**2. Background**

#### **2.1 Existing literature**

Previous literature does not provide a clear guidance as to the effects of hospital competition on health outcomes. For example, Propper et al. (2004) examined the relationship between hospital competition and mortality rates after emergency acute myocardial infraction (AMI). They find that higher competitive pressures in Britain's National Health Service actually increased mortality rates for AMI after controlling for patient mix and other observed characteristics of the hospital and its market area. Similarly, using a longer panel of

Can Competition Save Your Life? 157

 ���������)�� � �� � �������)� � �����������)� � ����������)� � �� � �� (1) where ∆EM represents changes in excess mortality between 2007 and 2003 for hospital i. Excess mortality is defined as the ratio between the observed mortality rate and the predicted rate. We estimate (1) using generalized linear model (GLM) separately for excess

measures hospital competition; the index is constructed based on available bed shares and CABG volume, to test whether our results are sensitive to the product definition. HHI is inversely related to competition and therefore we expect lower HHI to decrease excess mortality if competition indeed saves lives. We use hospital service area (HSA) as the relevant hospital market. Unlike market definitions based on geo-political borders (e.g., county), the

It is important to note that HHI can be calculated based on patient flows as well as political boundaries (such as county or MSA). However, a recent study by Wong et al. (2005) calculated hospital competition index using county, MSA, health service area (as done in this study), fixed radius and variable radius measures. They found that the method of constructing HHI did not affect their estimates of the effect of competition on hospital costs. We believe that our two alternative measures of HHI will capture the effect of competition on patient outcomes although the magnitude of the effect may be different, as was found by

We also control for hospital characteristics and other market characteristics, as follows: the hospital characteristics include ownership, church affiliation, teaching status, hospital size, ER, trauma unit and disproportionate share hospital status. Hospital size is measured as the number of staffed beds. Teaching status is defined as hospitals with some residents. We do control for the disproportionate share hospital status since hospitals with such designation serve a disproportionately high number of medically indigent patients. Hospital market characteristics reflect competitive pressures (HHI), area characteristics (percent uninsured and per capital income) and demand for community benefits (percent MediCal enrollment).

This study uses hospital data from the Hospital Annual Financial Disclosure Reports filed annually by all California hospitals with the Office of Statewide Health Planning and Development (OSHPD). Hospital data will be merged with risk adjusted mortality rates that are based on the California Coronary Artery Bypass Graft (CABG) Mortality Reporting Program (CCMRP). Mortality rates are adjusted for patient characteristics: age, gender, body mass index, acuity (elective, urgent, emergent or salvage) as well as for the secondary conditions (such as hypertension, diabetes, etc.). Two mortality rates have been calculated. Observed mortality rate is the ratio of the number of deaths and the CABG cases multiplied by 100. Expected mortality rate is the ratio of the number of expected deaths to CABG cases multiplied by 100; expected mortality rate is risk adjusted. In this study we calculate excess mortality as the ratio of observed to expected mortality rate. When excess mortality ratio is greater than 1.0, that indicates that there were more deaths at the hospital than would have been expected. A ratio less than 1.0 indicates that there were fewer deaths at a hospital than expected, given the case mix of patients treated at a hospital (CCMRP). Finally, we use risk

1 where *HHI* represents Herfindahl-Hirschman index (HHI) that

mortality and risk adjusted mortality rate relative to the state average.

Coefficient of interest is

HSA relies on patient flows.

Wong et al. (2005).

**3.2 Data** 

hospitals Propper et al. (2008) find a negative relationship between hospital competition and AMI mortality but they find that competition does decrease waiting times. The study concludes that hospitals in UK decreased unobservable dimensions of quality (such as patient outcomes) but improved observed waiting times. On the other hand, Shortell and Hughes (1988), Ho and Hamilton (2000) find no significant relationship between hospital competition and hospital quality. For example, Ho and Hamilton (2000) found that mergers increased 90-day readmission rates and early discharges for newborns for Mei-Cal patients but did not affect inpatient mortality rates.

In contrast to above mentioned findings, Sari (2002) and Kessler and McClellan (2000) show that hospital competition improves health outcomes along several important quality dimensions. Unlike UK studies above, Kessler and McClellan (2000) examined the effect of competition on AMI outcomes for Medicare patients and found higher AMI mortality rates when competition for Medicare patients decreased. Most recently, Gaynor and colleagues (2011) estimated the effect of moving from a market with 2.5 hospitals to one with 5 hospitals and find that such improvement in competition would result in about 55,000 more life years.

It is important to note that different patients may be affected differently by competition. Kessler and Geppert (2005) examined the effect of hospital competition on more and less severely ill patients. They show that higher competitive pressures improve health outcomes (and lead to higher costs of treatment) for severely ill patients but do not affect less severely ill patients. Finally, Gowrisankaran and Town (2003) and Schneider (2008) have found that hospital competition improves quality but these quality improvements are not always realized depending on Medicare penetration and hospital ownership status.

#### **2.2 Importance of the study**

CABG surgery is the most common surgical procedure for treating cardiovascular disease. Coronary artery disease is the leading cause of all adult non-maternal admissions to California hospitals, representing nearly 9% of all admissions (OSHPD). This study contributes to the existing research along several dimensions. First, we focus on 30 day mortality rates and risk adjusted mortality rates following CABG surgery, while the previous literature for this surgery only examined in-hospital mortality rates (Schneider 2008). Second, we use the hospital quality data from the time period when such reporting became mandatory rather than voluntary. Thus, relative to previous literature based on OSHPD data, there is less scope for hospital self-selection since all high and low quality hospitals have to report their mortality statistics to OSHPD. Lastly, the study attempts to examine the relationship between hospital-specific *changes* in CABG mortality rates and *changes* in market competition. We hope to gage whether changes in competitive pressures are responsible for changes in mortality statistics across hospitals.

## **3. Methods**

Empirical model and data sources are presented below.

#### **3.1 Empirical model**

The following model is used in this study:

hospitals Propper et al. (2008) find a negative relationship between hospital competition and AMI mortality but they find that competition does decrease waiting times. The study concludes that hospitals in UK decreased unobservable dimensions of quality (such as patient outcomes) but improved observed waiting times. On the other hand, Shortell and Hughes (1988), Ho and Hamilton (2000) find no significant relationship between hospital competition and hospital quality. For example, Ho and Hamilton (2000) found that mergers increased 90-day readmission rates and early discharges for newborns for Mei-Cal patients

In contrast to above mentioned findings, Sari (2002) and Kessler and McClellan (2000) show that hospital competition improves health outcomes along several important quality dimensions. Unlike UK studies above, Kessler and McClellan (2000) examined the effect of competition on AMI outcomes for Medicare patients and found higher AMI mortality rates when competition for Medicare patients decreased. Most recently, Gaynor and colleagues (2011) estimated the effect of moving from a market with 2.5 hospitals to one with 5 hospitals and find that such improvement in competition would result in about 55,000 more life years. It is important to note that different patients may be affected differently by competition. Kessler and Geppert (2005) examined the effect of hospital competition on more and less severely ill patients. They show that higher competitive pressures improve health outcomes (and lead to higher costs of treatment) for severely ill patients but do not affect less severely ill patients. Finally, Gowrisankaran and Town (2003) and Schneider (2008) have found that hospital competition improves quality but these quality improvements are not always

CABG surgery is the most common surgical procedure for treating cardiovascular disease. Coronary artery disease is the leading cause of all adult non-maternal admissions to California hospitals, representing nearly 9% of all admissions (OSHPD). This study contributes to the existing research along several dimensions. First, we focus on 30 day mortality rates and risk adjusted mortality rates following CABG surgery, while the previous literature for this surgery only examined in-hospital mortality rates (Schneider 2008). Second, we use the hospital quality data from the time period when such reporting became mandatory rather than voluntary. Thus, relative to previous literature based on OSHPD data, there is less scope for hospital self-selection since all high and low quality hospitals have to report their mortality statistics to OSHPD. Lastly, the study attempts to examine the relationship between hospital-specific *changes* in CABG mortality rates and *changes* in market competition. We hope to gage whether changes in competitive pressures

realized depending on Medicare penetration and hospital ownership status.

are responsible for changes in mortality statistics across hospitals.

Empirical model and data sources are presented below.

The following model is used in this study:

but did not affect inpatient mortality rates.

**2.2 Importance of the study** 

**3. Methods** 

**3.1 Empirical model** 

$$\log(\Delta EM\_l)\_{nt} = \beta\_0 + \beta\_1(\Delta HHI)\_l + \beta\_2(Hosplital)\_l + \beta\_3(\DeltaMarket)\_l + a\_l + \varepsilon\_l \tag{1}$$

where ∆EM represents changes in excess mortality between 2007 and 2003 for hospital i. Excess mortality is defined as the ratio between the observed mortality rate and the predicted rate. We estimate (1) using generalized linear model (GLM) separately for excess mortality and risk adjusted mortality rate relative to the state average.

Coefficient of interest is 1 where *HHI* represents Herfindahl-Hirschman index (HHI) that measures hospital competition; the index is constructed based on available bed shares and CABG volume, to test whether our results are sensitive to the product definition. HHI is inversely related to competition and therefore we expect lower HHI to decrease excess mortality if competition indeed saves lives. We use hospital service area (HSA) as the relevant hospital market. Unlike market definitions based on geo-political borders (e.g., county), the HSA relies on patient flows.

It is important to note that HHI can be calculated based on patient flows as well as political boundaries (such as county or MSA). However, a recent study by Wong et al. (2005) calculated hospital competition index using county, MSA, health service area (as done in this study), fixed radius and variable radius measures. They found that the method of constructing HHI did not affect their estimates of the effect of competition on hospital costs. We believe that our two alternative measures of HHI will capture the effect of competition on patient outcomes although the magnitude of the effect may be different, as was found by Wong et al. (2005).

We also control for hospital characteristics and other market characteristics, as follows: the hospital characteristics include ownership, church affiliation, teaching status, hospital size, ER, trauma unit and disproportionate share hospital status. Hospital size is measured as the number of staffed beds. Teaching status is defined as hospitals with some residents. We do control for the disproportionate share hospital status since hospitals with such designation serve a disproportionately high number of medically indigent patients. Hospital market characteristics reflect competitive pressures (HHI), area characteristics (percent uninsured and per capital income) and demand for community benefits (percent MediCal enrollment).

#### **3.2 Data**

This study uses hospital data from the Hospital Annual Financial Disclosure Reports filed annually by all California hospitals with the Office of Statewide Health Planning and Development (OSHPD). Hospital data will be merged with risk adjusted mortality rates that are based on the California Coronary Artery Bypass Graft (CABG) Mortality Reporting Program (CCMRP). Mortality rates are adjusted for patient characteristics: age, gender, body mass index, acuity (elective, urgent, emergent or salvage) as well as for the secondary conditions (such as hypertension, diabetes, etc.). Two mortality rates have been calculated. Observed mortality rate is the ratio of the number of deaths and the CABG cases multiplied by 100. Expected mortality rate is the ratio of the number of expected deaths to CABG cases multiplied by 100; expected mortality rate is risk adjusted. In this study we calculate excess mortality as the ratio of observed to expected mortality rate. When excess mortality ratio is greater than 1.0, that indicates that there were more deaths at the hospital than would have been expected. A ratio less than 1.0 indicates that there were fewer deaths at a hospital than expected, given the case mix of patients treated at a hospital (CCMRP). Finally, we use risk

Can Competition Save Your Life? 159

competition, on average the hospital competition index actually decreased between 2003 and 2007. This decrease indicates that hospital markets became more competitive despite the decrease in the number of hospitals that operate in California and this holds true for both measures of hospital competition. Since HHI does not simply measure the number of hospitals but rather hospital market shares, this means that with fewer hospitals in the market hospital shares became more equal. In this study we look at differences between 2003 and 2007; the

If competitive pressures indeed improve health outcomes, we expected changes in HHI to

Empirical results indicate that higher competitive pressures translate into better health outcomes as measured by decreases in excess mortality as well as decreases in risk adjusted mortality rates relative to the state average. Results are statistically significant at the 10% level for most specifications of the model. Although hospitals in the United States may not compete based on price due to high prevalence of public insurance and high prevalence of nonprofit hospitals in the industry, competition based on quality may be relatively more important, especially as public insurance coverage is being expanded after the passage of

Other important variables include hospital ownership status. Although there is no statistically significant relationship between mortality rates in for-profit and nonprofit hospitals, the nonprofit hospitals saw the same or smaller improvements in mortality than for-profits between 2003 and 2007. Municipal hospitals (city and county hospitals) show higher risk adjusted mortality rates and Table 2 shows that they saw smaller improvements

Although we do not see a statistically significant relationship between CABG volume and health outcomes, alternative specifications of the volume variable yield a better fit. For example, square root of volume has a statistically positive effect on mortality improvements (result not shown). This indicates that initial increases in CABG volume improve outcomes but this improvement, interestingly, increases at a decreasing rate. We also did not find a significant relationship between increases in the number of CABG surgeries performed and mortality statistics. This result is consistent with the previous research that shows that over time, the disparity in outcomes between low- and high-volume hospitals has narrowed as

Finally, the effect of hospital competition on health outcomes can be ambiguous since higher competitive pressures may potentially decrease the volume of surgeries that each hospital performs. If higher-volume hospitals, in fact, deliver better quality of care, competition may be undesirable. In our model we measure the effect of hospital competition holding CABG volume constant. Thus, we re-estimate our model without the effect of CABG volume. We find that even in this specification of the model the relationship between the hospital

in mortality rates even after we controlled for CABG volume and area characteristics.

higher the difference, the higher the increase in competition for the hospital market.

have a positive effect on changes in health outcomes.

Our regression results are presented in Table 2 below.

the Patient Protection and Affordable Care Act of 2010.

outcomes have improved significantly for all hospitals (Ho, 2000).

competition (as measured by HHI) holds.

**4. Empirical results** 

adjusted mortality rate which is obtained by multiplying the observed annual California state average mortality rate by a hospital's excess mortality ratio for that year.

In this study we used 479 California general acute care hospitals in 2003 and 456 hospitals in 2007 to compute HHI based on available beds. Then we reduced our sample to CABG performing hospitals to calculate HHI based on CABG surgeries. This definition of hospital competition uses a more narrow market definition since hospitals are assumed to compete only with other CABG performing hospitals. For our regression analysis, our sample was reduced to 109 hospitals that were performing CABG in both 2003 and 2007.


Table 1. Descriptive Statistics for Selected Variables

Table 1 shows that on average the hospitals in California experienced a decrease in both excess mortality and risk adjusted mortality rates between 2003 and 2007. Since our dependent variable measures the differences in excess mortality between 2003 and 2007, the higher the difference the greater the fall in excess mortality and risk adjusted mortality rates. Thus, our dependent variables measure improvements in health outcomes.

Between 2003 and 2007 the number of hospitals in California decreased by 4.8%. Although some hospitals saw an increase in competition and some hospitals saw a decrease in

adjusted mortality rate which is obtained by multiplying the observed annual California

In this study we used 479 California general acute care hospitals in 2003 and 456 hospitals in 2007 to compute HHI based on available beds. Then we reduced our sample to CABG performing hospitals to calculate HHI based on CABG surgeries. This definition of hospital competition uses a more narrow market definition since hospitals are assumed to compete only with other CABG performing hospitals. For our regression analysis, our sample was

(St. deviation) Minimum Maximum

(0.687) 0 3.0372

(0.955) 0 7.457

(1.062) -4.419 2.154

(1.995) 0 8.81

(2.246) 0 17.52

(2.685) -8.71 6.26

(621.272) 177.818 2545.682

(355.959) 181.108 1235.981

(344.853) -111.175 1352.938

(2058.170) 633.475 7847.276

(1204.323) 613.0733 5235.165

(1393.311) -331.8247 5252.218

0.621

185.611

512.0731

Table 1 shows that on average the hospitals in California experienced a decrease in both excess mortality and risk adjusted mortality rates between 2003 and 2007. Since our dependent variable measures the differences in excess mortality between 2003 and 2007, the higher the difference the greater the fall in excess mortality and risk adjusted mortality rates.

Between 2003 and 2007 the number of hospitals in California decreased by 4.8%. Although some hospitals saw an increase in competition and some hospitals saw a decrease in

Thus, our dependent variables measure improvements in health outcomes.

state average mortality rate by a hospital's excess mortality ratio for that year.

reduced to 109 hospitals that were performing CABG in both 2003 and 2007.

Variable Mean

Excess mortality in 2003 1.0869

Excess mortality in 2007 1.0570

Differences in excess mortality (2003-2007) 0.00936

Risk adjusted mortality rate, 2003 3.156

Risk adjusted mortality rate, 2007 2.463

HHI, based on available beds, 2003 790.939

HHI, based on available beds, 2007 603.411

HHI, based on CABG volume, 2003 2297.811

HHI, based on CABG volume, 2007 1783.738

Differences in risk adjusted mortality rates

HHI differences, HHI based on available beds

HHI differences, HHI based on CABG volume

Table 1. Descriptive Statistics for Selected Variables

(2003-2007)

(2003-2007)

(2003-2007)

competition, on average the hospital competition index actually decreased between 2003 and 2007. This decrease indicates that hospital markets became more competitive despite the decrease in the number of hospitals that operate in California and this holds true for both measures of hospital competition. Since HHI does not simply measure the number of hospitals but rather hospital market shares, this means that with fewer hospitals in the market hospital shares became more equal. In this study we look at differences between 2003 and 2007; the higher the difference, the higher the increase in competition for the hospital market.

If competitive pressures indeed improve health outcomes, we expected changes in HHI to have a positive effect on changes in health outcomes.
