The Effect of Health System Responsiveness on Health Outcome: Evidence from Spain

Background: For the last two decades, the health system responsiveness has gained attention in the health policy area. However, little is known about its effect within the healthcare system on health outcome. This study aims to investigate the influence of health system responsiveness on self-assessed health. Particularly it examines if self-assessed health is affected by satisfaction with communication, dignity and waiting time. Methods: The study used data from the Spanish Health Care Barometer Survey (SHBS) between 2011 and 2013. The Ordered Probit and the Hierarchical Ordered Probit (HOPIT) model was used to model anchoring vignettes and to control the problem of reporting heterogeneity arises from self-reported health. Results: The result suggests a strong positive association between reporting very good self-assessed health and most of the domains of health system responsiveness. Specifically, after adjusting for reporting heterogeneity, satisfaction with waiting time and communication were found to be statistically significant and positively associated with reporting better self-assessed health for respondents in primary care and hospital care settings, respectively in Spain. The marginal effect of a one unit increase in satisfaction with waiting time in primary care and communication in hospital care is associated with a 2% and 4% increase in the in the likelihood of reporting very good health status respectively, keeping other variables constant. Conclusions: Overall, the result suggests that improving patient’s satisfaction with health systems responsiveness may have a positive influence on patients’ health outcomes.


Introduction
Health System Responsiveness has been introduced by the World Health Organization (WHO) as one of the intrinsic goals of the healthcare systems, alongside health outcomes, and fairness of financial contributions [1]. Health system responsiveness as "the way in which individuals are treated and the environment in which they are being treated, encompassing the notion of an individual's experience of contact with the health system" [2]. Responsiveness is a multidimensional concept encompassing eight domains: prompt attention, dignity, communication, autonomy, choice, confidentiality, quality of basic amenities and access to family and community support networks. Prompt attention, communication, and dignity appear to be particularly relevant for the patients' satisfaction with health systems. For instance, they have been rated as the most important among the responsiveness domains by respondents to World Health Survey (a survey launched by the WHO in 2001 and comprising micro-data about 70 countries) [1,3]. This study aims to investigate the influence of these three domains of responsiveness on selfassessed health.
Prompt attention, communication and dignity have raised concern in the health policy area in recent years. As an example, waiting time (a facet of Prompt Attention) is a major health policy concern in several OECD countries in general [4][5][6], and in Spanish National Health System (NHS) in particular.
The increase in waiting times for surgery, diagnostic procedures and specialized visits are some of the challenges that the Spanish NHS face currently. For instance, average waiting times for cataract surgery increased -from 89 days in 2010 to 105.1 days in 2015 and for hip replacement increased from 136 to 150.1 days in the same period [4,7]. The increase in waiting lists observed in Spanish NHS is likely to be due to the austerity measures adopted by the Spanish government such as budgetary and supply cutbacks [7,8]. The Spanish economic recession began in 2008 due to falls in gross domestic products for two consecutive quarters and an increase in the unemployment rate [8]. The austerity measures in effect on emergency readmission rates. Some studies revealed a positive association between communication and health outcome [16,22,23]. People who treated with dignity and respect are positively associated with reporting a better health outcome [14,15].
Considering the multidimensionality of health system responsiveness, this paper considers the influence of a set of domains of health system responsiveness on health outcomes -and not just a single domainby exploiting a dataset which is still scarcely investigating, that is the Spanish Healthcare Barometer Survey. Moreover, the study addresses the potential issue of reporting heterogeneity that might affect the self-reported measure of health as a dependent variable. Since self-reported health is measured by an ordered and categorical scale, the interpretation of the response scale may systematically differ across populations or population sub-groups [18,24,25]. Such phenomenon has been termed as "reporting heterogeneity" [19,26,27] or differential item functioning [28,29]. This study addresses this issue by estimating a Hierarchical Ordered Probit (HOPIT) model and exploiting vignettes from the World Health Survey (WHS).
The objective of this study is to investigate the linkage between the health system responsiveness and health outcome. Specifically, the study examines if satisfaction with waiting time, dignity and communication affects self-assessed health.

The Ordered Probit Model
Since self-assessed health is measured via self-assessment of individual respondents on an ordinal and categorical scale, the Ordered Probit model can be used to model such a discrete dependent variable which assumes ordered outcomes, e.g., y = 1, 2, .. , m.
The ordered Probit model assumes that there is an observed latent variable * distributed with mean and variance 1 [30], where refers to the individual respondent. This can be expressed as: * ~( , 1), = 1, … … . . , ……….. (1) The mean level of the latent scale is a function for a given covariates.
Let be the observed categorical response of individual to the main self-report question the ordered probit model can be expressed as [30].
If the assumption of homogenous reporting behavior that is intrinsic in the ordered probit model arises from the constant cut-points does not hold, in particular, if the cut points vary according to some of the covariates, then imposing this restriction will lead to biased estimates of the coefficients in the latent health index since they will reflect both health effect and reporting effects" [31].

The Hierarchical Ordered Probit model (HOPIT)
The Hierarchical Ordered Probit model (HOPIT) is applied to identify and correct for reporting heterogeneity. The HOPIT model was developed by [30] and [25] and it is an extension of the Ordered Probit model. The HOPIT model is made of two parts. One part shows the reporting behavior equation If we denote * the observed self-assessed health by individual to vignette , the observation mechanism is defined as follows: The cut-points can be a function of covariates i.e. = ′ .
Let be the observed categorical responses on the self-report such that: It is assumed that the error terms in the vignette and the latent health equation, and are independent for all = 1, … 5 and = 1, … 5.

Data
The study sample is selected from the Spanish Healthcare Barometer Survey (SHBS). Every year the SHBS provides information on patient satisfaction with the Spanish National Healthcare System [33].
The SHBS is a population-based cross-sectional survey which collects information on patient satisfaction since 1995. The survey includes rich information on socio-demographic characteristics, occupation, political and health system responsiveness variables. The survey contains information for satisfaction on health system responsiveness in primary care, hospital care and specialized care on three domains of health system responsiveness including Prompt attention (waiting time), communication (clear explanation) and dignity (respectful treatment). For this study, a sample is taken for the year between 2011 and 2013.

Study variables
Self-Assessed Health (SAH) is the main outcome variable in this study. SAH is one of the most commonly employed measures of overall individual health [34]. SAH is often considered as a good predictor of objective health and mortality [34][35][36]. Self-assessed health has been used in the literature to examine the relationship between subjective health and a wide range of socio-economic factors including education, income, and employment [37]. In this study, the SHBS respondents were asked to rate their general health condition using the question "How would you describe your state of health in general?". Health is measured on five-point scales since the response categories are "very good", "good", "fair", "bad" and "very bad". For the regression analysis, four categories have been created for this variable by collapsing the "bad" and "very bad" responses categories together as "Bad" response.
The main regressors are the satisfaction with the health system responsive domains variables, which measured on a scale from 1 to 10, where 1 indicates "completely dissatisfied" and 10 "completely Year dummies and regional dummies have also been controlled. The description of all the variables used in this study is reported in Supplementary material S1 Table 1.
On the basis of the study of Harris et al. (2015) [24], vignettes collected through the World Health Survey 2003 [41] and SHARE wave 1 [42] and wave 2 [43] and combine them individually with data about selfreported health in the Spanish Healthcare Barometer Survey to adjust for reporting heterogeneity in selfassessed health. The main self-report assessment from the SHBS and the vignette samples from both the WHS and SHARE were merged by implicitly assuming reporting styles are similar across samples [24].
Vignette questions are asked respondents to evaluate, on the same response scale as the main self-report question. Anchoring vignettes can be used to adjust for reporting heterogeneity and to assess whether differences in rates of self-reports across individuals and socio-economic groups comparable [24,32,44,45]. The full description of the vignettes used in this study can be found in S2 Table 2.

Reporting behavior
This section presents an example of differential reporting behavior of the respondents by their socioeconomic and demographic characteristics. categories for such question are "none", "mild", "moderate", "severe" and "extreme".    When estimating the HOPIT model, which corrects for reporting heterogeneity, several of the coefficients related to the controls in the outcome equation lose their statistical significance and/or change the magnitude. For instance, being male in an Ordered Probit model is significant at 1% and 10% level for hospital care and specialized care, respectively, however it becomes insignificant for HOPIT models. The change in the magnitude and statistical significance of the coefficients in the outcome equation of the HOPIT model reflects the presence of reporting heterogeneity, which is shown in Table   2 (outcome equation). Remember that, since the response categories for self-assessed health and the vignette responses range from "very bad" to "very good", a negative sign for a coefficient related to a given individual character in the threshold equation means that individuals with such characteristic have lower expectation with regard to health and therefore tend to report better health than individuals who don't share such characteristic. When estimating the HOPIT model,  Table 2 (outcome equation) becomes larger. For example, after correcting for the reporting heterogeneity due to satisfaction with waiting times, the effect of waiting time on health appears to be larger for primary care.

Results
For hospital care, the reporting style of individuals appears to be affected by satisfaction with communication. Table 2 also shows that for hospital care, the coefficients for communication are statistically significant and positive and negative in threshold equations m1 and m2, respectively. The overall effect of the correction for this kind of reporting style is that the coefficient for communication in the outcome equation in Table 2 becomes larger.
To check whether correcting for reporting heterogeneity using external vignettes from the 2003 World Health Survey is affected by the specific vignettes a robustness check is run for HOPIT by using externally collected vignettes from the Spanish part of SHARE, wave 1 and 2. The results for the HOPIT model regression for the robustness check are mostly consistent with those obtained when using vignettes from the World Health Survey 2003. The HOPIT regression result using SHARE wave 1 and wave 2 vignettes are presented in S5 Table 5.    mu1  mu2  mu3  mu1  mu2  mu3  mu1  mu2  mu3 Gender ( Table 3 presents the marginal effect of satisfaction with health system responsiveness domains on the probability of reporting very good self-assessed health status for patients treated with primary care, hospital care, and specialized care models. The average individual who is taken as a reference to calculate these effects is male, aged 64, educated, employed and foreigner. All the marginal effects in Table 3 are positive and statistically significant, apart from satisfaction with waiting time in hospital care. After adjusting for reporting heterogeneity, there are some changes in the magnitude and significance of the coefficients. For instance, when looking at the Ordered Probit model, for the domains of satisfaction with communication, dignity and waiting time a one unit increase in satisfaction with these domains (which range from 1 to 10) is associated with a 0.9%, 0.44% and 0.5% increase in the likelihood of reporting very good health in primary care, and 0.5%, 0.6% and 0.4 % in specialized care respectively, keeping other variables constant. After adjusting for reporting heterogeneity, this percentage goes up to 2% for waiting time in primary care and to about 4% for communication in hospital care. Therefore, the positive effect of the satisfaction with these domains of responsiveness on health appears to be larger after correcting for reporting heterogeneity. The full description of the marginal and average partial effects for all the variables included in the regression model is available in S4 Table 4.

Discussion
Using data from the Spanish Healthcare Barometer Survey (SHBS), this paper investigates the effect of three domains of health system responsiveness (communication, dignity and waiting times) on selfassessed health. The self-reports of the respondents appear to be affected by their reporting style, which varies because of their individual and socioeconomic variables such as education, age, and gender. After adjusting for this phenomenon of reporting heterogeneity, significant and positive association between the domains of health system responsiveness and self-assessed health is observed for in primary, hospital and specialized care. The study results are consistent with some of the previous studies on a single domain of responsiveness. For example, satisfied patients regarding the information given by their doctors and those who have good communication with them are more likely to complete treatment regimens and to be compliant and cooperative [16,22,46]. Being treated with dignity and being involved in decisions are independently associated with positive health outcomes [14].
This study contributes to the literature in several ways. Most of the previous studies which have investigated the determinants of self-assessed health have mainly focused on its socioeconomic determinants. However, since the WHO introduced the concept of health system responsiveness in 2000, if responsiveness has an impact on health has not been very well investigated empirically. The study investigates this research question by exploiting a novel dataset, the Spanish Healthcare Barometer Survey (SHBS). Moreover, it addresses the issue of reporting heterogeneity that arises from using a selfreported measure of health as a dependent variable by exploiting anchoring vignettes and estimating a HOPIT model. This study is not without limitations. First, the transmission mechanisms through which the domains of responsiveness may influence health still need further investigation. Especially, the relationship between dignity and health outcome has received little attention from previous literature and the pathway linking dignity to health needs to study more thoroughly. Second, ideally, when using externally collected vignettes and merging them to a dataset to adjust for reporting heterogeneity, the vignettes should be collected in the same year and country as the main dataset [24]. This study used vignettes that, although collected in the same country as the main dataset, were collected in a different year. However, this study addressed this issue by performing, on the top of the main analysis conducted using vignettes from the WHS, robustness checks run by using two set of vignettes from SHARE, collected in wave 1 and wave 2.

Conclusions
Overall, this study suggests that improving the way in which patients are treated in the health care system have a positive health outcome. This can be an important implication for health policy. For instance, patients who satisfied with waiting time or older individuals who treated respectfully during medical treatment are more likely to have a better health outcome. This work contributes to a growing body of research examining the influence of health system responsiveness on health and provides a rich foundation for future research and health interventions.

Consent for publication
Not applicable

Availability of data and materials
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.