Bulletin of the World Health Organization

Can vouchers deliver? An evaluation of subsidies for maternal health care in Cambodia

Ellen Van de Poel a, Gabriela Flores b, Por Ir c, Owen O’Donnell d & Eddy Van Doorslaer a

a. Institute of Health Policy and Management, Erasmus University Rotterdam, PO Box 1738, 3000 DR Rotterdam, Netherlands.
b. Institute of Health Economics and Management, University of Lausanne, Lausanne, Switzerland.
c. Institute of Tropical Medicine, Antwerp, Belgium.
d. University of Macedonia, Thessaloniki, Greece.

Correspondence to Ellen Van de Poel (e-mail: vandepoel@bmg.eur.nl).

(Submitted: 19 August 2013 – Revised version received: 05 November 2013 – Accepted: 28 November 2013 – Published online: 17 March 2014.)

Bulletin of the World Health Organization 2014;92:331-339. doi: http://dx.doi.org/10.2471/BLT.13.129122

Introduction

Vouchers are increasingly being used to encourage the utilization of maternity services with the objective of bringing neonatal and maternal mortality rates closer to the targets set by Millennium Development Goals 4 and 5. Vouchers have been introduced in Bangladesh,13 China, India,4 Indonesia, Kenya,5,6 Pakistan7 and Uganda and have been associated with increased utilization of maternal health care.8,9 However, since other interventions have been implemented at the same time, one cannot be sure that the increase is attributable to vouchers alone. Given the presence of alternative interventions that can influence the demand for, and supply of, maternal health care,10 it is important to have evidence that voucher schemes of varying designs can be effective in different geographical and cultural contexts. The aim of this study was to examine the influence of vouchers on the utilization of maternal health care in Cambodia.

During the last decade, there has been a remarkable increase in the utilization of maternal health services in Cambodia. Coverage by skilled birth attendants has increased from 32% in 2000 to 71% in 2010, while the proportion of deliveries taking place in health-care facilities rose from 10% to 54%.11 Maternal health-care vouchers, which were introduced in 2007 and are now available in about one third of the country, may have contributed to these improvements. These vouchers can be used to pay for antenatal care, delivery and postnatal care at public health-care facilities whose costs are reimbursed by donor-financed agencies.

Unlike the voucher schemes operating in most other countries,9,12 those in Cambodia do not cover a range of providers but are, instead, restricted to subsidizing maternity care at public facilities, mainly health centres. As such, the vouchers function as fee waivers. In addition, the schemes have two other important features: health-care facilities are reimbursed for the care provided and women are given information encouraging them to use maternal health care in these facilities.

Between 2007 and 2010, voucher schemes were implemented in 22 of 77 operational health districts in Cambodia. In 14 districts, the voucher scheme was universal, whereas in 8 it targeted the poorest women (detailed information on the roll out of the voucher schemes is available from the authors on request). In both types of scheme, pregnant women were identified mainly by local health volunteers, who distributed the vouchers and provided advice on safe motherhood at village meetings with the aim of making women aware of the benefits of using maternal health care at public facilities.

At the end of each month, health centres were paid for each voucher coupon collected in accordance with the posted user fees. In principle, the universal voucher scheme provided reimbursement only when all components of a package of antenatal care, delivery and postnatal care had been completed. But, in practice, a health centre may have been paid for a delivery even though it did not provide proof that the woman had completed all the required antenatal and postnatal care visits and women may not have been reimbursed for fees they paid for antenatal care after completion of the care package. Table 1 provides details of the two types of voucher schemes.13

To isolate the effect of vouchers on the utilization of maternal health care, it is important to control for other interventions that could have an influence. Since 1999, various forms of performance-based financing have linked the funding of public health-care facilities in some operational districts to predefined targets, most of which involved the provision of maternal and child health services.14 Health equity funds and a government fee waiver scheme compensate facilities, mostly hospitals, for exempting poor patients from the need to pay fees.15,16 In addition, at the end of 2007, the government introduced the nationwide Midwife Incentive Scheme, which pays midwives 10 United States dollars (US$) for each live birth they attended in a referral hospital and US$ 15 for each delivery they attended in a health centre, on top of the fee charged to the patient.13,17

Methods

We used data from the 2010 Cambodian Demographic and Health Survey (DHS), a nationally representative sample of 18 754 women of reproductive age. The women were asked about their use of maternal health care for pregnancies in the previous 5 years. We concentrated on services covered by the voucher schemes: antenatal care, birth (i.e. delivery) in a public health-care facility and postnatal care from a skilled provider. Table 2 describes how, and for which births, outcome variables were assessed and reports the mean value of each outcome variable observed in the 2010 Cambodian DHS sample.

In estimating the effect of vouchers on the utilization of maternity services, we controlled for characteristics of the child, mother, household and operational district (Table 3). Covariates for the child included the child’s sex, the mother’s age at childbirth and indicators of a short birth interval and of birth order. Maternal characteristics were the mother’s age at first marriage, experience with modern contraception and educational level. Household socioeconomic status was estimated using quintiles of a wealth index that was devised by principal component analysis of a set of household assets and dwelling characteristics.18 Other household covariates included the age and sex of the head of the household. Time-invariant differences across districts were taken into account by including an indicator for each operational district. In addition, we included indicators of whether or not a health equity fund, a government fee-waiver scheme or performance-based financing was in place in the operational district at the time of each child’s birth and an indicator of whether the village location was urban or rural. We confirmed that the results of the study were not substantially affected by using a single indicator for any type of performance-based financing instead of separate indicators for the different types of performance-based financing that may have been in place during the study period. Data from the Cambodia Health Coverage Plan indicated that there were no systematic or substantial differences in the supply of health care at baseline between operational districts that implemented a voucher scheme and those that did not (details are available from the authors on request).

The Global Positioning System codes of the sample clusters in the 2010 Cambodian DHS were used to identify the operational district within which each commune was located. However, for reasons of confidentiality, the Global Positioning System codes were made slightly inaccurate and we were unable to identify the operational districts of 82 communes out of a total of 611. Since 27 of these 82 unmatched communes were located close to Phnom Penh, the mothers resident in them (i.e. 13% of the total) were somewhat better-off than average and more likely to live in an urban area. No other important difference in covariates was found between the matched and unmatched communes.

Statistical analysis

We assessed the effect of a voucher scheme by comparing changes in the utilization of maternal health care in operational districts in which the scheme was introduced in the period from 2005 to 2010 (i.e. intervention districts) with changes in districts that did not introduce vouchers during the same period (i.e. control districts). We assumed that, after taking covariates into account, the use of maternal health care in intervention districts would have evolved in the same way in the absence of vouchers as it did in the control districts – the common trends assumption. Under this assumption, the plausibility of which is assessed in the results section, the difference-in-differences strategy isolates the part of the before-and-after change that is attributable to the effect of vouchers.19 We implemented this strategy by estimating a logit model for each maternal health care utilization measure (i.e. antenatal care, delivery and postnatal care). The model included: an indicator of whether a voucher scheme was operating in the district at the time of pregnancy or delivery; indicators of the month and year of each child’s birth, which captured any time variation in utilization that was assumed to be common to intervention and control districts; a full set of district (fixed) effects that captured time-invariant differences between districts; and time-varying child, mother and district covariates that allowed for any differences in trends in these observable determinants between intervention and control districts and also increased precision. In examining the choice of place of delivery, we estimated a multinomial logit model with a birth year indicator instead of birth month and year indicators, as this model had difficulty converging with the larger number of time variables.

For each maternal health-care utilization measure, we present the estimated effect of the voucher scheme as the resulting change in the predicted probability of utilization averaged over all births taking place in intervention districts when the voucher scheme was operational. This corresponds to the estimated average effect of the voucher scheme among pregnant women and women giving birth in the intervention districts.20

We evaluated whether the effect of the voucher scheme on the behaviour of poor and non-poor women differed using, for each maternal health-care measure, an extended logit model that included the interaction between the voucher scheme indicator and an indicator of whether or not the household was in the bottom two wealth quintiles. Although the second indicator was intended to identify the poorest women, it did not use the same definition of poverty as targeted voucher schemes, which would probably, in any case, have made inclusion and exclusion errors in identifying the poor. The effect of the voucher scheme indicator was calculated for the poorest 40% of women and for the remaining 60%, to whom we refer as the “poor” and “non-poor”, respectively.

We assessed the differential effect of the two types of voucher scheme by replacing the single indicator for any voucher scheme by two indicators for the universal and targeted voucher schemes, respectively. In addition, we examined whether the effects of the two voucher schemes differed by poverty status. In four operational districts, there was a transition from a targeted scheme to a universal scheme during the study period. We decided which of the two voucher scheme indicators to use by comparing the child’s date of birth with the date on which the transition occurred.

Standard errors were adjusted for clustering at the operational district level to allow for any shocks to the outcome at that level which may be correlated over time, such as the outbreak of an infectious disease or fluctuations in the local economy. Such shocks could have resulted in an overstatement of the precision of the estimates.21,22 The significance of the effects was evaluated using Z-tests. All statistical calculations were performed using Stata version 12 (StataCorp. LP, College Station, United States of America).

Results

Overall, the use of maternal health care increased substantially between 2001 and 2009. Fig. 1 shows the trends in the use of antenatal care, delivery in a public health-care facility and postnatal care in both intervention and control districts.

Fig. 1. Trends in maternal health-care service utilization, by voucher scheme,a Cambodia, 2001–2009
Fig. 1. Trends in maternal health-care service utilization, by voucher scheme,<sup>a</sup> Cambodia, 2001–2009
a Voucher schemes were implemented gradually in intervention districts from 2007 onwards.

To examine the plausibility of our assumption that there was a common trend in the way the use of maternal health care would have changed in intervention and control districts in the absence of vouchers, we used data from both 2005 and 2010 Cambodian DHSs, which enabled us to examine data on births from as early as 2000. Fig. 1 shows that the trends in each outcome were reasonably parallel before the start of the voucher schemes in 2007. Thereafter, they began to diverge, especially the trend for postnatal care. In addition, the data presented in Table 3 also give credibility to the common trend assumption: the means of the household and mother covariates at baseline in 2005 were similar in the future intervention and control districts and the null hypothesis that there was no association between the introduction of a voucher scheme and the change in the mean of each covariate was never rejected (i.e. the P-values from t tests all exceeded 0.1).

Fig. 2 shows the sharp decrease in home births and the increase in births in public health-care facilities that occurred in both intervention and control districts between the 2005 and 2010 Cambodian DHSs. The proportion of home births fell more in districts in which a voucher scheme had been implemented by 2010 than in those with no voucher scheme.

Fig. 2. Place of delivery, by voucher scheme,a Cambodia, 2000–2010
Fig. 2. Place of delivery, by voucher scheme,<sup>a</sup> Cambodia, 2000–2010
DHS, Demographic and Health Survey.
a.Voucher schemes were implemented gradually in intervention districts from 2007 onwards.
b Percentages are based on births in the 5 years preceding each survey.

Table 4 shows the estimated average effect of voucher schemes on the utilization of antenatal care, delivery at a public health-care facility and postnatal care. The probability that delivery would take place in a public health-care facility was significantly and substantially increased, by 10.1 percentage points (pp), following the implementation of a voucher scheme. In addition, the probability of receiving postnatal care was increased by 5.3 pp. Voucher schemes had no significant effect on the probability that a woman would receive at least three antenatal care visits, although the estimate is positive. Voucher schemes increased the probability of delivery at a public health-care facility more for women in the poorest 40% of households than for non-poor women: the probability increase in the two groups was 15.6 pp and 5.3 pp, respectively. In addition, Table 4 also shows the specific effects of the universal and targeted voucher schemes. The universal scheme increased the probability that a woman would receive antenatal care by 5.4 pp; among the poor, the increase was 10.1 pp. The effect on the probability of delivery at a public health-care facility was also positive for both types of scheme, although it was larger and more significant for the universal scheme. For both types of scheme, the effect was larger among the poor: the probability of delivery at a public health-care facility for poor women was increased by 11.3 pp and 17.8 pp with the targeted and universal schemes, respectively. Moreover, the effect on the probability of receiving postnatal care was significant for both types of scheme only among the non-poor: the probability increase was 5.6 pp and 6.0 pp with the targeted and universal schemes, respectively.

Table 5 shows the estimated effect of voucher schemes on the probability of giving birth at one of three types of health-care facilities – public hospitals, public health centres and private facilities – as derived using the multinomial model. Voucher schemes increased the probability of delivery at a health centre by 7.4 pp; among the poor, the increase was 10.5 pp. In contrast, there was no significant effect on delivery at a public hospital or private facility. The main shift was, therefore, from delivery at home to delivery at a health centre. The estimated effect of vouchers on the location of delivery was similar for universal and targeted schemes.

Discussion

Vouchers for maternal health care at public facilities contributed to the substantial increase in the number of deliveries taking place in these facilities in Cambodia. The operation of voucher schemes raised the probability that a woman would give birth in a public facility by around 10 pp. This corresponds to about one fifth of the average increase during the study period in the proportion of births taking place in these facilities in districts with voucher schemes; the proportion increased from 17% in 2005 to 68% in 2010. Other interventions, such as incentive payments to midwives, were relatively more important at the national level.

Our study, like most similar studies, was not able to detect meaningful changes in maternal or neonatal mortality associated with the introduction of vouchers. The size of the sample would have enabled us to detect a minimum change in neonatal mortality of about 2 pp, which is two thirds of the average mortality rate in the sample. However, by using a central estimate from the literature that institutional delivery lowers the probability of neonatal death by 0.29 in low- and middle-income countries,23 we estimate that the introduction of vouchers would produce a 3% relative reduction in neonatal mortality.

In our study, voucher schemes were associated mainly with a shift from delivery at home to delivery in public health centres. There was no significant effect on the probability of delivery in public hospitals. Nor was there a significant effect on the caesarean section rate (data not reported). Although no data were available on the direct effect of vouchers on referrals, the fact that vouchers did not influence the probability of delivery in a public hospital or the caesarean section rate suggests that voucher schemes were not encouraging health centres either to refrain from referring problematic cases to hospitals so they could retain the reimbursement or to prefer normal deliveries and leave complicated deliveries unsupervised at home.24 Moreover, any unintended incentive to refrain from referring problematic deliveries to hospital is minimized with universal voucher schemes, which ensure that health centres are paid even when a referral is made. Furthermore, when we analysed deliveries at home according to whether the birth was supervised by a skilled attendant or an untrained attendant, usually a traditional birth attendant, we found no significant effects of vouchers on the former category (data not reported), which suggests that vouchers encouraged women who would otherwise have given birth at home without a skilled attendant to give birth in a health centre.

The effect of voucher schemes on the probability of delivery in a public health-care facility was significant for poor women but not for non-poor women. This observation was true even for universal schemes, which suggests that mainly the poorest women were encouraged by vouchers to give birth in a public facility, perhaps because of a perception that the quality of care is lower and staff attitudes are worse than in private facilities.13 Targeting may, therefore, increase the administrative costs of a voucher scheme without having an effect on how different population groups benefit from the vouchers.

Only universal voucher schemes had a significant impact on the probability of receiving antenatal care. This may have been because universal voucher schemes funded more antenatal care visits than targeted schemes (Table 1) and because they were designed to reimburse the package of care as a whole. However, both schemes did increase the probability of receiving postnatal care from a skilled provider, but the effect was significant only among non-poor women, perhaps because more women were classified as non-poor than poor. Moreover, we had no information on where postnatal care was received and it is possible that the existence of voucher schemes increased awareness among non-poor women and encouraged them to visit private facilities.

The only other study of a maternal health-care voucher scheme that was capable of quantifying its effects was carried out in Bangladesh. It found a slightly larger increase in the probability of institutional delivery than our study – 14 pp – and a much larger increase in the probability of receiving three antenatal care visits – 24 pp.1 However, the Bangladesh scheme was more generous and included a US$ 30 conditional cash transfer.

The nonrandomized implementation of voucher schemes in Cambodia prevents us from ruling out the possibility that our findings may have been biased by the existence of another programme, or some other confounder, that was present at the same time as the voucher schemes and may have influenced the utilization of maternity care. However, we reduced the risk of this occurring by controlling for the presence of various performance-based financing and fee waiver schemes. In addition, we also reduced the risk of bias by controlling for child, mother and household characteristics, which helped to ensure that the composition of the intervention and control groups was comparable over time. Our assumption that there was a common trend in the way the use of maternal health care would have changed in intervention and control districts in the absence of vouchers was supported by data showing that the trends in maternity care use and covariates were similar in intervention and control districts before the introduction of voucher schemes and by the observation that there was no difference between intervention and control districts in the trends in covariates after their introduction.

Since we estimated the effect of a voucher scheme on all pregnant women living in the district in which it was operating, the effect should be interpreted as an intention-to-treat effect. Consequently, if coverage of the target population was incomplete, the actual effect of the voucher scheme would have been greater.

Notwithstanding these limitations, our results suggest that voucher schemes for maternity care in public health-care facilities can increase deliveries in health centres and, to a lesser extent, improve antenatal and postnatal care. Unfortunately we were not able to assess the cost-effectiveness of voucher schemes in Cambodia because we had no data on costs. Cost information is important for comparing voucher schemes for public health care with conditional cash transfer schemes or schemes in which vouchers can be traded for care provided by the private sector with the aim of stimulating competition between providers and improving quality.


Acknowledgements

We thank Timothy Johnston and Emre Ozcan, previously at the Phnom Penh office of the World Bank, Joan Bastide and individuals who commented on our work at HEFPA workshops in Hanoi and Yogjakarta, the Second Global Symposium on Health Systems Research in Beijing, the PBF workshop in Bergen and at the 9th iHEA World Congress in Sydney.

Funding:

The study was funded through EU-FP7 research grant HEALTH-F2-2009-223166-HEFPA on Health Equity and Financial Protection in Asia. Ellen Van de Poel was supported by the Netherlands Organization for Scientific Research (Veni project 451-11-031).

Competing interests:

None declared.

References

Share