China Economic Review

Transcription

1 China Economic Review 22 (2011) Contents lists available at ScienceDirect China Economic Review The productive efficiency of Chinese hospitals Ying Chu NG Department of Economics, Hong Kong Baptist University, Kowloon Tong, Kowloon, Hong Kong, China article info abstract Article history: Received 23 November 2010 Received in revised form 6 April 2011 Accepted 1 June 2011 Available online 12 June 2011 JEL classification: I10 D22 L30 Keywords: Productive efficiency Data envelopment analysis Malmquist index Hospitals China Since the implementation of health care reforms, much has been documented regarding the practices and behavior of hospitals or health profession in China. While these documentations pointed to various problems arising from the reforms, a systematic analysis is rarely found. Based on five years' data, the present study aims at providing empirical evidence of inefficiency of hospitals in China. Using the data envelopment analysis, the sources of inefficiency were examined. Echoing the unnecessary care, over-prescription of drugs and the adoption of hightech treatments since the implementation of health care reforms, the sampled hospitals were found quite inefficient and pure technical inefficiency played a dominant role in driving the inefficiency of hospitals. Combining the panel nature of the data and the Malmquist Index computation, hospitals had experienced productivity growth between 2004 and Mirroring the behavior of hospitals, technological progress was the underlying force for the growth and the deterioration in efficiency change was found. The results summarized by region revealed that the stage of economic development and the efficiency performance of hospital did not necessarily go hand in hand Elsevier Inc. All rights reserved. 1. Introduction In parallel with its industrial and enterprise reforms, the Chinese government has implemented health care reform since Total national expenditure on health care was 3.17% of GDP in 1980 (Wong, Lo, & Tang, 2006) and 3.86% in 1995 (Liu, 2004). By 2002, as reported in The World Bank (2005), the percentage was 5.8% and the figure for 2004 was 5.5% (Hougaard, Osterdal, & Yu, 2008). At the same time, the government reduced its financial support for health care provision. As a percentage of total health spending, the government contributed 30% in the late 1980s, but the contribution had dropped to 15% by 2002 (Yip & Mahal, 2008). By 2005, the figure was 17.9% (Huang, Liang, Chu, Rutherford, & Geng, 2009). It is thus not surprising to find that out-ofpocket health care payments rose from 21% of the total in 1990 to nearly 60% in 2002 (Yip & Hsiao, 2008; Yip & Mahal, 2008). As reported by Eggleston, Wang, and Rao (2008), the percentage was still as high as 50% in Liu and Mills (2002) have pointed out that with the substantial cut in health care support from government sources, health care institutions in China have become revenue maximizers. In a comprehensive review of China's health care system, Eggleston, Li, Meng, Lindelow, and Wagstaff (2008) and Wagstaff, Yip, Lindelow, and Hsiao (2009) have recently identified the following issues: (1) inefficiency, (2) provision of unnecessary care, (3) poor quality, (4) excessive focus on drugs and high-tech care, (5) insufficient focus on public health, (6) rapid cost escalation, and (7) larger share of health spending financed out-of-pocket. During the course of reform, hospitals in China have been subject to (1) autonomization, (2) corporatization, and (3) privatization (Gu & Zhang, 2006). For privatization, hospitals are classified into for-profit or non-profit. By 2000, about 16% of hospitals nationwide were registered as for-profit(eggleston, Li, et al., 2008). As pointed out by Guo (2003), without fundamental changes in ownership and management style, such classification had little impact on hospital efficiency. Also, most private hospitals are rather small in size (Gu & Zhang, 2006). address: ycng@hkbu.edu.hk X/$ see front matter 2011 Elsevier Inc. All rights reserved. doi: /j.chieco

2 Y.C. Ng / China Economic Review 22 (2011) In pursuing autonomization and corporatization, hospitals have been granted greater autonomy to generate, retain and manage surplus revenue. In their assessment of China's health reform, Yip and Hsiao (2009) remarked that health care institutions are public in terms of ownership while they are private, for-profit in terms of behavior. It is well documented that hospitals are tempted to make up the fall in government subsidies by over-prescribing drugs and tests, adopting high-tech medical treatments, and delivering unnecessary medical care (Eggleston, Li, et al., 2008; Eggleston, Wang, & Rao, 2008; Herd, Hu, & Koen, 2010; Ma, Lu, & Quan, 2008; Meng, Sun, & Hearst, 2002; Wagstaff et al., 2009; Wong et al., 2006; Yip & Hsiao, 2008, 2009). Given the revenuebased bonus system, doctors in public hospitals have strong incentive to prescribe greater quantity or more expensive drugs (Meng et al., 2005). These authors also pointed out that drugs have become a major source for financing hospitals and the most profitable fee item in hospitals. Wong et al. (2006) report that drug income contributes about 47% of the income of a general hospital. In 38 government hospitals surveyed between 1985 and 1999, their investment in high-value equipment as a percentage of fixed assets was 36% by 1999 (Eggleston & Yip, 2004). Ma et al. (2008) report that by 2005, imported medical equipment accounted for more than 60% of large hospitals' fixed assets. Similarly, a survey of 22 hospitals in Zibo and 19 hospitals in Nantong indicated that the sale of drugs contributed 46% to 50% of each hospital's income in the 1990s (Meng et al., 2004). The total value of high-tech equipment in these two groups of hospitals was 120 to 130 million yuan. Liu, Liu, and Chen (2000) concluded that about 30% to 40% of drug consumption is inappropriately prescribed in an attempt to generate revenue for the hospital. It seems clear that any improvements in health care efficiency and quality available from market and economic incentives may be undermined by this revenue generation imperative (Ma et al., 2008). While much have been documented regarding the practices adopted by hospitals, there has been little prior work on the assessment of productivity of health care providers in China. This study was designed to fill that gap by analyzing the efficiency of a sample of Chinese hospitals. In parallel with the behavioral change of Chinese health care providers, the panel nature of the sample data also allowed the examination of hospital productivity change over time during the health care reform period. 2. Recent work on hospital efficiency and productivity growth In a Western context, health care efficiency studies by Hollingsworth (2003, 2008) and Worthington (2004), and DEA-based efficiency studies by O'Neill, Rauner, Heidenberger, and Karus (2008) and Emrouznejad, Parker, and Tavares (2008) provide a comprehensive efficiency baseline covering recent decades. In the United States, Brown and Pagán (2006), Harrison and Sexton (2006), and Harrison, Coppola, and Wakefield (2004) have analyzed hospitals nationwide in the 1990s using input-based data envelopment analysis (DEA) yielding efficiency scores of 0.68 to Ferrier, Rosko, and Valdmanis (2006) examined 170 hospitals in Pennsylvania in 2002 with output-oriented DEA and found that most inefficiency was attributed to pure technical inefficiency. The study by Chen, Hwang, and Shao (2005) concluded that there was a declining trend in overall technical efficiency for 89 acute care hospitals in California for the period and the fall in efficiency was attributed solely to a deterioration in pure technical efficiency. Examining DEA efficiency studies of hospitals in European countries published since 2000, again the majority of them 1 adopted input-based DEA calculations. The major exception was a study by Barbetta, Turati, and Zago (2007). In these studies, various types of staff and the number of beds (as a proxy for capital) were the inputs used in the efficiency estimation. In terms of outputs, inpatient and outpatient care (in days or cases), emergency room treatments, or specific hospital services were commonly used. These scholars adjusted the value of output with a case-mix factor whenever the data was available. Most European studies have concluded that the hospitals studied were quite inefficient in general. For example, the yearly efficiency scores for 48 Norwegian hospitals before reforms in the period were around 0.78; after the reforms ( ) the range was still 0.80 to 0.82 (Biørn, Hagen, Iversen, & Magnussen, 2003). Using two different country samples in two different periods, Steinmann, Dittrich, Karmann, and Zweifel (2004) reported a declining trend in efficiency. In Germany between 2000 and 2002, the average efficiency score fell from 0.82 to In Switzerland, efficiency fell from 61.34% in 1997 to 56.88% in 2000 based on a sample of 38 Swiss hospitals. Output-based constant returns to scale efficiency scores for hospitals in four districts of the Ukraine revealed that these hospitals were adjudged capable of producing 30% more output in 1997, but only 11% more by 2001 (Pilyavsky & Staat, 2006). In the developing world, three studies have addressed technical and scale efficiency in African hospitals. Kirigia, Emrouznejad, and Sambo (2002) found that 26% of the 54 district level hospitals in Kenya were technically inefficient, and their average efficiency score was 0.84 with a scale inefficiency estimate of A study relying on a relative small sample of 17 hospitals in Ghana in 2000 found that they suffered from serious pure technical inefficiency together with obvious scale inefficiency (Osei et al., 2005). The yearly overall efficiency scores computed between 1997/98 and 2000/01 for 26 district hospitals in Namibia revealed that about 62.7% to 74.3% of the actual inputs would have been sufficient to produce the actual output had these hospitals been efficient (Zere et al., 2006). On the other hand, a study by Hu and Huang (2004) of 80 hospitals with at least 250 beds in Taiwan in 2001 indicated that they were performing relatively well, using 92.7% of the inputs to produce the existing outputs had they been perfectly technically efficient, or 96.5% with perfect scale efficiency. Pilyavsky and his co-authors (Pilyavsky et al., 2006) took advantage of the political differences between eastern and western Ukraine to examine trends in regional differences among 65 hospitals. Their output-based efficiency estimates indicated that between 1997 and 2001, hospitals located in the east did not show any change in efficiency, but there was a 13.7% fall in efficiency for hospitals in the west. Another study of Taiwan hospitals by Chang, Cheng, and Das (2004) concluded that private regional and district hospitals were relatively more efficient than their 1 See Biørn et al. (2003), Gannon (2005), Hofmarcher, Paterson, and Riedel (2002), Kontodimopoulos, Bellali, Labiris, and Niakas (2006), Rebba and Rizzi (2006), Staat (2006), Steinmann and Zweifel (2003), and Steinmann et al. (2004).

3 430 Y.C. Ng / China Economic Review 22 (2011) public counterparts using 1996 and 1997 data. The input-based analysis of 92 public hospitals in Thailand revealed an average overall efficiency scores between 0.79 and 0.85 (Watcharariroj & Tang, 2004). Among the productive efficiency studies of health care services reviewed by Hollingsworth (2003, 2008), Worthington (2004), O'Neill et al. (2008) and Emrouznejad et al. (2008), there are only a handful which studied hospital productivity growth. Using panel data on 43 acute care hospitals in Finland for the period , Linna (2000) found productivity growth in the and periods. It was mostly attributable to technological change, whereas efficiency change played a minor role. Using an inputbased Malmquist index computation with two outputs and three inputs, Sommersguter-Reichmann (2000) revealed that the productivity growth of 22 Austrian hospitals in the late 1990s was based on a combination of technological improvement and a slight decline in efficiency. The efficiency decline originated from a decline in scale efficiency accompanied by a mild improvement in pure technical efficiency. Similar observations were reported by McCallion, Glass, Jackson, Kerr, and McKillop (2000) in their analysis of 23 Northern Ireland hospitals between 1986 and Also, a study of panel data on 75 Scottish hospitals for period 1991/2 to 1995/6 revealed productivity growth in general, but decomposition of the Malmquist index indicated that improvement in efficiency contributed most of the growth (Maniadakis & Thanassoulis, 2000). Scale efficiency played a minimal role. In contrast with most European studies, a Malmquist index computation indicated a 17% decline in the productivity of a sample of non-academic acute care hospitals in South Africa between 1992/3 and 1996/7. The source of the productivity decline was mainly technological regression, with year-to-year variations in efficiency (Zere, McIntyre, & Addison, 2001). A more recent study of African hospitals can be found in Kirigia, Emronznejad, Cassoma, Asbu, and Barry (2008). The municipal hospitals they studied in Angola were found to have a productivity growth of 4.5% on average between 2000 and The growth was attributed to improved efficiency. A study of hospitals from four districts in the Ukraine in the late 1990s concluded that on average the hospitals achieved only 1% growth in productivity over the period (Pilyavsky & Staat, 2006). An output-based Malmquist index decomposition indicated that those Ukrainian hospitals experienced technological regression but improved their efficiency over time. 3. Methodology and data 3.1. Methodology Data envelopment analysis is a technique commonly applied in efficiency studies. It is widely used because (1) no specific functional form is required in formulating the production function; (2) multiple-input, multiple-output analysis is allowed; and (3) the analysis does not require price information for technical efficiency analysis. DEA was the method chosen for this study of hospital efficiency in China. Since hospital management generally has greater control over inputs than over outputs, input-based DEA was applied. Suppose that there are a total of J hospitals and each of them uses N inputs to produce M outputs. Let Y j =(y 1j,y 2j,, y Mj ) and X j =(x 1j, x 2j,, x Nj ) be the observed output and input vectors of the jth hospital, respectively. The overall Farrell input-oriented technical efficiency measure (TE) can be computed using a typical nonparametric frontier, TE j = minλ subject to i z i y mi y mj ðm =1; ;MÞ i z i x ni λ j x nj ðn =1; ; NÞ i z i =1: ðz i N =0Þ: ð1þ Assume that the technology behind these relationships exhibits variable returns to scale (VRS). Alternatively, if the last constraint stated in (1) is removed, constant returns to scale (CRS) can be assumed in the DEA. Any differences in the efficiencies estimated assuming VRS and CRS are then assumed to reflect scale efficiencies. Formally, scale efficiency is calculated as the ratio of the CRS technical efficiency score to the VRS technical efficiency score. The overall technical efficiency score can then be decomposed into scale efficiency and pure technical efficiency. In addressing productivity change in Chinese hospitals, this study adopted the non-parametric Malmquist index. The index has been first introduced by Färe, Grosskopf, Yaisawarng, Li, and Wang (1990) and popularized by Färe, Grosskopf, and Lovell (1994). It measures changes in a production unit's efficiency in transforming inputs into outputs from time t to time t+1. The index can be expressed using various distance functions. Caves, Christenson, and Diewert (1982) defined the input-based Malmquist index in period t as D t M t o = o u t+1 ; x t+1 D t o ut ; x t : Alternatively, for period t+1, M t+1 o = o u t+1 ; x t+1 : u t ; x t D t+1 D t+1 o D o is the distant function and its superscripts indicate the time period within which the efficiency scores are calculated. The superscripts on u and x indicate the time period the data (outputs and inputs, respectively) used in the calculation of the efficiency

4 Y.C. Ng / China Economic Review 22 (2011) scores. For example, D o t+1 (u t,x t ) is computed using the sample of hospitals in period t+1 in reference to the production frontier based on period t. Based on the work of Färe, Grosskopf, and Weber (1989) and that of Färe, Grosskopf, Lindgren, and Roos (1992), the Malmquist index used in this study was the geometric mean of the above two indices. That is, M o D t u t +1 ; x t +1 ; u t ; x t o u t +1 ; x t Dt o u t +1 ; x t +1 = 4@ A@ D t o u t ; x t A5 u t ; x t D t +1 o 1 = 2 : ð2þ This index avoids arbitrarily selecting one of the time periods as the reference point. This index can then be decomposed into two components: u t +1 ; x t +1 ; u t ; x t = o u t +1 ; x t +1 D t o u t +1 ; x t D t o u t ; x t A Dt o u t ; x t A5 D t o +1 u t +1 ; x t +1 D t o +1 u t ; x t M o D t +1 1 = 2 : ð3þ The first component (the term in front of the square brackets) measures efficiency change, or how the position of the observed production unit has changed relative to the production frontier between time points t and t+1. The second component (the term in the square brackets) measures technical change, or how the production frontier shifted between the time points. Färe, Grosskopf, Norris, and Zhang (1994) have discussed the linear programming problems associated with the distance functions expressed in Eq. (2). Furthermore, these authors provide the decomposition of the efficiency change into two components, namely pure efficiency change and scale efficiency change (Eqs. (4) and (5), respectively) as follows: ov u t +1 ; x t +1 D t ov u t ; x t ð4þ D t +1 2 D t +1 ov u t +1 ; x t +1 =D t +1 oc u t +1 ; x t +1 4 u t ; x t =D t +1 u t ; x t D t +1 ov oc D t ov u t +1 ; x t +1 D t ov u t ; x t =D t oc u t +1 ; x t +1 =D t oc u t ; x t = 2 ð5þ where subscripts ov and oc relate to the VRS and CRS technologies, respectively. For ease of interpretation, the reciprocals of the computed indices are presented here. In other words, a value greater than one indicates productivity growth while a value less than one implies deterioration Data Hospitals, in China as elsewhere, do not directly measure outputs such as changes in health status. Studies of hospital efficiency must perforce rely on the intermediate good, health services, to quantify hospital output. The present study was no exception. Making reference to previous studies, especially those developing country studies, the present empirical analysis was designed based on the availability of data as well as measures commonly adopted in the literature. Hospital data for China's Guangdong province were extracted from a series of Guangdong Province's Health Statistical Yearbooks. Data for on 463 hospitals were used in both the efficiency and productivity growth analyses. All these hospitals are regulated by the Ministry of Health. According to the province's official classification, 264 of the hospitals were located in the Pearl River Delta region (PRD), 42 in the eastern region, 80 in the western region, and 77 in the province's mountain area. This regional classification has economic significance, as the PRD is by far the most developed region of the province, while the mountain region is the most remote region with the least economic development. As pointed out by Huang et al. (2009), the health care system in Guangdong province was similar to that in the rest of China. With distinct differences in the regional economic development in China as a whole and the imbalance development across regions in the Guangdong province, the analysis by region within the Guangdong province can thus serve as a good reference for regional performance of hospitals within the continent of China. Similar to studies in the literature, two outputs were used in the productive efficiency analysis: the number of outpatient cases and the number of inpatient cases. Labor and capital are both important in producing health services in a hospital. The focus here was on four labor indicators: (1) the number of doctors, (2) the number of nurses, (3) the number of pharmacists, and (4) the number of the other staff (other medical staff and administrative workers). A hospital's capital stock was proxied by the number of beds, as has been the practice in previous studies. All the analyses were based initially on the full sample of 463 hospitals. To highlight the relationship between performance and economic development, the results are summarized by region.

5 432 Y.C. Ng / China Economic Review 22 (2011) Results 4.1. An overview of the Guangdong hospitals Between 2004 and 2008, most inputs at all the hospital grew continuously, whichever region is considered (Table 1). Hospitals in the PRD, on average, consumed more health inputs than those in the other three regions during the period. Table 1 clearly shows Table 1 Hospital inputs by region, Pearl river delta area Eastern region Western region Mountain area 2004 Number of doctors (128.78) Number of nurses (147.40) Number of pharmacists (28.03) Number of other staff (52.87) Number of beds (302.42) (77.36) (102.41) (23.42) (41.74) (182.86) (79.07) (105.12) (18.29) (44.09) (234.32) (63.53) (84.26) (15.95) (28.72) (174.10) 2005 Number of doctors (135.13) Number of nurses (155.14) Number of pharmacists (28.53) Number of other staff (56.08) Number of beds (308.24) (79.05) (104.56) (25.21) (37.94) (190.89) (78.81) (109.38) (17.70) (50.69) (238.54) (64.68) (88.37) (16.47) (35.47) (159.09) 2006 Number of doctors (154.53) Number of nurses (200.48) Number of pharmacists (31.11) Number of other staff (66.54) Number of beds (329.27) (82.51) (106.32) (23.73) (34.82) (191.66) (83.00) (126.18) (18.54) (55.83) (250.90) (71.50) (89.84) (16.77) (30.51) (225.68) 2007 Number of doctors (183.45) Number of nurses (216.50) Number of pharmacists (34.14) Number of other staff (66.22) Number of beds (347.19) (90.53) (113.86) (20.08) (38.13) (194.31) (98.92) (133.16) (16.94) (49.36) (268.89) (82.92) (103.72) (16.15) (30.72) (235.03) 2008 Number of doctors (168.55) (86.94) (97.44) (81.77) Number of nurses (223.28) (111.12) (226.33) (111.39) Number of pharmacists (34.56) (21.35) (17.40) (17.66) Number of other staff (77.13) (42.57) (61.07) (35.52) Number of beds (363.67) (201.07) (293.32) (263.93) Sample size Note: Standard deviations are in parenthesis.

6 Y.C. Ng / China Economic Review 22 (2011) Table 2 Hospital outputs by region, Pearl river delta area Eastern region Western region Mountain area 2004 Number of outpatients treated ( ) Number of inpatients treated ( ) ( ) ( ) ( ) ( ) ( ) ( ) 2005 Number of outpatients treated ( ) Number of inpatients treated ( ) ( ) ( ) ( ) ( ) ( ) ( ) 2006 Number of outpatients treated ( ) Number of inpatients treated ( ) ( ) ( ) ( ) ( ) ( ) ( ) 2007 Number of outpatients treated ( ) Number of inpatients treated ( ) ( ) ( ) ( ) ( ) ( ) ( ) 2008 Number of outpatients treated ( ) ( ) ( ) ( ) Number of inpatients treated ( ) ( ) ( ) ( ) Sample size Note: Standard deviations are in parenthesis. that the percentage increase in the inputs to hospitals in the PRD was much the greatest. Less input was consumed by the hospitals in the western region, and the number of pharmacists employed there fell over time. In the most remote area, the mountain area, hospitals consumed slightly more of all inputs over the years, except for the number of pharmacists. Table 3 Efficiency measures by region, Pearl river delta area Eastern region Western region Mountain area 2004 Overall efficiency Scale efficiency Pure technical efficiency Overall efficiency Scale efficiency Pure technical efficiency Overall efficiency Scale efficiency Pure technical efficiency Overall efficiency Scale efficiency Pure technical efficiency Overall efficiency Scale efficiency Pure technical efficiency Sample size

7 434 Y.C. Ng / China Economic Review 22 (2011) Table 4 Distribution of overall efficiency scores by region, Overall efficiency score Pearl river delta area Eastern region Western region Mountain area (23.11%) 16 (38.10%) 44 (55.00%) 28 (36.36%) N0.25 and (43.56%) 21 (50.00%) 33 (41.25%) 48 (62.34%) N0.5 and (19.70%) 3 (7.14%) 3 (3.75%) 1 (1.30%) N0.75 and b1 21 (7.95%) 2 (4.76%) 0 (0%) 0 (0%) =1 15 (5.68%) 0 (0%) 0 (0%) 0 (0%) (19.70%) 17 (40.48%) 42 (52.50%) 31 (40.26%) N0.25 and (45.83%) 21 (50.00%) 35 (43.75%) 44 (57.14%) N0.5 and (19.70%) 2 (4.76%) 3 (3.75%) 1 (1.30%) N0.75 and b1 26 (9.85%) 1 (2.38%) 0 (0%) 1 (1.30%) =1 13 (4.92%) 1 (2.38%) 0 (0%) 0 (0%) (16.29%) 15 (35.71%) 42 (52.50%) 24 (31.17%) N0.25 and (46.59%) 15 (35.71%) 34 (42.50%) 49 (63.64%) N0.5 and (25.00%) 12 (28.57%) 3 (3.75%) 4 (5.19%) N0.75 and b1 22 (8.33%) 0 (0%) 1 (1.25%) 0 (0%) =1 10 (3.79%) 0 (0%) 0 (0%) 0 (0%) (53.79%) 32 (76.19%) 68 (85.00%) 68 (88.31%) N0.25 and (39.02%) 9 (21.43%) 11 (13.75%) 7 (9.09%) N0.5 and (4.17%) 1 (2.38%) 1 (1.25%) 2 (2.60%) N0.75 and b1 3 (1.14%) 0 (0%) 0 (0%) 0 (0%) =1 5 (1.89%) 0 (0%) 0 (0%) 0 (0%) (51.14%) 30 (71.43%) 60 (75.00%) 63 (81.82%) N0.25 and (42.42%) 10 (23.81%) 18 (22.50%) 12 (15.58%) N0.5 and (3.79%) 1 (2.38%) 2 (2.50%) 1 (1.30%) N0.75 and b1 3 (1.14%) 1 (2.38%) 0 (0%) 1 (1.30%) =1 4 (1.52%) 0 (0%) 0 (0%) 0 (0%) Sample size Table 2 shows the hospitals' outputs by region, generally demonstrating an upward trend similar to those of the inputs between 2004 and Hospitals treated more patients, except for a fall in outpatients in the eastern region between 2005 and 2006 and a drop in inpatients treated in the eastern region between 2007 and Another notable feature was that the number of patients treated was quite similar for hospitals in the eastern region and the mountain area before By 2008, hospitals of the mountain area even served more patients than those in the eastern region Efficiency Table 3 reports yearly efficiency estimates. Summary statistics (geometric means) for the overall efficiency (an efficiency measure with constant returns to scale) and its decomposition (scale efficiency and pure technical efficiency) are presented by region. 2 Hospitals in the Pearl River Delta consistently employed more inputs than those in other regions, and they were also relatively more efficient (overall efficiency). On the other hand, hospitals in the west were the least efficient despite the fact that they consumed the least inputs. The table asserts that in 2004, hospitals in Guangdong could have handled the same level of inpatient and outpatient cases with 21% 37% of the inputs they actually used had they been efficient. Notice that the regional pattern of efficiency scores holds for all five years. Notable fact is that over the period, hospitals did improve their efficiency. In particular, hospitals in both the PRD and western region showed continuous improvement in efficiency between 2004 and As shown in Table 4, there was an obvious decline in the number of inefficient hospitals with efficiency scores less than or equal to In the western region, the distribution by efficiency scores remained fairly stable between 2004 and 2006, while there was a general fall in percentage of hospitals with high efficiency performance in the eastern region. By 2007, improvement in efficiency of hospitals, regardless the region, reversed, although the regional pattern remained. One encouraging fact is that the overall efficiency improved slightly in 2008, particular hospitals in the non-pearl River Delta area (Table 3). The worsening in efficiency for hospitals in period was mainly a result of a higher percentage of hospitals attaining very low efficiency scores as shown in Table 4. In sum, after three years' overall efficiency improvement, hospitals of all regions experienced worsening in efficiency by , particular the case of hospitals in PRD and the eastern region. Examining the sources of inefficiency, the estimates of Table 3 indicate that pure technical inefficiency was the primary source of inefficiency. This is particular for hospitals in the PRD and the eastern region in period. Hospitals in Guangdong suffered relatively less from scale inefficiency than from pure technical inefficiency, except those in the western region. The 2 The non-parametric Kruskal Wallis test was adopted to test the null hypothesis that the efficiency scores across regions are indifferent for all years. For the three efficiency measures, the results indicated that the null hypothesis was rejected at 5% significant level for all five years.

8 Y.C. Ng / China Economic Review 22 (2011) Table 5 Distribution of scale efficiency scores by region, Scale efficiency score Pearl river delta area Eastern region Western region Mountain area (3.03%) 2 (4.76%) 14 (17.50%) 4 (5.19%) N0.25 and (14.02%) 4 (9.52%) 22 (27.50%) 14 (18.18%) N0.5 and (20.83%) 15 (35.71%) 17 (21.25%) 15 (19.48%) N0.75 and b1 149 (56.44%) 21 (50.00%) 27 (33.75%) 44 (57.14%) =1 15 (5.68%) 0 (0%) 0 (0%) 0 (0%) (1.89%) 4 (9.52%) 19 (23.75%) 2 (2.60%) N0.25 and (12.12%) 4 (9.52%) 13 (16.25%) 13 (16.88%) N0.5 and (21.97%) 10 (23.81%) 22 (27.50%) 17 (22.08%) N0.75 and b1 156 (59.09%) 23 (54.76%) 26 (32.50%) 44 (57.14%) =1 13 (4.92%) 1 (2.38%) 0 (0%) 1 (1.30%) (1.89%) 3 (7.14%) 15 (18.75%) 3 (3.90%) N0.25 and (10.98%) 2 (4.76%) 17 (21.25%) 12 (15.58%) N0.5 and (28.03%) 11 (26.19%) 18 (22.50%) 13 (16.88%) N0.75 and b1 146 (55.30%) 26 (61.90%) 29 (36.25%) 49 (63.64%) =1 10 (3.79%) 0 (0%) 1 (1.25%) 0 (0%) (3.79%) 4 (9.52%) 21 (26.25%) 9 (11.69%) N0.25 and (11.74%) 8 (19.05%) 17 (21.25%) 15 (19.48%) N0.5 and (21.59%) 12 (28.57%) 18 (22.50%) 20 (25.97%) N0.75 and b1 161 (60.98%) 17 (40.48%) 24 (30.00%) 33 (42.86%) =1 5 (1.89%) 1 (2.38%) 0 (0%) 0 (0%) (5.68%) 3 (7.14%) 21 (26.25%) 6 (7.79%) N0.25 and (20.45%) 8 (19.05%) 19 (23.75%) 18 (23.38%) N0.5 and (21.97%) 15 (35.71%) 20 (25.00%) 13 (16.88%) N0.75 and b1 133 (50.38%) 15 (35.71%) 20 (25.00%) 39 (50.65%) =1 4 (1.52%) 1 (2.38%) 0 (0%) 1 (1.30%) Sample size Table 6 Distribution of pure technical efficiency scores by region, Pure technical efficiency score Pearl river delta area Eastern region Western region Mountain area (5.68%) 5 (11.90%) 9 (11.25%) 3 (3.90%) N0.25 and (42.42%) 19 (45.24%) 40 (50.00%) 60 (77.92%) N0.5 and (24.24%) 9 (21.43%) 14 (17.50%) 13 (16.88%) N0.75 and b1 43 (16.29%) 4 (9.52%) 7 (8.75%) 1 (1.30%) =1 30 (11.36%) 5 (11.90%) 10 (12.50%) 0 (0%) (6.06%) 8 (19.05%) 9 (11.25%) 4 (5.19%) N0.25 and (39.39%) 19 (45.24%) 41 (51.25%) 59 (76.62%) N0.5 and (25.38%) 9 (21.43%) 14 (17.50%) 12 (15.58%) N0.75 and b1 45 (17.05%) 4 (9.52%) 4 (5.00%) 1 (1.30%) =1 32 (12.12%) 2 (4.76%) 12 (15.00%) 1 (1.30%) (5.68%) 4 (9.52%) 11 (13.75%) 4 (5.19%) N0.25 and (34.47%) 16 (38.10%) 37 (46.25%) 59 (76.62%) N0.5 and (28.41%) 15 (35.71%) 16 (20.00%) 8 (10.39%) N0.75 and b1 44 (16.67%) 4 (9.52%) 4 (5.00%) 5 (6.49%) =1 39 (14.77%) 3 (7.14%) 12 (15.00%) 1 (1.30%) (28.79%) 16 (38.10%) 34 (42.50%) 34 (44.16%) N0.25 and (53.41%) 18 (42.86%) 19 (23.75%) 36 (46.75%) N0.5 and (11.36%) 4 (9.52%) 12 (15.00%) 2 (2.60%) N0.75 and b1 9 (3.41%) 3 (7.14%) 2 (2.50%) 2 (2.60%) =1 8 (3.03%) 1 (2.38%) 13 (16.25%) 3 (3.90%) (23.48%) 13 (30.95%) 18 (22.50%) 30 (38.96%) N0.25 and (49.24%) 17 (40.48%) 32 (40.00%) 35 (45.45%) N0.5 and (15.53%) 8 (19.05%) 12 (15.00%) 8 (10.39%) N0.75 and b1 18 (6.82%) 2 (4.76%) 5 (6.25%) 1 (1.30%) =1 13 (4.92%) 2 (4.76%) 13 (16.25%) 3 (3.90%) Sample size

9 436 Y.C. Ng / China Economic Review 22 (2011) Table 7 Malmquist Productivity Index and Its Decomposition by Region, A. Pearl river delta area Malmquist index Technological change Change in efficiency Change in scale efficiency Change in pure technical efficiency Sample size B. Eastern region Malmquist index Technological CHANGE Change in efficiency Change in scale efficiency Change in pure technical efficiency Sample size C. Western region Malmquist index Technological change Change in efficiency Change in scale efficiency Change in pure technical efficiency Sample size D. Mountain area Malmquist index Technological change Change in efficiency Change in scale efficiency Change in pure technical efficiency Sample size frequency distributions of scale and pure technical efficiency scores presented in Tables 5 and 6 echo the pattern in the average scores shown in Table 3. As the tables make clear, less hospitals attained pure technical efficiency scores greater than 0.5. This is particularly the case in the eastern region. Hospitals in the PRD generally had relatively better scale efficiency than those in the other three regions (Table 3). The information shown in Table 5 indicates that more hospitals in the PRD fell into the upper ranges over the studied period. For hospitals in the eastern region and the mountain area, they, on average, experienced similar scale inefficiency between 2004 and Hospitals across regions experienced slightly improvement in scale efficiency before As shown in Table 3, scale inefficiency and pure technical inefficiency played more or less equal role in overall inefficiency for hospitals in the western region for the entire study period Productivity changes The five year panel data allowed computation of productivity changes in the hospitals. For ease of interpretation, the reciprocals of the computed input-based Malmquist indices are presented in Table 7. In other words, a value smaller than one means deterioration, whereas a value larger than one means improvement. As shown in the table, hospitals, on average, experienced productivity growth between 2004 and 2008 regardless their region. 3 There was substantial productivity growth for hospitals in the remote area, namely the western region and the mountain area. Between 2004 and 2008, the growth of these hospitals, on average, was 31% to 39%. For hospitals in other regions, the growth was much lower, 7% to 19% over the same period. Although the Malmquist indices are all greater than one, hospitals in the different regions showed different productivity growth patterns. As shown in Table 7, a downward trend in year-to-year productivity change was found for hospitals in the PRD. The overall productivity growth of hospitals in the western region and the mountain area was a result of a big jump in productivity between 2006 and On the other hand, the 19% average growth rate of hospitals in the eastern region was the good progress made in The decomposition of the Malmquist productivity index reveals that the hospitals' overall productivity growth originated from technological progress over the studied period. Probably due to the adoption of high-tech treatments, productivity of hospitals boosted by 68% to 94% between 2004 and 2008 (Table 7). On the contrary, hospitals of four regions experienced worsening in efficiency change of similar magnitude over time. Echoing the yearly efficiency measures, worsening in efficiency change over time mainly stemmed from the deterioration in pure technical efficiency change. As shown in Table 7, between 2004 and 2008, except hospitals in the western region, hospitals in 3 As with the yearly efficiency score by region, the non-parametric Kruskal Wallis test was applied to the Malmquist Index and its decomposition components for and the sub-periods. The statistical tests for sub-period indicated that the null hypothesis cannot be rejected while for other periods, the null hypothesis that there was no regional difference was rejected at 5% significant level.

10 Y.C. Ng / China Economic Review 22 (2011) other regions experienced similar pattern in terms of the worsening in scale efficiency change which is much less serious than that of change in pure technical efficiency. Another observation is that there was continuous improvement in change in year-to-year pure technical efficiency for hospitals in non-eastern region except during which all hospitals suffered a fall in change in pure technical efficiency. The year-to-year fluctuations in the change in scale efficiency may reflect the fluctuation in the demand for hospital services for hospitals across different locations. 5. Discussion The computed efficiency measures indicate that the efficiency of the sampled Guangdong hospitals was, in general, far below that of hospitals in developed countries as well as hospitals in other developing countries. 4 Estimates from previous study in the 2000s reveal that hospitals' overall efficiency normally ranges between 56% and 82%. Our results show that the overall efficiency of Guangdong hospitals was about 15% to 40%. The productivity growth found in the present study is comparable to findings for Finland, Austria and Northern Ireland. However, the sources of growth in the various countries are quite different. While hospitals in the Ukraine and in South Africa experienced technological regression combined with improvement in efficiency, the productivity growth of Guangdong hospitals encountered improvement in technology and efficiency over the studied period. Although the results show that the hospitals sampled were inefficient, they also show efficiency improvement between 2004 and 2006 and positive productivity growth between 2004 and This implies that, at least in Guangdong, health care provision was benefiting from the health care reforms that had been implemented in China over the previous two decades. The more competitive environment and better functioning market mechanisms probably exerted some effects on hospital efficiency. As is evident from the sources of efficiency improvement, improvement in scale efficiency played a key role in the first half of the period while pure technical efficiency took the lead in period. The observed productivity growth in these Guangdong hospitals was the result of the adoption of high-tech treatments reflecting in the technological change estimates of the Malmquist index. The sudden fall in efficiency of all Guangdong hospitals by 2007 is unexpected. This may reflect some weakness of the reforms implemented up to mid 2000s. That may be why a new set of reforms was announced in The revenue-based behavior of hospitals in responding to the cut in financial support from the governments can be seen from the patterns of technological change and change in efficiency estimates. As shown in Table 7, wherever there was technological improvement (shifting out of the production frontier), worsening in efficiency change was found (hospitals operated below the frontier), with the exception between 2004 and It is argued that, with the adoption of high-tech treatments and expensive drug therapies, the feasible and attainable output or outcome has expanded. However, due to inefficient usages of these medical inputs, targeted output level is far from reached, ending with worsening in change in efficiency. The low overall efficiency estimates of Table 3 provide supportive evidence to the argument. Imbalances in regional development in China are well documented, and economic performance varies by region accordingly. Examining the efficiency performance of these hospitals, hospitals in the relatively advanced Pearl River Delta area showed the best efficiency. However, hospitals in the most remote area, the mountain area, outperformed those in the western region. In terms of scale efficiency, hospitals in the mountain area even rank the second among the four regions. The latter phenomenon indicates the differences in the demand for hospital services by patients across regions. Huang et al. (2009) in analyzing Guangdong hospitals pointed out that patients tend to seek services from top-ranking hospitals. With most of these top-ranking hospitals located in more developed region, scale efficiency is more attainable for hospitals in the Pearl River Delta area. The comparable scale efficiency estimates of hospitals in the mountain area probably reflect the limited choice faced by patients due to geographic barriers. In sum, the economic environment, as in the case of Guangdong province, may not play such an important role in the performance of hospitals. The performance of non-profit organizations like hospitals can be addressed from various angles. The work of Culter and Berndt (2001) has provided an in-depth discussion of health care output and productivity. Hospital performance can be addressed from cost perspective by considering the amount of inputs per unit of output. Cost data for inputs were not available in the present study, so an input output relationship approach was adopted in analyzing the hospitals' performance. Similarly, we were not able to find information about the case-mix of each hospital or about patient outcome quality. DEA emphasizes the strict forward inputs and outputs relationship in computing the efficiency scores or Malmquist indices, but the lack of adjustment for case-mix or outcome quality implies that the results of the present study must be interpreted with caution. The empirical analyses can only serve as an example in understanding the performance of Chinese hospitals. The study focused on hospitals in one Chinese province. Although all the public hospitals in Guangdong Province were included in the study, the results generated are strictly applicable only to Guangdong. Due to data limitations, the study was not able to differentiate hospitals by type (such as teaching hospitals), and thus caution is called for when interpreting the results. Although the study has incorporated as many inputs as possible in carrying out the analysis, other relevant inputs such as material supplies (pharmaceutical and non-pharmaceutical supplies) and output, say length of hospital stay, have been left out in the estimation owing to a lack of data. Such omissions could mean that the efficiency estimates represent the lower bound of the efficiency scores given the fact that Chinese hospitals tend to over-prescribe drugs. On the other hand, DEA efficiency scores tend 4 Readers are reminded that this cross country comparison is only a causal comparison due to the fact that different studies may use slightly different input output mixes and hospitals in different countries may face different institutional constraints.

11 438 Y.C. Ng / China Economic Review 22 (2011) to improve with additional inputs or outputs. On balance, there is no clear reason to believe that the results are either upward or downward biased. 6. Conclusions Input-based DEA efficiency estimates demonstrate that hospitals in China's Guangdong Province were quite inefficient during Pure technical inefficiency is the driving force for pulling down the overall efficiency of these hospitals. As the Malmquist indices defined, the productivity growth in Guangdong hospitals during this period originated from technological change. This largely reflects the revenue-based behavior of hospitals in which unnecessary care, over-prescription of drugs, and the adoption of high-tech treatments are commonly found. The present framework of relating inputs and outputs as outlined in the data envelopment analysis (DEA) addresses efficiency issues from production view point. Owing to the lack of cost information, issues associating with allocative efficiency are left out. Together with other data limitations and the geographical coverage of the sample, further empirical investigation of Chinese hospitals is called for. Studies over a longer time span promise to help both policy makers and hospital managers understand the overall performance of hospitals in China. With the pros and cons of the DEA approach, verifying efficiency and productivity change of Chinese hospitals using stochastic method is another direction for further research. Acknowledgement The author acknowledges with gratitude the generous support of the Research Committee, Hong Kong Baptist University for the project (FRG/08-09/I-31), without which the timely production of the current publication would not have been feasible. References Barbetta, Gian Paolo, Turati, Gilberto, & Zago, Angelo M. (2007). Behavioral differences between public and private not-for-profit hospitals in the Italian national health service. Health Economics, 16(1), Biørn, Erik, Hagen, Terje P., Iversen, Tor, & Magnussen, Jon (2003). The effect of activity-based financing on hospital efficiency: A panel data analysis of DEA efficiency scores Health Care Management Science, 6(4), Brown, H. Shelton, III, & Pagán, Joe A. (2006). Managed care and the scale efficiency of US hospitals. International Journal of Health Care Finance and Economics, 6(4), Caves, Douglas W., Christenson, Laurits R., & Diewert, W. Erwin (1982). The economic theory of index numbers and the measurement of input, output and productivity. Econometrica, 50(6), Chang, Hsihiu, Cheng, Mei-Ai, & Das, Somnath (2004). Hospital ownership and operating efficiency: Evidence from Taiwan. European Journal of Operational Research, 159(2), Chen, Andrew, Hwang, Yuhchang, & Shao, Benjamin (2005). Measurement and sources of overall and input inefficiencies: Evidences and implications in hospital services. European Journal of Operational Research, 161(2), Culter, David M., & Berndt, Ernst R. (2001). Medical care output and productivity. Chicago, US: The University of Chicago Press. Eggleston, Karen, Li, Ling, Meng, Qingyue, Lindelow, Magnus, & Wagstaff, Adam (2008). Health service delivery in China: A literature review. Health Economics, 17(2), Eggleston, Karen, Wang, Jian, & Rao, Keqin (2008). From plan to market in the health sector? China's experience. Journal of Asian Economics, 19(5 6), Eggleston, Karen, & Yip, Winnie (2004). Hospital competition under regulated prices: Application to urban health sector reforms in China. International Journal of Health Care Finance and Economics, 4(4), Emrouznejad, Ali, Parker, Barnett R., & Tavares, Gabriel (2008). Evaluation of research in efficiency and productivity: A survey and analysis of the first 30 years of scholarly literature in DEA. Socio-Economic Planning Sciences, 42(3), Färe, Rolf, Grosskopf, Shawna, Lindgren, Björn, & Roos, P. (1992). Productivity changes in Swedish phamacies : A non-parametric Malmquist approach. Journal of Productivity Analysis, 3(1 2), Färe, Rolf, Grosskopf, Shawna, & Lovell, C. A. K. (1994). Production frontier. Cambridge, U.K.: Cambridge University Press. Färe, Rolf, Grosskopf, Shawna, Norris, Mary, & Zhang, Zhongyang (1994). Productivity growth, technical progress, and efficiency change in industrialized countries. The American Economic Review, 84(1), Färe, Rolf, Grosskopf, Shawna, & Weber, William L. (1989). Measuring school district performance. Public Finance Quarterly, 17(4), Färe, Rolf, Grosskopf, Shawna, Yaisawarng, S., Li, S. K., & Wang, Z. (1990). Productivity growth in Illinois electric utilities. Resources and Energy, 12, Ferrier, Gary D., Rosko, Michael D., & Valdmanis, Vivian G. (2006). Analysis of uncompensated hospital care using a DEA model of output congestion. Health Care Management Science, 9(2), Gannon, Brenda (2005). Testing for variation in technical efficiency of hospitals in Ireland. The Economic and Social Review, 36(3), Gu, Edward, & Zhang, Jianjun (2006). Health care regime change in urban China: Unmanaged marketization and reluctant privatization. Pacific Affairs, 79(1), Guo, Baogang (2003). Transforming China's urban health-care system. Asian Survey, 43(2), Harrison, Jeffrey P., Coppola, M. Nicholas, & Wakefield, Mark (2004). Efficiency of federal hospitals in the United States. Journal of Medical Systems, 28(5), Harrison, Jeffrey P., & Sexton, Christopher (2006). The improving efficiency frontier of religious not-for-profit hospitals. Hospital Topics, 84(1), Herd, Richard, Hu, Yu-Wei, & Koen, Vincent (2010). Improving China's health care system. Economics department working papers no. 75. : OECD. Hofmarcher, Maria M., Paterson, Iain, & Riedel, Monika (2002). Measuring hospital efficiency in Austria A DEA approach. Health Care Management Science, 5(1), Hollingsworth, Bruce (2003). Non-parametric and parametric applications measuring efficiency in health care. Health Care Management Science, 6(4), Hollingsworth, Bruce (2008). The measurement of efficiency and productivity of health care delivery. Health Economics, 17(10), Hougaard, Jens Leth, Osterdal, Lars Peter, & Yu, Yi (2008). The Chinese health care system: Structure, problems and challenges. Discussion Paper, No , Department of Economics, University of Copenhagen. Hu, Jin-Li, & Huang, Yuan-Fu (2004). Technical efficiencies in large hospitals: A managerial perspective. International Journal of Management, 21(4), Huang, Cunrui, Liang, Haocai, Chu, Cordia, Rutherford, Shannon, & Geng, Qingshan (2009). The emerging role of private health care provision in China: A critical analysis of the current health system. Asia Health Policy Program Working Paper #10. : Stanford University. Kirigia, Joses M., Emronznejad, Ali, Cassoma, Basilio, Asbu, Eyob Zere, & Barry, Saidou (2008). A performance assessment method for hospitals: The case of municipal hospitals in Angola. Journal of Medical Systems, 32(6),