From the perspective of community leaders and facility staff, the waiver card system was able to be implemented successfully in Afghanistan in 26 of 27 facilities across 10 provinces piloting the user fee system under the health financing pilot study. Given the flexibility built into the pilot design, as well as substantial local heterogeneity in local community structures in Afghanistan , the waiver card system implementation process was very context-specific and heterogeneous. Waiver cards were distributed to households within one year of pilot implementation and, in most cases, within six months, a relatively short time frame given the lack of additional human and financial resources devoted to the waiver card scheme. However, despite training and continued technical assistance for implementation, this lack of dedicated staff and resources made the waiver card system one of the more time-consuming and challenging aspects of the user fee pilot. Households in the catchment areas of pilot facilities had high awareness of fees, but relatively low awareness of the waiver card system (40.3%) and of exempt preventive/promotive services, indicating incomplete community mobilization and information dissemination.
Facility staff and community leaders involved in the user fee and waiver card systems reported that the targeting process was fair, transparent, and relatively accurate. However, they noted in many cases that more cards were needed as there were not enough to cover the large numbers of poor households in the area, and some mentioned occasional occurrences of leakage to wealthier households. It is important to keep in mind the perspectives and potential biases of the facility staff and village leaders, who were directly involved in the waiver card process and therefore have reason to present themselves and the process as fair and effective. Although there was limited information available to capture household's viewpoints, open-ended survey questions of card-holding and non-cardholding households indicated that they had mostly favorable perceptions of the targeting process among those aware of the waiver card system.
Household survey results using wealth scores derived from PCA on household assets revealed that although the card distribution was overall pro-poor, in the sense that the poor had a higher probably of receiving a card, there was significant under-coverage and leakage, a finding similar to that for the Kartu Sehat (health card) program for the poor in Indonesia [3, 33], where leakage and under-coverage were estimated to be 25.5% and 69.7%, respectively, considering the poorest 40% as "poor" (Authors' own calculations based on ). Even in Cambodia, which has largely been considered an example of successful targeting, one evaluation four years after the initial targeting found under-coverage and leakage to be about 40% each, reflecting imperfections in targeting and the dynamic nature of poverty over time . The goal is not necessarily perfect targeting, and efforts to increase the precision of targeting can have large costs as well, not only in terms of actual costs to more accurately target only the poor, but in terms of political economy as well, as broad-based support for the program across income groups may collapse and undermine the benefit, known as the "paradox of targeting" [8, 30, 41]. Part of the large under-coverage in Afghanistan may have been due to the suggested 15% eligibility for waiver cards. Although facilities and communities were instructed that this was flexible, they may have used this guide too stringently, leading to under-coverage in poorer areas. Interviews from some facilities also revealed that facility staff allowed community leaders to nominate households in each village, but capped the proportion of households in each village to be uniform across the catchment area, even though some villages are poorer than others, further exacerbating under-coverage problems. It was not clear why targeting of female-headed households did not seem to be effective.
A major strength of the waiver card system was pre-identification of households in their communities. Certain health equity funds operating in Cambodia have shown that eligibility screening and identification of poor patients after they reach the hospital results in low awareness of the fund among the target population and continued financial barriers to access among those unaware [42, 43]. However, compared to other pre-identification schemes documented in the literature, this pre-identification process was much less thorough and involved, and only four of the 25 facilities expended efforts to systematically verify the eligibility of nominated households through household visits. Among health equity funds in Cambodia that use pre-identification, although their exact processes for determining beneficiaries varies, nearly all followed up on initial lists of households gathered from government census and from village leaders to systematically evaluate nominated households in each village, and used dedicated staff and explicit scoring of proxy means tests (e.g., demographic and housing characteristics, education of family members, and ownership of various household assets as well as other factors) . Researchers working on evaluating health equity funds in Cambodia have recommended that pre-identification not be done only once but through a series of steps to evaluate the eligibility of households .
It appears that possession of a waiver card led to increased use of services among households, and this effect was even more pronounced for the poor, increasing their care-seeking by 6.2 percentage points. Additional analyses (not shown) indicated that it is likely that the card also led to increased use of delivery services at the facility, although it was not possible to tease apart this effect with accuracy due to relatively low use of delivery services overall and the limited sample size of the household survey. Even though increases in delivery use associated with the card were modest, at just over 3 percentage points, according to the simple probit model, this represents a 33% increase in delivery use compared to non-cardholders, an important gain in a country with high maternal mortality ratios and very low rates of institutional delivery [45, 46]. The findings are also relevant for other demand-side interventions and demonstrate the potential for mechanisms such as conditional cash transfers or other demand-side subsidies to increase use of underutilized services (e.g., delivery, immunizations), as these interventions can be even more comprehensive in addressing financial barriers to care and, in some cases, nonfinancial barriers such as transportation availability.
Cardholders also reported greater care-seeking autonomy among women. While this may reflect selection bias related to these households' propensity to value health and use care, it is also possible that ownership of the waiver card increased the autonomy of women in these households, as they no longer needed to consider the cost implications of seeking care as seriously and did not need to rely as much on the male household authority figures to give them money for user fees. Previous studies from low- and middle-income countries have found that women typically have the responsibility for taking sick children for care, but that men tend to control cash in the household . However, it is not possible to determine with certainty whether waiver cards were given to households with greater women's care-seeking autonomy, or whether they actually helped increase care-seeking autonomy among receiving households, as this variable was not measured at baseline.
The increase in service use from the waiver card is consistent with findings from Cambodia, where equity certificates for pre-identified households led to hospitalization rates among beneficiaries that were one-and-a-half to eight times greater than non-health-equity-fund recipients, and with findings from an evaluation of the Health Care Fund for the Poor in Vietnam [48–51]. Similar results were found in Indonesia, following distribution of health cards to vulnerable households to help counteract the Indonesian economic crisis .
Several important limitations apply to the results from this study. First, it is important to keep in mind that comparisons for the purposes of evaluating targeting accuracy were done without a strong measure of household consumption for comparison. Although they have been shown to be valid indicators of underlying wealth, proxy scores from asset indices created from PCA are just that: proxies. They may better reflect accumulated wealth than more liquid types of cash or assets that can be used to pay facility fees. The PCA methodology is most useful for identifying relative wealth, and not absolute wealth, such as wealth in relation to a poverty line created from household consumption. However, since the PCA was run separately for each province, the wealth scores should reflect local measures of relative wealth that are also important in targeting. Future research on targeting in Afghanistan should consider using both relative and absolute measures of wealth, as well as capturing more in-depth perspectives on wealth and targeting from community members.
In addition, although we did not find strong evidence of endogeneity of the card with the care-seeking outcome, we cannot rule out this possibility and determine definitively whether the card led to increased use of services. Evidence points in that direction, but without baseline data from the same households, it is difficult to conclude with certainty that the waiver card led to increased care-seeking.
Lessons learned and policy implications
If Afghanistan is to re-implement the waiver card system in the future, for example as a formal mechanism to increase access by the poor to services at higher-level hospitals, which still charge fees, what lessons can it learn from this pilot experience and those of its neighbors? Several targeting-related and operational changes should be considered.
First, Afghanistan should build upon the success it has had in implementing a community-nominated waiver card system that was perceived by community leaders and facility staff as relatively transparent and fair. This is an important accomplishment in a post-conflict setting, where trust in institutions, including at the community level, related to publicly financed services, has been worn down from years of fighting and ethnic rivalries. Involvement of community members in beneficiary nomination provides a way to increase community participation in health and contributes to rebuilding trust in institutions at the community level, a particularly important benefit in post-conflict settings .
As suggested in several interviews with facility staff and community leaders, a new waiver card system should consider more active involvement in nominating households by Mullahs/Imams, who are widely respected in communities and tend to have good knowledge of individual household conditions. Religious leaders, viewed as honest and trustworthy, have successfully been involved in community-based targeting in Cambodia, and researchers have pointed to the community-based management of targeting, including high involvement of the pagodas and their trusted volunteers and monks, as one of the important keys to success of the equity fund .
Another contributing factor to the successful community-based targeting used in several Cambodian health equity funds is the presence of clearly defined eligibility criteria , which were much looser in Afghanistan, given the wide geographic variability of the pilots and the absence of cash income in many areas. Although the individual NGOs implementing health equity funds in different regions of Cambodia ended up using different proxy means targeting criteria from one another, researchers have identified the relatively uniform social and demographic conditions in Cambodia as a key factor in the success of health equity funds . Afghanistan is much more diverse geographically, culturally, and socioeconomically than Cambodia. This indicates that a more diverse targeting strategy, including geographic targeting for high-poverty districts where the cost of individual household targeting does not make sense, may be warranted. This approach has been successfully implemented in Vietnam, which uses a mix of individual characteristic and geographical targeting to identify beneficiaries for its Health Care Fund for the Poor .
Despite the geographic, socioeconomic, and cultural diversity in Afghanistan, which may preclude the development of a uniform national set of targeting criteria, specific verifiable criteria should still be developed for future targeting schemes in Afghanistan. This will make eligibility easier to verify and the process more transparent. Similar to what is currently being piloted in Lao PDR, the process could begin in a similar manner as before with a broader list of poor households provided by village leaders, using pre-identification criteria, and then screened by dedicated staff or trained volunteers using the objective criteria . The objective criteria could--and certainly would be expected to--vary by geographic location. Close coordination with existing sources of data on poverty and vulnerability in Afghanistan, such as the biennial
National Risk and Vulnerability Assessment (NRVA) survey, could prove extremely valuable for implementation of successful geographic targeting and development of localized objectively verifiable proxy indicators for community targeting. It would also be worthwhile to pursue collaboration with other sectors, such as agriculture, livelihoods, education, and others, to examine the feasibility of developing a common waiver card or targeting scheme across multiple sectors where targeting can be an important aspect of service delivery.
The waiver card scheme was positively viewed by both community leaders and facility staff, despite the lack of reimbursement to facilities for the foregone user fee revenues. This reimbursement of foregone user fee revenues is one factor that has been cited as key to the success of the health equity funds in Cambodia [42, 49, 53] as well as for health financing programs for the poor in Vietnam . Even if fee revenues do not constitute a significant portion of staff income in Afghanistan, as they do in Cambodia and Vietnam , a waiver card scheme, especially at the hospital level where inpatient stays represent larger sources of revenue, should consider directly reimbursing facilities for use of care by cardholding patients.
Additionally, if waiver cards are used for hospital fees, it will be important to reimburse for indirect costs as well, including transportation, food, and other costs associated with seeking care . This research indicated that financial access still appears to be a barrier, even among cardholding households, and waiver card programs should be expanded to cover additional costs of seeking care. It is noteworthy that there was no evidence of "unofficial" charges to waiver cardholders at user fee facilities, but the reported financial barriers may have been due to other costs related to care-seeking, such as transportation fees, food, lodging, or seeking care at other providers (e.g., private providers). However, even more generous reimbursement packages may not be sufficient to remove all financial barriers to seeking care, as research in Cambodia found that health equity fund beneficiaries still were unable to afford the upfront costs of seeking hospital care .
Finally, it is critical to continue to invest in monitoring and evaluation efforts if a revised form of waiver cards is implemented in the future. This research contributes to the relatively scarce literature on community-identified targeting, demonstrating that it is possible to implement a pro-poor and favorably viewed waiver card scheme within a relatively short time and with limited additional support. However, targeting did not perform well according to PCA-derived relative wealth scores, indicating that these measures may not be the most appropriate for capturing disposable income and/or that community leaders are taking other important factors into consideration when nominating households.
The post-conflict setting in which the scheme was carried out provides encouraging evidence that beneficial health services delivery strategies can also contribute to community participation and rebuilding of institutions at the local level. These results lend support to the notion that waiver cards can increase utilization of care, even for services that have multiple barriers to use. Afghanistan abolished user fees for primary care in 2008, citing the results of the larger pilot study in which the waiver card scheme took place, but barriers to access remain. Further research should examine the effects of demand-side financing interventions to improve utilization of important public health services in Afghanistan, as well as a waiver card scheme for costly hospital-level services to prevent catastrophic expenditures and encourage use, building on the initial results from this study.