Scholarly article on topic 'Transformation of potential medical demand in China: A system dynamics simulation model'

Transformation of potential medical demand in China: A system dynamics simulation model Academic research paper on "Economics and business"

CC BY-NC-ND
0
0
Share paper
Academic journal
Journal of Biomedical Informatics
OECD Field of science
Keywords
{"Potential medical demand" / "System dynamics model" / "Health policy"}

Abstract of research paper on Economics and business, author of scientific article — Wenya Yu, Meina Li, Yang Ge, Ling Li, Yi Zhang, et al.

Abstract Objectives The increasing of potential medical demand in China has threatened the health of the population, the medical equity, accessibility to medical services, and has impeded the development of Chinese health delivery system. This study aims to understand the mechanism of the increasing potential medical demand and find some solutions. Methods We constructed a system dynamics model to analyze and simulate this problem, to predict the influences of health policies on the actual percentage of patients not seeking medical care (adjusting the quantity structure of hospitals and community health systems (CHSs), adjusting outpatient prices, and adjusting the level of health insurance). Results Decreasing the number of hospitals, increasing the number of CHSs, and raising the proportion of health insurance compensation would effectively increase the transformation of potential medical demand. But currently, changes of the outpatient prices didn’t play a role in the transformation of potential medical demand. Conclusions Combined with validation analysis and model simulation, we suggest some possible solutions. The main factors causing potential medical demand are accessibility to medical services and proportion of health insurance compensation. Thus, adjusting the number of hospitals and CHSs and increasing the proportion of health insurance compensation should decrease the actual percentage of patients not seeking medical care and accelerate the transformation of potential medical demand, which deserved being concerned in policymaking.

Academic research paper on topic "Transformation of potential medical demand in China: A system dynamics simulation model"

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

Journal of Biomedical Informatics xxx (2015) xxx-xxx

Contents lists available at ScienceDirect

Journal of Biomedical Informatics

journal homepage: www.elsevier.com/locate/yjbin

Transformation of potential medical demand in China: A system dynamics simulation model

Wenya Yu1, Meina Li1, Yang Ge, Ling Li, Yi Zhang, Yuan Liu, Lulu Zhang *

Institute of Military Health Management, CPLA, Faculty of Health Service, Second Military Medical University, Shanghai 200433, China

ARTICLE INFO ABSTRACT

Objectives: The increasing of potential medical demand in China has threatened the health of the population, the medical equity, accessibility to medical services, and has impeded the development of Chinese health delivery system. This study aims to understand the mechanism of the increasing potential medical demand and find some solutions.

Methods: We constructed a system dynamics model to analyze and simulate this problem, to predict the influences of health policies on the actual percentage of patients not seeking medical care (adjusting the quantity structure of hospitals and community health systems (CHSs), adjusting outpatient prices, and adjusting the level of health insurance).

Results: Decreasing the number of hospitals, increasing the number of CHSs, and raising the proportion of health insurance compensation would effectively increase the transformation of potential medical demand. But currently, changes of the outpatient prices didn't play a role in the transformation of potential medical demand.

Conclusions: Combined with validation analysis and model simulation, we suggest some possible solutions. The main factors causing potential medical demand are accessibility to medical services and proportion of health insurance compensation. Thus, adjusting the number of hospitals and CHSs and increasing the proportion of health insurance compensation should decrease the actual percentage of patients not seeking medical care and accelerate the transformation of potential medical demand, which deserved being concerned in policymaking.

© 2015 Published by Elsevier Inc.

Article history: Received 14 February 2015 Revised 20 July 2015 Accepted 12 August 2015 Available online xxxx

Keywords:

Potential medical demand System dynamics model Health policy

1. Introduction

1.1. Problem/significance

Currently, owing to the rapid growth of medical expenses and the unreasonable structure of China's health delivery system, there exists an unusual phenomenon, wherein as the supply of medical services increases, demand shows an unexpected decreasing trend. In fact, this potential medical demand (PMD) exists extensively in China. In this study, PMD is defined as medical need that has not been transformed into actual demand. PMD can be divided into one of two types: one that can be transformed into actual demand, and one that cannot. In China, the amount and structure of the supply and demand of medical services are out of balance, and this greatly influences the equal and effective use of medical services. Therefore, to understand and analyze the PMD problem in China,

* Corresponding author. Tel./fax: +86 02181871421.

E-mail address: zllrmit@163.com (L. Zhang).

1 The two authors made equal contributions to this study.

http://dx.doi.org/10.1016/jjbi.2015.08.015 1532-0464/© 2015 Published by Elsevier Inc.

this study sets forth a system dynamics (SD) simulation model by which to determine the main influencing factors and their mechanisms on PMD transformation; in this way, we can determine the best ways of transforming more PMD by making health policy adjustments.

Survey results vis-à-vis the demand for and use of medical services from 1993 to 2008 in urban and rural China are shown in Table 1 ; these are taken from An Analysis Report of National Health Services Survey in China [11]. As one can see, both of the medical services-use indexes show increasing trends; of these, the two-week medical consultation rate was in decline before 2003, but manifested a significant increase in both urban and rural areas in 2008. The annual rate of hospitalization, meanwhile, persistently increased between 1993 and 2008. These two PMD indexes generally showed an increasing trend, especially since the start of the 21st century, and although a slight declining tendency appeared in 2008, China's PMD continues to be large. The results of our analysis suggest that the growth of medical services supply and the decline in affordable medical demand had led to a low utilization rate, and the increased amount of PMD in China has dramatically

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

2 W. Yu et al./Journal of Biomedical Informatics xxx (2015) xxx-xxx

Table 1

Survey results of demand for and use of medical services from 1993 to 2008 in urban and rural China.

Index 1993 1998 2003 2008

Urban Rural Urban Rural Urban Rural Urban Rural

Two-week medical consultation rate (%) 19.9 16.0 16.2 16.5 11.8 13.9 12.7 15.2

Two-week untreated rate (%) 42.4 33.7 49.9 33.2 57.0 47.8 37.6 37.3

Annual rate of hospitalization (%) 5.0 3.1 4.8 3.1 4.2 3.4 7.1 6.8

Percentage of patients not being hospitalized (%) 26.2 40.6 29.5 35.5 27.8 30.3 26.0 24.7

influenced medical equity there. Therefore, it is necessary to analyze the reasons behind the growth of PMD and to seek reasonable health policies by which PMD can be transformed. At the same time, we need to highlight the factors that influence the two-week medical consultation rate, two-week untreated rate, annual rate of hospitalization, and the percentage of patients not being hospitalized; in urban areas, these are the provision of health insurance, family income [31], quality of care, and accessibility to medical care [72,64]. The factors that influence the four indexes in rural areas are the provision of health insurance [91], family income [98], medical costs, health status, socioeconomic status [60], and race [70]. Ultimately, no significant difference has been found in the demand for and use of medical services, between urban and rural areas.

The literature and survey results in China and elsewhere have explored the factors that influence medical demand, as well as those factors' underlying mechanisms. One of the most important factors refers to the patients themselves. Various studies have focused on the universal factors influencing patients' medical demand, such as health insurance (patients with health insurance were more likely to receive health care when it was needed) [10,52,21,72,48], economic status and monthly salary [30,60,73], health status [30,60,12], demographic factors (including gender, age, race, living settings, education level, and city size) [52,21,72,48,73], and patients' perceptions on medical need and medical use [65], types of health profiles [59], and social relations [81]. Another important influencing factor is the health care providers themselves. The quality of medical services [81], human resource structure, the provision of public and private services, reforms in health department, fee structure of health service system [68], and the degree of supplier-induced demand [84] will have an impact on medical decisions of health care providers.

In addition, more and more studies have focused on medical services demand in developing countries, such as those of the determinants of medical demand [63,1,19,26,79,2], the influencing factors of (and the changes to) medical demand, predictions of medical demand in the future [9,33,40,41,74,75,49], and the potential demand for AIDS vaccines in Thailand [87]. Additionally, some researchers have discussed the elasticity of medical demand in Ghana [50]; the elasticity of medical demand in a certain province of China [53]; the elasticity of medical demand among groups of infants, children, and low-income service recipients [79]; and the elasticity of medical demand among outpatients in China [98].

Additionally, some studies have explored the phenomenon of unmet medical demand. In exploring unmet medical demand and its influencing factors among various social groups, researchers have examined homeless adults [17], young adults [61], the disabled elderly [62], and children [42]. Some other studies explored the reasons behind an increase in unmet medical demand, and its solutions [80]; unmet medical demand regarding different types of disease [82]; and different methods of assessing unmet medical demand and the use of medical services [4].

Moreover, among studies of medical demand, a variety of methods and models have been used in analysis, estimation, and

determination. A two-part model and a discrete factor model were used to analyze a dataset that consisted of 6407 urban households in China, and to identify and estimate the determinants of medical demand based on the interaction between growth in medical demand and the shortage of medical care funds [31]. A stochastic dynamic programming model was used to research the influencing factors of medical demand [57]. The Grossman (1972) health capital model in a stochastic environment was used to investigate the effect of long-term care (LTC) insurance on medical care demand, mainly by comparing the influence of means-tested and health-based LTC programs on medical care consumption, decisions, and welfare [5]. A Probit regression model and a zero-truncated negative binomial regression model were used to test the effects of price and income on medical demand in rural China [98]. A Heckman-type model was employed to estimate medical demand and explore the influencing factors of medical demand in African countries [70]. A simplified version of a dynamic Grossman household production model was used to explore the effect of uncertainty of illness on medical demand [69], and Grossman's health production model and national survey data were also used to estimate the effect of digital health information on medical demand [20,38,89]. A generalized version of Grossman's health capital model was considered to examine medical demand, by testing the health capital model; it was also used to identify the key factors of the medical demand equation [45]. Finally, many health economists have developed medical demand models to predict medical demand [15,25,35,36,77,34].

There have also been some studies about medical demand in China. For example, some studies have analyzed medical demand, need, and service use in different regions in China [99,93]; examined a PMD-transformation system by constructing a logical model [54,55]; determined the influencing factors of medical demand by examining both theoretical and empirical elements and using an ordered Probit model [54,55,58,39,56,97]; and looked at the changing tendencies and characteristics of medical demand [92]. All these studies mainly concentrate on theoretical research and description analysis of the previous and current situation; however, they do have some limitations. For example, they cannot predict developing tendencies by simulating a system, and they cannot simulate whether or not a policy will exert a positive role in a health delivery system.

Therefore, to successfully transform PMD into actual demand, and in response to a proposal to bring about equity of medical services, we should make great efforts to identify the reasons behind PMD, recognize the mechanism by which PMD is transformed, pinpoint solutions to the problem of PMD transformation, and then improve the equity of medical services in China.

1.2. Objective

This study aims to determine the influencing factors of PMD and seek out transformation strategies, and thus provide reasonable evidence of scientific health resource allocation, balance the supply and demand of medical services, and ultimately bring about equity of medical services in China.

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

W. Yu et al./Journal of Biomedical Informatics xxx (2015) xxx-xxx

2. Methods

2.1. System dynamics model

The SD model is designed for use with large and complex systems; it is used to solve dynamic and complicated system issues, and it has been applied in a variety of fields. It has guided policymaking, helped find solutions to some specific problems, and assisted in assessing and optimizing interventions with regard to some socioeconomic issues.

In terms of the science and rationality of the SD model, more and more studies are leveraging it to examine various systems and problems; they mainly concentrate on fields in education, economy, policy, environment, medicine, and health, and the results thereof can be seen in the literature. Researchers have undertaken studies of supply and demand in a variety of fields, including medical service demand; other studies have touched upon water demand estimation [71], supply chain dynamics in health care services [78], supply and demand in LTC [43], forecasts of the demand for medical specialists [8], and the prediction of health care demands [88,18]. Previous studies on medical demand have mainly used theoretical analysis, empirical analysis, investigation analysis, and some other health economics models [20,38, 57,69,45,89,31,5,70,98]—all of which are mainly suitable for describing the current situation and analyzing the influencing factors of medical demand. However, they cannot predict or simulate long-term changes that will take place in the future. Both previous and current data can be used to construct and validate an SD model; by leveraging the simulation function of the SD model, one can not only describe the current situation, determine the influencing factors, and predict future tendencies, but also observe changes under a variety of policy scenarios [90]. Therefore, to achieve a long-term vision of the current system and increase the validity of policy-making and reform scenarios, it is feasible to study PMD transformation by using an SD simulation model.

Compared to a variety of other methods, the SD model offers obvious advantages. First, compared to mathematical models, the SD model can help us understand the various levels of influence of specific populations or issues—a tack frequently taken in research in overweight conditions and obesity, for example [23]. Second, unlike a "snapshot-in-time" study, studies that make use of the SD approach can examine circular feedback processes; this is a more practical and comprehensive method [27]. Third, the SD model allows one to explain and understand dynamic interactions further and better, by intermingling the influencing factors with some complex systems and problems [18]. Fourth, unlike some other approaches, the SD model overcomes acknowledged deficiencies, such as isolated, static, and one-sided analysis; as such, an SD model is a powerful tool in exploring difficult and complex management problems [76]. Fifth, compared to econometrics and input-output analysis, an SD model does not require high-quality statistics, and it stresses the conjunction structure and the feedback mechanism of the system. The realization of the SD model is easier, and its application extent is more extensive than that of other models [24]. Finally, it should be noted that in the current study, the SD simulation model is implemented through the use of Vensim DSS software.

2.2. Model description

A PMD-transformation system is one in which PMD is transformed into actual medical demand; when such a transformation occurs, the medical demands of patients whose demands cannot be realized (for various socioeconomic reasons) can be more easily met and, within the context of this study, the health conditions of

China's citizens can be improved. This system involves four sub-jects—namely, health service institutions, health insurance institutions, governments, and patients. Currently, health insurance compensation is far from comprehensive; there are three main types of health insurance in China, but only the basic medical insurance for urban workers has a relatively high compensation level. Most of those who hold the basic medical insurance for urban residents and the new rural cooperative medical insurance must pay a majority of medical expenses by themselves; only 62.5% of China's rural population with health insurance declared that part of their medical expenses had ever been compensated. In fact, the medical demands of these two subsets of the population have remained unmet, mainly on account of economic factors. The two-week morbidity rate of the whole of China's population in 2008 (18.9%) increased by 32.2 percentage points, compared to 2003 (14.3%); the morbidity rate of people with chronic disease in 2008 (20.0%) increased by 32.5 percentage points, compared to that of 2003 (15.1%). Additionally, 24.4% of China's population declared that because of their poor economic status and the high cost of medical services, they did not seek medical care when they needed it; 70.3% of patients who needed hospitalization had failed to be hospitalized because of the economic difficulty that doing so incurs [11]. All these data and evidence indicate that the health status of China's population is worsening, and that their volume of medical demand is greatly increasing. Meanwhile, the income structure of hospitals and community health systems (CHSs) is unreasonable, and the financial compensation mechanism therein is unsound. Among the total health expenditures of hospitals, the proportion of input from the government is always lower than that from residents and social groups: in 2012, for example, those figures were 30.0%, 34.4%, and 35.6%, respectively. Meanwhile, the proportion of the input from the government continues to decrease; among the total income of hospitals and CHSs, that brought in on account of medical services is always higher than that brought in from financial compensation from the government [11]. This incomplete medical expenditure compensation mechanism has pushed hospitals and CHSs to pursue economic benefits by increasing the prices that they charge for medical services. As mentioned, the deterioration of health status and the increasing medical demand of patients in China, in combination with the low-level health insurance system currently in place, the high medical expenses, and the lack of input and compensation from the government, have exacerbated PMD. Making improvements to this system would promote interaction, communication, and cooperation among health service institutions, patients, the health insurance system, and the government; optimize health resource allocation, the compensation system, and the medical service price system; and help satisfy medical demand while transforming PMD into real demand. Ultimately, China, its society, its hospitals, its CHSs, and its patients would benefit from a successful PMD-transformation system.

2.3. Causal loop diagram

Based on the conceptual model of a PMD-transformation system, and combined with factors that influence the behavior of the four major components, a causal loop diagram can be developed. According to factors that affect medical demand in developed countries—and which have been the topics of research (e.g., level of current health expenditure, number of medical acts, number of sick people, prices of available services, reimbursement arrangements, medical density, new patient requirements, technical progress, and medical advances) [28]—specific factors influence medical demand in China, including the proportion of individual health payment, health service expenses, visits of patients to primary health institutions, accessibility to health service, and the

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

W. Yu et al./Journal of Biomedical Informatics xxx (2015) xxx-xxx

health payment capabilities of residents. Fig. 1 illustrates a logic model that represents the interactions and interrelations among the key factors. The connotations of the causal loops in the diagram are explicated as follows.

Analysis of the high proportion of individual health payment. At present, both health insurance coverage and the reimbursements made by health insurance are at low levels, and the proportion of health payments made by individuals is still too high. Although the Chinese government has made great efforts to enlarge the coverage of its health insurance (i.e., the coverage rate reached 87.1%, as of 2008), many studies and statistics have shown that the current compensation system is inequitable and unreasonable [37,95]: the average compensation level, for example, is only 25.3%. Additionally, the proportion of individual health expenditure is still too high, comprising 34.4% of all health expenditures in 2012 [14]. All these circumstances have greatly and dramatically increased PMD in China.

Analysis of the increase in health service expenses. Health service expenses are highly, mainly on account of three factors. First, health service expenses are influenced by the development of chronic diseases and social aging. According to the definition of "aging," China entered a stage of societal aging at the beginning of the 21st century; the proportion of people older than 65 years reached 7.0% in 2000, and it had grew to 9.4% in 2012. Additionally, the morbidity rate of chronic diseases reached 15.7% in 2008, and increased by 22.7% in 1998 and by 27.6% in 2003 [11]. Second, induced demand and excessive medical treatment have played significant roles in the increase in health service expenses. Physician-induced demand exists extensively in health care systems worldwide—such as that for newborn treatment in Japan [84]—and it creates excessive medical treatment; this has been seen in some statistics and pieces of literature below. When one examines the statistics in An Analysis Report of National Health Services Survey in China [11], one can see that if the number of doctors in China were to increase by 10%, the number of outpatients would increase by 3.6% [96]. The incomplete state of China's health surveillance and management system makes it possible for induced demand and excessive medical treatment to continue unabated. Third, drug, material, and inspection expenses in China are too high; at present, the main compensation channel of hospitals and CHSs is health service expenses, a majority of which comes from the marketing of drugs and conducting unnecessary inspections. The ratios of drug and material expenses to total expenses have also increased too rapidly: in 2012, they accounted for 51.3% of all outpatient medical expenses and 41.3% of all hospitalization medical expenses. In addition, the current price policy is ineffective and ossified, and compensation from the government accounts for a relative small proportion (8.15%) [14]; this has led to high drug and material expenses and high inspection expenses with regards to expensive equipment, and thereby increased the financial burden placed on China's citizens.

Analysis of the decrease in visits of patients to primary health institutions, and of accessibility to health services. Presently, a majority of patients would like to seek medical care in hospitals rather than visit a primary health institution. The proportion of visits of patients to primary health institutions accounted for 61.87%, 60.68%, and 59.65% in 2010, 2011, and 2012, respectively; this is in keeping with a decreasing trend compared to previous years, and it indicates that the health care system is just as out of balance as ever [14]. As a result, too many people are crowded into hospitals, and this creates health care service bottlenecks there. This phenomenon occurs mainly on account of the unreasonable health resource structure, which assumes a reverse-triangle structure (i.e., much more health resources are allocated to hospitals, while primary health institutions are greatly lacking health resources). At the same time, this irrational structure has led to too much atten-

tion being paid to disease treatment, rather than to disease prevention. Meanwhile, the reasons behind the unreasonable resource structure and the unavailability of health services are that the main compensation channel continues to be health service expenses, and the rate of return of the medical market is too high.

Analysis of the poor health payment capability of some residents. There are still many poor people in China. National statistics indicate that poverty continues to be a serious problem, with the extremely low-income population accounting for 9.1% of the whole population (as well as 10.1% of the urban population and 8.7% of the rural population). Alarmingly, 24.4% of the population who should be receiving medical treatment decline to do so, on account of the high cost of medical services and their poor economic situation [11]. Therefore, owing to their adverse economic situation, individuals within the low-income population cannot afford to take on medical expenditures, and this ultimately restricts the necessary medical demand.

In consideration of the interactions inherent in China's health delivery system and the findings within the literature, the biggest obstacles to satisfying medical demand are the patients' economic conditions, the health service institutions themselves, China's health insurance system, and the involvement of the Chinese government. It is of great importance and significance that PMD be transformed into real demand, and the health conditions of the whole of China be improved; this can ultimately be brought about by implementing compensation system reforms and making changes to health financing, enhancing the internal power of hospitals, strengthening the responsibilities of the Chinese government, and alleviating the financial burden placed on patients. Given that more and more health system performance problems are being generated in China, the various levels of government need to seek out and enact solutions more rapidly, and create a balancing-feedback loop where more solutions lead to fewer problems. However, even when problems are solved temporarily, there can exist many unexpected consequences in the long term, the worst of which is the increased need to transform PMD. Such unexpected consequences could not only offset currently effective solutions, but also exacerbate current problems and make them more complex and difficult.

2.4. Stock-flow diagram

To make accurate and reasonable analyses and evaluations, it is necessary to quantify the various variables within a model. In this model, the indicator is ''the actual percentage of patients not seeking medical care"; this refers to patients who need medical care but who, for various reasons, fail to receive it. This phenomenon is a suitable starting point in reflecting on the state of PMD. In terms of medical care provision, the health delivery system comprises hospitals and CHSs; apart from the service providers, however, the issue of PMD transformation is simultaneously affected by patient behavior and by health policies established by the government (i.e., health insurance). Patient behavior can affect medical demand, and this behavior can in turn be affected by their own health conditions, financial situations, and self-treatment, inter alia—which can, in turn, affect the amount of medical services provided by virtue of supply-demand balance theory. Additionally, the behavior of health service institutions can affect patients' willingness to seek out medical care, on account of the medical expenses incurred, the technologies used, the convenience of transportation, and so on. Both the behavior of patients and of health service institutions, furthermore, can be restricted by the health insurance system and by policies issued by the government (e.g., the price system, the compensation system, etc.). We can see that the four major components are interrelated and interact with each other, and hence codetermine PMD transformation. In accor-

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

W. Yu et al./Journal of Biomedical Informatics xxx (2015) xxx-xxx 5

enhance the interal power of hospitals

social and economic factors

rate of return of the medical market is high —

the low-income population increases

unreasonable health resource structure: stress treatment and overlook prevention, reverse triangle structure

the health payment capability of part of

residents are poor

solution 1: compensation system reform

the main compensation avenue is health service expenses

price policy ossifies

health surveillance and management is not enoug+

solution 2: the way of health financing reform

the coverage of health insurance is low

influences of chronic diseases and social aging .

proportion of individual health payment is too high "

induced demand

and excessive medical treatment

+ health service +expenses increase

potential medical demand

arises in a great proportion

symptoms of issues of the health service system

Fig. 1. Casual loop diagram of a PMD-transformation system in China.

dance with PMD transformation, the system contains five subsystems (i.e., those of medical demand, outpatients in hospitals, inpatients in hospitals, outpatients in CHSs, and inpatients in CHSs). The five subsystems interact with each other through input and output variables contained within these subsystems and which are closely related. In this study, we focus on outpatient medical demand, and so the three major subsystems are those of medical demand, outpatients in hospitals, and outpatients in CHSs. These three subsystems interrelate with each other through the visiting rate and the number of visits. The visiting rate and the number of visits to hospitals and CHSs constitute the total number of visits, which jointly decides the observation variable ''the actual percentage of patients not seeking medical care" and the variable ''morbidity rate per year" in the subsystem of medical demand. In addition, given that referrals between hospitals and CHSs would overly complicate our analysis, this study interrelates the two subsystems through only one link. The subsystem of inpatients in hospitals and the subsystem of inpatients in CHSs interrelate with the whole system by linking with the subsystem of outpatients in hospitals and CHSs, respectively; the link between them is the variable ''income of hospitalization." Understanding these relationships and interactions can help us develop a SD model by which we can analyze a PMD-transformation system in China.

Fig. 2 represents the stock-flow diagram of the system, based on the causal loop diagram. This diagram summarizes the model by establishing some hypotheses, thus simplifying our analysis and evaluation and helping readers better understand the PMD-transformation system. The stock variables are state variables, which are depicted in the figure as rectangles. The inputs and outputs between stocks are connected with flow variables, which represent the rate variables and are depicted as double-lined arrows. Inter-relationships among various variables are linked with single-lined arrows.

Stock variables include total population, amount of hospital resources, and amount of CHS resources. Flow variables include number of births, number of deaths, increment of hospital resources, depreciation of hospital resources, increment of CHS resources, and depreciation of CHS resources. Auxiliary variables

include visiting rate to hospitals, visiting rate to CHSs, hospitalization rate at hospitals, hospitalization rate at CHSs, average outpatient price of hospitals, average outpatient price of CHSs, average hospitalization price of hospitals, average hospitalization price of CHSs, morbidity rate per year, number of patients per year, number of hospitals, number of CHSs, average individual income, health insurance expense per capita, input to hospitals from the government, input to CHSs from the government, and average value of general real estate. The initial variables include initial value of total population, initial average outpatient price of hospitals, initial average outpatient price of CHSs, initial average hospitalization price of hospitals, initial average hospitalization price of CHSs, initial number of hospitals, initial individual income, initial health insurance expense per capita, initial number of CHSs, initial input to hospital from the government, initial input to CHSs from the government, initial amount of hospital resources, and initial amount of CHS resources. The constant variables include birth rate, death rate, average ratio of input by hospitals per year, average ratio of input by CHSs per year, fixed depreciation rate, proportion of health expenditure in consumption expenditure, and growth rate of income. The observation indicators mainly include the actual percentage of patients not seeking medical care, number of visits to hospitals, and number of visits to CHSs. All these variables were selected in line with the literature, as clarified below.

The variables in the subsystem of medical demand were analyzed (Fig. 3). The behavior of patients is mainly influenced by age [44], multichannel communications with physicians [51], technical level, communication skills, respect shown by doctors to patients [32], social supports, and having a health-promoting lifestyle [85,94]. These patients' behaviors can be embodied by the health status of the population and the visiting rate, both of which are influenced by various socioeconomic factors. Thus, to simplify the SD model, we employed the variable ''number of patients per year" to reflect the health status of the population, and this variable is determined through the use of the morbidity rate per year and the total population [18]; the total population is further decided by the number of births, the birth rate, the number of deaths, and the death rate. In addition, the variable ''the actual per-

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

6 W. Yu et al./Journal of Biomedical Informatics xxx (2015) xxx-xxx

Fig. 2. The system dynamics model of a PMD-transformation system in China. Three level variables: amount of hospital resources, amount of CHS resources, total population. Six rate variables: number of births, number of deaths, increment of hospital resources, depreciation of hospital resources, increment of CHS resources, and depreciation of CHS resources.

untreated patients Fig. 3. Subsystem of medical demand.

521 centage of patients not seeking medical care" was chosen to reflect

522 PMD [61], and it is jointly determined by the number of patients

523 per year, the number of visits, and the number of untreated visits.

524 Here, we would like to state that, as discussed above, there is no

525 significant difference in the demand and use of medical services

526 between the urban and rural populations, and so to simplify the

527 whole model, we did not break out the population thus; rather,

528 we regarded it as a whole.

529 We analyzed the variables in the subsystems of outpatients in

530 hospitals and CHSs (Figs. 4 and 5). To simplify the entire system,

as mentioned, referrals between hospitals and CHSs were not taken 531

into consideration; therefore, the subsystem of outpatients in hos- 532

pitals and CHSs were related to the subsystem of medical demand 533

by virtue of only two variables—namely, the number of patients 534

per year and the number of visits. The number of patients per year 535

will influence the number of visits to hospitals and CHSs, through 536

three main factors—namely, accessibility to hospitals and CHSs, 537

technology level of hospitals and CHSs, and outpatient financing 538

of hospitals and CHSs. In turn, the number of visits to hospitals 539

and CHSs will ultimately determine the number of visits and influ- 540

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

W. Yu et al. /Journal of Biomedical Informatics xxx (2015) xxx-xxx 7

initial input to hospital by government

hospitalization market income of hospital>

<average outpatient price of hospital>

input to hospital by government

average ratio of input by hospital per year

fixed depreciation rate

increment of hospital resource

outpatient market income of hospital

number of visits to hospital

depreciation of hospital resource

average real estate of hospital

visiting rate of hospital i

factor of outpatient economy of hospital

technology level of hospital

accessibility to hospital

chart of outpatient economy of hospital

ratio of average real estate of hospital and average value of general real estate

chart of accessibility to hospital

density of hospital resource

initial number of hospital

<total population>

factor of outpatient payment capability of hospital

initial average outpatient price of hospital

proportion of health expenditure in consumption expenditure

initial individual income

initial health insurance expense per capita

growth rate of income

Fig. 4. Subsystem of outpatients in hospitals.

<lnme>

initial input t< community

hospitalization market income of CHS>

<Time> /

input to CHS by government

average ratio of input by CHS per year

outpatient market income of CHS

number of visits to CHS

<amount of patients per year>

total income of CHS

<fixed depreciation rate>

depreciation of CHS resource

<average outpatient chart of technology pnce of CHS> .eve, of CHS

ratio of average real estate of CHS and average value of general real estate

<average value of general real estate>

average real estate of CHS

technology level of CHS

density of CHS resource

number of

<total population>

initial number of CHS

change rate of CHS

visiting rate of CHS ^

accessibility to CHS

factor of outpatient economy of CHS

chart of accessibility to CHS

factor of outpatient payment capability of CHS

initial average outpatient price of CHS

average outpatient price of CHS

chart of outpatient economy of CHS

<individual payment of

medical expense>

<factor of outpatient>

<health insurance expense per capita>

Fig. 5. Subsystem of outpatients in CHSs.

541 ence the actual percentage of patients not seeking medical care. All

542 three factors affect the number of visits to hospitals and CHSs by

543 influencing the visiting rate to hospitals and CHSs; this has already

been demonstrated [14]. The drivers of seeking out medical care 544 among the people of China mainly include close location (56.0%), 545 good-quality services (16.0%), trust in doctors (9.0%), reasonable 546

<lime>

<lime>

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

W. Yu et al./Journal of Biomedical Informatics xxx (2015) xxx-xxx

medical expenses (4.9%), specific health institutions (3.7%), facilities being well equipped with devices and materials (3.6%), patients having relatives in health institutions (3.0%), facility personnel providing service in a good manner (1.2%), a sufficient amount of drugs (0.7%), and other reasons (1.8%) [14]. Through the use of brainstorming, we chose three such influencing factors: accessibility to hospitals and CHSs (in accordance with close location), technology level of hospitals and CHSs (in accordance with good-quality services, trust in doctors, specific health institutions, being well equipped with devices and materials, having relatives in health institutions, and sufficient drugs), and the factor of outpatient financing of hospitals and CHSs (in accordance with reasonable medical expenses). Accessibility to hospitals and CHSs can be estimated by using the accessibility chart, which is generated by considering the density of resources—something determined through field research and expert consultation [83,67]. The technology level of hospitals and CHSs can be estimated by using the technology-level chart, which is generated by considering the average real estate, average value of general real estate, number of hospitals and CHSs, the amount of resources, and the total income of hospitals and CHSs; the total income consists of input from the government [46] and outpatient and inpatient market income [16]. The factor of outpatient financing of hospitals and CHSs can be estimated by using the chart of outpatient financing and the factor of outpatient payment capability—of which, the latter is jointly determined by considering individual payment of medical expense, health insurance expense per capita [13], and average outpatient price [29]. The average outpatient price is related to average individual income (mainly influenced by the growth rate of income) [3] and the proportion of health expenditure in consumption expenditure.

We also analyzed the variables in the subsystems of inpatients in hospitals and CHSs (Fig. 6). Because the focus of this study is the subsystem of outpatients and its goal is to simplify the model, the two inpatient subsystems play a role only within their own circle of economic influence. The inpatient subsystem relates to the outpatient subsystem mainly through the variable ''hospitalization market income," which directly influences the total income and indirectly impacts the amount of resources, the technology level of hospitals and CHSs, the visiting rate of outpatients, and number of visits to hospitals and CHSs; this ultimately changes the actual percentage of patients who do not seek medical care. In addition, the hospitalization market income is decided by the number of hospitalizations and the average hospitalization price. The number of hospitalizations is in turn determined by considering the total population and the hospitalization rate. The hospitalization rate itself is similar to the visiting rate and is influenced by accessibility to hospitals and CHSs, the technology level of hospitals and CHSs, and the factor of inpatient financing of hospitals and CHSs. The average hospitalization price, together with the individual payment of medical expenses and the health insurance expense per capita, further influences the hospitalization rate by affecting the factor of hospitalization payment capability.

The following are some explanations for the aforementioned variables, as well as equations that reflect the relationships among these variables.

First, the initial values of the variables and constant variables in the model were determined using data in the Chinese Health Statistical Yearbook 2012 [14] and An Analysis Report of National Health Services Survey in China for 2008 [11]. The initial value for the total population, as per these sources, was 1,328,020,000; the initial average outpatient price for hospitals and CHSs was 138.3 RMB and 84 RMB, respectively. The initial average hospitalization price for hospitals and CHSs was 5234.1 RMB, and 2514.2 RMB, respectively. The initial number of hospitals was 21,979, and the initial number of CHSs was 32,860. The initial input to hospital from

the government was 10,127,700 RMB, and its initial input to CHSs was 1,921,950 RMB. The initial amount of hospital resources was 7590 / 21,979 CNY, and that of CHSs was 336 / 32,860 CNY. The initial individual income was 3437 RMB, and the initial health insurance expense per capita was 103.95 RMB. Meanwhile, the birth and death rates were 0.01214 and 0.00706, respectively. The average ratio of input by hospitals and by CHSs per year was 0.06 and 0.04, respectively. The fixed depreciation rate was 0.03, and the growth rate of income was 0.108. Finally, the proportion of health expenditure in consumption expenditure was 0.108.

Second, on the basis of these initial variables, we can construct some level variables. For example, in 2012, China's initial value for the total population was 1,328,020,000; taking that value and the constant variables (i.e., birth and death rates of 0.01214 and 0.00706, respectively), the equation by which to determine the total population is: initial total population + number of births - number of deaths. Additionally, the initial number of hospitals and CHSs were, respectively, 21,979 and 32,860, and the amounts of hospital and CHS resources were 75.9 and 3.36 million CNY, respectively; therefore, the initial amount of hospital and CHS resources was 7590 / 21,979 CNY and 336 / 32,860 CNY, respectively; the equations by which to calculate the amount of hospital and CHS resources are: amount of hospital resources = initial amount of hospital resources + increment of hospital resources - depreciation of hospital resources, and amount of CHS resources = initial amount of CHS resources + increment of CHS resources - depreciation of CHS resources.

Third, by combining these data with the initial variables, constant variables, and various change rates, we can construct some equations that refer to auxiliary variables. These functions take the general form of ''value of initial/constant variable / EXP (Time / change rate of the variable)"; the change rate of the variable was the average change rate over several years, and it was calculated by examining data from the Chinese Health Statistical Yearbook (2010-2013) and An Analysis Report of the National Health Services Survey in China (1993-2008). Thus, these formulations were determined as follows:

Morbidity rate per year = 0.1886 / 26 / EXP (Time / 0.005) Average outpatient price of hospitals = initial average outpatient price of hospitals / EXP (Time / 0.0959)

Average outpatient price of CHSs = initial average outpatient price of CHSs / EXP (Time / 0.017535)

Average hospitalization price of hospitals = initial average hospitalization price of hospitals / EXP (Time / 0.088046)

Average hospitalization price of CHSs = initial average hospitalization price of CHSs / EXP (Time / -0.01368)

Input to hospital from the government = initial input to hospital from the government / EXP (Time / 0.08)

Input to CHSs from the government = initial input to CHSs from the government / EXP (Time / 0.08)

Number of hospitals = initial number of hospitals / EXP (Time / 0.0084)

Number of CHSs = initial number of CHSs / EXP (Time / change rate of CHSs)

Average individual income = initial individual income / EXP (Time / growth rate of income)

Health insurance expense per capita = initial amount of insurance per person / EXP (Time / 0.08)

Average value of general real estate = 3000 / EXP (Time / 0.08) Fourth, to address some uncertain factors and the complexity of some influential mechanisms, we introduced the brainstorming method, the Delphi method (involving 15 experts), and a questionnaire survey (of 12,000 subjects, and more than 80,000 outpatient records) to determine the influencing factors and the relationships among them. Thus, the main influencing factors of the visiting rate and the hospitalization rate are accessibility to medical services,

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

W. Yu et al./Journal of Biomedical Informatics xxx (2015) xxx-xxx 9

initial average hospitalization price of hospital

<Time>

average hospitalization price of hospital

hospitalization

<individual payment of

hospitalization market income of hospital

hospitalization visits of hospital

chart of hospitalization economy of hospital

medical expense> w \ /

factor of hospitalization payment capability of hospital

factor of hospitalization economy of hospital

<health insurance expense per capita>

hospitalization rate of hospital

<total population>

hospitalization visits of CHS

hospitalization rati of CHS

hospitalization market income of CHS

<accessibility to hospital> <technology level of hospital>

<accessibility to CHS> <technology level of CHS>

factor of hospitalization economy of CHS

average hospitalization price of CHS /

initial average hospitalization price of CHS

chart of hospitalization economy of CHS factor of hospitalization payment capability of CHS

<individual payment of medical expense>

<factor of hospitalization>

<health insurance expense per capita> Fig. 6. Subsystems of inpatients in hospitals and CHSs.

factor of

<Time>

technology level, and factor of the outpatient economy. Meanwhile, the ratios of the weight of accessibility to medical service, technology level, and factor of the outpatient economy of outpatients in hospitals were 0.2, 0.6, and 0.2; those of inpatients in hospitals were 0.1, 0.5, and 0.4; those of outpatients in CHSs were 0.6, 0.1, and 0.3; and those of inpatients in CHSs were 0.3, 0.3, and 0.4. Then, according to the prototype of the Classic Douglas function, the power function equations can be constructed thus:

Visiting rate to hospitals = (accessibility to hospitals + 10-9)02 / (technology level of hospitals + 10-9)06 / (factor of the outpatient economy of hospitals + 10-9)02

Visiting rate to CHSs = (accessibility to CHSs + 10-9)06 / (technology level of CHSs + 10-9)01 / (factor of the outpatient economy of CHSs + 10-9)03

Hospitalization rate of hospitals = (accessibility to hospital + 10-9)02 / (technology level of hospital + 10-9)06 / (factor of the outpatient economy of hospitals + 10-9)02

Hospitalization rate of CHSs = (accessibility to CHSs + 10-9)06 / (technology level of CHSs + 10-9)03 / (factor of the outpatient economy of CHSs + 10-9)01

Fifth, we worked to explain table functions. While developing the model, we classified some nonlinear variables as various influencing factors; according to the results of field surveys, they can be expressed only as their changing tendency, rather than as mathematical functions. Thus, in line with SD modeling theory, we introduced table functions to determine such relations. The establishment of table functions was based on the description of the current tendency and expert consultation, and the determination of table functions was in line with a dozen or dozens of datasets. Variable values that cannot be determined by examining the survey data can be estimated by considering certain endpoints, intermediate points, and the slope. In this study, table functions were mainly employed to explain the influences of accessibility to medical services, technology level, and economic factors on patients' choices of visiting a health care facility and being

hospitalized. Data in these table functions were all derived from the survey results. In the table function, the meaning of each part of the function can be explained. In the first squared bracket behind the equal sign, the numerical values in the first rounded bracket represent the minimum values of x and y; the numerical values in the second rounded bracket represent the maximum values of x and y. The other rounded brackets following the first squared bracket contain every pairing of the numerical values of x and y. The table functions were shown in Appendix A.

2.5. Model validation

To validate the model, we set the initial value of variables to those of 2008 and simulated the model for the next four years (2009-2012). This would help us demonstrate whether the model aligns with the historical evolution of China's health delivery system. The validation results are shown in Table 2. The actual data were taken from historical records from the Chinese Health Statistical Yearbook [14], and the simulated data were the results of validation analysis. One can see that the deviation between the actual and simulated data was not significant—namely, between -1.73% and 9.16%, which is within a reasonable range. Thus, the model is considered validated and reasonable.

2.6. Sensitivity analysis

The aim of sensitivity analysis is to generate and compare outputs by changing the model parameters and construction, and thus determine their influence and assess the validity and reliability of the given model. To analyze parameter sensitivity, one should first pinpoint the sensitive parameters and restrict them to a reasonable range, to observe responses to variations in the model. Parameter sensitivity can be calculated in accordance with DY(t)/Y(t)—which represents the change in the output variable (i.e., in this study, the

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

W. Yu et al./Journal of Biomedical Informatics xxx (2015) xxx-xxx

Table 2

Validation analysis of the model.

Year 2009 2010 2011 2012

Number of visits to Actual data 1.922 2.040 2.259 2.542

hospital (billion)

Simulated data 2.098 2.16 2.22 2.587

(billion)

Deviation 9.16% 5.88% -1.73% 1.77%

actual percentage of patients not seeking medical care)—and AX(t)/X(t)—which represents the changing variable. Thus, the

formulation of sensitivity can be expressed as S(t) = |AY(tj/x(tj|.

In the current study, we chose four sensitive parameters with variation rates between -10% and 10%, to simulate changes in the actual percentage of patients not seeking medical care. The four variables chosen for sensitivity analysis were birth rate, death rate, proportion of health expenditure in consumption expenditure, and growth rate of income. All four variables were designed to concurrently change by between -10% and 10% of their current values, thus making the model operate 200 times in total; this procedure was conducted by leveraging the sensitivity simulation function in Vensim. The results of this sensitivity analysis are presented in Fig. 7. In so-called sensitivity analysis, making changes to one variable will cause some slight variations, but it will not lead to changes in the overall tendency. In light of this, in looking at Fig. 3, we can see that among 200 cases, 50% of them are in the yellow area, 75% are in the green area, 95% are in the blue area, and 100% are in the gray area. It can be seen that the total tendency of the actual percentage of patients not seeking medical care continued to decrease; only the numerical values showed some slight and reasonable changes. In analyzing the reasons behind these changes, we know that the birth and death rates will affect the total population, which will further influence the number of patients per year; in addition, the increase in health expenditure as a proportion of consumption expenditure, along with the growth rate of income, will increase the economic factor and hence further increase the visiting rate of patients. All these will ultimately cause fluctuations in the actual percentage of patients not seeking medical care. This result indicates that with changes to the constant variables, variations in the actual percentage of patients not seeking medical care will fall within a reasonable range. The output variable—which mainly represents the behavior of the model—did not show any extreme change as a result of slight changes to the constant variables. Therefore, the model is stable and reliable, and is considered suitable for simulation analysis.

2.7. Intervention scenarios

Upon undertaking sensitivity analysis and several simulations, we pinpointed the three variables that played the most significant role in influencing the actual percentage of patients not seeking medical care. The three constant parameters—namely, the number of hospitals and CHSs, the outpatient prices of hospitals and CHSs, and health insurance—were selected for use in the intervention trials. In the experiments detailed below, we reduced the number of hospitals and increased the number of CHSs to various extents, reduced the outpatient prices of hospitals and CHSs, and adjusted the standard of health insurance to various levels. The current situation was defined as the baseline, and four groups of experiments were conducted. The first group explored the effect of the number of hospitals and CHSs on the actual percentage of patients not seeking medical care (i.e., tests 1, 2, and 3). The second group explored the effect of the outpatient prices of hospitals and CHSs on the actual percentage of patients not seeking medical care

Fig. 7. Sensitivity analysis of constant parameters.

(i.e., tests 4, 5, and 6). The third group explored the effect of health insurance on the actual percentage of patients not seeking medical care (i.e., tests 7, 8, and 9). Finally, the fourth group of experiments explored the joint effect of the number of hospitals and CHSs and of health insurance on the actual percentage of patients not seeking medical care (i.e., tests 10, 11, and 12). The system simulation was set to a 20-year period. We compared the various experiment results to the baseline, to assess whether these changes would successfully transform PMD. Changes to the parameters were made as shown in Table 3.

3. Results

The model in this study was used in experiments that simulated policy interventions, all of which would have effects on medical demand and supply by virtue of changes to some of the parameters. In the intervention trials, by changing some key parameters, we could see and analyze changes to the effects on the actual percentage of patients not seeking medical care. The experiment results were compared to the current situation (Figs. 8-11), in which no intervention can be seen as the baseline (line 1 in each figure) through the use of any influencing factors. Therefore, noticeable changes in these intervention trials can be attributed strictly to the interventions in question. In addition, in line with the natural development of hospitals and CHSs, the economic growth of the whole of society, and the proliferation of supportive policies established by the Chinese government, the accessibility to medical care, technology level of medical providers, and economic factors of hospitals and CHSs will be promoted. Thus, even in the absence of any intervention, the baseline percentage of patients not seeking medical care will show a decreasing tendency. Based on this premise, this study mainly concentrated on how to achieve a low actual percentage of patients not seeking medical care, as quickly as possible, rather than achieve a much lower percentage. The specific results of the four intervention trials are shown in Figs. 8-11.

3.1. Effect of number of hospitals and CHSs on the actual percentage of patients not seeking medical care

In this experiment, only the number of hospitals and CHSs was changed; the other variables remained fixed. Fig. 8 shows the effect of the number of hospitals and CHSs on the actual percentage of patients not seeking medical care. As one can see from Fig. 4, although the actual percentage of patients not seeking medical care continued to decrease in the 20-year simulation under the status quo and under all three simulation situations, in terms of PMD transformation, the results of the three tests were more favorable

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

W. Yu et al./Journal of Biomedical Informatics xxx (2015) xxx-xxx

Table 3

Changes to the parameters of different intervention scenarios.

Group Test

Scenario

Parameter

Change 1 Value 1 Change 2 Value 2 Change 3 Value 3

Baseline Change the number of hospitals and CHSs Number of 21,979 Number of 32,860 _ -

hospitals CHSs

1 Decrease by 16,484 Increase by 41,075 - -

25% 25%

2 Decrease by 10,990 Increase by 49,290 _ -

50% 50%

3 Decrease by 5495 Increase by 57,505 _ -

75% 75%

Baseline Change the outpatient prices of hospitals and CHSs Prices of 138.3 Prices of 84 _ -

hospitals CHSs

4 Decrease by 103.73 Decrease by 63 - -

25% 25%

5 Decrease by 69.15 Decrease by 42 _ -

50% 50%

6 Decrease by 34.58 Decrease by 21 _ -

75% 75%

Baseline Change the health insurance expense per capita Expense per 103.95 И- - _ -

capita

7 Increased by 155.93 - - _ -

8 Increased by 207.9 - - _ -

9 Increased by 311.85 - - _ -

Baseline Change the number of hospitals and CHSs, and health insurance Number of 21,979 Number of 32,860 Expense per 103.95

expense per capita hospitals CHSs capita

10 Decrease by 16,484 Decrease by 41,075 Increased by 155.93

25% 25% 50%

11 Decrease by 10,990 Decrease by 49,290 Increased by 207.9

50% 50% 100%

12 Decrease by 5495 Decrease by 57,505 Increased by 311.85

75% 75% 200%

The actual percentage of patients not seeking medical care

0 2 4 6 8 10 12 14 16 18 20 Time (Year)

The actual percentage: current —1-1-1-1-1-1-1-f-

The actual percentage: testl -2-2-2-2-2-2-2-

The actual percentage: tcst2 -3-9-3-3-3-9-3-9-

The actual percentage: test3 —4-4-4-4-4-4-4-4—

Fig. 8. The variation trend of the actual percentage of patients not seeking medical care after the number of hospitals and CHSs was changed. The estimated actual percentage of patients not seeking medical care when adjusting the amount of hospitals and CHSs would change in 20 years.

842 than the current situation. With a reduction in the number of hos-

843 pitals and an increase in the number of CHSs, the rate at which the

844 actual percentage of patients not seeking medical care fell was

845 much higher than that seen in the current situation. In addition,

846 the ultimate percentage of patients not seeking medical care in

847 each of the three scenarios was lower than that in the current sit-

848 uation; this indicated that changes to the number of hospitals and

849 CHSs indeed worked to lower PMD. In comparing the results of the

850 three experiments, although the decreasing rate seen in test 3 was

851 obviously higher than that in test 2 or 1 in the short term, the total

852 decreasing range of the three tests were all similar. These results

853 suggested that cutting the number of hospitals and increasing

The actual percentage of patients not seeking medical care

0 2 4 6 8 10 12 14 16 18 20 Time (Year)

The actual percentage: current —1-1-1-1-1-1-1-1~

The actual percentage: test 1 -5-2-2-2-i-2-2-

The actual percentage: test2 -3-9-9-9-9-9-9-9-

The actual percentage: test3 —4-4-4-4-4-4-4-4—

Fig. 9. The variation trend of the actual percentage of patients not seeking medical care after the outpatient prices of hospitals and CHSs were changed. The estimated actual percentage of patients not seeking medical care when adjusting the outpatient prices of hospitals and CHSs would change in 20 years.

the number of CHSs would have a significant and beneficial influ- 854

ence on the actual percentage of patients not seeking medical care, 855 and thus lead to a larger, more effective, and faster decline in PMD 856

in both the short and long term. 857

3.2. Effect of outpatient prices of hospitals and CHSs on the actual 858

percentage of patients not seeking medical care 859

Assuming that the outpatient prices of hospitals and CHSs are 860

138.3 RMB and 84 RMB, respectively, we can derive the effect of 861

the outpatient prices of hospitals and CHSs on the actual percent- 862

age of patients not seeking medical care (Fig. 9); the other variables 863

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

W. Yu et al./Journal of Biomedical Informatics xxx (2015) xxx-xxx

Fig. 10. The variation trend of the actual percentage of patients not seeking medical care after health insurance was changed. The estimated actual percentage of patients not seeking medical care when adjusting health insurance would change in 20 years.

insurance would indeed have effects on the actual percentage of patients not seeking medical care. Therefore, this experiment combined these two influencing factors to explore changes to the actual percentage of patients not seeking medical care; the results are shown in Fig. 11, where one can see that the basic tendency is similar to that in the experiments featuring changes to the number of hospitals and CHSs. All the scenarios, including the current situation, showed a decreasing trend, but the rate at which the decrease was happening was much higher in the simulations than in the current situation, and the ultimate percentage in the simulations was lower than that in the current situation. The lower the number of hospitals is, the greater the number of CHSs would be, and the higher the level of health insurance is, the lower the percentage of patients not seeking medical demand in the long term would be, and the more quickly it would decrease in the short term. In addition, owing to the influence of health insurance, the changing range of the integrated plan was greater than that seen in the single-adjustment experiments.

The actual percentage of patients not seeking medical care

f— — — — — — — —

0 2 4 6 8 10 12 14 16 18 20 Tine (Year)

The actual percentage: current —1-1-1-1-1-1-1-1

The actual percentage: tcstl -2-2-2-it-2-2-2-

The actual percentage: tcst2 -3-3-3-3-3-3-3-3-

The actual percentage: test3 —--4-4-4-4-4-4-4—

Fig. 11. The variation trend of the actual percentage of patients not seeking medical care after number of hospitals and CHSs and health insurance was changed. The estimated actual percentage of patients not seeking medical care when adjusting number of hospitals and CHSs and health insurance would change in 20 years.

remained fixed. In the three experiments, the actual percentage of patients not seeking medical care all showed a slight declining trend in the first five years, but it lost its influence in the long term. However, during the 5-20 years of the simulation, the percentage across the three experiments was even higher than that in the current situation. These results suggested that changes to the outpatient prices of hospitals and CHSs would not play any role in reducing the actual percentage of patients not seeking medical care, and it would not increase the transformation of PMD, either.

3.3. Effect of health insurance on the actual percentage of patients not seeking medical care

The effect of health insurance on the actual percentage of patients not seeking medical care is shown in Fig. 10; the other variables remained fixed. As one can see from the figure, compared to the current situation (line 1), the higher the level of health insurance is, the lower the percentage of patients not seeking medical care would be. However, the changes were slight and smooth, thus indicating that the effect of health insurance was not apparent.

3.4. Effect of number of hospitals and CHSs and health insurance on the actual percentage of patients not seeking medical care

Considering the aforementioned experiment results, regulating the number of hospitals and CHSs and adjusting the level of health

4. Discussion and conclusions

Given the actual situation in China, the current health delivery system is out of balance and there are some limitations in reducing the PMD. At the moment, policies place more emphasis on hospitals and overlook the establishment and development of CHSs. Health insurance levels are still low, and individual-level health expenditures are too high; together, these circumstances give rise to a high proportion of PMD. In addition, as shown in the results of the aforementioned experiments, the current actual percentage of patients not seeking medical care is still too high. Adjusting the current health delivery system would play a positive role in reducing the actual percentage of patients not seeking medical care; this indicates that it is urgent to advance the transformation of PMD.

As can be seen from the simulation results, the influence of changing both the number of hospitals and CHSs and the level of health insurance was greatest in reducing the actual percentage of patients not seeking medical care. On the other hand, the single adjustment of the number of hospitals and CHSs played a relative large role; the single adjustment of health insurance coverage caused a slight and smooth change; and changes in the outpatient prices of hospitals and CHSs played no role in reducing the actual percentage of patients not seeking medical care. Consequently, the experiment results suggested that adjusting the structure of the number of hospitals and CHSs and improving the health insurance level could together play a role in advancing PMD transformation. These results indicated that China's current health delivery system requires reasonable adjustments to health resource allocations between hospitals and CHS, that more attention needs to be paid to CHSs, and a CHS-centered health delivery system needs to be established—all of which, taken together, would effectively reduce the medical costs of patients, increase the availability of medical services, reduce the actual percentage of patients who require medical care but do not seek it out, and alleviate the continuous growth of PMD. Ultimately, these changes would benefit the whole of the population of China.

The high level of PMD can be partially explained by the relative weakness of CHSs, unbalanced health resource allocations, and a low-level economic compensation system. If such trends continue, hospitals could become stronger and even more prevalent, thus driving medical expenses higher and exacerbating PMD. To address this problem, several measures can be taken, such as changing unjustified resource allocations [7], including changes to health care-oriented human resource allocations [6] and health budget allocations between hospitals and CHSs; providing more support to CHSs, including the provision of sufficient medical materials and

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

W. Yu et al./Journal of Biomedical Informatics xxx (2015) xxx-xxx

equipment [22], increasing the number and scale of CHSs, investing enough funds in the development and implementation of CHSs, and increasing the transparency vis-à-vis the fund allocation process [66]; and improving health insurance policies, including enlarging the coverage of health insurance, especially among the underprivileged population [47], raising the level of health insurance, providing a safe net for people (especially for those living in rural areas), and reducing the out-of-pocket health expenditures of households [86]. Since undertaking medical and health system reforms in 2009, China has made many achievements; for example, it has improved the coverage and compensation standards of health insurance, more than 33,000 basic medical and health institutions have been enhanced, and the service capability of CHSs has been improved. Nonetheless, more needs to be done to accelerate the transformation of PMD. All the aforementioned suggestions look to change the resource structure between hospitals and CHSs, enhance the health service capacity of CHSs, and improve the insurance of China's entire population, by transforming more PMD and ensuring that the basic health needs of citizens are met.

Use of the SD model enabled us to estimate the amount of PMD in the future, as well as trends thereof; its use allowed us to evaluate the influence of current policies on the basis of limited historical data. Generally speaking, the SD model is a methodological technique by which one can explore complicated scientific problems, and its use makes it possible to generate a more scientific and persuasive method of analyzing management problems in complex socioeconomic systems. Compared to some mathematical simulation models such as Monte Carlo—although both Monte Carlo and the SD model are appropriate for use with complicated systems—the SD model can simulate changing trends over a continuous period. What's more, the SD model overcomes several difficulties: it can use limited historical data to construct models, anticipate future development and trends with respect to some complex issues, and simulate the effectiveness of policies.

To the best of our knowledge, this study is the first to use an SD model to qualitatively analyze China's PMD-transformation system. This model was used to analyze various factors that influence PMD transformation in China; doing so allowed us to explore relationships between hospitals and CHSs, and it made it possible to calculate and estimate—solely on the basis of several limited historical data—the actual percentage of patients not seeking medical care in a changing environment. Furthermore, the model of ''PMD transformation" can be used to guide policy-making, and such guiding information can be garnered simply by changing some variables and simulating various scenarios. By using this model, one can answer questions about China's high level of PMD, and thereby deal with the inherent unfairness of the country's health delivery system in the long term. In essence, the use of this model can help China's policy-makers formulate a more efficient health care reform project.

In the first attempt to construct a SD model and address the issue of ''PMD transformation" while undertaking some intervention-oriented experiments, it became apparent that some limitations necessitate further research in the future. At the moment, this model can deal with various factors and variables, such as visiting rate to hospitals and CHSs, the hospitalization rate of hospitals and CHSs, population growth, changes to investment from the government, the health characteristics of the population, health insurance expenses, and changes in medical services demand. However, we must concede that an important limitation of SD models is their inability to provide estimates of uncertainty; to overcome this limitation and extend the current research, we would like in future studies to combine it with other methods. In addition, several assumptions and relationships inherent in the model have been simplified; for example, we looked at the population as a whole, and did not construct and simulate a model based

on different populations in urban and rural areas. Some influencing factors in the model have been condensed. And the current model does not take into account certain factors and elements, such as the health supervision system, individual characteristics, and preference, nor does it consider in intervention trials the specific ratio of health resources between hospitals and CHSs. All these deficiencies and problems can be addressed, and the established model can be further enhanced by making it more reasonable and practical.

In conclusion, the problems inherent in China's ''PMD-transformation system" are mainly caused by the country's defective health delivery system. Based on China's current systematic structure, health resource allocation structure, compensation and financing system, and health insurance system, the Chinese government should make great efforts to promote the internal power and structure of health delivery system mainly while concentrating on the reallocation of health resources between hospitals and CHSs and improving health insurance coverage; doing so would provide health institutions with a worthwhile and appropriate direction for development.

Conflict of interest

All authors declared that there are no conflicts of interest associated with this work.

Acknowledgments

The project was supported by National Natural Science Foundation of China (71233008, 71303248, 91224005), and Young Scientist Foundation of The Second Military Medical University (2012QN09).

Appendix A. Mathematical equations used by the model

Total population = INTEG (+number of births - number of deaths, 1,328,020,000)

Number of births = total population / birth rate Number of deaths = total population / death rate Birth rate = 0.01214 Death rate = 0.00706

Morbidity rate per year = 0.1886 / 26 / EXP (Time / 0.005) Number of patients per year = morbidity rate per year / total population

The actual percentage of patients not seeking medical care = number of untreated patients/number of visits per year

Number of untreated patients = number of patients per year-number of visits

Number of visits = number of visits to hospitals + number of visits to CHSs

Number of visits to hospitals = number of patients per year / visiting rate of hospitals

Outpatient market income of hospitals = number of visits to hospitals / average outpatient price of hospitals

Total income of hospitals = input to hospitals by government + outpatient market income of hospitals + hospitalization market income of hospitals

Input to hospitals from the government = initial input to hospitals from the government / EXP (Time / 0.08)

Initial input to hospitals from the government = 10,127,700 Hospitalization market income of hospitals = number of hospi-talizations to hospitals / average hospitalization price of hospitals Amount of hospital resources = INTEG (+increment of hospital resources - depreciation of hospital resources, 7590 / 21,979)

Increment of hospital resources = total income of hospitals / average ratio of input from hospitals per year

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

W. Yu et al./Journal of Biomedical Informatics xxx (2015) xxx-xxx

1080 1081 1082

1100 1101 1102

1110 1111 1112

1120 1121 1122

Depreciation of hospital resources = fixed depreciation rate / amount of hospital resources

Average ratio of input by hospitals per year = 0.06 Fixed depreciation rate = 0.03

Average real estate of hospitals = amount of hospital resources/ number of hospitals

Ratio of average real estate of hospitals and average value of general real estate = average real estate of hospitals/average value of general real estate

Average value of general real estate = 3000 / EXP (Time / 0.08) Chart of technology level of hospitals = [(0,0)-(10,000,1)], (0,0.156), (117.25,0.168), (232.75,0.234), (712.8,0.2937), (1167.25,0.336), (2051,0.4755), (2740.5,0.4898), (3150,0.528), (4320,0.651), (5763,0.727), (6428,0.7835), (7379,0.8402), (8462,0.933), (10,000,1)

Technology level of hospitals = chart of technology level of hospitals (''average real estate of hospitals/average value of general real estate")

Chart of accessibility to hospitals = [(0,0)-(0.3,1)], (0,0.05824), (0.0972477,0.06842), (0.2,0.18), (0.252294,0.52632), (0.273394,0.75264), (0.3,1)

Accessibility to hospitals = chart of accessibility to hospitals (density of hospital resources)

Number of hospitals = initial number of hospitals / EXP (Time / 0.0084)

Initial number of hospitals = 21,979

Density of hospital resources = number of hospitals/total population

Average outpatient price of hospitals = initial average outpatient price of hospitals / EXP (Time / 0.0959)

Initial average outpatient price of hospitals = 138.3 Factor of outpatient payment capability of hospitals = (individual payment of medical expense + health insurance expense per capita)/average outpatient price of hospitals

Factor of outpatient financing of hospitals = chart of outpatient financing of hospitals (factor of outpatient payment capability of hospitals)

Proportion of health expenditure in consumption expenditure = 0.108

Average individual income = initial individual income / EXP (growth rate of income / Time) Growth rate of income = 0.108 Initial individual income = 3437

Individual payment of medical expense = average individual income / proportion of health expenditure in consumption expenditure

Initial health insurance expense per capita = 103.95 Health insurance expense per capita = initial amount of insurance per person / EXP (Time / 0.08)

Chart of outpatient financing of hospitals = [(0,0)-(60,1)], (0.0244648,0.05386), (1.10092,0.0964912), (2.81346,0.7021), (3.60856,0.73635), (5.01529,0.74872), (6.75841,0.79559), (8.40979,0.82807), (9.90826,0.83682), (15,0.83946),

(30,0.99568), (45,1)

Visiting rate of hospitals = (accessibility to hospitals + 10-9)02 / (technology level of hospitals + 10-9)06 / (factor of outpatient financing of hospitals + 10-9)02

Number of visits to CHSs = number of patients per year / visiting rate of CHSs

Outpatient market income of CHSs = average outpatient price of CHSs / number of visits to CHSs

Total income of CHSs = input to CHS from the government + hospitalization market income of CHSs + outpatient market income of CHSs

Input to CHSs from the government = initial input to CHSs from the government / EXP (Time / 0.08)

Initial input to CHSs from the government = 1,921,950 1140

Amount of CHSs resources = INTEG (+increment of CHSs 1141

resources - depreciation of CHSs resources, 336 / 32,860) 1142

Increment of CHSs resources = total income of CHSs / average 1143

ratio of input from CHSs per year 1144

Depreciation of CHSs resources = fixed depreciation rate / 1145

amount of CHSs resources 1146

Average ratio of input from CHSs per year = 0.04 1147

Average real estate of CHSs = amount of CHSs resources/number 1148

of CHSs 1149

Ratio of average real estate of CHSs and average value of general 1150

real estate = average real estate of CHSs/average value of general 1151

real estate 1152

Chart of technology level of CHSs = [(0,0)-(1000,1)], (0,0.3377), 1153

(100.3,0.3575), (122.3,0.3837), (238.6,0.4605), (345.6,0.579), 1154

(474,0.628), (620.8,0.73627), (752.2,0.7982), (896,0.8197), 1155

(972.4,0.9708), (1000,1) 1156

Technology level of CHSs = chart of technology level of CHSs = 1157

(''average real estate of CHSs/average value of general real estate") 1158

Number of CHSs = initial number of CHSs / EXP (Time / change 1159

rate of CHSs) 1160

Initial number of CHSs = 32,860 1161

Chart of accessibility to CHSs = [(0,0)-(10,1)], (0,0.2724), 1162

(2.47706,0.3181), (3.48624,0.4106), (4.95413,0.5198), 1163

(6.20795,0.7829), (7.52294,0.8781), (8.92966,0.9316), (10,1) 1164

Density of CHSs resources = number of CHSs/total population 1165

Accessibility to CHSs = chart of accessibility to CHSs (density of 1166

CHSs resources) 1167

Average outpatient price of CHSs = initial average outpatient 1168

price of CHSs / EXP (Time / 0.017535) 1169

Initial average outpatient price of CHSs = 84 1170

Factor of outpatient payment capability of CHSs = (individual 1171

payment of medical expense + health insurance expense per cap- 1172

ita)/average outpatient price of CHSs 1173

Chart of outpatient financing of CHSs = [(0,0)-(400,1)], 1174

(0.0611621,0.09886), (2.07951,0.1453), (3.42508,0.175), (5.3211, 1175

0.2548), (6.78899,0.4052), (7.76758,0.4943), (8.74618,0.6017), 1176

(9.48012,0.6724), (11.2538,0.6785), (15.107,0.6847), (18.3486, 1177

0.723684), (18.6544,0.7239), (19.8777,0.7345), (23.5474, 1178

0.79386), (28.7462,0.820175), (33.945,0.837719), (39.1437, 1179

0.929825), (46.789,0.942982), (54.7401,0.973684), (62.0795, 1180

0.97807), (70,1), (355,1) 1181

Factor of outpatient financing of CHSs = chart of outpatient 1182

financing of CHSs (factor of outpatient payment capability of CHSs) 1183

Visiting rate of CHSs = (accessibility to CHSs + 10-9)0 6 / (tech- 1184

nology level of CHSs + 10-9)01 / (factor of outpatient financing of 1185

CHSs + 10-9)03 1186

Hospitalization visits of hospitals = total population / hospital- 1187

ization rate of hospitals 1188

Hospitalization rate of hospitals = (accessibility to hospitals + 1189

10-9)01 / (technology level of hospitals + 10-9)0 5 / (factor of 1190

hospitalization financing of hospitals + 10-9)0 4 1191

Factor of inpatient financing of hospitals = chart of inpatient 1192

financing of hospitals (factor of inpatient payment capability of 1193

hospitals) 1194

Chart of inpatient financing of hospitals = [(0,0)-(4,1)], (0,0), 1195

(0.718654,0.2527), (0.990826,0.42), (1.40673,0.5747), 1196

(1.90826,0.6316), (2.25076,0.741228), (3.24159,0.864035), (4,1), 1197

(4.05199,0.964912) 1198

Factor of inpatient payment capability of hospitals = (individual 1199

payment of medical expense + health insurance expense per cap- 1200

ita)/average hospitalization price of hospitals 1201

Average hospitalization price of hospitals = initial average hos- 1202

pitalization price of hospitals / EXP (Time / 0.088046) 1203

Initial average hospitalization price of hospitals = 5234.1 1204

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

W. Yu et al./Journal of Biomedical Informatics xxx (2015) xxx-xxx

Hospitalization visits of CHSs = total population / hospitaliza-tion rate of CHSs

Hospitalization market income of CHSs = hospitalization visits of CHSs / average hospitalization price of CHSs

Average hospitalization price of CHSs = initial average hospital-ization price of CHSs / EXP (Time / -0.01368)

Initial average hospitalization price of CHSs = 2514.2 Factor of hospitalization payment capability of CHSs = (individual payment of medical expense + health insurance expense per capita)/average hospitalization price of CHSs

Chart of inpatient financing of CHSs = [(-0.02,0)_(10,1)], (-0.0122324,0.007895), (0.880734,0.1816), (1.63914,0.2684), (2.54434,0.3947), (2.54434,0.3947), (3.22807,0.4728), (3.22807,0.4728), (3.98777,0.7071), (4.94404,0.8369), (6.8745,0.864035), (10,1)

Factor of inpatient financing of CHSs = chart of inpatient financing of CHSs (factor of inpatient payment capability of CHSs)

Hospitalization rate of CHSs = (accessibility to CHSs + 10-9)0 3 / (technology level of CHSs + 10-9)0 3 / (factor of inpatient financing of CHSs + 10-9)0 4

Ratio of number of patients between CHSs and hospitals = number of visits to CHSs/number of visits to hospitals.

References

J.S. Akin, C.C. Griffin, D.K. Guilkey, B.M. Popkin, The demand for primary health care services in the Bicol region of the Philippines, Econ. Dev. Cult. Change 34 (4) (1986) 755-782.

J.S. Akin, D.K. Guilkey, et al., Quality of services and demand for health care in Nigeria: a multinomial probit estimation, Soc. Sci. Med. (1982) 40 (11) (1995) 1527-1537.

Y. Albadr, T.M. Khan, Factors influencing community pharmacist decision to dispense generic or branded medicines; Eastern Province, Alahsa, Saudi Arabia, Saudi Pharm. J. (2014).

S. Allin, M. Grignon, et al., Subjective unmet need and utilization of health care services in Canada: what are the equity implications?, Soc Sci. Med. (1982) 70 (3) (2010) 465-472.

H. Ariizumi, Effect of public long-term care insurance on consumption, medical care demand, and welfare, J. Health Econ. 27 (6) (2008) 1423-1435. A.D. Asante, A.B. Zwi, Factors influencing resource allocation decisions and equity in the health system of Ghana, Public Health 123 (5) (2009) 371-377. A.D. Asante, A.B. Zwi, et al., Equity in resource allocation for health: a comparative study of the Ashanti and Northern Regions of Ghana, Health Policy (Amsterdam, Netherlands) 78 (2-3) (2006) 135-148. P. Barber, B.G. Lopez-Valcarcel, Forecasting the need for medical specialists in Spain: application of a system dynamics model, Human Resour. Health 8 (2010) 24.

C. Bezold, Vision for the 21st century health care, Futures Res. Q. Rep. 11 (2) (1995) 63-75.

C. Brindis, C. Kapphahn, et al., The impact of health insurance status on adolescents' utilization of school-based clinic services: implications for health care reform, J. Adolescent Health: Official Publ. Soc. Adolescent Med. 16 (1) (1995) 18-25.

M. Center for Health Statistics and Information, An Analysis Report of National Health Services Survey in China, Peking Union Medical College Press, 1993, 1998, 2003, 2008.

M.J. Chartier, J.R. Walker, et al., Separate and cumulative effects of adverse childhood experiences in predicting adult health and health care utilization, Child Abuse Negl. 34 (6) (2010) 454-464.

M.-L. Chen, Z.-Y. Su, et al., An empirical study on the factors influencing the turnover intention of dentists in hospitals in Taiwan, J. Dental Sci. 9 (4) (2014) 332-344.

N.H.a.F.P.C.o.t.P.s.R.o. China, Chinese Health Statistical Year Book 2013, Chinese Peking Union Medical College Press, 2013.

P. Ching, User fees, demand for children's health care and access across income groups: the Philippine case, Soc. Sci. Med. (1982) 41 (1) (1995) 37-46. K. De Jaegher, M. Jegers, A model of physician behaviour with demand inducement, J. Health Econ. 19 (2) (2000) 231-258.

M.M. Desai, R.A. Rosenheck, Unmet need for medical care among homeless adults with serious mental illness, Gen. Hosp. Psychiatry 27 (6) (2005) 418425.

R. Diaz, J.G. Behr, et al., A system dynamics model for simulating ambulatory health care demands, Simul. Healthcare - J. Soc. Simul. Healthcare 7 (4) (2012) 243-250.

A. Dor, P. Gertler, J. Van der Gaag, Nonprice rationing and the choice of medical care providers in rural Cote D'lvoire, J. Health Econ. 6 (3) (1987) 291-304.

D.S. Kenkel, Health behavior, health knowledge, and schooling, J. Polit. Econ. 99 (1991) 287-305.

[24 [25 [26

[28 [29 [30

[31 [32

[33 [34

[35 [36

[41 [42

[45 [46

[49 [50 [51

[53 [54

C. Fernandez-Olano, J.D.L.-T. Hidalgo, et al., Factors associated with health care utilization by the elderly in a public health care system, Health Policy (Amsterdam, Netherlands) 75 (2) (2006) 131-139.

S. Flessa, Priorities and allocation of health care resources in developing countries: a case-study from the Mtwara region, Tanzania, Eur. J. Oper. Res. 150(1) (2003) 67-80.

L.M. Frerichs, O.M. Araz, et al., Modeling social transmission dynamics of unhealthy behaviors for evaluating prevention and treatment interventions on childhood obesity, PLoS One 8 (12) (2013).

G. Xu, J. Zou, The method of system dynamics: principle, characteristics and new development, J. Hit 8 (4) (2006) 72-77.

P. Gertler, J. van der Gaag, The Willingness to Pay for Medical Care, Johns Hopkins University Press, Baltimore, 1998.

P. Gertler, L. Locay, W. Sanderson, Are user fees regressive? The welfare implications of health care financing proposals in Peru, J. Econ. 36 (1) (1987) 67-88.

N. Ghaffarzadegan, A.J. Epstein, et al., Practice variation, bias, and experiential learning in cesarean delivery: a data-based system dynamics approach, Health Serv. Res. 48 (2) (2013) 713-734.

L. Gille, T. Houy, Forecast of the supply of healthcare in France, Futures 51 (2013) 50-58.

J. Grytten, R. S0rensen, Type of contract and supplier-induced demand for primary physicians in Norway, J. Health Econ. 20 (3) (2001) 379-393. M.C. Gulliford, D. Mahabir, Utilisation of private care by public primary care clinic attenders with diabetes: relationship to health status and social factors, Soc. Sci. Med. (1982) 53 (8) (2001) 1045-1056.

H. Naci Mocan, Erdal Tekin, Jeffrey S. Zax, The demand for medical care in urban China, World Dev. 32 (2) (2004) 289-304.

N.D. Hamelin, A. Nikolis, et al., Evaluation of factors influencing confidence and trust in the patient-physician relationship: a survey of patient in a hand clinic, Chir. Main 31 (2) (2012) 83-90.

T. Hancock, Health care reform and reform for health: creating a health system for communities in the 21st century, Futures 31 (5) (1999) 417-436.

B. Hidayat, H. Thabrany, et al., The effects of mandatory health insurance on equity in access to outpatient care in Indonesia, Health Policy Plan. 19 (5) (2004) 322-335.

D.R. Hotchkiss, Expansion of rural health care and the use of maternal services in Nepal, Health & Place 7 (1) (2001) 39-45.

D.R. Hotchkiss, N.B. Mock, et al., The effect of the health care supply environment on children's nutritional status in rural Nepal, J. Biosoc. Sci. 34 (2) (2002) 173-192.

W.C. Hsiao, The Chinese health care system: lessons for other nations, Soc. Sci. Med. (1982) 41 (8) (1995) 1047-1055.

C.-R. Hsieh, S.-J. Lin, Health information and the demand for preventive care among the elderly in Taiwan, J. Human Resour. 32 (1997) 308-333.

X.Q. Huang, Analysis of influencing factors of medical demand of rural residents, Co-Oper. Econ. Sci. (2) (2010) 114-115.

S. Inayatullah, Dramatic transformations in health care and their implications for primary care, J. Futures Stud. 8 (4) (2004) 75-82. S. Inayatullah, Transforming Health Policy, Renaissance Universal, 2007. K.E. Jackson, S. Krishnaswami, et al., Unmet health care needs in children with cerebral palsy: a cross-sectional study, Res. Dev. Disabil. 32 (6) (2011) 27142723.

John P. Ansah, R.L. Eberlein, Sean R. Love, Mary Ann Bautista, James P. Thompson, Rahul Malhotra, David B. Matchar, Implications of long-term care capacity response policies for an aging population: a simulation analysis, Health Policy 116 (1) (2014) 105-113.

N. Kato, K. Kinugawa, et al., Adherence to self-care behavior and factors related to this behavior among patients with heart failure in Japan, Heart Lung 38 (5) (2009) 398-409.

M.-K. Kong, H.-K. Lee, Demand for medical care, consumption and cointegration, Econ. Lett. 62 (3) (1999) 325-330.

P. Kostagiolas, N. Korfiatis, et al., Work-related factors influencing doctors search behaviors and trust toward medical information resources, Int. J. Inf. Manage. 34 (2) (2014) 80-88.

V.B. Kreng, C.-T. Yang, The equality of resource allocation in health care under the National Health Insurance System in Taiwan, Health Policy 100 (2-3) (2011)203-210.

P.-T. Kung, W.-C. Tsai, et al., Determining factors for utilization of preventive health services among adults with disabilities in Taiwan, Res. Dev. Disabil. 33 (1) (2012) 205-213.

T.H. Laurent Gille, Forecast of the supply of healthcare in France, Futures 51 (2013) 50-58.

V. Lavy, J.M. Quigley, Willingness to Pay for the Quality and Intensity of

Medical Care, Low-income Households in Ghana, 1993.

C.J.W. Ledford, C.C. Ledford, et al., Extending Physician ReACH: influencing

patient activation and behavior through multichannel physician

communication, Patient Educ. Couns. 91 (1) (2013) 72-78.

L.K. Leslie, J. Landsverk, et al., Children in foster care: factors influencing

outpatient mental health service use, Child Abuse Negl. 24 (4) (2000)

465-476.

L. Li, S. Yang, et al., Forecast model for medical expenses and insurance factor analysis, Chin. Health Service Manage. 5 (1994) 235-240. L. Li, L.L. Zhang, et al., Establishment of logical model of potential medical demand transformation, Acad. J. Second Military Med. Univ. 26 (11) (2006) 1213-1216.

YJBIN 2416 ARTICLE IN PRESS No. of Pages 16, Model 5G

22 August 2015

W. Yu et al./Journal of Biomedical Informatics xxx (2015) xxx-xxx

1367 [55

1370 [56

1372 [57

1374 [58

1377 [59

1380 [60

1382 [61

1384 [62

1386 [63

1388 [64

1390 [65

1392 [66

1395 [67

1398 [68

1401 [69

1403 [70

1405 [71

1408 [72

1411 [73

1415 [74

1417 [75

1419 [76

1422 [77

L. Li, L.L. Zhang, et al., Influence factor analysis and countermeasure study of [78 the potential medical demand of Chinese residents, Chin. Health Econ. Mag. 25 (7) (2006) 36-39. [79

Y.C. Li, J.H. Fu, Analysis of influencing factors of medical demand, Jiangxi Acad. Soc. Sci. 1 (2013) 186-189.

B. Liljas, The demand for health with uncertainty and insurance, J. Health Econ. [80 17 (2) (1998) 153-170.

X.S. Lin, C.R. Ai, Determinants of Chinese residents' demand for medical care - [81 an application of semiparametric estimation of ordered probit model, Stat. Res. 25 (11) (2008) 40-45. [82

L.-F. Liu, W.-H. Tian, et al., Utilization of health care services by elderly people with National Health Insurance in Taiwan: the heterogeneous health profile [83 approach, Health Policy 108 (2-3) (2012) 246-255.

Y.-L. Liu, H.-M. Wang, et al., Factors influencing the care needs and demands of [84 rural outpatients in Shaanxi, China, Public Health 120 (11) (2006) 1052-1054. E.G. Marshall, Do young adults have unmet healthcare needs?, J Adolesc. [85 Health 49 (5) (2011) 490-497.

Y.A. Momtaz, T.A. Hamid, et al., Unmet needs among disabled elderly Malaysians, Soc. Sci. Med. 75 (5) (2012) 859-863. [86

P. Musgrove, Family health care spending in Latin America, J. Health Econ. 2(3) (1983) 245-257.

T. Neutens, Accessibility, equity and health care: review and research [87 directions for transport geographers, J. Transp. Geogr. 43 (2015) 14-27. M. Nichter, Vaccinations in the Third World: a consideration of community demand, Soc. Sci. Med. (1982) 41 (5) (1995) 617-632. [88

R. O'Loughlin, A. Kelly, Equity in resource allocation in the Irish health service. A policy Delphi study, Health Policy (Amsterdam, Netherlands) 67 (3) (2004) 271-280. [89

A. Paez, J. Esita, et al., Exploring resource allocation and alternate clinic accessibility landscapes for improved blood donor turnout, Appl. Geogr. 45 (2013) 89-97. [90

J.O. Parkhurst, L. Penn-Kekana, et al., Health systems factors influencing maternal health services: a four-country comparison, Health Policy (Amsterdam, Netherlands) 73 (2) (2005) 127-138. [91

G. Picone, M. Uribe, et al., The effect of uncertainty on the demand for medical care, health capital and wealth, J. Health Econ. 17 (2) (1998) 171-185. [92

S. Pokhrel, M. De Allegri, et al., Illness reporting and demand for medical care in rural Burkina Faso, Soc. Sci. Med. (1982) 70 (11) (2010) 1693-1700.

C. Qi, N.-B. Chang, System dynamics modeling for municipal water demand [93 estimation in an urban region under uncertain economic impacts, J. Environ. Manage. 92 (6) (2011) 1628-1641.

D. Qian, H. Lucas, et al., Determinants of the use of different types of health [94 care provider in urban China: a tracer illness study of URTI, Health Policy 98

(2) (2010) 227-235. [95

R.K. Rai, P.K. Singh, et al., Utilization of maternal health care services among married adolescent women: insights from the Nigeria Demographic and [96 Health Survey, 2008, Women's Health Issues: Official Publ. Jacobs Inst. Women's Health 22 (4) (2012) e407-e414. [97

I.f.A.F. Report, Health and Health Care in 2032, Report from the RWJF Futures Symposium, 2012. [98

I.f.A.F. Report, Primary Care 2025: A Scenario Exploration, Alexandria, VA, 2012. [99

A.S. Rwashana, D.W. Williams, et al., System dynamics approach to immunization healthcare issues in developing countries: a case study of Uganda, Health Inf. J. 15 (2) (2009) 95-107.

D. Sahn, S. Younder, G. Genicot, The demand for health care services in rural Tanzania, Oxford Bull. Econ. Stat. 65 (2003) 241-259.

C. Samuel, K. Gonapa, et al., Supply chain dynamics in healthcare services, Int. 1424

J. Health Care Qual. Assurance 23 (7) (2010) 631-642. 1425

R. Sauerborn, A. Nougtara, et al., The elasticity of demand for health care in 1426

Burkina Faso: differences across age and income groups, Health Policy Plan. 9 1427

(2) (1994) 185-192. 1428 E.F. Schmid, D.A. Smith, Pharmaceutical R&D in the spotlight: why is there still 1429 unmet medical need?, Drug Discovery Today 12 (23-24) (2007) 998-1006 1430 J. Schooley, C. Mundt, et al., Factors influencing health care-seeking behaviours 1431 among Mayan women in Guatemala, Midwifery 25 (4) (2009) 411-421. 1432 M.A. Shah, N.M. Shah, et al., Unmet need for contraception in Kuwait: issues 1433 for health care providers, Soc. Sci. Med. (1982) 59 (8) (2004) 1573-1580. 1434 P. Sharan, V. Krishnan, Principles: mental health resources and services, Int. 1435 Encyclopedia Public Health (2008) 418-430. 1436 H. Shigeoka, K. Fushimi, Supplier-induced demand for newborn treatment: 1437 evidence from Japan, J. Health Econ. 35 (2014) 162-178. 1438 E.S. Shin, S.Y. Hwang, et al., Relationships of factors affecting self-care 1439 compliance in acute coronary syndrome patients following percutaneous 1440 coronary intervention, Asian Nurs. Res. 7 (4) (2013) 205-211. 1441 E.M. Sidze, J. Pradhan, et al., Reproductive health financing in Kenya: an 1442 analysis of national commitments, donor assistance, and the resources 1443 tracking process, Reprod. Health Matters 21 (42) (2013) 139-150. 1444 V. Tangcharoensathien, W. Phoolcharoen, et al., The potential demand for an 1445 AIDS vaccine in Thailand, Health Policy (Amsterdam, Netherlands) 57 (2) 1446 (2001)111-139. 1447 K. Taylor, B. Dangerfield, et al., Simulation analysis of the consequences of 1448 shifting the balance of healthcare: a system dynamics approach, J. Health Serv. 1449 Res. Policy 10 (4) (2005) 196-202. 1450 T.H. Wagner, H.B. Jimison, Computerized health information and the demand 1451 for medical care, Value Health: J. Int. Soc. Pharmacoecon. Outcomes Res. 6 (1) 1452 (2003) 29-39. 1453 H.J. Wong, R.C. Wu, et al., Smoothing inpatient discharges decreases 1454 emergency department congestion: a system dynamics simulation model, 1455 Emergency Med. J. 27 (8) (2010) 593-598. 1456 G. Xing-Yuan, T. Sheng-Lan, Reform of the Chinese health care financing 1457 system, Health Policy 32 (1) (1995) 181-191. 1458 L. Yang, Y. Cui, et al., A study on the changing trend and characteristics of rural 1459 resident's needs and demands of health service, Chin. Health Econ. Mag. 24 (9) 1460 (2006) 28-31. 1461 L. Yang, Y.R. Zhou, et al., Health service need and demand of urban and rural 1462 residents in Jiangxi province and the utilization analysis, Chin. J. Evidence- 1463 Based Med. 11 (10) (2011) 1115-1119. 1464 M. Yi, J. Kim, Factors influencing health-promoting behaviors in Korean breast 1465 cancer survivors, Eur. J. Oncol. Nurs. 17 (2) (2013) 138-145. 1466 W.C.-M. Yip, W.C. Hsiao, et al., Early appraisal of China's huge and complex 1467 health-care reforms, Lancet 379 (9818) (2012) 833-842. 1468 Z. Mao, Estimating quantity of induced demand for medical services in China, 1469 Chin. Health Econ. 25 (1) (2006) 65-67. 1470 W.D. Zhang, H.F. Kong, The empirical research of the influence factors of urban 1471 residents' medical demand in China, Chin. Health Econ. 4 (2014) 021. 1472 Z. Zhou, Y. Su, et al., New estimates of elasticity of demand for healthcare in 1473 rural China, Health Policy 103 (2) (2011) 255-265. 1474 G.H. Zhu, Q.Y. Cao, et al., Analysis of health service need, demand and 1475 utilization of urban and rural residents in Jiangxi province, Chin. J. Health Stat. 1476

(3) (2010) 297-298. 1477