Geographic Variations in Stroke Inpatient Healthcare Utilization and Hospital Expenditure Equity in China

Background: Stroke is a devastating disease that imposes a heavy nancial burden on patients and their families and a signicant economic cost on a nation’s healthcare system. Under China’s fragmented health insurance system, it was unclear whether geographic variations in healthcare utilization and hospital expenditures lead to healthcare inequities for stroke inpatients. This study assesses the geographic variations in stroke patients’ healthcare utilization and hospital expenses and the potential geographical inuencing factors. Methods: Covering all municipalities and provinces in mainland China, our main data were a 5% random sample of stroke claims from the Urban Employees Basic Medical Insurance (UEBMI) and Urban Residents Basic Medical Insurance (URBMI) schemes from 2013 to 2016, totaling 217969 inpatients and 280804 admissions. The Theil index was employed to evaluate the (in)equity in healthcare utilization and hospital expenditures across all 31 mainland Chinese provinces. Using multiple linear regression analysis, the geographic inuencing factors, comprising ability to deliver healthcare, geographical accessibility of health services, healthcare-seeking and economic factors, were explored. Results: UEBMI stroke inpatients had higher hospital costs and longer ALOS, but lower OOP expenses than those with URBMI. UEBMI insurance had a lower Theil index value than the URBMI scheme. Stroke patients’ healthcare utilization and hospital expenditures showed signicant differences both within and between regions. The intra-region Theil (in)equality index value was higher than the inter-region Theil index, with the Theil index highest within eastern China, China’s richest and most developed region. The ability to deliver healthcare, the enabling factors and the provincial-level economic factors had signicant effects (P<0.05) on healthcare utilization and hospital expenses. Conclusions: Our data revealed signicant geographic variation in healthcare utilization and hospital expenditures for stroke patients. In addition to differences in the coverage and reimbursements of the UEBMI-URBMI schemes, disparities within regions were associated with the ability to deliver healthcare (open hospital beds per 100 patients), the enabling factors (regional reimbursement rate and regional education level) and the provincial-level economic factors (GDP per capita). China’s fragmented urban health insurance schemes require further reform to ensure better equity in healthcare utilization and hospital


Introduction
A devastating disease for individuals and families, stoke is the second most common cause of death globally, and one of the major causes of disability worldwide [1]. Across a variety of countries, between 2-4% of total healthcare expenditures areaccounted for by stroke, imposing a heavy nancial burden on the healthcare systems in many countries [2]. Compared to all other countries, China has the highest incidence of stroke [3], withstroke costingabout $US5.9 billion toChina's healthcare system annually in 2010 [4]. Imposing a economic burden on stoke patients, stroke family caregivers and stroke familes [5][6][7], manyface the risk of nancial catastrophe and bankruptcy [8]. A previous study reported that 71% of the sample of 4739 stroke patients in China had suffered catastrophic out-of-pocket (OOP) expenditures, and workers with health insurance were seven times less likely to experience catastrophic payments than those without health insurance [9]. Clearly, health insurance can effectively promote access to healthcare and reduce the incidence of catastrophic medical payment, especially for poor families [10].
In China, there were two health insurance schemes designed exclusively for urban residents: Urban Employee Basic Medical Insurance scheme (UEBMI) for the urban employed and theUrban Resident Basic Medical Insurance scheme (URBMI) for the urban unemployed, retired, children and students,who were not covered by the UEBMI [11]. Covering more than 95% of the residents in urban China during the 2013-2016 period, the two schemes covered roughly 750 million, or 54%, of the total Chinese population [12].The basic differences between UEBMI and URBMI are set out in Table 1.UEBMI is funded both by employers (2% of wages) and employees (6% of annual wages) [13], while the revenue for the URBMI is from individual premium contributions and subsides from central and local government.As shown in Table 1, UEBMIprovides better nancial protection and offers a more comprehensive coverage than URBMI, which focuses on inpatient services, catastrophic illness insurance, but withlimited coverage of basic outpatient services [14]. Patients with UEBMI usually utilize more health services and have higher expense than those with URBMI [15].
From aggregate studies of health insurance utilization, we know that the elderly's access to different insurance schemes results in different utilization rates between rural insurance schemes and urban insurance schemes within a single city, with disparities in health insurance bene ts, health service expenses and healthcare utilization equity [16][17][18]. These ndings are consistent with the major differences between UEBMI and URBMI in Information Database to examine geographic variation in morbidity and mortality of cerebrovascular diseases (CVDs), which revealed moderate geographic correlation between CVD morbidity and mortality [20]. Third, research has identi ed the geographic variation in healthcare services, where, for example, there exist marked geographic variations in the use of drugs for patients with prostate cancer [20] and health service expenses [21][22][23]. Other factors promoting geographic variations includedifferences in family income andthe economic developmentlevel that variedacross China's 31 provinces, municipalities and autonomous regions.For the 2013-2016 period, we conduct a cross-sectional assessment of the geographic inequities in UEBMI and URBMI inpatient stroke healthcare utilization and hospital expenses within and between three regions in China and explore the economic and health resource factors contributing to these geographic inequities.

Methods
Data resources and regional division We collecteda 5% random sample of UEBMI and URBMI insured stoke inpatients' claim data between January

Measuring tool
The Theil index is a relative indicator measure of economic and other types o nequality [24].The advantage of the Theil index is that it can calculate the contribution of intra-group and inter-group inequities to total inequities, thus avoiding the calculation of absolute values. The Theil index rangesbetween 0 and 1, where smaller values point to more equitable distribution of some economic phenomena, such as income, across a population [25].While originally used to measure inequalities in economic data, the Theil index is increasing used to evaluate (in)equities in health services, including healthcare utilization and health service expenses.
The Theil index formula is: where P i is theproportion of insured population in one province accounting for the overall China insured population; andY i is the proportion of healthcare utilization/expenditure in one province accounting for the total utilization/expenses nation wide.
Since we divided the 31 provinces into three regions, the Theil index in formula (1) can also be decomposed into the T intra , which measures utilization/expenditureinequality "within region" where P g is the proportion of insured population in one region accounting for the total insured population and T g is the Theil index of one region (eastern, centraland western China).The Theil index (1) can also be decomposed into T inter , which measures utilization/expenditure inequality "between regions" where Y g is the proportion of healthcare utilization/expense in one region accounting for the total healthcare utilization/expense. Higher T intra and T inter index values mean greater inequality, while lower index values mean more equality.

Main indicators and multiple linear regression
The Theil index was calculated with two indicators of expenses, hospital costs and OOP expenditures,and healthcare utilization, measured as average length of stay (ALOS). Ordinary least squares (OLS) was used to explore the potential geographical factors in uencing geographic variation in healthcare utilization.To deal with the skewness of data, we converted the hospital costs, OOP expenses and ALOS to natural logarithms.
For the independent variables, standard behavioral healthcare models [26]suggest measures impacting healthcare utilization and hospital expenses,such as the ability to deliver healthcare, geographical accessibility of health services,enabling factors facilitating healthcare-seeking and economic factors [27][28][29][30]. In our study, the ability to deliver healthcare contains two aspects: the number of open beds per 1000 residents and whether the hospital is certi ed by the National Stroke Center, since certi edcenters usually have strong stroke medical care capacity. These data were obtained from the China Health Statistics Yearbook 2014-2017 and the website of National Stroke Center.The enabling factors facilitating or impeding the use of healthcareservices comprised the regional reimbursement rate, regional education level and regional employment rate.Provincial-level economic factors were represented by the insurancefund per capita,measured by the total health insurance revenue divided by the insured population in that year,and GDP per capita. These data were collected from the China Labor Statistics Yearbook 2014-2017 and the China Statistical Yearbook 2014-2017.
Descriptive statistics were employed to illustratethe regional distribution and time trend of healthcare utilization for stroke inpatients, and statistical analyses were conducted using STATA version 14.0 (Stata Corp LP, College Station, TX), with statistical signi cance α = 0.05.

Results
Healthcare utilization and hospital costs of stroke inpatients from 2013 to 2016 RMB3969.0 (US$597.5). Re ecting the bene t and reimbursement schedule in Table 1, UEBMI inpatients had higher hospital costs and longer ALOS, but lower OOP expenses than those with URBMI. This con rms the inequalities in expenses and utilization between the two insurance schemes.

Theil (in)equality index
The Theil inequality index values in Table 4shows that there were signi cant variations in hospital costs, OOP expenses and ALOS under both UEBMI and URBMI. Table 4 shows that Theil index values were signi cantly higher than the Theil index of health expenditure per capita in China, which was reported to range from 0.0583 to 0.0686 between 2013 and 2016 [31].As shown in Table 4, the year-by-year UEBMI Theil hospital cost index  Table 4 calculates the difference in the Theil UEBMI and URBMI index values, which shows that URBMI inpatients suffered higherexpenses and utilization than UEBMI inpatients, with higher numbers showing higher inequality. Second, the differences in Table 4 also show that these differences varied signi cantly year-by-year. An alternative illustration of inequalities in expenses and utilization is shown in Figure 1, which plots for each region the year-by-year Theil index for URBMI and UEBMI expenses and utilization.In the UEBMI and URBMI

Results of regression analysis
The regression analysis exploredthe geographical factors in uencing the geographical variation of patients' healthcare utilization and expenditure. We found that the ability to deliver healthcare (open beds), the enabling factors (regional reimbursement rate and regional education level) and the provincial-level economic factors (GDP per capita) all had a signi cant in uence on inpatients hospital expenses and healthcare utilization.Importantly, in uencing factors varied by insurance type. As shown in

Discussion
For urban insurance schemes, this study provides a comprehensive nationwide exploration ofequity in stroke inpatients' healthcare utilization and expenses.Using the Theil (in)equality index to measure "horizontal equity", we found signi cant geographic variationin stroke inpatients' healthcare utilization and hospital expenses. Furthermore, the ability to deliver healthcare (open beds), the enabling factors (regional reimbursement rate and regional education level) and the provincial-level economic factors (GDP per capita) were found to have a signi cant impact on healthcare equity.
A steady rise in the Theil index of ALOS both in UEBMI and URBMI group, indicated the growing inequity in ALOS. For all patients in China, Xie et al [32] also found that the overall geographical inequity in healthcare utilization rose from 2011 to 2015. We speculate that the increasingly geographical difference in UEBMI and URBMI ALOS might be attributable to policy factors. Since the ALOS was one of the important performance assessment indexes, public hospitals in China actively reduced patients' ALOS, by encouraging, for example, patients to be discharged early, to improve their performance [33]. Previous studies report that multimorbidity, speci cally hypertension, was a strong predictor of longer ALOS for stroke patients [34]. There were considerable geographic variations in prevalence of hypertension in China, and high hypertension prevalence zones, which extended from parts of the southeast to the northern and northeast [35,36]. These different geographical hypertension prevalence zones would impact the geographical utilization rates of UEBMI and URBMI in Table 3. We also expect that different regional stroke prevention and treatment systems were an important factor leading to the geographic variation in patients' ALOS. A well-functioning stroke prevention and treatment system could send stroke patients into a quali ed hospital for a shorter time and provide more effective treatment than regular hospitals. Delayed medical checks and treatment, and the limited treatment capacity of a hospital, could dramatically contribute to longer ALOS for stroke patients [37,38]. Previous studies noted the large geographic variation in stroke prevention and treatment system in China [39], which may explain some of the regional variation in stroke patient healthcare utilization. We recommend that public hospitals in China establish a uni ed and more scienti c assessment index system for stroke patients and improve their capacity for treating stroke patients. The government should further strengthen the stroke prevention and treatment system, especially in poor areas with diminished healthcare delivery.
Our empirical evidence clearly revealed that the UEBMI group had an overall smaller Theil index of hospital costs, OOP expenses and ALOS than the URBMI group from 2013 to 2016. That means stroke inpatients with URBMI experienced greater inequity in healthcare utilization and expenditure than those with UEBMI. Examining the equity of health services utilization in different regions, Zhang [40] also found patients covered by UEBMI had greater geographical variation in healthcare utilization than those covered by URBMI. As shown in Table 1, The UEBMI scheme provided more generous bene ts, more comprehensive service coverage, as well as stronger nancial protection [41]. Since the UEBMI scheme provided stronger nancial protection than the URBMI scheme, UEBMI patients would seek a more comprehensive treatment than URBMI patients [42]. With lower levels of bene ts and nancial protection, URBMI patients would economize on their level of health services utilization subject to their family's nancial status [42]. This would contribute to geographical variations in healthcare utilization and hospital expenses. Importantly, patients covered by UEBMI had stable incomes due to their employment status and usually had a better nancial situation than URBMI unemployed, retiree, student and childinpatients. Without worrying about the occurrence of catastrophic health expenditure, patients in different regions would receive treatment as required, but UEBMI patients could incur higher OOP expenses and longer hospital stays [43]. Furthermore, a large proportion of patients covered by the URBMI were unemployed and children, with lower education levels than UEBMI patients. Education level was considered an important factor which would affect the ALOS for stroke inpatients [38]. There were signi cant regional variations in education level, with people in eastern China having the highest education level, followed by the central and western regions 59 .These regional education level differences probably interacted with the type of insured patient to contribute to inequity in UEBMI-URBMI healthcare utilization and hospital expenses. Figure 1 displays the inequity in healthcare utilization and health service expenses. We found that the internal differences within regions were the major factors contributed to inequity of healthcare utilization and health service expenses. Internal differences in the eastern region accounted for the largest part of the Theil index.
When the gap of socio-economic development level between the richer eastern region and poorer western and northern regions was signi cant, it seemed counterintuitive that the richest region had the greatest inequity in healthcare utilization and health service expenses. The coastal areas and strong economic zones in eastern China had advantages of export-linked and foreign investment industry, enhanced infrastructure and bene ted most from economic policy reforms that transformed China's economy, but widened the economicgap between different cities in eastern China. Differences between cities inthe central and western provinces were relatively smaller than within the eastern region. Therefore, socio-economic factors may be an important reason for the greaterhealthcare inequity within the eastern region. Another possible reason for the inequalities within the eastern and between the eastern and other regions was healthcare resource allocation. Previous studies reported that the eastern region had been experiencing the worst equity in health resource allocation [19], which was re ected in 20,62 inequities inhealthcare delivery. We recommend that the government should not only make policies to improve the medical system in central and western regions, but also take the less developed provinces and prefectures in eastern China into account.
Our results demonstrated that the ability of healthcare delivery (represented by number of actual open beds per 1000 residents) had a signi cant negative impact on the hospital expenses both in the UEBMI and URBMI group, where more beds improved the equity of hospital expenses. Similarly, Kim et al [44] found that better nurse sta ng levels had a signi cantlynegative association with the ALOS and medical expenses of patients with hip or knee procedures. We believe that patients in areas with a stronger ability to deliver healthcare have easier access to high quality healthcare. To improve delivery, we suggest that an effective treatment protocol could reduce ALOS and hospital expenses [45].The enabling factors (regional reimbursement rate and regional education level) were identi ed as important factors in uencing healthcare utilization and hospital expenses in the URBMI group, but not in the UEBMI group. First, higher reimbursement signi ed an insurance fund's stronger nancial protection, encouraging patients to utilize more health services and bene t more from central subsidies.This is re ected in our data that showsUEBMI patients enjoying higher reimbursement rates and usually higher total expense, but fewer OOP cost than URBMI patients [42]. Second, education level had a signi cant negative impact on URBMI inpatients. Sinceeducation level was considered an important factor causing prehospital delay in stroke treatment,better educated UEBMI patients sought stroke treatment earlier,with reduced ALOS, than URBMI patients [46]. This is consistent with Milagros et al [47]who reported that low levels of education hada strong relationship with longer length of hospitalstays.Our results also revealed that GDP per capita was signi cantly associated with hospital costs and OOP.Economic factors are universally acknowledged as a main factor affecting healthcare utilization. We speculate patients with UEBMI had high incomes, consequently, the reimbursement rate and GDP per capitawere not signi cant explanatory variables forhealthcare utilization and hospital expenses. URBMI patients with higher incomeswere probably more willing to follow recommendations from doctors in the utilization of health services, thus GDP per capita was associated with higher health expenditure [48].
This study has a number of limitations. Our data applies to healthcare utilization and hospital expenses of urban stroke inpatients, not the whole stroke patient population in China.Due to data limitations, we used ALOS to re ect healthcare utilization of stroke patients, but healthcare utilization based on need and demand cannot be easily divided, therefore, the results should be interpreted with care. We recommend that other potential factors in uencing the equity in healthcare utilization and hospital expenses, such as regional health investment be considered in future studies.

Conclusion
Under the fragmentedhealth insurance system in China, there were signi cant geographical variations in health service utilization and hospital expenses within and between regions for stroke inpatients. The UEBMI group enjoyed greater equity in healthcare utilization and hospital expenses than the URBMI group. The differences were mainly caused by the disparities within regions and not between regions. The ability to deliver healthcare, the enabling factors and the provincial-level economic factors were found to be signi cant.We recommend that the government improves the regional ability to deliver healthcare, establish strong regional stroke prevention and treatment networks and further develop the capacity of stroke treatment. We also recommend strengthening health education on stroke, especially for URBMI patients, to shorten the prehospital delay by stroke victims. It is also essential to further reform the urban health insurance schemes, such as increasing the reimbursement rate for URBMI stroke patients to narrow the considerable gap in healthcare utilization and expenditure between the UEBMI and URBMI group.

Consent for publication
Not applicable