A Tale of Transition: Trends of Catastrophic Health Expenditure and Impoverishment in Urban China, 1986-2009

ABSTRACT From 1986 to 2009, China’s health system reform first adopted a market-oriented approach and later reemphasized the role of the government starting from 2002. China’s oscillating health care financing policies present us a unique opportunity to examine the consequences of government-led financing and market-oriented financing measures. This study uses the Urban Household Survey, a diary data in China that covers the period of 1986 to 2009, to examine the long-run trends in the incidence and intensity of catastrophic health expenditure and medical impoverishment. Four major findings emerge. First, the incidence and intensity of catastrophic health expenditure in urban Chinese households increased rapidly between 1986 and 2002, whereas they stabilized after 2002. Second, the incidence of medical impoverishment and its depth in the poverty gap remained stable before 2002 and decreased rapidly after 2002. Third, income and regional inequality in measures of catastrophic health expenditure widened from 1986 to 2002. They narrowed in the 2000s but remain wide. Fourth, income and regional inequality in medical impoverishment remained unchanged between 1986 and 2002 and narrowed substantially after 2002. All these results suggest that China’s two cycles of health care reform generated significantly different outcomes in financial protection, holding lessons for the ongoing health care reform in China and other countries.


Introduction
As a key component in Universal Health Coverage (UHC), financial protection against outof-pocket health expenditure is what many lowand middle-income countries are trying to accomplish through health reforms. 1,2When governments design healthcare financing policies, they often wrestle with the role of the government and the market in the health sector. 3 Progovernment advocates emphasize on market failures in the health sector and give equity high priority. 4Pro-market advocates argue that the government also suffers serious failures and market forces still work in improving efficiency in the health sector. 5,6Despite heated debates, there is little real-world evidence that shows how financial risk protection changes in the long run as a nation's healthcare financing policies oscillate between government-led and market-oriented measures.
China's oscillating health financing policies between 1986 and 2009 present us a unique opportunity to shed light on the consequences of government-led financing and market-oriented health financing measures.During this period, China was the largest transitional economy around the world, experiencing a rapid transition from a centrallyplanned economy to a market-oriented one with fast economic growth.Concurrently, China's health system went through two cycles of reform: a market-oriented reform since 1986 and a shift toward reemphasizing the government's role after 2002. 3sing data from the Urban Household Survey in China, this study examines trends of catastrophic health expenditure (CHE) and medical impoverishment (MI) in urban China between 1986 and 2009.Our major finding is that China's two major cycles of health care reform generated significantly different policy outcomes in financial protection.The first cycle of the reform worsened the condition of financial risk protection and widened regional and income inequity while the second cycle ceased the deterioration.This finding is based on four key results.First, we observe a peak for CHE incidence and intensity in urban China at around 2002, which coincides with the time of the policy shift.Second, we observe a sharp reduction in the incidence of MI as well as its depth in the poverty gap after 2002, while they remained largely unchanged between 1986 and 2002.Third, inequality in CHE incidence and intensity across income quintiles and geographic regions greatly increased between 1986 and 2002 and experienced a moderate decrease in the 2000s.Despite improvements, a high degree of regional disparities and income inequality in CHE measures still existed in 2009.Fourth, we observe large regional disparities and income inequality in MI with small improvements between 1986 and 2002, and substantial improvements in reducing income inequality and regional disparities after 2002.
Our findings mainly contribute to two bodies of literature.First, our findings contribute to the literature on financial protection and its influencing factors by adding the evidence on how financial protection change under government-led financing and marketoriented health care financing policies.][9][10][11][12][13] In addition, extensive studies have investigated the effects of health insurance expansions on improving financial protection.The introduction of Medicare in the USA substantially reduced the exposure to out-of-pocket spending for the elderly. 14As for developing countries, Limwattananon et al. find the legislation of universal health insurance decreased out-of-pocket payment. 15ing et al. find that Mexico's health insurance expansion reduced overall catastrophic expenditures and benefited the most for the poorest individuals. 16ietnam's health care fund for the poor substantially reduced out-of-pocket expenditure. 17Indonesia's health insurance expansion for the informal sector and the poor increased out-of-pocket spending in urban areas. 18Early assessment on China's rural insurance scheme showed no effect in reducing outof-pocket expenditure, 19 while the expansion of China's urban insurance scheme increased the risk of catastrophic spending. 20However, few studies examine how financial risk protection changes in the long run as a nation's healthcare financing policies oscillate between government-led and market-oriented ones.Second, our study advances the literature on trends of financial protection in China.Most existing studies focus on the trends after the 21st century but few studies have examined the trends in the 1980s and 1990s when China experienced rapid economic and societal transformations. 12,21,22

The First Cycle of Reform: 1986-2002
Before the economic reform in 1978, the Chinese government created a socialistic health care system that prioritized equity.In urban areas, the majority of residents were covered by two insurance schemes: the Government Insurance Scheme (GIS) and the Labor Insurance Scheme (LIS).Financed by the government budget, the GIS covered public servants and government entity employees.The LIS, financed by each stateowned enterprise's (SOE) welfare fund, covered employees, their dependents, and retirees of SOEs.In addition, China created a public health care delivery system that generates most of its revenue from government subsidies and the rest from out-of-pocket health spending under a strict price schedule.Under this health system, most of the urban residents enjoyed equal access to basic health services. 3nfluenced by the market-oriented economic reform, the Chinese urban health care system had experienced a dramatic transition since 1986.The Chinese leaders adopted a laissez-faire policy that defaulted health care to market forces and selfreliance. 23With the idea that rising tides raise all boats, the Chinese leaders believed that economic growth would presumably make health care affordable with individuals' increased income. 3When China's economic reforms led to a dramatic reduction in health insurance coverage, the government adopted a benign neglect position in the beginning.Although the Chinese government created the Urban Employee Basic Medical Insurance scheme (UEBMI) in 1998 to replace the LIS and GIS, it only covered formal employees in state-owned and private enterprises but no longer covered their dependents.As shown in Table 1, approximately 50% of urban residents were left without effective social health insurance coverage in the late 1990s. 24ven for those covered by UEBMI, their health insurance benefits were less generous. 25hina's health delivery system also went through dramatic reforms from a centrally-planned system to a heavily market-based one.Like other transitional economies, the Chinese government faced serious fiscal difficulties in the initial reform era due to the shrinking public sector.Government subsidies for public health care facilities fell to a mere 10% in the early 1990s. 23,26Meanwhile, with the intention of making services affordable for patients, the government continued to fix service charges below actual costs.However, public health care facilities had to survive financially.The government allowed health facilities to make profits from prescribing/dispensing drugs and providing high-technology tests.It led to widespread over-prescription of drugs and overuse of diagnostic tests. 26The consequence of reduced government subsidies and distorted provider incentives was burgeoning health cost escalation. 3

The Second Cycle of Reform: 2003-2009
The widespread discontent and the outbreak of SARS prompted Chinese leaders to reexamine China's health system.A series of measures have been introduced since the early 2000s.The first was the expansion of the UEBMI at a faster pace.Although the central government launched the UEBMI in 1998, its implementation was slow before 2003 due to practical difficulties.For example, many young workers in private enterprises were less willing to participate in the UEBMI. 25As a result, the insurance coverage of urban residents in 2003 was even slightly smaller than that in 1998.The central government acknowledged these problems and gradually adjusted its policies.Beginning in 2003, the central government set a target on coverage rate for local governments, leading to strict implementation in insurance enrollment, especially for private enterprises.As of 2009, UEBMI covered 220 million urban residents, nearly doubling the number in 2002. 27he second measure was the introduction of the Urban Resident Basic Medical Insurance scheme (URBMI), which covered children, the elderly, the disabled, and the unemployed among urban residents.This new insurance scheme was introduced in 2006 and was largely funded by government subsidies with low individual premiums.By the end of 2009, around 180 million people were enrolled in the scheme and more than 90% of urban residents were covered by either UEBMI or URBMI. 27In addition, the Chinese government introduced the Medical Assistance scheme in 2005 to provide a health expense safety net for the poorest and most vulnerable population in urban and rural areas.As shown in Figure 1, a reemphasis on the government's role in financing contributed to a sharp decrease in the share of out-of-pocket expenses in total health expenditures from 57.7% in 2002 to 37.5% in 2009.
However, little was changed on the supply side.Hospitals remained to be profit-driven to survive financially.While the government implemented and piloted policies to increase health care accessibility and affordability, the incentives to overprescribe and over-test for health facilities, especially for public hospitals, remained fundamentally unchanged.Massive fiscal inputs into the health sector were not transformed into cost-effective services. 28The annual real rate of health expenditure growth was as high as 14.8% between 2003 and 2009, which outpaced the number between 1990 and 2000.

Method
We use catastrophic health expenditure and medical impoverishment as our main indicators to analyze the trends of financial risk protection in urban China.We use difference-in-means tests to evaluate overall trends and compare subsample results by income groups and regions to examine changes in inequality.

Dataset
We use the Urban Household Survey (UHS) as our main data source to examine the trends of financial protection in urban China.We also use data from the National Health Service Survey (NHSS) to provide supplementary evidence.
The UHS has been conducted annually in approximately 30 Chinese provinces since 1986 by China National Bureau of Statistics (NBS).It includes households with urban residence registration (hukou), providing extensive information on income, consumption, expenditure, as well as demographic characteristics.0][31] The UHS uses a multi-stage stratified design with a rotationsampling scheme that replaces one-third of the households yearly to ensure representativeness and balance households' burden. 32The Urban Household Survey collects information on disposable income and expenditure using the diary method.Each member of the household keeps a diary of disposable income and expenditure, which is aggregated at the household level.The categories for expenditures include food, clothing, housing, household goods, transportation and communication, education and entertainment, health, and others.Health expenditure consists of expenditures on medical supplies, medicine, and health care services.The categories for income include salary, operational net income, capital net income, and transfer net income.Transfer net income is calculated by subtracting transfer expenditure from transfer income. 1 The staff from the local bureau of statistics visit selected households monthly or bi-weekly to ensure diary quality and collect monthly records.
The UHS data used in this study covers 18 of 30 provinces from 1986 to 2009.As shown in Figure 2, these 18 provinces locate in different geographic regions, enabling us to examine the regional difference across China. 2 In Figure 3, we show that the subsample of UHS in 18 provinces used in this study is consistent with NBS aggregate data. 33Panel A shows that trends of consumption expenditure per capita between 1986 and 2009 calculated from our UHS sample are consistent with aggregate data reported by the NBS.Panel B shows that the scale and trends of medical expenditure per capita in the UHS 18 province sample data and NBS national aggregate data are similar.These results verify that our dataset can be representative of the urban population in China.

Indicators
Following the literature, we use four indicators to measure financial protection in health: the incidence  of CHE, the intensity of CHE, the incidence of MI, and the intensity of MI. 34 Appendix describes in detail how to calculate each indicator.
The incidence of CHE in this study mainly refers to the proportion of households whose health expenditure is at least 40% of the total household expenditure net of food consumption.We use this definition of CHE as our main indicator following both the international literature 7,8 and the Chinaspecific literature 12,21,22 so that our estimates can be compared to previous studies.The intensity of CHE is measured by the average gap of CHE, which is the proportion of health spending in ability-to-pay exceeding the CHE threshold.Considering that CHE incidence and intensity may be sensitive to thresholds, we explore the robustness of our results with alternative specifications of thresholds such as 10% and 25% with food expenditure included in total expenditure. 10e measure impoverishment incidence as the change in poverty headcount with and without out-of-pocket health care spending in total. 11,22sing the poverty line set by the World Bank  We use the poverty gap to measure the intensity of MI.The poverty gap measures the change under the poverty line incurred by out-of-pocket health spending.The poverty gap incorporates both households impoverished by medical expenses and households in poverty before medical expenses so that it holistically reflects the impoverishing effect of medical expenses.This is especially relevant in our study since 16% of the urban households were living below the poverty line in urban China in the 1980s.We use the poverty gap divided by the poverty line to generate a normalized poverty gap, which is the main indicator for MI intensity in our study.
It should be noted that CHE and MI capture different aspects of financial protection.CHE reflects the concern on a sharp decline in living standards.By contrast, MI reflects the concern that out-ofpocket health spending may drive households into poverty.Previous studies also showed that MI does not imply CHE and vice versa. 11,34

Statistical Analysis
We examine the trends in financial protection by comparing results between 1986 and 2009.We present variable means with 95% confidence intervals (CIs) and differences between survey years, and test for statistical significance using χ2 tests for categorical variables and t-tests for continuous variables.
Given that China exhibits regional differences in social and economic development, we examine trends by regions (Northeast, East, Central, and West) to show regional inequality in financial protection.We also construct income quintiles by household income per capita to investigate changes in income inequality; an increase in quintiles (Q1, Q2, Q3, Q4 to Q5) corresponds to a progressive increase in household income per capita.We perform regression analyses to investigate inequity across regions and income groups.Our estimation model is as follows: The subscript i indicates a household.Y i represents dependent variables: indicators of CHE and MI.We use CHE indicators derived by the 40% threshold.X i is a vector of control variables, including age, family structure, the gender of the household head (HH), education level of the HH, employment status, etc.The key variables of interest are Region i and IncomeGroup i .Region i is a vector of dummy variables indicating different regions.IncomeGroup i is a vector of dummy variables representing income quintiles by household income per capita.The coefficients δ 1 and δ 2 capture inequality between different regions and inequality by income groups.We run the regression on each respective wave and examine the trend in the coefficients δ 1 and δ 2 .

Catastrophic Health Expenditure
The black lines in Figure 4 show national trends in the incidence and intensity of CHE between 1986 and 2009.We observe that the incidence and intensity of CHE increased rapidly until 2002 and remained stable afterward, which is concurrent with the timing of the shift from market-oriented health care financing policies toward governmentled financing.
Table 3 shows the incidence and intensity of CHE in urban China in selected waves between 1986 and 2009 by three threshold definitions.Incidence of CHE measured with the 40% threshold increased from 0.256% in 1986 to 1.05% in 1992, to 2.75% in 1997, and to 5.07% in 2002.The differences between 1986 and 1992, 1992 and 1997, 1997 and 2002 are all statistically significant.However, the change between 2002 and 2009 was not statistically significant.This finding is consistent with the trend observed in  Like the incidence of CHE, trends in CHE intensity show a turning point at around 2002.At the 40% threshold, the differences in CHE intensity between selected waves before 2002 were all positive and statistically significant, while there was no statistically significant change between 2002 and 2009.Results are similar when we use thresholds of 25% and 10%.  4 show trends in CHE incidence and intensity by regions.Regional differences were relatively small in the 1980s and the early 1990s.In the mid and late 1990s, the incidence of CHE in Northeast China increased sharply, reaching as high as 8.15% in   4 exhibit income inequality in CHE incidence and intensity.Similar to the overall trend, the incidence and intensity of CHE for all income groups experienced a substantial increase in the last 15 years of the 20 th century.For Q1, CHE incidence increased by 5.6 pp during this period, compared to increases of 5.6 pp, 5.0 pp, 4.5 pp, 3.2 pp for Q2, Q3, Q4, and Q5, respectively.Households in poorer quintiles experienced larger increases, and the wealthiest quintile group always received better financial protection in most years.After entering the 21 st century, income inequality in CHE incidence and intensity stabilized over time, remaining largely unchanged.As of 2009, the CHE incidence for Q1 was 5.8%, in contrast with 4.1% for Q5.Their difference was almost the same as the number in 2002.
Regression results from Table 4 confirm the observations above.Controlling for individual and household characteristics, we find that the four regions exhibited no difference in CHE incidence and intensity measure in 1986.Regional disparities emerged after 1992.As of 2009, regional disparities had been reduced to a large extent, though the gap between the Northeast and the East region in CHE incidence is still positive and statistically significant.Income inequality exhibits a similar story of emerging in 1992 and kept expanding until 2002.Compared to Q1, which is the poorest group, the likelihood of incurring CHE and its depth in wealthier groups were at a lower level.Even after 2002, income inequality remained almost unchanged.

Medical Impoverishment
The black lines in Figure 5 show the national trends in the incidence and intensity of MI between 1986 and 2009.Concurrent with the timing of the policy shift, we observe downward trends in the incidence and intensity of MI after 2002.
Table 5 shows the MI incidence and intensity in selected years.Results in the first two rows for headcount measure suggest that the share of households below the poverty line was constantly decreasing during the period from 1986 to 2009, regardless of including out-of-pocket health care spending in total household expenditures or not.However, the incidence of MI does not exhibit a steady downward trend.The share of urban households with MI was approximately 0.9 percent in 1986 and increased to 1.25% in 1992 and 1997.Even in 2002, the incidence of MI was more than one percent.However, China experienced a sharp reduction in the incidence of MI afterward.As of 2009, it was as low as 0.29%.The intensity of MI measured by the changes in the normalized poverty gap shows the same trend as the incidence measure.In 1986, 1992, 1997, and 2002, the normalized poverty gaps due to out of pocket health care expenditure all exceeded 0.2%, whereas it substantially declined to 0.04% in 2009.
Results on regional disparity and income inequality in MI are shown in Figure 5. Panel A and B display regional disparities.Similar to the results for CHE, the East was always better off with much lower incidence and intensity of MI throughout this period.In the 1980s and the early 1990s, the West suffered the highest incidence and intensity while Northeast China became the hardest-hit region in the mid and late 1990s.Entering the 21st century, all regions exhibit a downward trend in both incidence and intensity measures.Between 2002 and 2009, the incidence of MI in the East, Northeast, Central and the West region experienced decreases of 0.27 pp, 1.29 pp, 1.22 pp, and 0.92 pp, respectively.The greater reductions in Northeast, Central and the West regions resulted in narrowing regional gaps.The incidence gap between East China and Northeast China shrank from 1.27 pp in 2002 to 0.26 pp in 2009.The intensity gap was also narrowed, consistent with the incidence measure.
Trends of income inequality in MI are shown in Panel C and D of Figure 5.The poorest quintile is the most vulnerable group to MI.The incidence of MI in the poorest quintile fluctuated from 4% to 5.5% in the 1990s while the incidences for Q4 and   Q5 never exceeded 0.2% and 0.1% over the same period.The gap between Q1 and Q5 in MI incidence was as high as 5.4 pp at the beginning of the 21st century.However, income inequality became narrower as Q1 and Q2 experienced substantial declines in incidence and intensity measures of  Variable means with 95% confidence interval in parentheses are shown in the table.***, **, and * denote statistical significance at the 1%, 5%, and 10% level, respectively.H Pre is the share of households in which total consumption per capita (including health expenditure) is below the poverty line; H Post is the share of households in which consumption per capita excluding health expenditure is below the poverty line; P I is the incidence of medical impoverishment.NG Pre is the normalized poverty gap when including health expenditure.NG Post is the normalized poverty gap when excluding health expenditure.PI NG is the intensity of medical impoverishment measured by the change in poverty gaps.The poverty line in this study is defined as 1866 RMB per person-year in 2010 and is adjusted by the urban Consumer Price Index (CPI) for each year.Data source: Urban Household Survey 1986-2009 MI. Between 2002 and 2009, the reductions in the incidence measure for Q1, Q2, Q3, Q4, and Q5 were 3.6 pp, 0.60 pp, 0.19 pp, 0.17 pp, and 0.02 pp, respectively.As a result, the gap between Q1 and Q5 in the incidence measure shrank to 0.63 pp in 2009.The differences between any pairs of Q2, Q3, Q4 and Q5 were reduced to nearly zero.Controlling for individual and household characteristics, regression results from Table 6 show the same story on the inequity of MI as descriptive results above.

Discussion and Conclusion
The transition in China's health care system since the 1980s offers us a good chance to examine how a government-led financing or a market-oriented health care financing approach affects the performance of financial protection in the health system.Using unique data from the Urban Household Survey, this study examines the trends of CHE and MI between 1986 and 2009 in urban China.We find a turning point around 2002 for the trends of financial protection.Before 2002, the CHE incidence and intensity were both growing.Our results show that 5.07% of the urban population (25 million people) in China experienced financial catastrophe in 2002, compared to 0.26% of the urban population in 1986.Increasing regional disparities and income inequality accompanied this upward trend.After 2002, however, the upward trends in the CHE incidence and intensity and their widening inequality ceased.Instead, the CHE incidence and intensity remained largely unchanged between 2002 and 2009 with narrowing regional disparities and non-widening income inequality.
The trends for MI measures exhibited a similar turning point at around 2002.Despite admirable achievements in poverty reductions in urban China between 1986 and 2002, there was no concurrent decline in MI indicators.Nor do we observe decreasing inequality across income groups and regions.The incidence of MI was 1.00% in 2002, even slightly higher than that in 1986.It suggests that more than 5 million urban population suffered from MI in 2002.However, after 2002, both the incidence and intensity of MI exhibited downward trends, accompanied by decreasing gaps in income groups and regions.As of 2009, only approximately 1.85 million urban population were exposed to MI.All these results suggest that China's two cycles of health care reform generated significantly different outcomes in financial protection.
Our findings provide several insights into the debate between government-led and marketoriented health financing policies.First, China's experience suggests that the government needs to play a significant role in the health sector.Our results demonstrate that a laissez-faire policy that defaulted health care to self-reliance and out-of-pocket payment did not generate a desirable outcome as the economic theory of the market predicted.Instead, it resulted in unaffordable health care services, health expenditure inflation, as well as great inequality in the last 15 years of the 20 th century in China.China ranked the fourth from the bottom in terms of health finance equity in the year 2000. 35However, we observe that the deterioration in financial protection was curbed once the government became more proactive in health financing after 2002.These facts imply that public financing is a necessary condition for better financial protection.
Second, China's trends in CHE and MI measures between 1986 and 2002 suggest that improvements in financial protection do not automatically follow economic growth, though it theoretically would expand resources available in the health sector. 3 Improvements in financial protection need efficient allocations of these economic gains into the health sector.These findings also reinforce the advocacy for balancing economic developments and human developments. 36revious studies offer many examples where progress in education, health, gender equity and other human development indicators do not follow economic growth. 37,38Especially in transitional economies, gains from market-oriented reform may be accompanied by greater income inequality and shrinking government revenues, resulting in less accessible and affordable public services. 39hird, the unchanged incidence and intensity of CHE between 2002 and 2009 suggest that the increasing share of public financing alone may   not be a sufficient condition for better financial protection.During this period, the government substantially increased fiscal investments in the health sector, leading to a rapid expansion of social insurance as well as a decreasing share of out-ofpocket health spending.Despite increased fiscal investments, we do not observe a decreasing incidence of CHE.It contradicts the evidence based on cross-sectional data, which shows a negative correlation between the share of public financing and the incidence of CHE. 7,9One explanation for this trend is that China had limited progress in reforming its inefficient and wasteful delivery system.The distorted incentives constrainted China from transforming money into cost-effective health services. 40Instead, huge government investments further stimulated the inflation of health expenditure.These findings are consistent with the previous evidence, showing that the expansion of health insurance in China had limited effects on reducing the incidence of CHE. 20It implies that comprehensive policy measures that restructure incentives on the supply side should be in place.
Compared with previous studies, the incidences of CHE and MI estimated in this study are lower than those in previous studies.Results based on NHSS with the same threshold definition show the incidences of CHE and MI in urban China in 2008 were 9.0% and 3.3% respectively, much larger than the incidences in this study. 12,21These differences may be attributable to the method of data collection.Past research is inconclusive about the reliability of diary data for expenditure reporting.2][43] In household surveys such as NHSS, the total expenditure is likely to be under-reported when it is collected by a single recall question. 44,45The health expenditures may be overstated if the respondents cannot distinguish the total expenditure and out-ofpocket spending.In contrast, diary data in the UHS is likely to be more accurate because it requires the respondents to keep records of household expenditure.As for the issue of fatigue in diary keeping, the UHS attempts to minimize the problem by sending staff to visit selected households monthly or bi-weekly and by sample rotation to ensure the quality of the diary.
This study has three limitations.First, trends shown in this study should be interpreted as a reflection of the overall impact of all health policies as well as social developments during the period from 1986 to 2009.We cannot isolate the independent effect of any specific policy throughout this period.Second, due to limited data availability, we are unable to examine trends between 2010 and 2018, which is during the period of China's new round of health care reform starting in 2009.Analyzing trends in this period, if data become available, would provide richer information on the influencing factors of financial protection.Third, we only consider households incurring health care expenditure.However, there was a considerable share of households who does not seek treatment because they could not afford it.We cannot examine its trends in the same period.These are important issues to be addressed in future research.

Notes
the collection, analysis and interpretation of data; in the writing of the articles; and in the decision to submit it for publication.The incidence of medical impoverishment We measure impoverishment incidence as the change in poverty headcount with and without including out-of-pocket health spending in total consumption.Formally it could be written as follows:

equals to
where H Pre POV denotes is the incidence (Headcount) of poverty when including health spending; P Pre i equal to 1, if Expenditure per capita < Poverty line and zero otherwise.H Pre POV denotes is the incidence (Headcount) of poverty when excluding health spending; P Post i equal to 1, if Expenditure per capita À Medical Expenditure < Poverty line and zero otherwise.PI is the incidence of medical impoverishment, measured by the change in the incidence (Headcount) of poverty.

The intensity of medical impoverishment
We use the changes in poverty gaps due to health spending to measure the intensity of medical impoverishment.Formally, they could be written as follows: Where PI G is the change in poverty gaps and PI NG
(1.25 USD per person-day in 2005) and the purchasing power parity between China and the United States (4.09 RMB/USD in 2005), this study defines the poverty line as 1866 RMB per person-year in 2010 and adjusts this threshold by the urban Consumer Price Index (CPI) for each year.

Figure 2 .
Figure 2. Map of regions in the sample.Shaded areas are provinces with Urban Household Survey data that the study has access to.

Figure 3 .
Figure 3.Comparison between UHS and NBS aggregate data.Panel A compares real consumption per capita in the UHS and NBS aggregate data; Panel B compares health consumption per capita in the UHS and NBS aggregate data.Data Source: Data source: Urban Household Survey 1986-2009, National Bureau of statistics.

Figure 4 .
The incidences of CHE by alternative thresholds exhibit similar trends of continuous increase between 1986 and 2002 and statistically insignificant changes between 2002 and 2009.

Figure 4
Figure 4 also displays trends in CHE indicators by regions and income quintiles using the 40% threshold.Panel A and B in Figure4show trends in CHE incidence and intensity by regions.Regional differences were relatively small in the 1980s and the early 1990s.In the mid and late 1990s, the incidence of CHE in Northeast China increased sharply, reaching as high as 8.15% in

Figure 4 .
Figure 4.The incidence and intensity of catastrophic health expenditure (CHE) by regions and income groups, 1986-2009.Panel A and B shows the incidence and intensity of CHE by regions.Panel C and D shows the incidence and intensity of CHE by income groups.Data source: Urban Household Survey 1986-2009, National bureau of statistics.

Figure 5 .
Figure 5.The incidence and intensity of medical impoverishment by regions and income groups, 1986-2009.Panel A and B shows the incidence and intensity of medical impoverishment by regions.Panel C and D shows results by income groups.Data source: Urban Household Survey 1986-2009, National bureau of statistics.

Table 2
shows summary statistics for demographic variables and household expenditure in the UHS.Household income and expenditures experienced tremendous increases between 1986 and 2009, reflecting China's achievements in economic developments.The average household size declined from 3.76 person in 1986 to 2.84 person in 2009.The average age of household heads increased from 42.68 years old in 1986 to 49.45 years old in 2009, reflecting population aging in urban China.
Variable means and standard deviations are reported in this table.Income and expenditure variables are not adjusted by CPI.Data source: Urban Household Survey 1986-2009

Table 3 .
The incidence and intensity of catastrophic health expenditure in urban China, 1986-2009.

Table 4 .
Regression results on catastrophic health expenditure.
**, **, and * denote statistical significance at the 1%, 5%, and 10% level, respectively.Q1 and the East region are used as reference groups.Age_5 =1 if there is at least one child aged below 5 in the household.Age_5_14 = 1 if there is at least one child aged between 5 and 14.Age_65 = 1 if there is at least one family member aged above 65.We include age, age squared, marital status, education level, gender of the household head and household size as controls.Robust standard errors are reported in parentheses.Data source: Urban Household Survey 1986-2009. *

Table 6 .
Regression results on medical impoverishment.
**, **, and * denote statistical significance at the 1%, 5%, and 10% level, respectively.Q1 and the East region are used as reference groups.Age_5 =1 if there is at least one child aged below 5 in the household.Age_5_14 = 1 if there is at least one child aged between 5 and 14.Age_65 = 1 if there is at least one family member aged above 65.We include age, age squared, marriage status, education level, gender of the household head and family size as controls.Robust standard errors are reported in parentheses.Data source: Urban Household Survey 1986-2009. * CHE treshold and zero otherwise; N is the sample size.
is the change in normalized poverty gaps; g Pre i denotes the poverty gap when including health spending, which equals to Total consumption per capita À Poverty line if Total consumption per capita < Poverty line and zero otherwise.gPosti denotes the poverty gap when excluding health spending, which equals to Expenditure per capita without health spending À Poverty line if Expenditure per capita without health spending < Poverty line and zero otherwise.G PostPOV and G Pre POV denote the average post-payment poverty gap and the average pre-payment poverty gap.PL denotes the poverty line.