Menu
Menu
Menu
Menu
Substantial investments have been made in clinical social franchising to improve quality of care of private facilities in low- and middle-income countries but concerns have emerged that the benefits fail to reach poorer groups. We assessed the distribution of franchise utilization and content of care by socio-economic status (SES) in three maternal healthcare social franchises in Uganda and India (Uttar Pradesh and Rajasthan). We surveyed 2179 women who had received antenatal care (ANC) and/or delivery services at franchise clinics (in Uttar Pradesh only ANC services were offered). Women were allocated to national (Uganda) or state (India) SES quintiles. Franchise users were concentrated in the higher SES quintiles in all settings. The percent in the top two quintiles was highest in Uganda (over 98% for both ANC and delivery), followed by Rajasthan (62.8% for ANC, 72.1% for delivery) and Uttar Pradesh (48.5% for ANC). The percent of clients in the lowest two quintiles was zero in Uganda, 7.1 and 3.1% for ANC and delivery, respectively, in Rajasthan and 16.3% in Uttar Pradesh. Differences in SES distribution across the programmes may reflect variation in user fees, the average SES of the national/state populations and the range of services covered. We found little variation in content of care by SES. Key factors limiting the ability of such maternal health social franchises to reach poorer groups may include the lack of suitable facilities in the poorest areas, the inability of the poorest women to afford any private sector fees and competition with free or even incentivized public sector services. Moreover, there are tensions between targeting poorer groups, and franchise objectives of improving quality and business performance and enhancing financial sustainability, meaning that middle income and poorer groups are unlikely to be reached in large numbers in the absence of additional subsidies.
In each setting, we undertook a cross-sectional survey of women who had attended a social franchise facility for ANC and/or delivery care and who had delivered at the time of the survey. We aimed to survey a total of 760 women in each of the three study settings. The sample size was estimated to allow detection of a difference of 50% between two equally sized groups (e.g. wealthiest and poorest) for a proportion of 50%, with power of 80%, significance level of 5% and an estimated design effect of two to account for clustering at facility level. In each facility, we contacted more women than we targeted to interview to account for an estimated non-response rate of 20% and low utilization of some facilities. In Uganda, we randomly selected 15 out of the 140 ProFam facilities. The sample was stratified by whether the facility provided C-sections or not (selecting 4 out of 16 C-section facilities and 11 out of 124 other facilities). Of the total sample of 15 facilities, eight were PFP and seven private not-for-profit (PNFP). Similarly, in Rajasthan, we randomly selected 15 out of 57 Merrygold facilities stratified by level (10 out of 19 urban facilities and 5 out of 38 rural/peri-urban facilities). In Uttar Pradesh, out of the 50 SkyHealth facilities we randomly selected 12. For each program, women eligible for the survey were identified through available records, together with their contact details. Target numbers of women to recruit from each facility were set in proportion to estimated utilization, as reported by the implementing NGO. In Uganda, all facilities kept records using standard Health Management Information System (HMIS) books and it was possible to randomly select our sample from these records. In India, the data we could access from the facilities were limited: they varied by format, content and completeness. Sometimes they were even not available at all. As a result, in Rajasthan we obtained women’s details from the Merrygold registers maintained by Outreach District Coordinators, who are implementing agency staff based at the clinics. In Uttar Pradesh, we had to rely on a mix of data from facility records (4 facilities), implementing agency district coordinators (7 facilities) and community health workers (1 facility). In both Rajasthan and Uttar Pradesh, we requested the names of all clients who had delivered in the year prior to the survey but in both cases were only provided a sub-set of these, and it was not clear how this sub-set had been selected. For all facilities in Rajasthan and three in Uttar Pradesh, we aimed to interview all women from the lists provided since the numbers were close to our targets for these facilities. In the other Uttar Pradesh facilities, where the lists were larger than our target, we randomly selected from these lists. We contacted women by telephone to arrange appointments and community health workers often assisted the team in identifying their addresses. We requested informed written consent (or oral witnessed consent in the case of illiterate participants) from all women located, and if they agreed, we carried out an interview. Women were assured about the confidentiality of their answers. In Uganda, based on facility utilization and target numbers of women per facility, we interviewed women who had delivered in the last 4 months prior to the survey, giving a time lag of 6–10 months between the first ANC visit and the survey. In India, because of poor record keeping and low patient volumes in some facilities, we extended the recruitment period to 1 year prior to the survey, giving a time lag of 6–18 months from first ANC to survey. To aid women’s recall, in Uganda enumerators asked to see the women’s ANC cards where available, which contain detailed information about their pregnancy (such cards were not available in India). Data were collected from July to November 2015 in Uganda and from March to June 2016 in India. Response rates in Uganda, Rajasthan and Uttar Pradesh were 74.5, 76.2 and 71.7%, respectively, with the main reasons for non-response being that mobile numbers were either missing from the records or were wrong (incomplete number of digits) or nobody answered the call. In Uttar Pradesh, we managed to contact most women from the sampling lists from all but one facility, which had a particularly low response rate (3/56), which reflected a high number of inaccuracies in those records. Data were double entered and analysis was conducted in Stata 14. The analysis was weighted to reflect variation in sampling probability across facilities and across women within facilities, thereby producing estimates that were representative of all women using the social franchise network. Each woman was given a specific weight relevant to (1) the stratum-specific probability of the facility she visited being sampled and (2) her probability of being selected within that facility. The second probability varied depending on whether the analysis concerned women attending for ANC, delivery or either service. To assign the sampled women to SES groups, in each setting we derived asset weights and SES quintile cut-offs from an existing household survey that was representative of the whole country (Uganda) or State (India). In Uganda, we used the 2011 Ugandan Demographic Health Survey (DHS). The most recent Indian DHS was quite dated (2005–06) so we used the 2012 Indian Human Development Survey (IHDS) for Rajasthan State for Merrygold, and for Uttar Pradesh State for the Sky network. In each setting, our survey included all the questions on household characteristics and asset ownership in the DHS/IHDS, e.g. ownership of televisions and bicycles, materials used for housing construction and types of water access and sanitation. For Uganda, we used the asset weights provided on the DHS website (www.measuredhs.com), while asset weights for the IHDS were calculated using principal component analysis. These weights were applied to the assets of each woman’s household in our survey and summed to calculate the wealth score for each woman. Using the SES quintile cut-offs for the asset scores from the DHS/IHDS, we then allocated each woman to a national wealth quintile (Uganda) or state wealth quintile (India). The full list of assets used in each setting and their weights is presented in the Supplementary Appendix S1. Given the construction of wealth quintiles as a relative measure of wealth using national/state populations as references, the meaning of belonging to a specific wealth category will differ according to the average wealth of the population of reference. For example, in 2015 Gross Domestic Product (GDP) per capita ranged from 1300 USD in Rajasthan to 770 USD in Uttar Pradesh and 609 USD in Uganda (Economic and Statistical Organization Government of Punjab, 2016; www.esopb.gov.in) (www.knoema.com), meaning that a household in Quintile 5 in Rajasthan will be substantially wealthier in absolute term than a household in the fifth quintile in Uganda. It should also be noted that in poorer countries even some of those in higher quintiles would still be considered as poor in absolute terms. For example, 34.6% of the Ugandan population lived on less than the standard poverty lines of 1.9 USD per day in 2012 and 69.4% on <3.1 USD per day (2011 Purchasing Power Parity (PPP)) (www.data.worldbank.org), meaning that even the fourth quintile will include people living under accepted poverty thresholds. We also assessed the coverage and equity of the content of ANC and delivery care received. We included only care received at franchise facilities, though we recognize that women may have received some components of ANC at other public or private providers during the course of their pregnancy. For ANC, we assessed all components of care included in the DHS: weight measurement, blood pressure measurement, urine test, blood test, discussion about previous pregnancy complications, iron supplementation, malaria prophylaxis and deworming tablets (the latter two are only relevant for Uganda, as they were not included in government ANC guidelines in India). For delivery content, we selected the DHS indicators which we felt women could reasonably be expected to recall during a household survey: blood pressure taken upon arrival at the facility, presence of a person for support during labour, type of delivery, baby immediately dried and wrapped, baby weighed at birth. We also included two indicators on disrespect and abuse. Concentration indices were used to summarize the socio-economic distribution for each content of care indicator. The concentration index, ranging from −1.0 to +1.0, captures the extent to which a health variable is distributed among the economically worse off as compared with the better off. The convention is that the index takes a negative value when the health variable is disproportionately concentrated among the poor and a positive value when it is concentrated among the better off (O’Donnell and Van Doorslaer, 2008). We used the methods proposed by Erreygers (2009) to derive the concentration indexes. Ethical clearance was obtained from the London School of Hygiene and Tropical Medicine (LSHTM) Makerere University and Gene Bandhu (NGO) Ethics Committees.
