The impact of user charges on health outcomes in low-income and middle-income countries: a systematic review

Search strategy

In September 2017, we conducted searches of electronic medical and economics databases (MEDLINE, Econlit, Scopus, JSTOR, WHO Library Database, World Bank e-Library) to locate studies about the impact of user charges on health outcomes in LMICs. We included all types of health outcomes with quantitative measures, with the search strategy based on a combination of three sets of keywords—(1) health, (2) synonyms of user charges and (3) a list of LMICs (detailed search strategy can be found in online appendix—database search strategy):

• Synonyms of health.

• User charges: “reimbursement”, “copayment”, “cost sharing”, “coinsurance”, “deductible”, “user charge”, “user fee”, “out-of-pocket”, “health insurance”, “medical insurance”.

• LMICs: “Low and middle income country”, “Asia”, “South East Asia”, “Central Asia”, “sub-Saharan”, “Africa”, “South America”, “Latin”, “low-income country”, “middle-income country”, “developing country”, “under developed” and all LMICs listed in the World Bank website in year of 2016.15

The searches were restricted to studies written in English (peer-reviewed articles, working papers, conference papers and reports) published from January 1990 to September 2017. We also carried out additional literature searches by appraising reference lists of the studies identified.

Inclusion and exclusion criteria

User charges were defined as direct payments made at the time of health service use,16 with any possible combination of fees from registration, consultation, drugs and medical supplies, treatment, hospitalisation, delivery fees, laboratory tests or other health services provided in public or publicly subsidised sectors. The charges could be paid based on each visit to a healthcare provider or for treatment of the whole episode of illness.7

Studies were screened based on the inclusion and exclusion criteria shown in table 1. All study populations were eligible. For outcomes, we considered both self-reported and clinically measured health outcomes in relation to both increases and decreases in user charges. As eligible interventions, only studies which examined changes in the levels of user charges (in either direction or magnitude) were included, while studies focusing on the impact of health insurance without explicitly examining changes in user charges were excluded.

To synthesise findings from robust evidence, we only included studies with either QE or RCT study designs to control for confounding and bias. For example, estimated policy effects could be biassed if self-selection exists, when individuals who expect to have high healthcare use choose insurance schemes with lower user charges.7 Therefore, the relationship between the levels of user charges, healthcare use and health outcomes may have elements of reverse causality.17 A broad definition of QE was considered which included difference-in-differences (DID), propensity score matching (PSM), instrumental variable (IV), regression discontinuity (RD) and interrupted time series (ITS).18–20 To isolate changes in health outcomes attributing to user charges, we removed studies that evaluated multifaceted policy changes (from both demand or supply sides) or consisted of several concurrent policy changes which precluded assessment of individual policy impacts.

One reviewer (VMQ) independently reviewed titles and abstracts and discussed with another reviewer (JTL) on the uncertain studies. Subsequently, full texts were screened for eligibility by two independent reviewers (VMQ and JTL). Any disagreements were resolved through discussion with a third reviewer (TH).

Quality assessment

Quality assessments were dependent on risk of bias for each study. We used a modified ROBINS-I (Risk of Bias in Non-Randomised Studies—of Interventions) tool21 for studies adopting a QE design. First, we assessed the risk of biases (two relating to pre-intervention, one relating to at-intervention, four relating to post-intervention) for each QE to reach an overall risk of bias (ie, low/moderate/serious/critical/no information). We graded the quality of each QE as high (low risk of bias), moderate (moderate risk of bias) or low (serious risk of bias or below) based on the overall risk of bias.

We used the Cochrane Risk of Bias tool to assess seven domains of biases for studies adopting an RCT design. The quality of the RCT was then graded high (low risk of bias for more than five domains), moderate (high risk of bias for two domains) or low (high or unclear risk of bias for more than two domains).22

Data extraction and synthesis

The data extracted from selected articles consisted of the study setting, detail on the change in user charge policy, study design, data sources, follow-up period and key findings on health outcomes. We also examined secondary outcomes, such as access to healthcare and levels of financial protection, where possible.

Due to heterogeneity in policy settings, countries, study designs, intervention and outcomes, meta-analysis was not feasible. We conducted a narrative review and reported effects, stratified by different types of health outcomes. We also outlined effects for secondary outcomes to explore possible mechanisms of action between user charges and health outcomes. We analysed the impact of user charges policies on health by different population groups, such as low-income populations and children, to explore whether impact varied by population groups.

Additionally, we undertook an analysis to understand the evidence gap by graphically displaying the knowledge gap of the findings in type of health outcomes and population studied, and quality of the studies.

Study characteristics

We identified 6902 citations from bibliographic databases and an additional 73 from other sources. After removal of duplicates, 5683 unique citations were screening by title and abstract, and 336 full texts were sourced. Of these studies, 319 studies were excluded for the following reasons: a policy resulting in user charge change not evaluated (56 studies); health outcomes not studied (100 studies); QE or RCT study designs not employed (123 studies); not based in LMICs (26 studies); and other reasons such as duplicate studies, unrelated topics, under review or no full text available (14 studies). Seventeen studies met the final inclusion criteria. Figure 1 provides details of the process of study identification.

Request permission

The 17 included studies were conducted in 12 LMICs: five upper-income to middle-income countries—China (two studies), Georgia (one study), Jamaica (one study), South Africa (one study) and Mexico (two studies); five LMICs—India (three studies), Vietnam (three studies), the Philippines (one study), Ghana (one study) and Kenya (one study); and two low-income countries—Senegal (one study) and Nepal (one study). Of the 17 studies, 14 were published after 2010. One study was an RCT, nine used DID design, two employed RD, three used PSM, one used DID design with PSM (DID-PSM) and one used IV regression. More details of study characteristics are shown in table 2.

A range of health outcomes were studied and categorised into five groups (figure 2): general health outcomes (nine studies), mortality (four studies), infectious disease–related outcomes (three studies), chronic condition–related outcomes (three studies), and nutritional and anthropometric outcomes (two studies).

Request permission

Changes in user charge policies were classified as removing user charges (15 studies), reducing user charges (one study) or increasing user charges (one study). Thirteen studies examined user charges in primary or outpatient services, while four examined user charges in secondary and tertiary care. The majority (14/17) of the studies examined relevant secondary outcomes on access to healthcare and levels of financial protection. The median follow-up period from intervention to the last observation was 45 months, with a range of 12–144 months. Studies on chronic condition–related outcomes had the longest median follow-up period of 72 months.

Quality of included studies

Overall study quality was moderate, with four studies of high quality, 10 of moderate quality and three of low quality. Areas of potential bias that many QEs failed to address were potential selection bias (six studies) with the possibility selection could be related to intervention status or outcome, and potential recall or misclassification bias (seven studies) as using self-reported health outcomes may have affected the ‘measurement of outcomes’. The quality of the only RCT study included was rated high with nearly all domains assessed at low risk except the domain ‘performance bias’, which was high risk as participants were un-blinded to the intervention which may contaminate the results (online supplementary table 1).

Findings on the relationships between user charges and health outcomes

Differential impact of user charges and explanatory factors for improvement in health outcomes

The relationship between reduced user charges and improved health outcomes was more evident in studies focusing on children and lower-income populations. Six out of seven studies26 30 32 33 41 44 focusing on children and infants found improved health following a reduction in user charges, including two studies on improved general health, two on reduced neonatal mortality, one on infectious disease–related outcomes and one on nutritional outcomes.

Among the nine studies24 25 27 29 33 40 42–44 evaluating the impact of reducing user charges for the poor (including eight studies on economically poor and one on people working in informal sectors), seven studies (78%) found improved health outcomes following user charges removal. In comparison, of the other eight studies23 26 28 30–32 39 41 focusing on the impact of user charge policy for the general population, five (63%) found improved health outcomes following user charge removal. Two studies assessed and compared the differential effect of user charges by population groups, with mixed findings. One analysis found that the magnitude of reduction in neonatal mortality was larger among women from the lower castes or indigenous ethnic groups,32 while the other study found similar policy effect among various income groups.28

Among the 14 studies reporting healthcare use, 12 found access to and use of healthcare increased following reduction in user charges, and nine further found the increase in healthcare use along with improved health outcomes. Increased access to healthcare due to reduced user charges of care was posited as the possible explanatory factor for better health outcomes in five quasi-experimental studies.29 32 40 41 44 For example, Lamichhane et al, a high-quality study, evaluated the impact of free birth delivery programme on neonatal mortality in Nepal,32 and found a 4%–6.9% reduction in neonatal death in the treatment group compared with control group. The reduced neonatal mortality was consistent with a 6.1%–8.2% increase in women delivering with assistance from skilled birth assistants and an increased use of public facilities, which are possible factors to explain the association between reduced user charges and reduced neonatal mortality.

Main findings and interpretation

Our study is the first to examine and synthesise RCT and QE evidence on the relationship between user charges and population health outcomes in LMICs. Our findings are broadly consistent with evidence from HICs which show that user charges may have adverse impacts on health outcomes.11 34–36 45 A systematic review by Goldman et al 36 found that increased cost sharing for prescription drugs was associated with reduced drug treatment and adherence, in particular for chronically ill patients, resulting in more use of expensive medical services such as hospitalisation and emergency visits. This study also suggested that such an adverse effect was likely to be magnified among the poor, who are more price sensitive. However, our study findings contrast with a recent systematic review which focused on maternal health outcomes in LMICs. Dzakpasu et al synthesised evidence and found that reduced user charges were associated with increased facility deliveries, yet with little effect on maternal health outcomes.37 The inconsistency between our study findings on health outcomes is likely due to the broader inclusion criteria by Dzakpasu et al, which included weaker study designs (mainly cross-sectional studies) that are potentially subject to bias. In contrast, our review only included evidence from studies adopting RCT or QE study designs that would reduce bias arising from potential confounders, therefore strengthening causal inference.37

Our review includes several important caveats. First, most studies included here were conducted in middle-income countries, specifically in China, India and Mexico. There were only two studies conducted in low-income countries (Nepal and Senegal). Therefore, it is unclear as to whether the results could be extended in other LMICs at different levels of economic and health system development. Second, a majority of the studies included have a follow-up period of health outcomes of less than 5 years. Future studies should examine the long-term health effects of changes in user charge policies. Although user charges were introduced to raise funding for health systems in several LMICs in 1980s, our review only included one study that evaluated the impact of increased user charges on health outcomes. These may reflect the fact that fewer studies adopted QE study designs until more recently or many user charge policies were not properly evaluated at the time.

Furthermore, the type of evidence contained in RCTs and QEs is unable to support clarity on a range of important questions about the mechanisms by which user fees and health outcomes are inversely related. Likely mechanisms include improved access to effective healthcare and reduced impoverishment from healthcare payments with their associated health effects. But health markets are complex, and the price of accessing public services is only one of a set of key variables in those markets. Implications for the mix of public and private, effective and ineffective and higher and lower priced healthcare options sought, the timeliness of healthcare-seeking decisions and quality of care across all the providers in the market among many other variables are contingent on a host of other contextually specific factors such as the pre-existing public–private mix in any health system, the economic capacity of populations affected, the price levels of other providers and other goods and services, and the regulatory and governance environments. Some measurements of health may not be possible to discern whether user charges had limited effects on health outcomes, or that the conditions measured or the populations studied were not sensitive to the changes.38 Future studies should also use comprehensive measures for patients’ health such as patients’ use of secondary care, clinical and self-reported health outcomes, and mortality rates. Despite the complexity of the mechanisms of effect and the contextual variability of the factors involved, RCTs and QEs suggest quite clearly that user charges are a population health–damaging policy. Few health system policy assessments can produce such a clear result.

Policy and research implications

An important finding of our review is that decreased user charges lead to improvements in health outcomes, especially for children and low-income groups. A potential reason behind this association is that reducing user charges increases access to healthcare. These findings highlight the importance of moving away from user charges to finance UHC, and towards contributory schemes based on prepayment through taxation and insurance contributions with large-scale risk pools that enable cross-subsidisation from the healthy and wealthy to the sick and low-income groups.4

This evidences the importance of public finance for subsidising the costs of healthcare for low-income and disadvantaged populations, and as an effective policy lever to reduce inequities in access and improve health outcomes.4 46 47 While all stand to benefit from enhanced financial protection brought about by greater reliance on prepayment and cross-subsidisation, the lowest-income and less healthy populations will benefit most, as these groups are more likely to face financial hardship due to ill health. Replacing user charges with public funding for these disadvantaged populations should help to reduce financial barriers to accessing care, in turn, improving health outcomes for these groups and promoting equity in health.46 48