Access to basic health services of acceptable quality is still denied to many of the world's poorest people. (1) Against a backdrop of severely underfunded health systems, (1,2) governments are faced with a dilemma. Payments for health services, in the form of user charges, are likely to present a barrier to access. Yet, a shortage of resources at the facility level is a contributor to failure to deliver quality services, and this also presents a barrier to access. Some have argued that user charges can generate vital resources at the local level and help provide good quality services; (3-5) others have highlighted their negative effects, particularly on equity; (6-9) Recently, several international campaigns have advocated the removal of user fees, especially for primary care services. (1,10)
Some recent articles have underlined the paucity of evidence on the effectiveness of policy interventions in low-income countries; (11,12) others have noted the importance of systematic reviews for understanding health systems. (13) Despite the central importance of the user-fee debate, no systematic review has examined the quality of the empirical evidence on this topic. To redress this imbalance, this review set out to assess the quality of the existing evidence on the impact of user fees on health service utilization, household expenditures and health outcomes in low- and middle- income countries.
Scope of the review
User lees refer to a financing mechanism that has two main characteristics: payment is made at the point of service use and there is no risk sharing. User fees can entail any combination of drug costs, supply and medical material costs, entrance fees of consultation fees. They are typically paid for each visit to a health service provider, although in some cases follow-up visits for the same episode of illness can be covered by the initial payment. This review aimed to assess the effect on health service utilization of introducing, removing, increasing or decreasing user fees in low- and middle-income countries.
Search strategy and inclusion criteria
We searched 25 databases covering the social science, economics and health literature. We also searched the reference lists of all relevant articles, the web sites of related research centres or institutions (lists of sources searched are available from the authors upon request) and existing reviews. (14-19) The search strategy combined looking for terms in subject headings and within the text pertaining to health financing ("health financing", "user charges", "user fees", "cost recovery", "direct payment", "drug revolving fund", "fee") and outcomes ("utilization", "access to services", "health expenditures", etc.). No limitation on date or publication language was applied. Only studies from low- and middle-income countries, as defined by the World Bank, were included.
Only experimental or quasi-experimental study designs were included --cluster randomized controlled trials (C-RCTs), controlled "before and after" (CBA) studies and interrupted time series (ITS) studies (Table 1)--as suggested by the Effective Practice and Organisation of Care (EPOC) group of the Cochrane Collaboration, where this review was registered. Indeed, such designs are known to provide the most reliable measures of effect. Papers were assessed only if the effect of the intervention was measured in terms of either changes in utilization, household expenditure, health outcome or equity. Both authors independently sifted the titles and abstracts of publications for retrieval. In case of disagreement, full-text articles were retrieved and examined. All retrieved articles were then independently reviewed by the two authors, and agreement was reached over whether they fulfilled the criteria for inclusion in the review.
Reanalysis of data
We found several studies that had longitudinal data on utilization but had not performed a time series analysis, (20-26) To be able to include these, we relaxed the original definition of ITS (27) (Table 1) and set out to reanalyse the data appropriately. When they were not directly reported in the paper, original data series were requested from the authors. Whenever the authors could not be found or did not respond, we attempted to reconstruct data series by scanning graphs. (1)
Data series were then examined with the following segmented regression model to control for secular trends and potential serial correlation of data, and to detect any significant changes after the introduction of the new policy:
[Y.sub.t] = [[beta].sub.0] + [[beta].sub.1] x Preslope + [[beta].sub.2] x Intervention + [[beta].sub.3] x Postslope + [[epsilon].sub.t]
where [Y.sub.t] is the outcome variable at time t. Intervention is coded 0 for preintervention time points and 1 for postintervention time points; its coefficient [[beta].sub.2] reflects the immediate impact of the intervention on the dependent variable. Preslope is a continuous variable indicating time from the start of the study up to the intervention (if the intervention occurred at the nth period, preslope is coded sequentially from 1 to n before the intervention and remains equal to n for the rest of the series). It thereby captures the structural trend that has started before the intervention. Postslope is coded 0 up to the last point before the intervention phase and coded sequentially from 1 thereafter. Its coefficient [[beta].sub.3] therefore reflects the trend or growth rate in outcome after the intervention. When auto-correlation was detected by the Durbin-Watson test, it was corrected with a PraisWinsten regression.
In addition, to provide more comparable results, we computed price elasticities ([[epsilon].sub.p]) for studies reporting changes in user fees, and "net" elasticities for those with a control site. We also computed the statistical significance of the observed effects if it was not reported in the original paper.
Data extraction and quality assessment
Quality criteria were adapted from those suggested by the EPOC group of the Cochrane Collaboration (Table 2). When a study presented unsatisfactory or unclear elements for two or more criteria, it was scored as being of "low" quality. When only one criterion was unclear or unmet, it was scored as being of "moderate" quality, and when all elements were satisfied, the study was considered as being of "high" quality. For each included study, both authors extracted data and assessed quality. They then reviewed one another's conclusions. Discrepancies were resolved by discussion.
Description of studies
The initial database search generated over 24 000 references. An initial sift of tides and abstracts led to the inclusion of 243 documents for further investigation (Fig. 1 provides more details on the search strategy). Sixteen studies met our inclusion criteria. We only found studies reporting effects on health service utilization. None reported an effect on expenditures or health outcomes, and two reported effects on utilization by different socioeconomic groups.
Eight papers presented data on the effects of introducing user fees (Table 3), five on the effects of removing fees (Table 4) and five on the effects of decreasing or increasing fees (Table 5 and Table 6). Some papers reported results from specifically designed studies, (4,28-32) while others sought to analyse the effect of nationally- implemented strategies using routine data. (20-26,33-35)
Study settings varied considerably (type of service, type of facility, type of payment). A range of utilization measures were reported as outcomes, including new visits, registrations, weekly/ quarterly/monthly attendance, outpatient and/or inpatient attendance. As a result, a narrative approach to reporting the results has been adopted.
[FIGURE 1 OMITTED]
Impact of removing user fees
Five studies used longitudinal data to report the effects of abolishing user fees on utilization. These were all reanalysed. …