barriers to professionally attended delivery.
METHODS
Our main data source was the Demographic and Health Surveys (DHS) conducted from 1996 onwards. Our dependent variable was whether a woman’s most recent birth within 5 years before the survey was conducted by a skilled birth attendant: a doctor, nurse or midwife. Individual-level characteristics included were the woman’s age, previous delivery (prior to the index delivery), education (any education versus none), and household wealth (relative wealth quintile constructed using an index of household assets in each country).3,4,6,7 The community-level variable was urban versus rural residence.8,9 The national-level variables were gross national income per capita (GNI), the proportion of the total health expenditure financed by the government,10,11 Gini coefficient, land area (in km2), health care workers (doctors, nurses, and midwives) per 1000, hospital beds per 1000, and the population size of the country. These variables were based on previous literature in the field.10,12–16
Individual- and community-level characteristics and the dependent variable were taken from the DHS survey database. Country-level variables were obtained from the World Bank Development Indicators and the World Health Organization National Health Accounts. All country-level data were lagged 1 year or the closest year available within 5 years. Because of the limited availability of the health care resource variables the value closest to the year of the DHS survey was used within 10 years.
Complete data were available for 31 countries. Where necessary to improve linearity, variables were log transformed. We estimated 3 separate 3-level logistic random intercept models: a null model without covariates, an individual-factor only model, and a full model with individual, cluster, and national factors.9,17 Multilevel models have the advantage of adjusting standard errors for the complex sample design of DHS surveys.18 We calculated the proportion of variability in odds of professional delivery explained by model variables at each level.
Finally, to illustrate the incremental effect of national income and health workers, we compared the predicted probabilities of attended delivery for individuals with most adverse predisposing individual determinants as calculated using the parameters in the multilevel model: aged 17 years, lowest income quintile, no education, delivered previously, urban. The probability of delivery for a woman with these characteristics was modeled at increasing values of national GNI per capita and health care workers per 1000. All modeling and calculation of predicted probabilities were performed using Stata/IC version 10.0 (College Station, TX).
RESULTS
Table 1 lists descriptive statistics for the variables in the models. Table 2 lists the characteristics of the 31 countries in the analysis.
TABLE 1—
Sample Characteristics at Randomization, by Study Group: Michigan, 1997–2001
| Characteristic | Community Care Group (n = 264), No. (%) | Nurse–Community Health Worker Group (n = 266), No. (%) | χ2 P |
| Age, y | .22 | ||
| < 20 | 90 (34.1) | 73 (27.4) | |
| 20–25 | 123 (46.6) | 141 (53.0) | |
| > 25 | 51 (19.3) | 52 (19.6) | |
| Race | .95 | ||
| African American | 72 (27.2) | 71 (26.7) | |
| Hispanic | 62 (23.5) | 62 (23.3) | |
| White | 110 (41.7) | 109 (41.0) | |
| Other | 20 (7.6) | 24 (9.0) | |
| < 12 y of education | 156 (59.1) | 147 (55.3) | .37 |
| Unmarried | 220 (83.3) | 218 (82.0) | .68 |
| Unemployed | 148 (56.1) | 154 (57.9) | .67 |
| Prior live birth | 143 (54.2) | 153 (57.5) | .44 |
| Unplanned pregnancy | 208 (78.8) | 208 (78.2) | .87 |
| Current tobacco use | 88 (33.3) | 85 (32.0) | .74 |
| Current drug use | 22 (8.33) | 16 (6.0) | .3 |
| Current alcohol use | 12 (4.6) | 10 (3.8) | .65 |
| History of physical abuse | 141 (53.4) | 133 (50.0) | .43 |
| Depressive symptomsa | 154 (58.3) | 145 (54.5) | .38 |
As indicated by a score of 16 or higher on the Center for Epidemiologic Studies Depression Scale.
TABLE 2—
Types of Perceived Help Reported by Respondents, Ranked in Decreasing Order of Prevalence: Michigan, 1997–2001
| Item | Community Care Group (n = 249), Rank (%) | Nurse–Community Health Worker Group (n = 249), Rank (%) |
| Gave you things to read | 1 (59.51) | 1 (71.95) |
| Helped you learn about child development | 2 (49.39) | 2 (68.72) |
| Taught about birth control | 3 (48.58) | 3 (66.39) |
| Helped keep clinic appointments | 4 (39.11) | 9 (45.71) |
| Gave you a feeling that you belonged | 5 (39.11) | 4 (66.13) |
| Chance to get feelings out | 6 (36.69) | 5 (59.51) |
| Helped with transportation | 7 (32.13) | 12 (37.40) |
| Helped give children better start | 8 (28.96) | 8 (47.09) |
| Helped to have confidence in self | 9 (26.12) | 6 (50.21) |
| Helped have a happier life | 10 (21.46) | 10 (39.84) |
| Person to talk to who cares | 11 (20.97) | 7 (47.97) |
| Helped understand self | 12 (19.76) | 11 (38.37) |
| Provided wake-up (or other reminder) calls | 13 (17.34) | 17 (25.10) |
| Helped with child care | 14 (16.13) | 22 (17.50) |
| Helped with emergency | 15 (16.13) | 14 (29.46) |
| Made phone calls to advocate for you | 16 (15.45) | 13 (29.80) |
| Helped getting a doctor | 17 (15.38) | 21 (17.96) |
| Helped get furniture | 18 (15.04) | 16 (26.23) |
| Helped getting along with family | 19 (12.90) | 23 (17.28) |
| Helped with bad habit (e.g., smoking, eating too much) | 20 (11.69) | 31 (13.11) |
| Helped plan daily schedule | 21 (11.69) | 19 (21.90) |
| Helped planning for future | 22 (11.29) | 15 (27.05) |
| Helped learn homemaking skill | 23 (9.27) | 30 (14.17) |
| Helped going back to school | 24 (9.24) | 25 (16.46) |
| Helped get along with partner | 25 (8.94) | 20 (19.17) |
| Helped understand others better | 26 (8.50) | 18 (22.04) |
| Helped find housing | 27 (7.29) | 29 (15.00) |
| Helped make new friends | 28 (5.65) | 26 (16.33) |
| Helped find job | 29 (5.65) | 28 (15.10) |
| Helped with budget | 30 (4.84) | 27 (15.16) |
| Helped with moving | 31 (4.42) | 32 (7.79) |
| Went to agency with you | 32 (3.23) | 24 (16.94) |
Table 3 shows the results of regression models. In all models older women, those without any education, those women from lower wealth groups, and those who had previously given birth were less likely to have a professionally attended birth. Adjusting for individual- and community-level effects, a woman who lived in an urban community, a country with higher levels of public health spending, greater national wealth, less income inequality, and more health resources had significantly greater odds of having a professionally attended birth. Based on the variance of the random effects, community and national factors explained 66% of the country-level variance whereas individual factors explained 16%. The community and national factors explained 3% of the cluster-level variance whereas the individual factors explained 56% of the cluster-level variance.
TABLE 3—
Descriptive Statistics of Survey Sample: Demographic and Health Surveys, 1997–2008
| Mean (SE) | |
| Level 1: individuals (n = 165 774) | |
| Maternal age, y | 28.7 (0.21) |
| People per income quintile for country, % | |
| Lowest | 22 (0.3) |
| Lower middle | 21 (0.3) |
| Middle | 21 (0.4) |
| Higher middle | 19 (0.2) |
| Highest | 17 (0.3) |
| Mother’s who had some primary education, % | 64 (6) |
| Women who have had a previous birth, % | 77 (1) |
| Level 2: clusters (n = 18235) | |
| Urban, % | 38 (4) |
| Level 3: countries (n = 31) | |
| Public health expenditure as a percentage of the total health expenditure | 49 (4) |
| GNI per capita, PPP | 1892 (342) |
| Average land area, km2 | 475 941 (91 895) |
| Gini index | 0.45 (0.02) |
| Health care workers per 1000 | 1.62 (0.29) |
| Hospital beds per 1000 | 1.16 (0.16) |
| Population, no. | 20 800 000 (4 072 060) |
| Population % that received skilled birth attendance for their last birth within 5 y | 59 (4.6) |
Note. GNI = gross national income; PPP = purchasing power parity. Values were calculated using the survey design and DHS sampling weights. The Gini index is a measure of income inequality, in which 0 is perfectly equal income distribution and 1 is perfectly unequal income distribution.
Figure 1 demonstrates the predicted probabilities of women with factors associated with low probability of professional delivery at varying country-level characteristics.
FIGURE 1—
Predicted probability of professionally attended delivery among women with adverse individual predictors at increasing levels of national income and health workers: Demographic and Health Surveys, 1997–2008.
Note. GNI = gross national income. Predicted probabilities were calculated from the full multilevel model. Individuals for all predictions were women with the least likely probability of a professional delivery (poor, no education, delivered previously, aged 17 years, and had previously given birth). The percentiles of GNI per capita and health care worker per 1000 were calculated as equally distant segments between the minimum and maximum values.
DISCUSSION
We found that several national-level factors were important determinants of the likelihood of professionally attended delivery for a woman living in a low- or middle-income country. As expected, women in wealthier countries had higher odds of professionally attended delivery, whereas greater wealth inequality within countries was associated with lower odds of attended delivery. In countries where wealth is concentrated among a few, more women will be left without resources to seek professional assistance in childbirth, compared with countries with similar but more equitably distributed income.19–21
Health system structural factors were also important: for every 10% increase in health care workers per 1000 the odds of a woman having a safe delivery increased by 8%. This finding corresponds to ecological studies that found positive associations between country wealth, health resources and overall proportion of skilled birth attendance.12 In addition to improving health outcomes, better resourced health systems encourage greater utilization.22 Unlike in other studies, government’s role in health spending was not associated with the outcome, likely because countries with greater government spending also have more health workers and hospital beds, which were separately included here.10,23
Living in a larger and more populous country tended to reduce the odds of safe delivery, although these characteristics were not significant at the 5% level in this analysis, unlike in other studies.13,24 Larger and more populous countries require proportionally greater health system and other investments, such as education and transportation, to increase coverage of professional delivery.
Urban women were 4 times as likely to have professional assistance at delivery, as were their rural counterparts. This is likely a result of easier access to health professionals in cities (supply side) and potentially a change in health care seeking resulting from urbanization that favors modern health care (demand side).12,14 This urban utilization advantage is consistent with data from individual low-income countries, such as Nigeria, and ecological studies.12,14
All individual-level variables were associated with facility delivery in the expected directions. Having had more than 1 previous childbirth before the index delivery reduced the odds of safe delivery by half whereas having any education doubled the odds. Particularly notable was the dose-response nature of the wealth variable with women in the highest income quintile having a 9-fold likelihood of attended delivery compared with women in the poorest quintile.
To aid interpretation of the impact of contextual versus individual factors on the likelihood of safe delivery, we modeled the change in odds of safe delivery for a woman with the most adverse individual predictors for different levels of 2 national variables: gross national income and health workers per 1000 population. We found that in a rural area, that woman’s individual probability of attended delivery would increase from 9% in a country at the 10th percentile of GNI and health workers (668 purchasing power parities [PPPs] per capita and 0.5 health workers per 1000) to 76% in a country at the 70th percentile (3027 PPPs and 4.6 health workers). In this sense, context can substantially mitigate negative individual determinants to increase the odds of professionally attended delivery for women. It is plausible that the rapid urbanization seen today in Asia, Africa, and other low-income regions, if accompanied by economic growth and requisite health system investments, will bring about rapid changes in uptake of professional birth attendants.
Our study had several limitations. Because we used cross-sectional DHS data, we cannot infer causality. Although we lagged our national economic variables, the actual time lag for the effects of GNI or Gini coefficient on our outcome of interest is not known. Although unobserved error was incorporated into the model because of the multilevel framework, it is possible that unobserved variables may confound our estimates. These limitations emphasize the need for more longitudinal research to explore causal associations.
In conclusion, these findings provide insight into the disparate progress in reducing maternal mortality among low- and middle-income countries, confirming the important role of economic factors and national health systems in promoting safe delivery. While individual-level factors remain important, this work highlights the substantial contribution of structural changes to improving population health. Our work supports the recent WHO Commission on the Social Determinants of Health that argues for more focus on contextual factors in addressing population health needs.20 Macrosocial and structural solutions, such as health system investments and policies promoting economic growth and income redistribution, may be both effective and efficient in promoting utilization of life-saving maternal health care.
Acknowledgments
We gratefully acknowledge the contribution of Sandro Galea to initial conceptualization of this article.
Human Participant Protection
No human participant protection was required because all data used in the analysis were secondary, de-identified data.
References
- 1.Bryce J, Daelmans B, Dwivedi Aet al. Countdown to 2015 for maternal, newborn, and child survival: the 2008 report on tracking coverage of interventions. Lancet. 2008;371(9620):1247–1258 [DOI] [PubMed] [Google Scholar]
- 2.Dyer O. Some countries will not achieve millennium development goals for health by 2015, WHO says. BMJ. 2005;330(7496):863. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Filippi V, Ronsmans C, Campbell OMet al. Maternal health in poor countries: the broader context and a call for action. Lancet. 2006;368(9546):1535–1541 [DOI] [PubMed] [Google Scholar]
- 4.Koblinsky M, Matthews Z, Hussein Jet al. Going to scale with professional skilled care. Lancet. 2006;368(9544):1377–1386 [DOI] [PubMed] [Google Scholar]
- 5.Magadi MA, Agwanda AO, Obare FO. A comparative analysis of the use of maternal health services between teenagers and older mothers in sub-Saharan Africa: evidence from Demographic and Health Surveys (DHS). Soc Sci Med. 2007;64(6):1311–1325 [DOI] [PubMed] [Google Scholar]
- 6.Koblinsky M. Reproductive and newborn health. J Health Popul Nutr. 2006;24(4):377–379 [PMC free article] [PubMed] [Google Scholar]
- 7.Magadi MA, Agwanda AO, Obare FO. A comparative analysis of the use of maternal health services between teenagers and older mothers in sub-Saharan Africa: evidence from Demographic and Health Surveys (DHS). Social science & medicine (1982) 2007;64(6):1311–1325 [DOI] [PubMed] [Google Scholar]
- 8.Boyle MH, Racine Y, Georgiades Ket al. The influence of economic development level, household wealth and maternal education on child health in the developing world. Soc Sci Med. 2006;63(8):2242–2254 [DOI] [PubMed] [Google Scholar]
- 9.Sepehri A, Sarma S, Simpson W, Moshiri S. How important are individual, household and commune characteristics in explaining utilization of maternal health services in Vietnam? Soc Sci Med. 2008;67(6):1009–1017 [DOI] [PubMed] [Google Scholar]
- 10.Kruk ME, Galea S, Prescott M, Freedman LP. Health care financing and utilization of maternal health services in developing countries. Health Policy Plan. 2007;22(5):303–310 [DOI] [PubMed] [Google Scholar]
- 11.Castro-Leal F, Dayton J, Demery L, Mehra K. Public spending on health care in Africa: do the poor benefit? Bull World Health Organ. 2000;78(1):66–74 [PMC free article] [PubMed] [Google Scholar]
- 12.Anand S, Barnighausen T. Human resources and health outcomes: cross-country econometric study. Lancet. 2004;364(9445):1603–1609 [DOI] [PubMed] [Google Scholar]
- 13.Anand S, Barnighausen T. Health workers and vaccination coverage in developing countries: an econometric analysis. Lancet. 2007;369(9569):1277–1285 [DOI] [PubMed] [Google Scholar]
- 14.Babalola S, Fatusi A. Determinants of use of maternal health services in Nigeria–looking beyond individual and household factors. BMC Pregnancy Childbirth. 2009;9(43):43. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Fraser AM, Brockert JE, Ward RH. Association of young maternal age with adverse reproductive outcomes. N Engl J Med. 1995;332(17):1113. [DOI] [PubMed] [Google Scholar]
- 16.Hotchkiss DR, Krasovec K, El-Idrissi MD, Eckert E, Karim AM. The role of user charges and structural attributes of quality on the use of maternal health services in Morocco. Int J Health Plann Manage. 2005;20(2):113–135 [DOI] [PubMed] [Google Scholar]
- 17.Rabe-Hesketh S, Skrondal A. Multilevel and Longitudinal modeling Using Stata. College Station, Texas: Stata Press; 2008.
- 18.Uthman OA, Moradi T, Lawoko S. The independent contribution of individual-, neighbourhood-, and country-level socioeconomic position on attitudes towards intimate partner violence against women in sub-Saharan Africa: a multilevel model of direct and moderating effects. Soc Sci Med. 2009;68(10):1801–1809 [DOI] [PubMed] [Google Scholar]
- 19.Kondo N, Sembajwe G, Kawachi I, van Dam RM, Subramanian SV, Yamagata Z. Income inequality, mortality, and self rated health: meta-analysis of multilevel studies. BMJ. 2009;339:b4471 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Marmot M, Friel S, Bell R, Houweling TA, Taylor S. Closing the gap in a generation: health equity through action on the social determinants of health. Lancet. 2008;372(9650):1661–1669 [DOI] [PubMed] [Google Scholar]
- 21.Houweling TA, Ronsmans C, Campbell OM, Kunst AE. Huge poor-rich inequalities in maternity care: an international comparative study of maternity and child care in developing countries. Bull World Health Organ. 2007;85(10):745–754 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Kruk ME, Paczkowski M, Mbaruku G, de Pinho H, Galea S. Women’s preferences for place of delivery in rural Tanzania: a population-based discrete choice experiment. Am J Public Health. 2009;99(9):1666–1672 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Rannan-Eliya R, Somanathan A. Access of the Very Poor to Health Services in Asia: Evidence on the Role of Health Systems From Equitap. London: DFID Health Systems Resource Centre; 2005. Available at: http://www.equinetafrica.org/bibl/docs/RANehs.pdf. Accessed February 1, 2012.
- 24.Gage AJ, Guirlene Calixte M. Effects of the physical accessibility of maternal health services on their use in rural Haiti. Popul Stud (Camb). 2006;60(3):271–288 [DOI] [PubMed] [Google Scholar]