Financial conditions, health care provision, and patient outcomes: Evidence from Chinese public hospitals

Journal Pre-proof Financial conditions, health care provision, and patient outcomes: Evidence from Chinese public hospitals Mengna Luan, Xiang Shao, Fengman Dou

PII: DOI: Reference:

S0165-1765(19)30442-2 https://doi.org/10.1016/j.econlet.2019.108876 ECOLET 108876

To appear in:

Economics Letters

Received date : 14 August 2019 Revised date : 26 November 2019 Accepted date : 27 November 2019 Please cite this article as: M. Luan, X. Shao and F. Dou, Financial conditions, health care provision, and patient outcomes: Evidence from Chinese public hospitals. Economics Letters (2019), doi: https://doi.org/10.1016/j.econlet.2019.108876. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

© 2019 Published by Elsevier B.V.

Journal Pre-proof Using a comprehensive dataset of 117 Chinese public hospitals, the paper tests how financial conditions affect the provision of health care and patient outcomes. We find Chinese Zero Markup Drug Policy (ZMDP) reduces the profits that hospitals obtain from prescribing and dispensing drugs.

pro of

Hospitals also increase revenues from imaging tests after ZMDP.

Jo

urn a

lP

re-

We find hospitals increase capital expenditure on medical equipment but find no effect of the policy on health care quality.

Journal Pre-proof

Financial Conditions, Health Care Provision, and Patient Outcomes:

∗

Mengna Luan

pro of

Evidence from Chinese Public Hospitals

Xiang Shao

Fengman Dou

November 29, 2019

re-

Abstract

Jo

urn a

lP

This paper investigates the impact of health care providers' nancial conditions on the provision of health care, using a policy shock in China  Zero Markup Drug Policy (ZMDP)  which reduces the prots that hospitals obtain from prescribing and dispensing drugs. The study utilizes a comprehensive data set of 117 public hospitals in a major Chinese city from 2007 to 2015 and adopts the dierence-in-dierences identication strategy based on the staggered adoption of ZMDP. Our ndings show that the policy puts nancial pressure on hospitals, as their revenue from prescription drugs per patient decreases signicantly. In response to the nancial shock, hospitals increase revenues from other treatments and procedures per patient, such as imaging tests. Moreover, as hospital revenues become more dependent on medical tests after the policy shock, hospitals increase their capital expenditure on medical equipment. Nevertheless, the study nds no eect of ZMDP on the quality of health care.

∗ Mengna Luan: Southwestern University of Finance and Economics, [email protected]; Xiang Shao: Fudan University, [email protected]; Fengman Dou (corresponding author): Chengdu Institute for Medical Information, [email protected]. Mengna Luan is grateful for the support by Research Seed Fund from Southwestern University of Finance and Economics. Xiang Shao is grateful for the support by Shanghai Pujiang Program (No.2019PJC014).

1

Journal Pre-proof

1

Introduction

Financing health care has been increasingly challenging even in some of the world's richest countries. With mounting pressure on public expenditure, governments in developed and developing economies are adopting various cost-control policies, ranging from payment reform to enhancement of competition. However, these

pro of

cost-control methods are probably aecting health care providers' nancial conditions. Hence, it is of critical importance to investigate whether and how nancial conditions aect the provision of health care and patient outcomes. This investigation is particularly urgent, given that patients face great diculty in assessing the necessity and the quality of the health care services they receive.

We study the impact of nancial conditions on health care provision in the context of the hospital industry in China for two reasons. First, Chinese hospitals, including public ones, receive little in direct

re-

government subsidies and need to make ends meet. Therefore, self-sucient hospitals are sensitive and responsive to negative shocks to their nancial condition. Second, in 2012, the Chinese government implemented the Zero Markup Drug Policy (ZMDP) that had a large negative eect on public hospitals' nancial health. In the pre-ZMDP period, Chinese hospitals could legally make prots by prescribing and

lP

dispensing drugs. On the one hand, hospitals had a strong nancial incentive to prescribe drugs because of the allowed 15 percent prot margin (Fu, Li, and Yip, 2018); on the other hand, the government sets the fees for many medical services below the actual costs (Yip and Hsiao, 2008). Consequently, prescription

urn a

drug sales became the largest source of revenue for hospitals. Through phasing out the previously allowed markup, ZMDP substantially reduced hospitals' revenues from prescription drug sales. Using a comprehensive data set of 117 public hospitals in a major city in China from 2007 to 2015, we take advantage of the roll-out of ZMDP in the city and implement a dierence-in-dierences (DD) identication framework to study the impact of nancial conditions on hospitals. We nd a decrease of 8.38% in prescription drug expenses per patient and a decrease of 4.30% in the proportion of prescription

Jo

drug sales in the hospitals' total revenues after ZMDP. However, we do not nd a signicant decrease in total medical expense per patient after ZMDP. Hospital revenues from other treatments and procedures increased to compensate the removed prots from prescription drug sales. Specically, in response to the negative nancial shock caused by ZMDP, the hospitals increased imaging test expenses per inpatient. We also nd that the hospitals increased capital investment proxied by the total value of medical equipment after ZMDP, as increased medical tests required more diagnostic equipment. Finally, we nd no eect of ZMDP on the quality of care proxied by the emergency department mortality rate, in-hospital all-cause

Journal Pre-proof

mortality rate, and the nurse/doctor-to-bed stang ratios. Our work contributes to the growing literature on the impact of nancial conditions on hospitals. For example, Adelino et al. (2018) and Dranove et al. (2017) nd that there was a drop in hospitals' capital investment and their provision of service-related community benets following the sharp reductions in

pro of

their assets caused by the 2008 stock market collapse. Our study diers by showing that, when facing a negative shock to their nancial condition, hospitals increase treatments or procedures that are more protable. Such response (shifting from prescription drug to other services) is similar to cost shifting (from one group of patients to another group) in hospitals in the literature (e.g., Zwanziger and Bamezai, 2006).1 Moreover, Fu, Li, and Yip (2018) study ZMDP and nd that physicians are incentivized to increase expenditures for medical services at county hospitals in China. With the help of a more comprehensive data set, ours is the rst paper to not only study ZMDP's direct eect on drug expenses, but also assess

re-

the impact of ZMDP on hospitals' cost shifting behaviors between treatments (procedures) and capital expenditure, and patient outcomes.

Data and Empirical Strategy

lP

2

2.1 Data

Our sample covers all 117 public general hospitals in a major Chinese city during 2007-2015.2 The data

urn a

are collected and audited annually by the Health and Family Planning Commission of the city. In this study, we focus on public hospitals for two reasons. First, public hospitals' fee schedules are regulated by the government and are potentially subject to the negative nancial shock caused by ZMDP.3 Second, private hospitals may play a dierent role in medical expenditure compared with their public counterparts (Liu et al., 2009).

We have information on individual hospital characteristics (e.g., hospital grade). We also observe

Jo

the nancial performance of each hospital, including total and itemized revenues. Moreover, we have information on hospital operations, including inputs such as doctors, nurses, and medical equipment; outputs such as the numbers of outpatient visits and inpatient admissions; and medical performance measures of quality. Based on the hospital-level data, we rst construct a pair of dependent variables to illustrate how drug prescriptions were directly aected by the policy shock. We then construct a battery 1

See more detailed discussions on cost-shifting behaviors in online Supplementary Appendix B. In 2014, there were 40 million patient visits and RMB 26 billion in revenue in the sample city, which has a population of 14 million. 3 In contrast, private hospitals have much more exibility in pricing and they are not subject to ZMDP. 2

Journal Pre-proof

of outcome variables to reect hospitals' responses in the provision of protable services (i.e., imaging tests and medical consumables), capital investment, and the quality of care. Table 1 provides detailed

2.2 Estimation Framework

pro of

information on variable construction, and Table 2 presents the descriptive statistics of these variables.

ZMDP had a large negative impact on nancial conditions of hospitals in the sample city. The hospitals had relied heavily on drug prescriptions, with drug sales being the largest source of hospital revenue (40%).4 To identify the impact of the nancial shock, we utilize the staggered adoption of ZMDP. Among 117 public hospitals, 36 were required to adopt ZMDP by the government in 2013, and another 9 were required to adopt the policy in 2014. The remaining 72 hospitals serve as control group. Specically, we

re-

estimate the following dierence-in-dierences model:

yi,t = βZM DPi,t + γi + δt + θt,g + εi,t

(1)

where yi,t is our hospital performance measure in year t for hospital i. ZM DPi,t equals one if hospital i

lP

was subjected to ZMDP in year t, and zero otherwise. γi and δt are the hospital and year xed eects, respectively. θg,t is the grade-year cross xed eects. εi,t is the error term. Robust standard errors are clustered at the administrative district level.

urn a

Our empirical strategy allows us to address the following concerns in identifying the eects of ZMDP. First, there may be concern that hospital-level dierences such as location or bureaucratic aliation may be correlated with hospital behaviors. We control for hospital xed eects, so time-invariant dierences across hospitals cannot drive our estimation result. Second, there may be concern that factors which aect the evolution of health care development might bias our estimation. Therefore, we control for time trends of health care development by adding year xed eects. Third, confounding policy changes, such as the

Jo

insurance policy change in 2009, may have dierential eects on hospitals of dierent grades.5 Such policy changes could also bias our results; therefore, by including grade-year cross xed eects, we control for time-varying variation across hospitals of dierent grades.6 4

See Figure SA1 for the revenue breakdown of the sample hospitals. Patient cost sharing was reduced asymmetrically among hospital grades in 2009, which made it cheaper for patients to go to lower-grade hospitals. This insurance policy change generated faster growth in patient volume for lower-grade hospitals. 6 Although time varying factors at hospital level, such as the numbers of doctors and beds, do not aect the randomization of policy assignment, we also run the tests with these control variables. The results are qualitatively the same. 5

Journal Pre-proof

3

Results

3.1 Eects on Patient Medical Expenses and Hospitals' Capital Expenditure We construct two measures to study the eect of ZMDP on drug prescriptions. The rst measure is the

pro of

share of prescription drug in total medical expenses, which proxies the hospital's dependence on drug sales. The second measure is drug expenses per patient. Column 1 in Table 3 shows that after ZMDP is implemented, the proportion of drug sales in hospital revenue decreases by 0.044 log points, that is, a -4.3% (e−0.044 -1) change in drug sales share. Similarly, column 2 in Table 3 shows that ZMDP reduces drug expenses per patient by 0.0875 log points, that is, a -8.38% (e−0.0875 -1) change in drug expenses per patient. The average drug expenses per patient in our sample is 194.46 RMB; therefore, the economic magnitude for average drug expense is -16.29 RMB (194.46*8.38%), which is about 2.65 in 2014 USD

re-

value. The results conrm that ZMDP delivers its policy goal and signicantly decreases patient drug expenses. The two results remain statistically signicant after we jointly adjust the p-values for multiple testing based on Romano-Wolf stepdown p-values.

lP

We further examine total medical expenses. Column 1 in Table 4 shows that the eect of ZMDP on total medical expenses per patient is not statistically signicant. We suspect that hospitals turn to other treatments or procedures to compensate the loss of drug revenue. However, to help compensate the loss of drug revenue, the government also adjusted the prices of some medical services, which makes it dicult

urn a

to identify whether hospitals have increased the quantity per patient of these medical services, as we only have revenue data. Therefore, we focus on two medical expenses for which the prices were not adjusted after the policy: imaging tests and medical consumables. First, we examine the eect of the nancial shock by ZMDP on imaging test expenses (e.g., computed tomography imaging). Expenses for imaging tests and prescription drugs have long been viewed as the major sources of protability in the pre-ZMDP period for hospitals in China. If the imaging test expenses

Jo

per patient of the treated hospitals increase compared with those of the non-treated hospitals, this would indicate that ZMDP has a positive eect on the intensity of imaging tests. Columns 2 and 3 in Table 4 show that the imaging test expenses per patient increase after ZMDP, especially in the inpatient sector. This implies that when hospitals face a reduction in revenue from prescription drugs, they tend to compensate the loss by increasing the intensity of imaging tests. We also nd that the value of equipment increases after ZMDP, which indicates hospitals increase investments in equipment. This is consistent with our ndings that hospitals try to make more money from imaging tests after ZMDP.

Journal Pre-proof

Second, we examine hospital revenue from medical consumables. Consumables includes disposable medical supplies and surgical instruments, such as medical gauze and vascular stents used in surgeries. Column 5 and 6 in Table 4 show that the expenses for consumables per patient increase signicantly after ZMDP, especially in the inpatient sector, which relies more on revenue from consumables. However, the

pro of

results should be taken with caution as it is not statistically signicant after we adjust the p-values for multiple testing of 6 hypotheses in Table 4 based on Romano-Wolf stepdown p-values. Last, we check the validity of our result with the parallel trend test and the random assignment placebo test. See online Supplement Appendix A for details.

3.2 Eects on Hospital Quality

A crucial issue in health care policy is its consequences for the quality of care. Therefore, we test whether

re-

ZMDP aects hospital quality through imposing nancial pressure on hospitals. We use two measures of mortality: mortality rate in the emergency department, which is similar to the quality measure used by Bloom et al. (2015), and all-cause in-hospital mortality rate. We test the same empirical model as in

lP

Equation 1. Panel A in Table 5 shows no evidence that ZMDP aects hospital mortality rates, whether it is the emergency department mortality rate or the all-cause mortality rate. The coecients of ZMDP in columns 1 and 2 are both statistically insignicant.7

We also run the tests using some other quality measures related to hospital operations. In columns 1

urn a

and 2 in panel B in Table 5, we use nurse-to-bed and doctor-to-bed ratios as hospital quality measures concerning the stang level, following Lin (2015) and Lu and Lu (2017). The results are also statistically insignicant, which is consistent with our previous results with the mortality rates.

4

Conclusion

Jo

Using the data on Chinese hospitals, this paper documents that hospitals increase revenue from other treatments and procedures when facing a drop in prescription drug sales as a nancial shock. The paper shows that under the current fee-for-service model in the Chinese health care system, price control is not an eective tool to deal with the rising medical expenditure. Policy makers should be aware that, given the information asymmetry between heath care providers and patients, hospitals can always nd other ways to compensate their revenue loss due to price control. 7 The p-values in Table 5 are not adjusted for multiple testing based on Romano-Wolf stepdown p-values, as the original p-values are not signicant.

Journal Pre-proof

References [1] Adelino, M., Lewellen, K. and McCartney, W.B., 2018. Financial condition and product quality: the case of nonprot hospitals. Working paper.

pro of

[2] Bloom, N., Propper, C., Seiler, S., Van Reenen, J., 2015. The impact of competition on management quality: evidence from public hospitals, The Review of Economic Studies, Vol.82, pp.457~489. [3] Dafny, L.S., 2005. How do hospitals respond to price changes? American Economic Review, 95(5): 1525-1547.

[4] Dranove, D., Garthwaite, C. and Ody, C., 2017. How do nonprots respond to negative wealth shocks? The impact of the 2008 stock market collapse on hospitals. The Rand Journal of Economics, 48(2),

re-

pp.485-525.

[5] Fu, H., Li, L. and Yip, W., 2018. Intended and unintended impacts of price changes for drugs and

lP

medical services: evidence from China, Social Science & Medicine, 211, 114-122. [6] Lin, H., 2015. Quality choice and market structure: A dynamic analysis of nursing home oligopolies. International Economic Review, 56(4), 1261-1290. [7] Liu, G.G., Li, L., Hou, X., Xu, J. and Hyslop, D., 2009. The role of for-prot hospitals in medical

urn a

expenditures: evidence from aggregate data in China. China Economic Review, 20(4), pp.625-633. [8] Lu, S.F. and Lu, L.X., 2016. Do mandatory overtime laws improve quality? Stang decisions and operational exibility of nursing homes. Management Science, 63(11), pp.3566-3585. [9] Yip, W. and Hsiao, W.C., 2008. The Chinese health system at a crossroads. Health Aairs, 27(2),

Jo

pp.460-468.

[10] Zwanziger, J. and Bamezai, A., 2006. Evidence of cost shifting in California hospitals. Health Aairs, 25(1), pp.197-203.

Journal Pre-proof Table 1 Variable Definitions Variable Name

Description Equals one after hospitals started to be subject to ZMDP. 36 out

ZMDP

of 117 hospitals were treated in 2013 and 9 were treated in 2014. The proportion of drug sales in total hospital revenue

Drug_PerPatient

Logarithm of drug sales (in thousands) per patient

TotalExpense_PerPatient

Logarithm of total medical expense (in thousands) per patient

Imaging_PerPatient

Logarithm of imaging test expenses (in thousands) per patient

(Outpatient)

for outpatients

Imaging_PerPatient (Inpatient)

Logarithm of imaging test expenses (in thousands) per patient for inpatients

Consum_PerPatient(Outpatient)

pro of

Drug_Share

Logarithm of total medical consumables expense (in thousands) per patient for outpatients (available from 2012)

Consum_PerPatient(Inpatient)

Logarithm of total medical consumables expense (in thousands) per patient for inpatients (available from 2012)

Logarithm of hospital equipment value (in ten thousands)

Mortality_Emergency

Mortality rate in Emergency Department

Mortality_All

In-hospital mortality rate for all causes (available from 2011)

Nurse_Bed

The ratio of number of nurses to number of beds

Doc_Bed

The ratio of number of doctors to number of beds

lP

re-

Equipment_Value

Table 2 Descriptive Statistics (1)

(2)

(3)

(4)

(5)

N

Mean

Std

Min

Max

809

0.135

0.342

0

1

Drug_Share

777

0.441

0.108

0.144

0.816

Drug_PerPatient

777

-1.885

0.764

-6.145

0.459

TotalExpense_PerPatient

783

-1.039

0.728

-4.365

1.480

ImgFee_PerPatient (Outpatient)

795

-3.981

1.078

-9.134

-1.810

ImgFee_PerPatient (Inpatient)

771

-1.166

0.923

-5.918

1.785

Consum_PerPatient (Outpatient)

291

-5.605

1.390

-11.845

-2.667

Consum_PerPatient (Inpatient)

299

-0.751

1.145

-4.133

1.914

Equipment_Value

786

7.178

1.785

1.552

11.99

Mortality_Emergency

706

0.00138

0.00322

0

0.0371

Mortality_All

392

0.00449

0.00389

0

0.0314

Nurse_Bed

802

0.478

0.162

0.028

1.171

Doc_Bed

802

0.391

0.147

0.020

1.650

Variables

Medical Expenses

Medical Consumables

Quality

Jo

Capital Investment

ZMDP (dummy)

urn a

Measure of Financial Conditions

Note: There are fewer observations for some variables due to data constraints (e.g., missing data and/or a shorter sample period available).

Journal Pre-proof

Table 3 Effect on Prescription Drugs

Drug_Share

Drug_PerPatient

(1)

(2)

-0.044***

-0.0875**

(-5.380)

(-2.281)

ZMDP

0.000

Fixed effects

Hospital，Year

Hospital Grade*Year Fixed Effect

Yes

Observations

777

R-squared

0.771

Mean in Monetary Value

-

re-

Romano Wolf p-value

pro of

VARIABLES

Change in Monetary Value

-

0.036

Hospital，Year Yes 777

0.802

¥ 194.49

- ¥ 16.29

Jo

urn a

lP

Note: This table reports the difference-in-differences estimates of the effects of ZMDP on drug prescriptions. The dependent variable in column 1 is the proportion of drug sales in total hospital revenue and the dependent variable in column 2 is the logarithm of prescription drug expense per patient. The last two rows report the change in monetary value (RMB) with respect to the mean of original monetary value. Standard errors are clustered at the district level, and the t-statistics are reported in parentheses. ** and *** indicate significance at the 5% and 1% level for Romano Wolf pvalue, respectively.

0.952

Hospital，Year

Yes 803 0.813 ¥27.91

0.932

Hospital，Year

Yes

783

0.935

¥ 452.53

¥433.80

0.772

698

Yes

¥7.00

0.860

291

Yes

Hospital，Year

0.952

(0.376)

0.0730

(5)

(Outpatient)

Consum_PerPatient

pro of

¥52,830,980

0.942

786

Yes

Hospital，Year

0.02

(3.383)

0.317**

(4)

Equipment_Value

re-

Hospital，Year

0.02

(3.080)

0.320**

(3)

(Inpatient)

Imaging_PerPatient

lP

(0.310)

(0.380)

(2)

0.234

(1)

0.0272

(Outpatient)

¥814.18

0.914

299

Yes

Hospital，Year

0.207

(1.650)

0.260

(6)

(Inpatient)

Consum_PerPatient

+¥ 12.47 +¥7.36 +¥163.60 +¥19,706,091 +¥0.53 +¥241.75 Note: In column 1, the dependent variable is the logarithm of the total medical expenses per patient. In columns 2 and 3, the dependent variables are the logarithm of imaging test expenses per outpatient and the logarithm of imaging expenses per inpatient, respectively. In column 4, the dependent variable is the logarithm of the total value of medical equipment. In columns 5 and 6, the dependent variables are the logarithm of medical consumables expenses per outpatient and the logarithm of medical consumables expenses per inpatient (only available from 2012), respectively. The last two rows report the change in monetary value (RMB) with respect to the mean of original monetary value. Standard errors are clustered at the district level, and the t-statistics are reported in parentheses. ** indicate significance at the 5% level for Romano Wolf p-value.

Romano Wolf pvalue Fixed effects Hospital Grade*Year Fixed Effect Observations R-squared Mean in Monetary Value Change in Monetary Value

ZMDP

Imaging_PerPatient

PerPatient

urn a

Jo

VARIABLES

TotalExpense_

Table 4 Effects on Medical Revenue and Capital Expenditure

Journal Pre-proof

Journal Pre-proof

Table 5 Effect on Hospital Quality Panel A Mortality_Emergency

Mortality_All

VARIABLES (1)

(2)

0. 000291

-0. 000248

(0.5)

(-0.66)

Fixed effects

Hospital，Year

Hospital，Year

Hospital Grade*Year Fixed Effect

Yes

Yes

Observations

706

392

R-squared

0.567

0.815

re-

Panel B

pro of

ZMDP

Nurse_Bed

Doc_Bed

(1)

(2)

0.0159

0.00165

(1.55)

(0.55)

Fixed effects

Hospital，Year

Hospital， Year

Hospital Grade*Year Fixed Effect

Yes

Yes

Observations

802

802

lP

VARIABLES

urn a

ZMDP

R-squared

Jo

0.587 0.276 Note: In Panel A, the dependent variables in columns 1 and 2 are Emergency Department mortality and all-cause mortality (only available from 2011), respectively. In Panel B, the dependent variables in columns 1 and 2 are nurse-to-bed staffing ratio and doctor-to-bed staffing ratio, respectively. Standard errors are clustered at the district level, and the t-statistics are reported in parentheses.