Does Categorization Method Matter in Exploring Volume-Outcome Relation? A Multiple Categorization Methods Comparison in Coronary Artery Bypass Graft Surgery Surgical Site Infection

Abstract

Background:

Volume-infection relation studies have been published for high-risk surgical procedures, although the conclusions remain controversial. Inconsistent results may be caused by inconsistent categorization methods, the definitions of service volume, and different statistical approaches. The purpose of this study was to examine whether a relation exists between provider volume and coronary artery bypass graft (CABG) surgical site infection (SSI) using different categorization methods.

Methods:

A population-based cross-sectional multi-level study was conducted. A total of 10,405 patients who received CABG surgery between 2006 and 2008 in Taiwan were recruited. The outcome of interest was surgical site infection for CABG surgery. The associations among several patient, surgeon, and hospital characteristics was examined. The definition of surgeons' and hospitals' service volume was the cumulative CABG service volumes in the previous year for each CABG operation and categorized by three types of approaches: Continuous, quartile, and k-means clustering.

Results:

The results of multi-level mixed effects modeling showed that hospital volume had no association with SSI. Although the relation between surgeon volume and surgical site infection was negative, it was inconsistent among the different categorization methods.

Conclusions:

Categorization of service volume is an important issue in volume-infection study. The findings of the current study suggest that different categorization methods might influence the relation between volume and SSI. The selection of an optimal cutoff point should be taken into account for future research.

Since Luft et al. [1] published the article “Should Operations be Regionalized? The Empirical Relation between Surgical Volume and Mortality” in 1979, researchers' interests in studying the volume-outcome relation has increased and further research on several related topics has been triggered. In the past decade, the outcome measure used in volume-outcome studies has gradually changed from mortality to other complications such as surgical site infection (SSI). The relation between service volumes and clinical outcomes is based on two assumptions. The first is that “practice makes perfect,” the belief that the accumulated experiences of hospitals or surgeons contribute to better treatment outcomes. The second is the “selective referral” effect, which suggests that if a particular hospital is chosen or recommended, it will lead to a higher service volume because of the better treatment outcomes for patients treated at that hospital, thus establishing the relation between high service volumes and good clinical outcomes [1].

However, the findings of previous volume-infection studies are controversial. Some studies have found a negative relation between volume and infection [2 –6]. Some found that the medium-volume group has better performance [7], whereas other studies have found no relation between provider's volume and infection [8,9]. Previous studies have shown the heterogeneity of defining what is high volume and what is low volume [10,11]. In addition, the neglect of cluster characterization of data that should use hierarchical models may result in biased estimation of the variation [12].

In recent years, most studies have adopted multi-level analysis to fix the data clustering issue, however, the heterogeneity of service volume definitions is still an issue that needs attention. Therefore, the purpose of this study was to compare the results of volume-infection studies under different service volume categorization methods, using coronary artery bypass graft (CABG) surgery as an example.

Patients and Methods

Study design

This retrospective study used a multi-level design to examine the relation between provider service volumes and SSI for CABG patients.

Database

We used the 2005–2009 Taiwan National Health Insurance Research Database. The database covers every episode of care provided to its 23 million Taiwanese enrollees (approximately 98% of the population). It is a de-identified secondary database containing patient-level demographic and administrative information. It is released for public access for research purposes [13].

Ethics statement

The protocol for this study was approved by the Institutional Review Board of the National Taiwan University Hospital (protocol #201001027R).

Study population

The study included all patients who underwent CABG surgery (ICD-9-OP code: 36.1x-36.2) surgery between 2006 and 2008 in all hospitals in Taiwan.

Exclusion criteria

Patients were excluded from analysis if they were: (1) Aged <20 y, (2) had post-operative surgical site due to prior operation, (3) death after surgery in hospital, and (4) patient's characteristics were incomplete.

Definition of Variables

Dependent variables

The SSI cases were identified by the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes. The SSI cases are divided into two categories: Index hospitalization events and post-discharge events (i.e., SSIs that occur within 1 y after discharge and require re-admission to a hospital and/or the use of ambulatory services). The ICD-9-CM codes for hospitalization events are 996.03, 996.61, 996.72, and 998.5. The ICD-9-CM codes for ambulatory services events are 038.0–038.4, 038.8, 038.9, 682.6, 682.9, 780.6, 790.7, 875.0, 875.1, 891.0, 891.1, 996.03, 996.61, 996.72, 998.3, and 998.5. Following the work by Wu et al. [6], this study used the secondary ICD-9-CM diagnosis codes for index hospitalization events, and the primary and secondary diagnosis codes for post-discharge events as criteria for SSI in order to avoid cases in which infection developed prior to hospitalization [6].

Independent variables

The definition of surgeons' and hospitals' service volume was the surgeons' and hospitals' cumulative CABG service volumes in the previous year for each CABG operation. We analyzed service volume as three different variables: (1) A continuous variable; (2) a subjective categorical variable based on first quartile and third quartile as cutoff points; and (3) a data-driven categorical variable based on k-means clustering approach. The subjective categorical variable was the most common method to categorize the service volume [5,14 –16]. This study pre-determined the service volume be categorized into low, medium, and high groups. The cutoff points of surgeon's volume were 11 and 48 cases, and of hospital's volume were 73 and 176 cases. The data-driven categorical variable based on k-means clustering approach had been applied by Kuo et al. [12] to explore the relations among volume, mortality, and recurrence in breast cancer. This study also pre-determined the surgeon and hospital service volume into low-, medium-, and high-volume groups by k-means clustering method. The cutoff points determined by the statistical model were 25 and 64 cases for surgeon's volume, and 107 and181 cases for hospital's volume.

Control variables

In addition to service volume and SSI, the study collected patient-level data such as age, gender, Charlson/Romano comorbidity index, numbers of vessels obstructed (as a proxy indicator of duration of operation [17]), and underlying diseases, such as diabetes mellitus (ICD-9-CM code: 250), chronic obstructive pulmonary disease (ICD-9-CM code: 490-496), heart failure (ICD-9-CM code: 584-586), renal failure and renal insufficiency (ICD-9-CM code: 428), which are associated with surgical infections [6]. Regarding the surgeon-level data, the study collected surgeon's gender and age. For hospital level, this study also collected hospital ownership and accreditation level as control variables.

Statistical Analysis

The data were analyzed using multi-level logistic regression. Most previous studies have used conventional regression models, which ignore the possible correlations between outcomes for a given surgeon or hospital, resulting in the underestimation of a variable's standard error and produce incorrect statistical significance [13,18,19]. Therefore, multi-level logistic regression was applied to explore the relation between provider volume in CABG and SSI, adjusted for patient severity, comorbidities, and demographics, and characteristics of the surgeon/hospital. Patients (level 1) were considered to be nested within surgeons (level 2) and then within hospitals (level 3). A stepwise variable selection procedure was performed for avoiding multicollinearity. All statistical analyses were performed using SAS (version 9.2, SAS Inc., Cary, NC).

Results

The results showed that there were 10,405 CABG procedures performed by 214 surgeons in 54 hospitals during 2006–2008 in Taiwan. Most of the patients were male (>75%), with a mean age of 65 y. Regarding the status of the four diseases associated with surgical infections, more than 50% of the patients had diabetes mellitus, one-third had heart failure, approximately 15% had renal failure and renal insufficiency and more than 20% had chronic obstructive pulmonary disease. The mean number of vessels obstructed was approximately 1.6. The average comorbidity score was approximately 1.5. There were 403 patients (3.87%) identified as having SSIs during hospitalization and 1,079 patients (10.37%) were identified as index hospitalization and post-discharge event (Table 1).

Table 1.

Description of Patient Characteristics by Year (n=10,405)

Characteristics	2006		2007		2008
n	3,438		3,478		3,489
Age^a	65.19	(10.93)	65.50	(10.90)	65.29	(11.08)
Length of stay^a	21.36	(11.44)	21.15	(11.77)	20.79	(11.15)
Numbers of vessels obstructed^a	1.57	(0.50)	1.60	(0.49)	1.58	(0.50)
Comorbidity index^a	1.57	(1.09)	1.54	(1.10)	1.50	(1.06)
Gender: Male^b	2,612	(75.97)	2,645	(76.05)	2,699	(77.36)
Index hospitalization event^b	138	(4.01)	131	(3.77)	134	(3.84)
Index hospitalization and post-discharge event^b	362	(10.53)	382	(10.98)	335	(9.60)
Diabetes mellitus^b	1,804	(52.47)	1,779	(51.15)	1,809	(51.85)
Heart failure^b	1,170	(34.03)	1,168	(33.58)	1,134	(32.50)
Renal failure and renal insufficiency^b	495	(14.40)	541	(15.55)	512	(14.67)
Chronic obstructive pulmonary disease^b	805	(23.41)	769	(22.11)	770	(22.07)

Mean (standard deviation).

n (%).

Table 2 shows the characteristics of hospitals and surgeons that had ever performed CABG operations during 2006–2008. Among 54 hospitals, most were not-for-profit hospitals; 19 were medical centers (35.19%); 34 were regional hospitals (62.96%), and others were community hospitals (1.85%). The annual average service volume was from 149.84 cases in 2006 to 152.66 cases in 2008; the maximum volume was from 294 cases in 2006, to 392 cases in 2008. In terms of surgeons, more than 90% were male, with a mean age of approximately 42 y. The average service volume was from 52.53 cases in 2006, to 50.38 cases in 2008, the maximum volume was from 120 cases in 2006, to 169 cases in 2008.

Table 2.

Provider Characteristics by Year

Characteristics	2006	2007	2008
Hospital characteristics
n	51	53	54
Ownership^a
Public	15 (29.41)	13 (24.53)	13 (24.07)
Not-for-profit	28 (54.90)	32 (60.38)	36 (66.67)
Private	8 (15.69)	8 (15.09)	5 (9.26)
Level^a
Medical center	19 (37.25)	19 (35.85)	19 (35.19)
Regional hospital	31 (60.78)	33 (62.26)	34 (62.96)
Community hospital	1 (1.96)	1 (1.89)	1 (1.85)
Annual average volume^b	149.84 (89.78)	139.81 (86.64)	152.66 (99.84)
Maximum and minimum volume	1,294	1,328	1,392
Surgeon characteristics
n	187	214	198
Age^b	41.74 (7.43)	41.85 (7.48)	43.21 (8.35)
Gender: Male^a	175 (94)	197 (92)	186 (94)
Annual average volume^b	52.53 (33.95)	48.22 (33.09)	50.38 (36.59)
Maximum and minimum volume	1,120	1,126	1,169

n (%).

Mean (standard deviation).

As shown in Table 3, the incidence of SSI among female patients with four major comorbidities was significantly greater than that among males and those patients without four major comorbidities. The length of stay and the Charlson/Romano comorbidity index of SSI cases were also higher than non-SSI cases. Most patients underwent CABG surgery in medical centers and not-for-profit hospitals. Community hospitals had a lower incidence (5.88%) than medical centers (10.3%) and regional hospitals (10.53%). Not-for-profit hospitals had a greater incidence (10.84%) than other hospitals. However, the differences did not reach statistical significance. In terms of hospitals' and surgeons' service volume, the average of hospitals' service volume in the non-SSI group and the SSI group were 147.50 and 145.80 (p=0.717), the average of surgeons' service volume in non-SSI group and the SSI group were 50.62 and 44.14 (p<0.0001). The results also showed that the SSI incidence of high-volume hospitals and surgeons was the lowest among the groups, except for hospitals' volume categorized by the quartile method.

Table 3.

Characteristics of Patients and Providers and the Association with Surgical Site Infections

Characteristic	Non-SSI (n=9,326)	SSI (n=1,079)	p
Patient characteristics
Age^a	65.21 (10.97)	66.28 (10.96)	0.0024
Gender^b			0.0039
Female	2,157 (88.08)	292 (11.92)
Male	7,169 (90.11)	787 (9.89)
Numbers of vessels obstructed^a	1.59 (0.50)	1.58 (0.50)	0.5067
Comorbidity index^a	1.51 (1.07)	1.79 (1.12)	<0.0001
Length of stay, days^a	19.82 (10.35)	32.17 (14.23)	<0.0001
Heart failure^b			<0.0001
Yes	3,019 (86.95)	453 (13.05)
No	6,307 (90.97)	626 (9.03)
COPD^b			<0.0001
Yes	2,045 (87.24)	299 (12.76)
No	7,281 (90.32)	780 (9.68)
Diabetes mellitus^b			<0.0001
Yes	4,737 (87.85)	655 (12.15)
No	4,589 (91.54)	424 (8.46)
Renal insufficiency^b			<0.0001
Yes	1,322 (85.4)	226 (14.6)
No	8,004 (90.37)	853 (9.63)
Surgeon characteristics
Volume: Continuous	50.62 (34.70)	44.14 (32.04)	<0.0001
Volume: Categorized by quartile^b			<0.0001
Low (<11)	2,397 (88.16)	322 (11.84)
Medium (11–47)	4,678 (89.36)	557 (10.64)
High (>48)	2,251 (91.84)	200 (8.16)
Volume: categorized by k-means^b			0.0033
Low (>25)	3,483 (89.06)	428 (10.94)
Medium (25–63)	4,220 (89.27)	507 (10.73)
High (>64)	1,623 (91.85)	144 (8.15)
Hospital characteristics
Accreditation level^b			0.7813
Medical center	6,285 (89.7)	722 (10.3)
Regional hospital	3,025 (89.47)	356 (10.53)
Community hospital	16 (94.12)	1 (5.88)
Owner^b			0.1064
Public	3,380 (90.18)	368 (9.82)
Not-for-profit	5,544 (89.16)	674 (10.84)
Private	402 (91.57)	37 (8.43)
Volume: Continuous	147.50 (92.65)	145.80 (86.89)	0.7177
Volume: Categorized by quartile^b			0.0238
Low (<73)	2,440 (88.44)	319 (11.56)
Medium (73–175)	4,855 (90.36)	518 (9.64)
High (>176)	2,031 (89.35)	242 (10.65)
Volume: Categorized by k-means^b			0.0411
Low (<107)	2,623 (88.44)	343 (11.56)
Medium (107–180)	4,072 (90.07)	449 (9.93)
High (>181)	2,631 (90.16)	287 (9.84)

Mean (standard deviation).

n (%).

Hospital volume is defined as the number of coronary artery bypass graft operations performed at a particular hospital in the previous year before each operation.

Surgeon volume is defined as the number of coronary artery bypass graft operations performed by a particular surgeon in the previous year before each operation.

SSI=surgical site infection; COPD=chronic obstructive pulmonary disease.

All control variables were put into the model. After stepwise selection, diabetes mellitus, chronic obstructive pulmonary disease, heart failure, and renal failure and renal insufficiency were kept in the model. After that, this study put the independent variable with different definitions into the multi-level model. Table 4 summarizes our multi-level mixed effect model. Model 1 found that surgeon's volume had a negative relation with infection, however, hospital volume had no relation with infection. Model 2 found that patients treated by low-volume (odds ratio [OR]=1.458, p=0.017) and medium-volume surgeons (OR=1.308, p=0.059) had greater risk than by high-volume surgeons. In addition, the risk among high-, medium-, and low-volume hospitals was not statistically significant. Model 3 found that there was no relation between hospital volume and SSI infection, and also surgeon volume. In addition, four comorbidities were found to have a positive relation with SSI in all models. Also, model 3 had better performance in model fitting.

Table 4.

Results of the Multi-Level Analysis Among Three Types of Service Volume Categorization Methods

	Model 1 (continuous)				Model 2 (quartile)				Model 3 (k-means)
		95% CI				95% CI				95% CI
n=10,405	OR	LL	UL	p	OR	LL	UL	p	OR	LL	UL	p
Fixed effects
Hospital-level
Cumulated service volume	1.000	0.999	1.002	0.471
Cumulated service volume (reference: High volume)
Low volume					0.884	0.661	1.181	0.396	1.060	0.815	1.380	0.659
Medium volume					0.791	0.613	1.022	0.072	0.937	0.749	1.171	0.563
Surgeon-level
Cumulated service	0.996	0.992	0.999	0.013
Cumulated service volume (reference: High volume)
Low volume					1.458	1.073	1.981	0.017	1.246	0.886	1.752	0.201
Medium volume					1.308	0.989	1.729	0.059	1.252	0.914	1.715	0.158
Random effects	Variance	SE			Variance	SE			Variance	SE
Intercept: surgeon (hospital)	0.052	0.185			0.049	0.183			0.033	0.166
Intercept: hospital^a	0.127	0.175			0.126	0.173			0.146	0.157
AIC	2,455.34				2,455.78				2,453.42
BIC	2,463.84				2,466.17				2,459.47

Adjusted by heart failure (reference: with heart failure), COPD (reference: with COPD), diabetes mellitus (reference: with diabetes), and renal insufficiency (reference: with renal insufficiency), all control variables were significant (p<0.001).

OR=Odds Ratio; CI=confidence interval; LL=lower limit; UL=upper limit; COPD=chronic obstructive pulmonary disease; SE=standard error. AIC=Akaike Information Criterion, one of the popular measures for mode fitting (small is better); BIC=Bayesian Information Criterion, one of the popular measures for model fitting (small is better).

Discussion

This nationwide population-based study was the first, to our knowledge, to use an uncategorized subjective method and a data-driven method to categorize service volume to explore the volume-infection relation. The surgeon and hospital annual average service volume were approximately 50 and 150, the service volume were less than that of the United States [20], but similar to previous experience in Taiwan [21]. In addition, unlike previous studies that calculated the accumulated service volume of hospitals or surgeons during the study period [3,5,6,8], this study considered the cumulative CABG service volumes in the previous year for each CABG operation, which was a similar design to that used by Solomon et al. [14] and Yasunaga et al. [22]. Therefore, this study could better identify whether “practice makes perfect” applies to the relation between CABG service volume and SSI.

There were three major findings of this study. First, the relation between hospital volume and SSI did not exist in these data; the relation between surgeon volume and SSI was partially consistent among different categorization methods. In addition, the data-driven categorization method had better model fitting than other methods. Second, the phenomenon of centralized service supply might exist. The hospital and surgeon annual average volume decreased slightly in 2006–2008, however, we also found that the standard deviation and maximum volume had increased. It might indicate that the phenomenon of CABG service centralization existed in Taiwan during these 3 y. Whereas we calculated the coefficient of variance, we found the level of deviation for surgeons was greater than hospitals (Table 2). The phenomenon of CABG service centralization was more obvious in surgeons than in hospitals. Third, patient's comorbidities might be more important than provider characteristics. In this study, patient, surgeon, and hospital characteristics were used as covariates, however, only four comorbidities remained after stepwise selection. These four comorbidities were also significant in all multi-level mixed effect models. It appeared the patient's comorbidities might be more important than surgeon and hospital characteristics.

In this study, we found the relation between provider's volume and SSI was ambiguous. In terms of surgeon's volume, the relation between surgeon volume and SSI was not consistent after adjusting other covariates. The result of the continuous model (model 1) showed that there was a negative relation between surgeon volume and SSI, whereas the result of quartile method model (model 2) demonstrated that the patients operated on by low-volume surgeons had higher risk than by high-volume surgeons. The risk of medium-volume surgeons was borderline significant higher than high-volume surgeons. However, the result of k-means model (model 3) revealed there was no relation between surgeon volume and SSI.

The controversial finding might be caused by the definition of service volume; it might also indicate that the cutoff point could be an important issue in volume-outcome studies [10,11,23]. Previous volume-outcome studies have categorized service volume in a subjective manner. Some of these studies have categorized service volume into three [6,8,24] or four groups [15,16,25] and others have used a specific case number [26] or percentage [2] for categorization. The categorization methods of previous studies and the current work were based on the linear assumption existed. Whereas a non-linear relation existed between volume and outcome, how to choose the optimal cutoff point should be dealt with carefully in future research.

In terms of hospital volume, there was no relation found in our study. Furthermore, our study contradicts that of Wu et al. [6] who found that the hospital's volume had a greater effect than surgeon's volume, and claimed that the findings may imply that a hospital's teamwork was more important than its individual surgeons. The data of our study were 5 y later and longer than that of Wu et al. [6]. During these 5 y, the severe acute respiratory syndrome (SARS) outbreak occurred in Taiwan in 2003. Subsequently, the hospital infection control system in Taiwan was reviewed and re-designed. These efforts might have also improved SSI control in hospitals and thus led to the inconsistent results with the study by Wu et al. [6].

Although a multi-level analysis was applied to manage the nesting issue in the data and three different methods were used to examine the relation between service volume and SSI, two common limitations of studies that use secondary data impacted this study. The first limitation was the accuracy of the outcome variable because data from health care claims data are not designed for infection surveillance. Previous studies indicated that the accuracy issue might exist in claim data because of an inaccurate code list or be distorted by payment scheme or other factors [27 –29]. It was unavoidable in this study. The second limitation relates to unmeasured variables, such as length of stay before operation, infection condition, hair removal, duration of operation, the environment, surgical skills, use of post-operative drains, number of operations involved, and the surgical site and wound care [17]. Furthermore, the information regarding type (elective or urgent) and incision site of surgery was not available in Taiwan NHI claims data.

Conclusion

The findings of this study suggest that surgeon volume and patient comorbidity may play more important roles than hospital volume in SSI. However, because the relation between surgeon volume and infection was not consistent, how to choose the optimal cutoff point should be an important issue for future research.

Footnotes

Acknowledgments

This study was supported by the Taiwan Centers for Disease Control (DOH100-DC-1020).

T.H.Y developed study concept, analyzed the data, and drafted the manuscript; Y.C.T. was involved in collecting data and drafting the manuscript; and K.P.C. collected data, coordinated the study, and revised the manuscript. All of the authors have prepared, read, and approved the manuscript.

The authors would like to thank the Centers for Disease Control of Taiwan for financially supporting this project.

Author Disclosure Statement

The authors certify that there is no conflict of interest with any financial organization regarding the material discussed in the manuscript.

References

Luft

, Bunker

, Enthoven

. Should operations be regionalized? The empirical relation between surgical volume and mortality. N Engl J Med, 1979; 301:1364–1369.

Vogel

, Dombrovskiy

, Graham

, Lowry

. The impact of hospital volume on the development of infectious complications after elective abdominal aortic surgery in the Medicare population. Vasc Endovascular Surg, 2011; 45:317–324.

Meyer

, Weitzel-Kage

, Sohr

, Gastmeier

. Impact of department volume on surgical site infections following arthroscopy, knee replacement or hip replacement. BMJ Qual Saf, 2011; 20:1069–1074

Shahian

, O'Brien

, Normand

, et al. Association of hospital coronary artery bypass volume with processes of care, mortality, morbidity, and the Society of Thoracic Surgeons composite quality score. J Thorac Cardiovasc Surg, 2010; 139:273–282.

Wei

, Lin

, Shi

, Chiu

. Effects of provider patient volume and comorbidity on clinical and economic outcomes for total knee arthroplasty: A population-based study. J Arthroplasty, 2010; 25:906–912.e901.

, Chen

, Ng

, Chu

. The relationship between surgical site infection and volume of coronary artery bypass graft surgeries: Taiwan experience. Infect Control Hosp Epidemiol, 2006; 27:308–311.

Anderson

, Hartwig

, Pappas

, et al. Surgical volume and the risk of surgical site infection in community hospitals: Size matters. Ann Surg, 2008; 247:343–349.

Huang

, Cheu

, Ying

, Wei

. Association between provider volume and comorbidity on hospital utilization and outcomes of total hip arthroplasty among National Health Insurance enrollees. J Formos Med Assoc, 2011; 110:401–409.

Kreder

, Grosso

, Williams

, et al. Provider volume and other predictors of outcome after total knee arthroplasty: A population study in Ontario. Can J Surg, 2003; 46:15–22.

10.

Chowdhury

, Dagash

, Pierro

. A systematic review of the impact of volume of surgery and specialization on patient outcome. Br J Surg, 2007; 94:145–161.

11.

Hodgson

, Zhang

, Zaslavsky

, et al. Relation of hospital volume to colostomy rates and survival for patients with rectal cancer. J Natl Cancer Inst, 2003; 95:708–716.

12.

Kuo

, Chung

, Lai

. Re-examining the significance of surgical volume to breast cancer survival and recurrence versus process quality of care in Taiwan. Health Serv Res, 2013; 48:26–46.

13.

Tung

, Chang

, Chen

. Associations of physician volume and weekend admissions with ischemic stroke outcome in Taiwan: A nationwide population-based study. Med Care, 2009; 47:1018–1025.

14.

Solomon

, Losina

, Baron

, et al. Contribution of hospital characteristics to the volume-outcome relationship: Dislocation and infection following total hip replacement surgery. Arthritis Rheum, 2002; 46:2436–2444.

15.

Hervey

, Purves

, Guller

, et al. Provider volume of total knee arthroplasties and patient outcomes in the HCUP-Nationwide Inpatient Sample. J Bone Joint Surg Am, 2003; 85-A:1775–1783.

16.

Al-Khatib

, Lucas

, Jollis

, et al. The relation between patients' outcomes and the volume of cardioverter-defibrillator implantation procedures performed by physicians treating Medicare beneficiaries. J Am Coll Cardiol, 2005; 46:1536–1540.

17.

Mangram

, Horan

, Pearson

, et al. Guideline for prevention of surgical site infection, 1999. Infect Control Hosp Epidemiol, 1999; 20:250–278.

18.

Raudenbush

, Bryk

. Hierarchical Linear Models: Applications and Data Analysis Methods. Newbury Park, CA: Sage, 2002.

19.

Hofmann

. An overview of the logic and rationale of hierarchical linear models. J Manage, 1997; 23:723–744.

20.

Peterson

, Coombs

, DeLong

, et al. Procedural volume as a marker of quality for CABG surgery. JAMA, 2004; 291:195–201.

21.

Hockenberry

, Lien

, Chou

. Surgeon and hospital volume as quality indicators for CABG in Taiwan: Examining hazard to mortality and accounting for unobserved heterogeneity. Health Serv Res, 2010; 45:1168–1187.

22.

Yasunaga

, Matsuyama

, Ohe

. Volume-outcome relationship in rectal cancer surgery: A new perspective. Surg Today, 2009; 39:663–668.

23.

Hollenbeck

, Hong

, Zaojun

, Birkmeyer

. Misclassification of hospital volume with Surveillance, Epidemiology, and End Results Medicare data. Surg Innov, 2007; 14:192–198.

24.

Browne

, Pietrobon

, Olson

. Hip fracture outcomes: Does surgeon or hospital volume really matter?. J Trauma, 2009; 66:809–814.

25.

Shah

, Wainess

, Karunakar

. Hemiarthroplasty for femoral neck fracture in the elderly surgeon and hospital volume-related outcomes. J Arthroplasty, 2005; 20:503–508.

26.

Katz

, Losina

, Barrett

, et al. Association between hospital and surgeon procedure volume and outcomes of total hip replacement in the United States medicare population. J Bone Joint Surg Am, 2001; 83-A:1622–1629.

27.

Jhung

, Banerjee

. Administrative coding data and health care-associated infections. Clin Infect Dis, 2009; 49:949–955.

28.

Stevenson

, Khan

, Dickman

, et al. Administrative coding data, compared with CDC/NHSN criteria, are poor indicators of health care-associated infections. Am J Infect Control, 2008; 36:155–164.

29.

Sands

, Yokoe

, Hooper

, et al. Detection of postoperative surgical-site infections: Comparison of health plan-based surveillance with hospital-based programs. Infect Control Hosp Epidemiol, 2003; 24:741–743.