Social Science & Medicine 133 (2015) 21e27

Contents lists available at ScienceDirect

Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed

Patient access to health care and medicines across low-income countries Divya Srivastava a, b, *, Alistair McGuire b a b

OECD, Health Division, 2 Rue Andr e Pascal, Paris, 75016, France LSE Health, London School of Economics and Political Science, Houghton Street, London, WC2A 2AE, United Kingdom

a r t i c l e i n f o

a b s t r a c t

Article history: Available online 11 March 2015

This study explores the issue of demand for health care and medicines in low-income country settings. Using the World Health Survey, multivariate analysis of cross-sectional household data from 35 lowincome countries found that when ill, patient demand for health care to visit a clinic or hospital is inelastic ranging from
0.19 to 0.11. The main determinants of health seeking behaviour include having insurance, having a chronic condition, high household expenditure, and marital status. Women, the educated and those living in urban settings are more likely to seek care in a clinic. These findings suggest low-income patients will experience access problems, raising important policy implications to improve access to health care and medicines in these settings. © 2015 Elsevier Ltd. All rights reserved.

Keywords: Medicines Health-care access Determinants of health Low-income countries Pharmaceutical policy Health policy

1. Introduction Access to health care and medicines is an important public health issue, particularly for those that can least afford to purchase them. Patients use medicines to either improve or maintain their quality of life and health. Typically, patients living in low-income countries require medicines to treat infectious disease but now many of these countries are also experiencing a rise in the prevalence of patients suffering from chronic conditions (e.g. diabetes, cardiovascular disease), which requires regular use of medicines for treatment. According to the World Health Organization (WHO), “expenditure on medicines accounts for a major proportion of health costs in developing countries and therefore access to treatment is heavily dependent on the availability of affordable medicines” (WHO, 2007). This demand for medicines in low-income countries raises important policy implications because health systems are cash constrained and medicines are not typically subsidised as they are in high-income countries. High medicine costs may undermine the decision to seek care. Furthermore, there are equity implications for patients that cannot afford the cost of care. Raising the price of health services is a means to generate revenue for the cash strapped

 Pascal, Paris, 75016, * Corresponding author. OECD, Health Division, 2 Rue Andre France. E-mail address: [email protected] (D. Srivastava). http://dx.doi.org/10.1016/j.socscimed.2015.03.021 0277-9536/© 2015 Elsevier Ltd. All rights reserved.

health care sector. A revenue generating mechanism will undoubtedly lead to welfare loss for patients who cannot afford the cost of care and the extent of the welfare loss is a function of the price elasticity. 2. Background Individuals seek medical care once they perceive that they are ill but the recognition of an illness and the decision to seek medical care are individual discrete sequential choices (Pokhrel et al., 2010). The choice to seek care is therefore used as a measure of access to health care. It is important to note that access to health care is a complex concept (Donebedian, 1972; Aday and Andersen, 1974) and a number of factors affect it including time and money costs in seeking health services (Le Grand, 1978; Mooney, 1983); consideration of specific services, quality, and inconvenience (Goddard and Smith, 2001). Access should reflect a “degree of fit” between supply and demand related factors or more broadly as the “freedom to use health services” (Oliver and Mossialos, 2004; Thiede et al., 2007). The theoretical framework in this study assumes that the demand for medical care is derived from the demand for health (Grossman, 1972; Jack, 1999). It is assumed that individuals generate utility from the consumption of commodities. Some commodities are purchased (e.g. medicines, diagnostic services) while others are produced by individuals through a process that combines their own time with other inputs which can be purchased

22

D. Srivastava, A. McGuire / Social Science & Medicine 133 (2015) 21e27

in the market where ‘health’’ is such a commodity (Pokhrel et al., 2010). Therefore, it is assumed that individuals decide to consume medical care to the extent to which it maximises their utility. A large body of empirical work has documented demand for health care in low-income country settings including systematic reviews (Creese, 1991; McPake, 1993; Gilson, 1997; Sepehri and Chernomas, 2001; Hutton, 2004; Palmer et al., 2004; James et al., 2006; Lagarde and Palmer, 2008). Earlier work on estimating demand for health care in low-income settings found small negative price elasticities
0.02 to
0.003 (Heller, 1982; Akin et al., 1986) while recent studies found prices/costs to be significant with a wider range of estimates from
10.2 (Ojeda et al., 1994) to zero (Borah, 2006). The more recent empirical work has also improved the analysis by relaxing the assumptions of restrictive choice models such as the multinomial logit model (MNL). This is to correct for the violation of two of the MNL properties: the independence of irrelevant alternatives (IIA) and the independence of error terms (IID). More recent work has tried to account for endogeneity in the estimation of price elasticities (Asfaw et al., 2004; Borah, 2006; Sarma, 2009; Qian et al., 2009). A fundamental issue in analysing the relationship between utilisation and health expenditure is the direction of causality. That is, whether utilisation explains health expenditure or whether expenditures explain utilisation levels will be partly determined by the price elasticity of demand. To the extent that the volume/utilisation response incorporated within expenditure is defined by price elasticity it is difficult to determine the direction of causation. One approach in the literature uses predicted, rather than actual health expenditure to account for this endogeneity (Asfaw et al., 2004). A relatively small number of studies have studied price responsiveness for medicines, with estimates ranging from
3.6 (Abdu et al., 2004) to zero (Akin et al., 1986). Laxminarayan et al. (2006) simulated the welfare effects for malaria treatment with elasticities ranging from
1.05 to
0.49. Empirical work suggests that the determinants of health seeking behaviour include the price (e.g. or cost of the visit) which is generally negatively related to the decision to seek care (Dzator and Asafu-Adjaye, 2004; Soyibo et al., 2004; Borah, 2006; Anyanwu, 2007). Income (or some measure of household wealth) has a positive effect (Bolduc and Lacroix, 1996; Asfaw et al., 2004). Those in poor health are more likely to seek care (Schwartz et al., 1988; Qian et al., 2009). Education and insurance both have a positive effect in seeking care, while distance to access care has a negative relationship (Gertler et al., 1987; Asfaw et al., 2004; Sarma, 2009; Qian et al., 2009). Dzator and Asafu-Adjaye (2004) found that a 1% increase in distance will reduce the probability of demand for treatment by
0.36% at a public provider. Travel time, and treatment time are also highlighted as important determinants of the decision to seek care (Heller, 1982; Bolduc and Lacroix, 1996; Borah, 2006). Findings related to time and distance provide useful information but not all studies collect detailed information at the household level and more rely on information at the aggregate level (e.g. village level) to compute this information. The empirical work also distinguishes between the type of health care received (e.g. drug related, family planning) reflecting the wide heterogeneity of the provision of health care in outpatient and inpatient settings (Dzator and Asafu-Adjaye, 2004; Borah, 2006; Qian et al., 2009). The literature also emphasises that the availability of traditional forms of care and self-treatment are equally important to the provision of modernised medical care offered in public or private settings (Sarma, 2009). Age has been shown to have a mixed effect: some studies suggest that older age groups seek most care (Sauerborn et al., 1994; Asesno-Okyere et al., 1996). Other empirical work indicates that it

is the younger ages that utilise health care more, possibly reflecting the predictions of human capital theory where families invest in the younger more productive members of the family (Gertler et al., 1987; Sarma, 2009). Sex has shown to have differing effects depending on the type of health care: men are more likely to seek care in general (Gertler and van der Gaag, 1990), but in specific cases, most obviously relating to childbirth, women are more likely to seek care (Dzator and Asafu-Adjaye, 2004). Household size is also an important determinant of the decision to seek care (Borah, 2006). Quality may be estimated from a structural dimension such as drug availability, physician availability, machine availability and qualification of staff (Sepehri and Chernomas, 2001). Quality is a difficult factor to capture in both cross-section and time series analysis because important dimensions such as process and outcome are not easily captured (McPake, 1993; Sepehri and Chernomas, 2001). Marital status, not commonly modelled, has had a positive effect on the decision to seek care (Qian et al., 2009). The evidence base of price elasticities could be improved as not all studies have controlled for covariates. This has also been confirmed by reviews in the literature of limited evidence on price responsiveness, small sample sizes and confounding factors (Sepheri et al., 2001; Palmer et al., 2004). Unlike high-income settings where estimation of price elasticities and determinants of health seeking behaviour (Cutler, 2002) come from well-funded and developed databases, a serious limitation is the availability of data from low-income country settings. Furthermore empirical work has not always considered the policy environment. Some work has more broadly considered implications for health policy financing, but not specifically related to pharmaceuticals. To understand the demand for medicines, this study investigates the determinants of access to medicines and health care in low-income and lower middle-income countries, focussing on the calculation of price-elasticities. It is important to note that data constraints in these settings make it impractical to calculate price elasticities directly since these require information on both prices and volume, which is generally unavailable. For this reason, studies on the demand for health care in these settings have often used patient health expenditure data to compute price elasticities. This method is based on a less direct method of demand estimation, but does provide some information on demand structures in these settings. The aim of this paper is to contribute to the evidence base by drawing on larger data sets for the analysis of demand for health care than has been used previously and across rural and urban settings. This paper contributes to the evidence base to address endogeneity issues related to health expenditure and health seeking behaviour. Finally, information on price responsiveness has implications for policy. This paper contributes to this topic by considering the pharmaceutical policy making environment. The paper hypothesises that income is an important determinant of access to medicines and health care and that access is low for low-income individuals. Second, it hypothesises that an expectation of a high level of expenditure on medicines reduces the propensity to consume (which implies a negative price elasticity). The paper's research objectives are as follows: to explore whether income has an effect on access; to investigate whether regulation has an effect on access; and third to compute price elasticities. 3. Methods Discrete choice models are typically applied to model the decision to seek care and to understand the determinants of health care demand and we follow this trend. Discrete choice models, including the logit model and the multinomial logit model (MNL) are used to model the decision to seek care. These models are used

D. Srivastava, A. McGuire / Social Science & Medicine 133 (2015) 21e27

when outcomes are qualitative and unordered. A logit model is used when there are two qualitative outcomes (e.g. decision to see the GP or not to visit the GP). A MNL model is applied when there are more than two outcomes: for instance, a patient visits a doctor, visits a specialist, or visits a traditional practitioner. The theory allows for consumer (i) to make a choice (j) where the consumer's utility Uij is the maximum among the j choices (Greene, 2008). The statistical model estimates the probability that choice j is made using log likelihood estimation. The outcome variable is an unordered categorical variable and involved one of four choices on whether the patient sought care within the past year in a hospital setting, a clinic setting, did nothing even though ill, or did not access health care because they were not sick. These four choices are considered using a multinomial logit model. Consequently the dependent variable took the following values: hospital care chosen ¼ 1, clinic care chosen ¼ 2, doing nothing even though ill ¼ 3 and no access to care as not sick ¼ 4.

ProbðYi ¼ jjwi Þ ¼ Pij ¼


 exp w0i aj  j ¼ 1; 2; 3; or 4;  P 1 þ 4k¼1 exp w0i ak

a4 ¼ 0 An individual not sick was set as the base outcome. The following regression model was run for the ith individual across j alternatives where j ¼ 1, 2, 3, 4 in country p. The estimated coefficients measure the change relative to the base case. While patients also seek traditional care, self-medication, and spiritual care in these settings the data do not capture this information for analysis. A limitation with this approach is that the choice model is influenced not only by the patient's access to care but other factors such as social stigma, individual habit as well as salience of symptoms. Therefore coefficients on predictors of the probability of seeking care may pick up a combination of the predictors' effect on access as well as the predictors' effect on choice. The selected regressors aim to minimise this effect. In a multinominal logit model, it is recommended to calculate the marginal effects which indicate the marginal impact on the probability of the dependent variable‘s choice outcome (Greene, 2008). If the marginal effect of having good health is 0.02, it reduces the probability of going to the hospital (the choice outcome) by 0.02 (the marginal effect). Data from the WHO World Health Survey (WHS) was used for the analysis (WHO, 2006). This cross-sectional survey was undertaken in 2003 (with the exception of China carried out 2002, Pakistan carried out in 2003/2004, and Kenya carried out in 2004) and collected information on socio-demographic characteristics, health state descriptions, health state valuations, risk factors, mortality, health care utilisation, health system responsiveness and health goals and social capital in 38 low-income and lower-middle income countries. While the data were collected before for China, after for Kenya and partly overlapped for Pakistan with the year 2003, this is not likely to pose large problems in the model estimation. Low-income countries were defined as having gross national income (GNI) per capita of US $765 and lower middle-income were defined as GNI per capita US $766 to $3035 for the year 2003 (World Bank, 2005). Many of these countries have seen an increase in their GNI [Table A1]. The regressors were chosen to capture information on the patient's utilisation, as a measure of access, the patient's health status, as measure for need, and socioeconomic information. The age, sex and marital status of the patient were used as confounders. Risk scores are frequently used in high-income countries but this requires substantial claims data and many low-income countries do

23

not have such data systems in place. While self-reported measures do have better predictive ability over age and sex, they are less powerful than health measures that draw on diagnosis (Van de Ven and Ellis, 2000). Two variables about the patient's health were drawn from self-reported health and whether the patient was diagnosed with any of the following chronic conditions: arthritis, angina, asthma, depression, schizophrenia or psychosis, diabetes, and tuberculosis. Socioeconomic information included whether the patient had education, was employed, whether the patient lived in a rural or urban setting, whether the patient had private health insurance and the number of members living in the household. An indicator for wealth or income was drawn from the households' expenditure as no direct income information was collected. As the price of health services is lacking in these country settings, expenditure data was used in the analysis. Expenditure data is used as a proxy as in other studies where price data are readily unavailable (Asfaw et al., 2004). Expenditure data has its limitation as the data could represent changes in quantity, changes in prices or changes in both. The selected regressors aim to minimise this impact through the computation of the predicted expenditure. Data on the previous month's household expenditure was collected and included food expenditure, utilities, education, health expenditure, health insurance premium expenditure and other related household expenditure. Household health expenditure data included hospital expenditure, health professionals, traditional healers, dentist, medicine, health products, diagnostics and other related health expenditure. The data consist of only the previous month's expenditure and double counting is unlikely as the data cover a visit to the doctor within the past month. For a small set of observations this may have happened earlier than a month. Weighted predicted health expenditure data was used to partly address this issue. Household expenditure data contained some observations with extremely large values of expenditure that exceeded even average per capita monthly expenditures when compared with World Bank development indicators (World Bank, 2005). Data with high values that were not in line with World Bank data estimates were dropped. These are likely measurement errors. The common approach in the literature is to standardise expenditure data (Jones et al., 2007). The data were converted into US$ denoted as purchasing power parities (PPP) for the year 2003 and then transformed into logs (World Bank, 2005). To smooth the data, two times plus or minus the standard deviation based on a log normal distribution of household expenditure and a log normal distribution of health expenditure per visit was included in the analysis. This process dropped extreme values, amounting to 6572 observations (5.2% of the sample) of household expenditure data and 8140 observations (6.4% of the sample) of health expenditure data from the analysis. Patients reported on their out-of-pocket (OOP) costs, as related to their visits and included doctor's fees, medicine costs, diagnostic tests, transportation costs and other related expenditure. OOP were transformed into logs to account for non-linearities in the data for clinic and hospital visits. This information was entered in the model through the predicted amount of expenditure based on the patient's characteristics. Predicted health expenditure per visit averaged over rural and urban settings within each country was estimated to correct for endogeneity as used previously in the literature (Asfaw et al., 2004). The predicted health expenditure was calculated by regressing the log expenditure (in a clinic or hospital) against age, sex, employed, education, urban or rural setting, log household expenditure and the reason for the visit. The reason for the visit included the following health categories:

24

D. Srivastava, A. McGuire / Social Science & Medicine 133 (2015) 21e27

high fever, severe diarrhea or cough, immunisation, antenatal consultation, family planning, childbirth, dental care, arthritis, asthma, heart disease, bodily injury, minor surgery or other. The multinomial regression equation was subsequently specified as:

yijp ¼ bijp þ Xageijp þ Xage2ijp þ Xsexijp þ Xselfreportedhealthijp þ Xchronicijp þ Xeducationijp þ Xemployedijp þ Xmaritalstatusijp þ Xhouseholdsizeijp þ Xurbanijp þ Xhealthinsuranceijp þ Xloghouseholdexpenditureijp þ Xlogpredictedexpenditureijp þ Xcountrydummyijp þ Xreasonforvisitijp þ 2ijp

to 66% of the sample are married and employed, approximately 30% have a chronic condition, 40% live in urban settings and less than 20% have private health insurance. Among those who reported being ill within the past year of the survey, the majority (93%) of these sought care, with most seeking outpatient care (86%), while a smaller percentage visited a hospital (6%) and 8% did nothing. Medicine expenditure accounted for the largest share of OOP expenditure with an average of 57% in clinic setting and 51% in hospital setting. Average expenditure in a clinic was higher than in inpatient settings as most patients sought outpatient care with an average expenditure of US$ 5.40 in a clinic and US$ 0.41 in a hospital. OOP expenditure by country ranged from US$ 12 in China to US$ 0.24 in Mali with most countries spending on average US$ 6 [Table A2]. 4. Results

The age coefficient is expected to be positive. There may be nonlinearities with age so this term is also squared. Health status variables should suggest that those with good self-reported health are less likely to seek care, while those with chronic conditions are more likely to seek care. Higher education and those that are employed could be linked to better health, which would lead to a negative coefficient. Previous research, typically points to the effect of education and being employed to have a positive effect on the probability of seeking care. The insurance variable is treated as exogenous as, given that we have controlled for income, urban dwelling and education we would expect the demand for insurance to be reflective of the probability of illness. This probability of illness will not affect health care utilisation per se, however the purchase of insurance will ex post affect the utilisation level of health care, and we merely expect insurance therefore to have a positive effect on seeking care. High OOP expenditures are predicted to have a negative effect on the probability of seeking care, while household expenditures (e.g. a measure of household wealth) are predicted to have a positive effect on the probability. Women, all else equal, are predicted to be more likely to seek care due to their health needs relating to child health and child delivery, however the literature points to mixed evidence suggesting that men can be more likely to seek care. Similarly, the effect of marital status is ambiguous on the probability of seeking care. Household size is ambiguous, although in this setting it may serve as a proxy for capturing wealth of a household. The effect of the urban dummy variable is ambiguous as well. Individuals living in urban settings may be more likely to seek care because there are likely to be more facilities available in urban settings. Alternatively, rural patients may be more likely to seek care if this variable correlates with need: poor rural patients may struggle with health conditions and could be more likely to seek care. The country dummies aim to account for system heterogeneity, in part reflecting the regulatory environment. Dummies that capture the reason for visit were also included. The data set contained a cross section survey of 38 developing countries with 126,806 observations. After data cleaning and accounting for missing values, 35 countries remained for available for analysis. The cross-sectional dataset contained a total of 42,668 observations for analysis. Approximately 20% of the sample reported being ill within the past year of the survey. The regressions were run using STATA software. Estimations were run with and without sampling weights but the comparative results were consistent. Estimates without sampling weights are presented in this study. The cross-sectional sample has an average adult age of 42 years, and is fairly evenly split between men and women, as well as whether the individual has at least primary education or not. Close

Results with both sets of dummies are presented below and are generally in line with a priori expectations. Multinomial regression coefficients are not easily interpretable, so the marginal effects were calculated, which indicate the marginal impact on the probability of the dependent variable (Greene, 2008). Each column presents the marginal effect for patients who reported being ill relative to not being sick which is the base outcome. For example, in the hospital column, the marginal effect of having good health reduces the probability of visiting a hospital by 0.00825 [Table A3]. These results suggest that those with health insurance are more likely to seek care at a hospital or clinic and less likely to do nothing. Income and having insurance are important determinants and imply that the poor have access problems. Married individuals and adults with a chronic condition are more likely to seek care at a hospital or clinic. Those with a chronic condition are less likely to do nothing when unwell. Women are more likely to seek care in a clinic. Those in urban settings are likely to seek care in a clinic and less likely to do nothing. Adults with good self-reported health are, unsurprisingly, less likely to need care. Both the educated and employed are less likely to seek care at a hospital. The more educated are likely to seek care at a clinic rather than doing nothing when unwell. Households with fewer family members are less likely to seek care at a clinic, while households with large monthly expenditures are more likely to seek care at a clinic and less likely to do nothing when unwell. The age variable was negative for all choices relative to not being sick, while the age-squared term revealed mixed significance. These results do not give a clear pattern with regard to the importance of age and may be specific to the sample, but suggest that age is not a driving factor relative to the other regressors when seeking care. The reason for visit variable is computed relative to those who have fever, cough or severe diarrhea. These revealed that individuals who are more likely to choose a hospital go for reasons related to child birth, asthma, heart disease, bodily injury, minor surgery or other reason not specified. These responses seem intuitive and seem to capture the main types of hospital services. Those visiting a clinic are more likely to go for antenatal or dental care reasons. Country dummies were used as control variables (not shown). The marginal effects of Namibia, Sri Lanka, and Vietnam dummies were positive on seeking care in a hospital or clinic and negative on doing nothing. The results suggest that typically most country dummies had a positive effect on seeking care in a clinic (21 out of 35 country dummies) while only 10 had a positive effect on seeking care in a hospital [Table A4]. The marginal effects for the clinic results are overall greater in magnitude than the marginal effects for the hospital results. Again, these results should be interpreted

D. Srivastava, A. McGuire / Social Science & Medicine 133 (2015) 21e27

with caution as the country dummies in part reflect the regulatory environment and aim to account more broadly for system heterogeneity. It should be highlighted that secondary data poorly records multiple health visits, so utilisation data is limited to one health care visit in these settings. Data on prices of medicines and volume consumed is lacking in these settings to directly estimate price elasticities, which necessitates imputation of price elasticities from patient or household level health care expenditure data. The elasticity of the predicted expenditure variable was calculated. The elasticity here is defined as the percentage change in the predicted probability of whether choosing to seek care at a provider as a result of a 1% increase in the expenditure of the same provider evaluated at the sample means. The elasticity results are mixed. The result for hospital care is
0.19 which implies that a 1% increase in expenditure at the hospital will reduce the probability of seeking care by 0.19% but the result is not significant (p-value 0.482). The result for clinic is 0.11 is significant at the 5% level (p-value 0.015). 5. Conclusion and policy implications This paper studied determinants of access to medicines and health care at the patient level and to estimate price responsiveness. The dataset covers a large cross section of countries and includes urban and rural settings and allowed for more robust estimates to contribute to the evidence base. The empirical methods aimed to correct for endogeneity between the cost of the care and the decision to seek care, using predicted expenditure levels as an instrument. The findings indicate that certain variables affect the decision to seek care and these include gender, marital status, health status, insurance, urban settings, education, employment, and households with large monthly expenditures. These findings are consistent with the literature and suggest that the poor will have access problems relative to those who can afford insurance. Elasticities ranged from
0.19 (hospital) to 0.11 (clinic). The results for hospital are fairly close to estimates in the literature even though the result was not significant. The counterintuitive price elasticity estimate for clinic suggests that price elasticities may not be strictly negative and inelastic. This could in part be due to model misspecification relating to missing indirect expenditure information or bias in the recall period. Respondents were asked to provide information on their most recent visit within the past year. The potential bias should be low as most of the respondents indicated that their most recent visit occurred within the past month. Other reasons for these counterintuitive results could relate to cultural factors, where additional payments are expected as a form of gratitude (Falkingham, 2004), or perceptions of improved quality of care through additional payments once they are at the facility (Kondo and McPake, 2007). This might occur more at the clinic than hospital level, accounting for the differential sign across these elasticity estimates. The findings indicate that system regulation has an effect on seeking care. In the MNL model, the results suggest that typically most countries have a positive effect on seeking care in a clinic (21 out of 35) while only 10 have a positive effect on seeking care in a hospital. There are limitations with this study which should be noted. The data used direct health expenditure and did not include indirect expenditure for constructing the predicted health expenditure price variable. Reviews of the WHS estimated reveal that health expenditure is overestimated relative to other international surveys (Ustun et al., 2003; Xu et al., 2009). Furthermore, while household expenditures are the common proxy for income, this variable may not appropriately capture differences in true income between

25

households. This model does not capture information relating to travel costs or waiting times which would have been preferable to include in the analysis. As a result of missing observations, it would have been preferable to have a larger cross-sectional dataset for analysis. There are policy implications from these results. The mixed results of the elasticity estimates do not necessarily imply that high user fees could be a policy response for revenue generation. While insurance schemes have other tools that are used to reduce the adverse effects of user fees such as benefit information or to encourage take-up of preventive care, they are unlikely to play a large role in the countries covered in this study and is more applicable in high-income country settings. Early empirical work initially provided the basis for widespread promotion of user fee policy due to estimates of inelastic demand for health care. The negative effects on utilisation and equity, with mixed impact on quality, modest impact on revenue generation have caused a shift in the international debate on user fees found in the WHO resolution 58.33 and the World Bank's strategy to support the removal of user fees for children and pregnant women (Preker et al., 2002; Palmer, 2004; WHO, 2005; James et al., 2006; Lagarde, 2008; Meessen et al., 2007). A decade of research showing such negative impacts has moved towards prioritising work on equity and the importance of quality along with increased emphasis on exemption schemes (Meessen et al., 2011; Ridde et al., 2011; Orem et al., 2011; Sekabaraga et al., 2011; Nimpagaritse et al., 2011; Ponsar et al., 2011; Witter et al., 2011; Steinhardt et al., 2011; McPake et al., 2011). A recent multi-country study documented on how countries implemented user fee removal (Meessen et al., 2011). The introduction of subsidies led to gaming behaviour among health professionals, abolition of user fees led to a rise in OOP, but there are system-wide effects relating to capacity due to increased utilisation such as drug-stockouts or the capacity of facilities to recruit local staff (Nimpagaritse et al., 2011; Orem et al., 2011; Ponsar et al., 2011; Ridde et al., 2011; Witter et al., 2011). One alternative implication is that government policies should aim to lessen the burden of OOP payments faced by households in these settings. A number of regulatory hurdles exist, including unregulated mark-ups along the supply chain which result in high retail prices of medicines (Levison and Laing, 2003), presence of private sector procurers, poor public sector governance, and lack of enforcement capacity (Seiter, 2010). The recent WHO/HAI survey provided an important first step in data collection to analyse the medicine supply chain. Of particular importance was that mark-ups typically are not regulated and in some countries can vary from 20% to 150% (WHO/HAI, 2006), contributing to a larger share of a medicine's overall price than the manufacturer's price. For example in Malaysia, mark-ups were higher for generics (46%e150%) versus branded drugs (27%e80%) and greater mark-ups were noted for dispensing doctors (129% for originator and 234% for generic) (WHO/HAI, 2006). Furthermore, mark-ups in the private sector exceeded those in the public sector. Better regulation and enforcement of mark-ups could reduce the overall retail price of medicines to patients thereby improving their affordability and access. Another key challenge is that governments are not the only procurers of medicines. The private sector plays a key role as well (Russo and McPake, 2010). Some countries could have very high procurement prices and therefore it would be useful to understand the factors which underpin high procurement prices. Another policy challenge for governments relates to corruption. While most large pharmaceutical firms have explicit policies against corruption and unethical business practices based on the international codes for ethical marketing, these are less likely to be enforced in countries with weak overall governance (Seiter, 2010).

26

D. Srivastava, A. McGuire / Social Science & Medicine 133 (2015) 21e27

This is a particular problem where activities take place on a local level, where smaller firms are less exposed to oversight and more likely to resort to unethical practices. Some examples include using cheaper, lower-quality raw materials, eliminating labour-intensive in-process controls; switching off electricity-consuming airhandling systems and reducing other activities that are part of good manufacturing practices (Seiter, 2010). There is evidence that importers collude with foreign suppliers to misreport procurement prices, and with retailers not to apply statutory margins. Pharmacists are reported to adjust prices according to market demand (Russo and McPake, 2010; Seiter, 2010). These findings are consistent with the body of economic literature which maintains that price controls are not effective policy tools, especially in low-income countries (Hongoro and Kumaranayake, 2000). Nevertheless, there is scope for government policies to strengthen procurement and pricing and reimbursement policies. Explicit pricing policies are not common place in low-income countries. Such policies are involved and incur administration costs (WHO, 2004). The WHO report noted that such costs contribute to the low uptake of adopting pricing policies with only half of all lower income countries have any pricing policy in place (WHO, 2004). In the short run, the arrival of international organisations involved in drug procurement could offer countries the possibility to draw on existing price data as they could benefit from this information to improve their skills in demand forecasting for medicines. Formalised agreements with these institutions could assist countries to improve their knowledge on prices. Similarly they could draw on existing cost-effectiveness data for drugs that are common in both high and low-income settings (e.g. prevalence of chronic diseases). Cost-effectiveness analysis could be applied to reflect these settings where factors such as morbidity, mortality and disease prevalence could inform such discussions. International organisations could draw on such tools to inform their pricing decisions in the short run. The long term solution is for countries to develop this capacity, as they become more skilled in procurement negotiations. Even though pharmacoeconomic analysis currently does not play an important role in policy-making in many low-income countries, external pressures could potentially encourage its uptake (Babar and Scahill, 2010). The increasing presence of international organisations that work on behalf of these countries to purchase medicines, assist with procurement, provide donations in the form of freely available medicines, are required to follow appropriate accountability and transparency policies. As more multinational firms operate in countries which recognise their international property rights, this will have implications for rationing of health spending. Pharmacoeconomic analysis could play a greater role in informing government decisions on the pricing and reimbursement of medicines. At the provider level, the right incentives are required for pharmacies and hospitals to procure medicines such as financial incentives to encourage cost effective prescribing (e.g. flat payment to prescribe the cost effective drug) (Homedes et al., 2001). There is scope for policies to target physicians, and pharmacists to improve rational prescribing practices in these settings. Financial and nonfinancial incentives should complement one another. Financial incentives could be designed to reward cost effective prescribing or through pay for performance activities. Financial policies have the risk of leading to gaming behaviour and should not be the only policy tool. Non-financial incentives such as continuing education activities, clinical guidelines, licensure and accreditation and revalidation should reward clinical behaviour. Evidence shows that multi-pronged approaches achieved an improvement in rational prescribing practices (Pagnoni et al., 1997; Chaudhury et al., 2005).

Future research could study patterns of demand over time. Data on patient preferences for traditional medicine could help to better understand unmet need as well as better classifying households with respect to need an access. While there is some research on the impact of cost-sharing and medicine utilisation in high-income settings, less has been done in lower income settings and the possible non-compliance issues relating to costs and efficacy (Soumerai, 1994; Hughes et al., 2001; Motheral, 2001; Landsman, 2005; Goldman et al., 2007). Information on treatment settings (public or private setting) or at home would supplement this analysis. Another approach would be to estimate price elasticities for each country and to include interaction effects rather than computing one overall estimate. This paper has shown that access to medicines is a pressing yet complex public health issue. Research in this area is needed to continue to build evidence to inform the design of effective pharmaceutical policy and to contribute to improving access to medicines for people in low-income countries. Acknowledgements We would like to thank Martin Knapp, Barbara McPake and Simeon Thornton and three anonymous reviewers for their helpful comments. Appendix A. Supplementary data Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.socscimed.2015.03.021. References Abdu, Z., Mohammed, Z., et al., 2004. The impact of user fee exemption on service utilization and treatment seeking behaviour: the case of malaria in Sudan. Int. J. Health Plan. Manag. 19 (S1), S95eS106. Aday, L.A., Andersen, R., 1974. A framework for the study of access to medical care. Health Serv. Res. 9 (3), 208e220. Akin, J., Griffin, C., et al., 1986. The demand for primary health services in the Bicol region of Philippines. Econ. Dev. Cult. Change 34 (4), 755e782. Anyanwu, J.C., 2007. Demand for health care Institutions' services: evidence from malaria fever treatment in Nigeria. Afr. Dev. Rev. 19 (2), 304e330. Asenso-Okyere, W.K., Dzator, J.A., et al., 1996. The behaviour towards malaria care-a multinomial logit approach. Soc. lndicators Res. 39, 167e186. Asfaw, A., v. Braun, J., et al., 2004. How big is the crowding-out effect of user fees in the rural areas of Ethiopia? Implications for equity and resources mobilization. World Dev. 32 (12), 2065e2081. Babar, Z., Scahill, S., 2010. Is there a role for pharmacoeconomics in developing countries? Pharmacoeconomics 28 (12), 1069e1074. Bolduc, D., Lacroix, G., et al., 1996. The choice of medical providers in rural Benin: a comparison of discrete choice models. J. Health Econ. 15 (4), 477e498. Borah, B.J., 2006. A mixed logit model of health care provider choice: analysis of NSS data for rural India. Health Econ. 15 (9), 915e932. Chaudhury, R.R., Parameswar, R., et al., 2005. Quality medicines for the poor: experience of the Delhi programme on rational use of drugs. Health Policy Plan. 20 (2), 124e136. Creese, A.L., 1991. User charges for health care: a review of recent experience. Health Policy Plan. 6 (4), 309e319. Cutler, D.M., 2002. Health care and the public sector. In: Auerbach, A.J., Feldstein, M. (Eds.), Handbook of Public Economics, vol. 4. Elsevier Science, Amsterdam, pp. 2142e2243. Donabedian, A., 1972. Models for organising the delivery of personal health services and criteria for evaluating them. Millbank Mem. Fund. Q. 50 (4), 103. Dzator, J., Asafu-Adjaye, J., 2004. A study of malaria care provider choice in Ghana. Health Policy 69 (3), 389e401. Falkingham, J., 2004. .Poverty, out-of-pocket payments and access to health care: evidence from Tajikistan. Soc. Sci. Med. 58 (2), 247e258. Gertler, P., Locay, L., et al., 1987. Are user fees regressive?: the welfare implications of health care financing proposals in Peru. J. Econ. 36 (1e2), 67e88. Gertler, P., van der Gaag, J., 1990. The Willingness to pay for Medical Care: Evidence from two Developing Countries. Johns Hopkins University Press, Baltimore. Gilson, L., 1997. The lessons of user fee experience in Africa. Health Policy Plan. 12 (3), 273e285. Goddard, M., Smith, P., 2001. Equity of access to health care services: theory and evidence from the UK. Soc. Sci. & Med. 53 (9), 1149e1162. Goldman, D.P., Joyce, G.F., et al., 2007. Prescription drug cost sharing: associations with medication and medical utilization and spending and health. JAMA 298

D. Srivastava, A. McGuire / Social Science & Medicine 133 (2015) 21e27 (1), 61e69. Greene, W.H., 2008. Econometric Analysis, sixth ed. Pearson Prentice Hall, Upper Saddle River. Grossman, M., 1972. On the concept of health capital and the demand for health. J. Polit. Econ. 80 (2), 223e255. Heller, P., 1982. A model of the demand for medical and health services in Peninsular Malaysia. Soc. Sci. & Med. 16 (3), 267e284. Homedes, N., Ugalde, A., et al., 2001. Scientific evaluations of interventions to improve the adequate use of pharmaceuticals in Third World countries. Public Health Rev. 29 (2e4), 207e230. Hongoro, C., Kumaranayake, L., 2000. Do they work? Regulating for-profit providers in Zimbabwe. Health Policy Plan. 15 (4), 368e377. Hughes, D.A., Bagust, A., Haycox, A., Walley, T., 2001. The impact of non-compliance on the cost-effectiveness of pharmaceuticals: a review of the literature. Health Econ. 10, 601e615. Hutton, G., 2004. Charting the Path to the World Bank's “No Blanket Policy on User Fees”: a Look Over the Past 25 Years at the Shifting Support for User Fees in Health and Education and Reflections on the Future. DFID Health Systems Resource Centre, London. Jack, W., 1999. Principles of Health Economics for Developing Countries. The World Bank, Washington, DC. James, C.D., Hanson, K., et al., 2006. To retain or remove user fees?: reflections on the current debate in low- and middle-income countries. Appl. Health Econ. Health Policy 5 (3), 137e153. Jones, A.M., Rice, N., et al., 2007. Applied Health Economics. Routledge, Abingdon. Kondo, M., McPake, B., 2007. Making choices between prepayment and user charges in Zambia. What are the results for equity? Health Econ. 16 (12), 1371e1387. Laxminarayan, R., Parry, I.W.H., et al., 2006. Should new Anti-malarial Drugs be Subsidized? Resources for the Future (Discussion Papers, Washington DC). Lagarde, M., Palmer, N., 2008. The impact of user fees on health service utilization in low- and middle-income countries: how strong is the evidence? Bull. World Health Organ. 86 (11), 839e848. Landsman, P.B., Yu, W., Liu, X., Teutsch, S.M., Berger, M.L., 2005. Impact of 3-tier pharmacy benefit design and increased consumer cost sharing on drug utilization. Am. J. Manag. Care 11 (10), 621e628. Le Grand, J., 1978. The distribution of public expenditure: the case of health care. Economica 45 (178), 125e142. Levison, L., Laing, R., 2003. The hidden costs of essential medicines. Essent. Drugs Monit. 33, 20e21. McPake, B., 1993. User charges for health services in developing countries: a review of the economic literature. Soc. Sci. & Med. 36 (11), 1397e1405. Mcpake, B., Brikci, N., Cometto, G., Schmidt, A., Araujo, E., 2011. Removing user fees: learning from international experience to support the process. Health Policy Plan. 26 (Suppl. 2), ii104eii116. Meessen, B., Kashala, J.P., et al., 2007. Output-based payment to boost staff productivity in public health centres: contracting in Kabutare district, Rwanda. Bull. World Health Organ. 85 (2), 108e115. Meessen, B., Hercot, D., et al., 2011. Removing user fees in the health sector: a review of policy processes in six sub-Saharan African countries. Health Policy Plan. 26 (Suppl. 2), ii16eii29. Mooney, G.H., 1983. Equity in health care: confronting the confusion. Eff. Health Care 1 (4), 179e185. Motheral, B., Fairman, K., 2001. Effect of a three-tier prescription copay on pharmaceutical and other medical utilization. Med. Care 39 (12), 1293e1304. Nimpagaritse, M., Bertone, M., 2011. The sudden removal of user fees: the perspective of a frontline manager in Burundi. Health Policy Plan. 26 (Suppl. 2), ii63eii71. Ojeda, G., Murad, R., et al., 1994. Testing pricing payment systems to improve access and cost-recovery from Norplant. In: Association Pro-Bienestar de la Familia Colombiana (PROFAMILIA), PCII, International Planned Parenthood Federation Western Hemisphere Region (IPPF/WHR). USAID, Washington DC. Oliver, A., Mossialos, E., 2004. Equity of access to health care: outlining the foundations for action. J. Epidemiol. Commun. Health 58 (8), 655e658. Orem, J., Mugisha, F., Kirunga, C., Macq, J., Criel, B., 2011. Abolition of user fees: the Uganda paradox. Health Policy Plan. 26 (Suppl. 2), ii41eii62. Pagnoni, F., Convelbo, N., et al., 1997. A community based program to provide prompt and adequate treatment of presumptive malaria in children. Trans. R. Soc. Trop. Med. Hyg. 91, 512e517. Palmer, N., Mueller, D., et al., 2004. Health financing to promote access in low income settings-how much do we know? Lancet 364 (9442), 1365e1370. Pokhrel, S., De Allegri, M., Gbangou, A., Sauerborn, R., 2010. Illness reporting and

27

demand for medical care in rural Burkina Faso. Soc. Sci. & Med. 70 (11), 1693e1700. Ponsar, F., Van Herp, M., Gerard, S., Philips, M., Jouquet, G., 2011. Abolishing user fees for children and pregnant women trebled uptake of malaria-related interventions in Kangaba, Mali. Health Policy Plan. 26 (Suppl. 2), ii72eii82. Preker, A.S., Carrin, G., et al., 2002. Effectiveness of community health financing in meeting the cost of illness. Bull. World Health Organ. 80 (2), 143e150. Qian, D., Pong, R.W., et al., 2009. Determinants of health care demand in poor, rural China: the case of Gansu Province. Health Policy Plan. 24 (5), 324e334. Ridde, V., Richard, F., Bicaba, A., Queuille, L., Conombo, G., 2011. .The national subsidy for deliveries and emergency obstetric care in Burkina Faso. Health Policy Plan. 26 (Suppl. 2), ii30eii40. Russo, G., McPake, B., 2010. Medicine prices in urban Mozambique. A public health and economic study of pharmaceutical markets and price determinants in lowincome settings. Health Policy Plan. 25 (1), 70e84. Sarma, S., 2009. Demand for outpatient healthcare, empirical findings from rural India. Appl. Health Econ. Health Policy 7 (4), 265e277. Sauerborn, R., Nougtata, A., et al., 1994. The elasticity of demand for health care in Burkina Faso: differences across age and income groups. Health Policy Plan. 9 (2), 185e192. Schwartz, J.B., Akin, J.S., et al., 1988. Price and income elasticities of demand for modern health care: the case of infant delivery in the Philippines. World Bank Econ. Rev. 2 (1), 49e76. Seiter, A., 2010. A Practical Approach to Pharmaceutical Policy. World Bank, Washington DC. Sekabaraga, C., Diop, F., Soucat, A., 2011. Can innovative health financing policies increase access to MDG-related services? Evidence from Rwanda. Health Policy Plan. 26 (Suppl. 2), ii52eii62. Sepehri, A., Chernomas, R., 2001. Are user charges efficiency- and equityenhancing? A critical review of economic literature with particular reference to experience from developing countries. J. Int. Dev. 13 (2), 183e209. Soumerai, S.B., McLaughlin, T., Ross-Degnan, D., Casteris, C.S., Bollini, P., 1994. Effects of a limit on medicaid drugreimbursement benefits on the use of psychotropic agents and acute mental health services by patients with schizophrenia. N. Engl. J. Med. 331 (10), 650e655. Soyibo, A., Mbanefoh, G.F., Anyanwu, J.C., 2004. Health care demand in Nigeria: emerging policy issues and recommendations. In: Mwabu, Germano, Wang'ombe, Joseph, Okello, David, Munishi, Gaspar (Eds.), Improving Health Policy in Africa. University of Nairobi Press, Kenya, pp. 57e76. Steinhardt, L., Aman, I., Pakzad, I., Kumar, B., Singh, L., Peters, D., 2011. Removing user fees for basic health services: a pilot study and national roll-out in Afghanistan. Health Policy Plan. 26 (Suppl. 2), ii103eii192. Thiede, M., Akweongo, P., et al., 2007. Exploring the dimensions of access. In: McIntyre, D., Mooney, G.M. (Eds.), The Economics of Health Equity. Cambridge University Press, Cambridge, pp. 103e123. Üstün, T.B., et al., 2003. WHO multi-country survey study on health and responsiveness 2000e2001. In: Murray, C.J.L., Evans, D.B. (Eds.), Health Systems Performance Assessment: Debates, Methods and Empiricism. World Health Organization, Geneva. Van de Ven, P.M.M., Ellis, R.P., 2000. Risk adjustment in competitive health plan markets. In: Culyer, A.J., Newhouse, J.P. (Eds.), Handbook of Health Economics, vol. 1(A). Elsevier Science B.V., Amsterdam, pp. 755e845. Witter, S., Khadka, S., Nath, H., Tiwari, S., 2011. The national free delivery policy in Nepal: early evidence of its effects on health facilities. Health Policy Plan. 26 (Suppl. 2), ii84eii91. World Bank, 2005. World Bank Development Indicators. World Bank, Washington DC. WHO/HAI, 2006. Price, Availability and Affordability: an International Comparison of Chronic Disease Medicines. Health Action International, Geneva. WHO, 2004. WHO Medicines Strategy: Countries at the Core 2004e2007. WHO, 2005. WHA58.33 Sustainable Health Financing, Universal Coverage and Social Health Insurance. Fifty-eight World Health Assembly WHO, Geneva. Available at: http://uhcforward.org/sites/uhcforward.org/files/WHO_WHA58_ 33_Resolution_for_Sustainable_health_financing_universal_coverage_and_ social_health_insurance.pdf. WHO., 2006. World Health Survey. World Health Organization, Geneva. WHO, 2007. Access to Medicines. Retrieved July, 2007, from. http://www.who.int/ trade/glossary/story002/en/. Xu, K., Ravndal, F., Evans, D.B., Carrin, G., 2009. Assessing the reliability of household expenditure data: results of the World Health Survey. Health Policy 91 (3), 297e305.