Pediatr Cardiol (2015) 36:205–213 DOI 10.1007/s00246-014-0987-2
ORIGINAL ARTICLE
Surgical Volume, Hospital Quality, and Hospitalization Cost in Congenital Heart Surgery in the United States Titus Chan • Jaewhan Kim • L. LuAnn Minich Nelangi M. Pinto • Norman J. Waitzman
•
Received: 12 May 2014 / Accepted: 22 July 2014 / Published online: 7 August 2014 Ó Springer Science+Business Media New York 2014
Abstract Hospital volume has been associated with improved outcomes in congenital cardiac surgery. However, the relationship between hospital volume and hospitalization cost remains unclear. This study examines the relationship between hospital surgical volume and hospitalization costs, while accounting for measures of quality, in children undergoing congenital heart surgery. A retrospective, repeated cross-sectional analysis was performed, using discharges from the 2006 and 2009 Kids’ Inpatient Database. All pediatric admissions (\18 years) with a Risk Adjustment for Congenital Heart Surgery procedure and hospitalization cost/charge data were included. Multivariate, linear mixed regression models were run on hospitalization costs, with and without adjustment for indicators of quality (hospital mortality rate and complication rate). Both medium and high-volume hospitals (200–400 cases/ year and [400 cases/year, respectively) were associated with lower odds of mortality but not occurrence of a
complication. Hospital mortality was associated with the largest increase in hospitalization costs. High-volume hospitals ([400 cases/year) were associated with the lowest hospitalization costs per discharge ($37,775, p \ 0.01) when compared to low-($43,270) and medium($41,085)volume hospitals, prior to adjusting for quality indicators. However, when adjusting for hospital mortality rate, highvolume hospitals no longer demonstrated significant cost savings. When adjusting for hospital complication rate, high-volume hospitals continued to have the lowest hospitalization costs. High-volume hospitals are associated with a reduction in hospitalization costs that appear to be mediated through improvements in quality. Keywords Health economics Health services research Healthcare quality
Introduction T. Chan (&) Division of Critical Care Medicine and the Heart Center, Department of Pediatrics, Seattle Children’s Hospital, University of Washington, 4800 Sand Point Way NE, Seattle, WA 98105, USA e-mail: [email protected] J. Kim Department of Family and Preventive Medicine, University of Utah, Salt Lake City, UT, USA L. L. Minich N. M. Pinto Division of Pediatric Cardiology, Department of Pediatrics, Primary Children’s Hospital, University of Utah, Salt Lake City, UT, USA N. J. Waitzman Department of Economics, University of Utah, Salt Lake City, UT, USA
The increasing focus on value in health care delivery in the United States has led to a growing demand for costeffective health care. Recently, health investigators and policy officials have shifted their attention to the effect of hospital characteristics on both outcomes and costs. In several fields, including pediatric cardiac surgery, higher hospital and provider volume are associated with improved health care outcomes [8, 18]. However, the relationship between hospital volume and hospitalization costs remains unclear [2, 12] and may be confounded by the association between hospital volume and quality of care. The principle of economies of scale, in simplest terms, states that increasing production volume is associated with a disproportionate decrease in average costs. If this principle applied to congenital heart surgery, high-volume
123
206
hospitals would utilize fewer resources per patient than low-volume hospitals. However, because hospital surgical volume is associated with improved outcomes and poor outcomes have been associated with increased costs [1, 4], hospital-level quality of care may confound the relationship between volume and cost. Thus, the relationship between hospital surgical volume and cost must also account for variation in hospital-level quality. If highvolume hospitals are associated with improved clinical outcomes as well as lower hospitalization costs, there would be clinical and economic incentives to regionalize this highly specialized service. Using a US national health care dataset, a comprehensive surgical risk adjustment system and measures for both mortality and morbidity, we explored the influence of hospital surgical volume on cost in pediatric cardiac surgery.
Methods Kids’ Inpatient Database We utilized the 2006 and 2009 Kids’ Inpatient Database (KID), which is generated every 3 years by the Healthcare Cost and Utilization Project (HCUP) [6]. The KID is comprised of approximately three million de-identified pediatric discharge abstracts from 38 states in 2006 and 44 states in 2009. Each discharge abstract includes patientlevel demographics, diagnostic and procedure codes, indicators of co-morbidities, and hospital-level characteristics, such as teaching status and bed size. The sample for the analysis included discharges for all children (B18 years) who underwent congenital cardiac surgery as determined by qualifying International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9) codes in each KID hospitalization record. Risk Adjustment, Volume, and Cost Variables The Risk Adjustment for Congenital Heart Surgery-1 (RACHS) was used to stratify and adjust for the complexity of surgical procedures among the qualifying diagnostic and procedural codes [9]. While the RACHS categories are consensus based, higher RACHS categories (range 1–6) have been associated with increased surgical complexity, increased risk of mortality and morbidity, and higher inpatient hospitalization costs [2]. For admissions with multiple procedures, the highest RACHS category was assigned and only admissions that were associated with a RACHS surgical procedure were included for analysis. In some comparisons, RACHS categories were grouped into low-(RACHS 1 and 2), medium-(RACHS 3 and 4) and high-complexity (RACHS 5 and 6) procedures.
123
Pediatr Cardiol (2015) 36:205–213
The primary independent variable of interest was annual hospital surgical volume, computed by totaling the number of separate RACHS procedures in each individual hospital for each study year. The dependent variable of interest was the total hospitalization cost for each hospital admission. Each hospitalization record contains edited total charges, which is the total amount of money that the hospital charges any payer, exclusive of professional fees. Hospitalization cost was derived by multiplying each patient’s hospitalization charge by the hospital-specific cost-tocharge ratio (CCR). The CCR is constructed from allpayer, inpatient cost and charge data that hospitals supply to Centers for Medicare and Medicaid Services and has been used in a wide range of cost analyses [3, 7, 14–16]. In hospitals where the hospital-specific CCR is unavailable (\10 % of hospitals in the KID), a group-weighted average CCR (based on state, urban status, ownership, and bed size) was used to calculate hospitalization costs. All 2006 charges and costs were converted to equivalent 2009 US dollars using the Medical Care component of the Consumer Price Index. Patient-Level Variables The analyses adjusted for major, non-cardiac structural anomalies, prematurity and age at admission, consistent with previously published RACHS models [9]. The presence of a major, non-cardiac structural defect, and prematurity (gestational age \37 weeks) were denoted with dichotomous variables. Age at admission was divided into three groups: \30 days, 1 month to 1 year, and [1 year. We also examined race/ethnicity (categorized as non-Hispanic White, non-Hispanic Black, Hispanic ethnicity, and Other race/ethnicity) and expected source of payment (categorized as private insurance, Medicaid insurance, and Other forms of payment). Emergent admission was also included as a dichotomous variable since non-elective or emergent admissions (as classified by the KID) requiring cardiac surgery may be an indicator of severity of illness and a risk factor for increased resource utilization. In other pediatric studies, patients experiencing death or complications have high resource utilization [1, 4]. Thus, hospital mortality and complication status were included in the analysis. Using ICD-9 diagnosis codes for complications from the Wisconsin Medical Injuries Prevention Program [10, 11], the number of complications experienced by a patient was computed. Hospital-Level Variables Hospital-level characteristics incorporated in the analyses included location in a metropolitan statistical area (urban) versus location outside a metropolitan statistical area
Pediatr Cardiol (2015) 36:205–213
(non-urban). Designations for pediatric versus general hospitals and teaching versus non-teaching institutions were taken from the KID (based on data from the American Hospital Association Annual Survey of Hospitals and the Children’s Hospital Association). Separate measures of hospital-level quality were constructed for hospitals that performed [10 procedures/year: standardized mortality rate (SMR) and complication rate. The SMR adjusts an institution’s mortality rate for the complexity of each individual hospital’s surgical caseload. This is done by weighting the mortality rate for each RACHS category at each hospital by the national distribution of RACHS categories. A hospital complication rate was calculated by dividing the total number of complications that occurred in the individual hospital in each study year by the number of cardiac surgeries and multiplying this number by 100, resulting in the number of complications per 100 surgeries. Statistical Analysis Median costs for demographics, medical and surgical features, hospital characteristics, and socioeconomic variables were compared. The Wilcoxon–Mann–Whitney and the Kruskal–Wallis tests were used for the comparison of nonnormally distributed cost data. Data are presented as median values with the interquartile range (IQR; 25th and 75th quartiles). Because of the low numbers and the similarity in the distribution of costs and mortality rates, RACHS 5 and 6 were combined for parts of the analysis. Multivariate, hierarchical logistic regression models examining hospital volume, mortality, and complications were constructed, accounting for clustering of outcomes among patients admitted to the same hospital. Based on preliminary mortality models, surgical volume categories were simplified into low-volume (\200 cases/year), medium-volume (200–400 cases/year), and high-volume hospitals ([400 cases/year). The relationship between surgical volume and cost was analyzed by constructing a multivariate, generalized linear mixed model, accounting for nesting of patients within hospitals and using a gamma family distribution with log link function. A p value of \0.05 in the univariate analysis was considered statistically significant and was used as criteria for consideration in the multivariate model. An area wage index, constructed by the Centers for Medicare and Medicaid, was included in all of the cost models to adjust for geographical variations in cost and price factors unrelated to inpatient care. An initial model was constructed, strictly examining the relationship between surgical volume and hospitalization costs without adjusting for markers of quality (SMR or hospital complication rate). Successive models included quality indicators as fixed effects. Recycled predictions were used
207
to model the fixed-effect estimates of cost for each characteristic in the regression models. These fixed-effect estimates of cost were used as the adjusted hospitalization costs. Because unadjusted costs do not account for covariates (such as surgical complexity) that may vary with hospital volume and also affect hospitalization costs, unadjusted cost estimates may cause omitted variable bias that may lead to inaccurate estimates of the association between surgical volume and hospitalization costs. The adjusted hospitalization costs in the multivariate analysis represent the predicted cost for hospitalization if all other covariates are held constant at the national average. Thus, the adjusted hospitalization costs in the first multivariate model represent the difference in costs when an identical, ‘‘average’’ patient is hospitalized at low-, medium-, and high-volume hospitals. The subsequent models that adjust for SMR and complication rates predict the hospitalization costs for an ‘‘average’’ patient admitted to low-, medium-, and high-volume hospitals that possess identical mortality and complication rates, respectively. Finally, to determine if the cost-volume relationship varied with mortality and complication status, separate cost models stratified on mortality status (survivor versus non-survivor) and experiencing a complication (no complications versus any complication) were constructed.
Results Univariate Analysis From both the 2006 and 2009 KID dataset, there were 24,992 records of children (B18 years of age) who underwent a RACHS surgical procedure with hospitalization charge and cost data available (Table 1). Patients were most commonly \1 year of age, non-Hispanic White, covered by private insurance or Medicaid and underwent low- or medium-complexity surgical procedures. A very small proportion of patients (1.1 %) underwent surgery (all procedures were RACHS categories 1–3) at a non-urban hospital, likely reflecting the preponderant placement of pediatric cardiac surgery centers in metropolitan areas. The overall mortality rate of 2.2 % was associated with increased surgical complexity (RACHS 1 mortality rate: 0.4 %, RACHS 2: 0.8 %, RACHS 3: 2.0 %, RACHS 4: 4.1 %, RACHS 5 and 6: 17.8 %). The majority of hospitals (Table 2) were in the lowest volume category (\200 cases/year) and approximately 68 % of these hospitals performed \10 operations/year (147 in 2006 and 209 in 2009). Despite a small number of hospitals, high-volume hospitals ([400 cases/year) performed a large proportion of higher complexity procedures (26 and 32 % of all medium- and high-complexity
123
208 Table 1 Baseline characteristics
a
Some data missing Risk Adjustment for Congenital Heart Surgery c Interquartile range b
Pediatr Cardiol (2015) 36:205–213
2006 N = 12,913 (%) Died 315 (2.4) Age category [1 year 5,537 (42.9) 1 month to 1 year 5,190 (40.2) \1 month 2,186 (16.9) Female 5,850 (45.3) Insurancea Private 6,061 (46.9) Medicaid 5,615 (43.5) Other 1,231 (9.5) Race/ethnicity Non-Hispanic white 5,219 (40.4) Non-Hispanic black 1,107 (8.6) Hispanic ethnicity 2,232 (17.3) Other race/ethnicity 1,193 (9.2) Missing data 3,162 (24.5) RACHS categoryb 1 1,659 (12.9) 2 4,434 (34.3) 3 4,855 (37.6) 4 1,506 (11.7) 5 19 (0.1) 6 440 (3.4) Premature birth 320 (2.5) Non-cardiac structural 581 (4.5) defects present Hospital type General hospital 7,058 (54.7) Pediatric hospital 5,855 (45.3) Emergency admission 4,003 (31.0) Unadjusted cost (2009 USD) 40,483 (25,923–71,564) Median (IQRc) Length of stay (days) 7 (4,14) Median (IQRc)
procedures, respectively). Complexity-specific volume largely paralleled total hospital volume (correlation coefficient: RACHS 1 and 2: 0.966; RACHS 3 and 4: 0.986; RACHS 5 and 6: 0.863, p \ 0.001 for all comparisons). Across volume categories, there were global differences in complication rates (p \ 0.001) and median unadjusted hospitalization costs (p \ 0.001), while no difference was noted in standardized mortality (SMR range: 0–22 %; p = 0.778). The median facility charge for inpatient hospitalization requiring pediatric cardiac surgery was $106,844 (2009 US Dollars), while the median unadjusted cost was $39,627 USD. Median unadjusted hospitalization costs were lower at low-volume hospitals without adjustment for other covariates (Table 2). In the univariate analysis of patient-level variables, hospital mortality, male sex,
123
2009 N = 12,079 (%)
Total N = 24,992 (%)
224 (1.9)
539 (2.2)
6,355 (52.6) 3,641 (30.1) 2,083(17.2) 5,620 (46.5)
11,892 (47.6) 8,831 (35.3) 4,269 (17.1) 11,470 (45.9)
5,707 (47.3) 5,339 (44.2) 1,023 (8.5)
11,768 (47.1) 10,954 (43.8) 2,254 (9.0)
5,588 1,157 2,126 1,231 1,977
10,807 (43.2) 2,264 (9.1) 4,358 (17.4) 2,424 (9.7) 5,139 (20.6)
(46.2) (9.6) (17.6) (10.2) (16.4)
1,497 (12.4) 4,193 (34.7) 4,667 (38.7) 1,343 (11.1) 12 (0.1) 367 (3.0) 297 (2.5) 699 (5.8)
3,156 (12.7) 8,627 (34.5) 9,522 (38.1) 2,849 (11.4) 31 (0.1) 807 (3.2) 617 (2.5) 1,280 (5.1)
6,849 (56.7) 5,230 (43.3) 2,698 (22.3)
13,907 (55.7) 11,085 (44.3) 6,701 (26.8)
38,685 (25,153–68,770)
39,627 (25,570–70,083)
6 (4,13)
6 (4,14)
premature birth, younger age, non-white race/ethnicity, Medicaid insurance, non-cardiac structural defects, emergent admission, increasing number of complications, and greater surgical complexity were associated with increased hospitalization costs (Fig. 1, p \ 0.05). In terms of hospital characteristics, admissions to a pediatric, teaching, or urban hospital were associated with increased hospitalization costs (p \ 0.05). Hospital SMR and complication rate were correlated with individual hospitalization costs (r = -0.0396 and 0.0188, respectively; p \ 0.05). Multivariate Analysis Baseline logistic regression models examining the association between hospital volume and mortality and
Pediatr Cardiol (2015) 36:205–213
209
Table 2 Hospital characteristics by volume categories
Hospital volume category Low
Medium
High
\200
200–400
[400
2006
219
29
10
2009
302
22
9
2006
4,583
5,341
3,286
2009
5,341
3,801
2,937
26 (14–48)e
82 (70–94)
138 (110–158)
e
83 (66–99) 7 (4–10)
170 (125–209) 14 (9–20)
1.7 (0–4.1)e
1.6 (0.96–2.8)
1.5 (0.93–2.3)
63 (39–93)e
53 (44–87)
75 (65–104)
43,140 (34,685–54,172)
42,178 (36,908–48,248)
Cases/year Number of hospitals
Number of admissions
Surgical volume Median (IQR) RACHSa 1 and 2 a
a
Risk Adjustment for Congenital Heart Surgery
b c
Standardized mortality rate Interquartile range
d
Number of complications/100 surgeries
e
Excluding hospitals with surgical volume B10 cases/year
RACHS 3 and 4 RACHSa 5 and 6
25 (12–35) 0 (0–3)e
SMRb (%) Median (IQRc) Complication rate Median (IQRc)
d
Unadjusted cost (2009 USD) Median (IQRc)
30,908e (24,418–40,287)
occurrence of one or more complications were constructed, adjusting for surgical complexity, age category, insurance, race/ethnicity, prematurity, non-cardiac structural defects, hospital type, emergency admission, hospital urban status, and teaching status (Table 3). In the hospital mortality model, low surgical volume (\200 cases/year) was associated with increased odds of mortality compared to hospitals with high surgical volume ([400 cases/year) (odds ratio (OR) 1.83, 95 % confidence interval (CI) 1.20–2.79). The odds of mortality were not statistically different between medium- and high-volume hospitals. In contrast to its relationship with mortality, surgical volume was not associated with the occurrence of a complication. The relationship between hospitalization costs, surgical volume, and hospital-level quality indicators, adjusting for patient- and hospital-level covariates, are presented in Fig. 2. Without quality indicators (no indicators) and with all other covariates held at the national average, high-volume hospitals were associated with the lowest adjusted hospitalization costs among all hospitals ($37,775; p \ 0.01), and medium-sized hospitals trended toward lower costs ($41,085) than low-volume hospitals ($43,270). In the models adjusting for SMR and hospital complication rates, high-volume hospitals continued to have the lowest adjusted costs of all hospitals. However, in the SMR-adjusted model, this difference was no longer significant. In all of the quality-adjusted models, medium volume hospitals had the highest costs of all volume groups. The predicted costs associated with the other
covariates in the regression model without quality indicators are depicted in Fig. 1. Of all patient-level variables, hospital mortality was associated with the greatest increase in cost ($39,405/hospital admission). In separate models restricted to non-survivors and patients experiencing a complication, high-volume hospitals were not associated with reductions in cost (Fig. 2). Additionally, in the regression models restricted to survivors and patients having no complication, high-volume hospitals had adjusted higher costs compared to mediumand low-volume hospitals (p \ 0.05).
Discussion In this study, hospitals performing [400 surgical cases/ year had the lowest hospitalization costs for children undergoing congenital heart surgery. These savings appear to be due to improvements in quality afforded by highvolume hospitals. By reducing hospital mortality, the risk factor associated with the greatest increase in cost, highvolume hospitals were able to offer improvements in quality with concomitant reductions in cost. While low-volume hospitals may have lower unadjusted costs in the univariate analysis, this may be confounded by the patient mix and severity of lesions dealt with at hospitals with different surgical volumes. However, when these other covariates are held constant, high-volume hospitals demonstrate lower hospitalization costs when
123
210
Pediatr Cardiol (2015) 36:205–213
Characteristic
Univariate
Multivariate
Survived to Hospital Discharge Died in Hospital Age >1year Age 1month−1year Age
ORIGINAL ARTICLE
Surgical Volume, Hospital Quality, and Hospitalization Cost in Congenital Heart Surgery in the United States Titus Chan • Jaewhan Kim • L. LuAnn Minich Nelangi M. Pinto • Norman J. Waitzman
•
Received: 12 May 2014 / Accepted: 22 July 2014 / Published online: 7 August 2014 Ó Springer Science+Business Media New York 2014
Abstract Hospital volume has been associated with improved outcomes in congenital cardiac surgery. However, the relationship between hospital volume and hospitalization cost remains unclear. This study examines the relationship between hospital surgical volume and hospitalization costs, while accounting for measures of quality, in children undergoing congenital heart surgery. A retrospective, repeated cross-sectional analysis was performed, using discharges from the 2006 and 2009 Kids’ Inpatient Database. All pediatric admissions (\18 years) with a Risk Adjustment for Congenital Heart Surgery procedure and hospitalization cost/charge data were included. Multivariate, linear mixed regression models were run on hospitalization costs, with and without adjustment for indicators of quality (hospital mortality rate and complication rate). Both medium and high-volume hospitals (200–400 cases/ year and [400 cases/year, respectively) were associated with lower odds of mortality but not occurrence of a
complication. Hospital mortality was associated with the largest increase in hospitalization costs. High-volume hospitals ([400 cases/year) were associated with the lowest hospitalization costs per discharge ($37,775, p \ 0.01) when compared to low-($43,270) and medium($41,085)volume hospitals, prior to adjusting for quality indicators. However, when adjusting for hospital mortality rate, highvolume hospitals no longer demonstrated significant cost savings. When adjusting for hospital complication rate, high-volume hospitals continued to have the lowest hospitalization costs. High-volume hospitals are associated with a reduction in hospitalization costs that appear to be mediated through improvements in quality. Keywords Health economics Health services research Healthcare quality
Introduction T. Chan (&) Division of Critical Care Medicine and the Heart Center, Department of Pediatrics, Seattle Children’s Hospital, University of Washington, 4800 Sand Point Way NE, Seattle, WA 98105, USA e-mail: [email protected] J. Kim Department of Family and Preventive Medicine, University of Utah, Salt Lake City, UT, USA L. L. Minich N. M. Pinto Division of Pediatric Cardiology, Department of Pediatrics, Primary Children’s Hospital, University of Utah, Salt Lake City, UT, USA N. J. Waitzman Department of Economics, University of Utah, Salt Lake City, UT, USA
The increasing focus on value in health care delivery in the United States has led to a growing demand for costeffective health care. Recently, health investigators and policy officials have shifted their attention to the effect of hospital characteristics on both outcomes and costs. In several fields, including pediatric cardiac surgery, higher hospital and provider volume are associated with improved health care outcomes [8, 18]. However, the relationship between hospital volume and hospitalization costs remains unclear [2, 12] and may be confounded by the association between hospital volume and quality of care. The principle of economies of scale, in simplest terms, states that increasing production volume is associated with a disproportionate decrease in average costs. If this principle applied to congenital heart surgery, high-volume
123
206
hospitals would utilize fewer resources per patient than low-volume hospitals. However, because hospital surgical volume is associated with improved outcomes and poor outcomes have been associated with increased costs [1, 4], hospital-level quality of care may confound the relationship between volume and cost. Thus, the relationship between hospital surgical volume and cost must also account for variation in hospital-level quality. If highvolume hospitals are associated with improved clinical outcomes as well as lower hospitalization costs, there would be clinical and economic incentives to regionalize this highly specialized service. Using a US national health care dataset, a comprehensive surgical risk adjustment system and measures for both mortality and morbidity, we explored the influence of hospital surgical volume on cost in pediatric cardiac surgery.
Methods Kids’ Inpatient Database We utilized the 2006 and 2009 Kids’ Inpatient Database (KID), which is generated every 3 years by the Healthcare Cost and Utilization Project (HCUP) [6]. The KID is comprised of approximately three million de-identified pediatric discharge abstracts from 38 states in 2006 and 44 states in 2009. Each discharge abstract includes patientlevel demographics, diagnostic and procedure codes, indicators of co-morbidities, and hospital-level characteristics, such as teaching status and bed size. The sample for the analysis included discharges for all children (B18 years) who underwent congenital cardiac surgery as determined by qualifying International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9) codes in each KID hospitalization record. Risk Adjustment, Volume, and Cost Variables The Risk Adjustment for Congenital Heart Surgery-1 (RACHS) was used to stratify and adjust for the complexity of surgical procedures among the qualifying diagnostic and procedural codes [9]. While the RACHS categories are consensus based, higher RACHS categories (range 1–6) have been associated with increased surgical complexity, increased risk of mortality and morbidity, and higher inpatient hospitalization costs [2]. For admissions with multiple procedures, the highest RACHS category was assigned and only admissions that were associated with a RACHS surgical procedure were included for analysis. In some comparisons, RACHS categories were grouped into low-(RACHS 1 and 2), medium-(RACHS 3 and 4) and high-complexity (RACHS 5 and 6) procedures.
123
Pediatr Cardiol (2015) 36:205–213
The primary independent variable of interest was annual hospital surgical volume, computed by totaling the number of separate RACHS procedures in each individual hospital for each study year. The dependent variable of interest was the total hospitalization cost for each hospital admission. Each hospitalization record contains edited total charges, which is the total amount of money that the hospital charges any payer, exclusive of professional fees. Hospitalization cost was derived by multiplying each patient’s hospitalization charge by the hospital-specific cost-tocharge ratio (CCR). The CCR is constructed from allpayer, inpatient cost and charge data that hospitals supply to Centers for Medicare and Medicaid Services and has been used in a wide range of cost analyses [3, 7, 14–16]. In hospitals where the hospital-specific CCR is unavailable (\10 % of hospitals in the KID), a group-weighted average CCR (based on state, urban status, ownership, and bed size) was used to calculate hospitalization costs. All 2006 charges and costs were converted to equivalent 2009 US dollars using the Medical Care component of the Consumer Price Index. Patient-Level Variables The analyses adjusted for major, non-cardiac structural anomalies, prematurity and age at admission, consistent with previously published RACHS models [9]. The presence of a major, non-cardiac structural defect, and prematurity (gestational age \37 weeks) were denoted with dichotomous variables. Age at admission was divided into three groups: \30 days, 1 month to 1 year, and [1 year. We also examined race/ethnicity (categorized as non-Hispanic White, non-Hispanic Black, Hispanic ethnicity, and Other race/ethnicity) and expected source of payment (categorized as private insurance, Medicaid insurance, and Other forms of payment). Emergent admission was also included as a dichotomous variable since non-elective or emergent admissions (as classified by the KID) requiring cardiac surgery may be an indicator of severity of illness and a risk factor for increased resource utilization. In other pediatric studies, patients experiencing death or complications have high resource utilization [1, 4]. Thus, hospital mortality and complication status were included in the analysis. Using ICD-9 diagnosis codes for complications from the Wisconsin Medical Injuries Prevention Program [10, 11], the number of complications experienced by a patient was computed. Hospital-Level Variables Hospital-level characteristics incorporated in the analyses included location in a metropolitan statistical area (urban) versus location outside a metropolitan statistical area
Pediatr Cardiol (2015) 36:205–213
(non-urban). Designations for pediatric versus general hospitals and teaching versus non-teaching institutions were taken from the KID (based on data from the American Hospital Association Annual Survey of Hospitals and the Children’s Hospital Association). Separate measures of hospital-level quality were constructed for hospitals that performed [10 procedures/year: standardized mortality rate (SMR) and complication rate. The SMR adjusts an institution’s mortality rate for the complexity of each individual hospital’s surgical caseload. This is done by weighting the mortality rate for each RACHS category at each hospital by the national distribution of RACHS categories. A hospital complication rate was calculated by dividing the total number of complications that occurred in the individual hospital in each study year by the number of cardiac surgeries and multiplying this number by 100, resulting in the number of complications per 100 surgeries. Statistical Analysis Median costs for demographics, medical and surgical features, hospital characteristics, and socioeconomic variables were compared. The Wilcoxon–Mann–Whitney and the Kruskal–Wallis tests were used for the comparison of nonnormally distributed cost data. Data are presented as median values with the interquartile range (IQR; 25th and 75th quartiles). Because of the low numbers and the similarity in the distribution of costs and mortality rates, RACHS 5 and 6 were combined for parts of the analysis. Multivariate, hierarchical logistic regression models examining hospital volume, mortality, and complications were constructed, accounting for clustering of outcomes among patients admitted to the same hospital. Based on preliminary mortality models, surgical volume categories were simplified into low-volume (\200 cases/year), medium-volume (200–400 cases/year), and high-volume hospitals ([400 cases/year). The relationship between surgical volume and cost was analyzed by constructing a multivariate, generalized linear mixed model, accounting for nesting of patients within hospitals and using a gamma family distribution with log link function. A p value of \0.05 in the univariate analysis was considered statistically significant and was used as criteria for consideration in the multivariate model. An area wage index, constructed by the Centers for Medicare and Medicaid, was included in all of the cost models to adjust for geographical variations in cost and price factors unrelated to inpatient care. An initial model was constructed, strictly examining the relationship between surgical volume and hospitalization costs without adjusting for markers of quality (SMR or hospital complication rate). Successive models included quality indicators as fixed effects. Recycled predictions were used
207
to model the fixed-effect estimates of cost for each characteristic in the regression models. These fixed-effect estimates of cost were used as the adjusted hospitalization costs. Because unadjusted costs do not account for covariates (such as surgical complexity) that may vary with hospital volume and also affect hospitalization costs, unadjusted cost estimates may cause omitted variable bias that may lead to inaccurate estimates of the association between surgical volume and hospitalization costs. The adjusted hospitalization costs in the multivariate analysis represent the predicted cost for hospitalization if all other covariates are held constant at the national average. Thus, the adjusted hospitalization costs in the first multivariate model represent the difference in costs when an identical, ‘‘average’’ patient is hospitalized at low-, medium-, and high-volume hospitals. The subsequent models that adjust for SMR and complication rates predict the hospitalization costs for an ‘‘average’’ patient admitted to low-, medium-, and high-volume hospitals that possess identical mortality and complication rates, respectively. Finally, to determine if the cost-volume relationship varied with mortality and complication status, separate cost models stratified on mortality status (survivor versus non-survivor) and experiencing a complication (no complications versus any complication) were constructed.
Results Univariate Analysis From both the 2006 and 2009 KID dataset, there were 24,992 records of children (B18 years of age) who underwent a RACHS surgical procedure with hospitalization charge and cost data available (Table 1). Patients were most commonly \1 year of age, non-Hispanic White, covered by private insurance or Medicaid and underwent low- or medium-complexity surgical procedures. A very small proportion of patients (1.1 %) underwent surgery (all procedures were RACHS categories 1–3) at a non-urban hospital, likely reflecting the preponderant placement of pediatric cardiac surgery centers in metropolitan areas. The overall mortality rate of 2.2 % was associated with increased surgical complexity (RACHS 1 mortality rate: 0.4 %, RACHS 2: 0.8 %, RACHS 3: 2.0 %, RACHS 4: 4.1 %, RACHS 5 and 6: 17.8 %). The majority of hospitals (Table 2) were in the lowest volume category (\200 cases/year) and approximately 68 % of these hospitals performed \10 operations/year (147 in 2006 and 209 in 2009). Despite a small number of hospitals, high-volume hospitals ([400 cases/year) performed a large proportion of higher complexity procedures (26 and 32 % of all medium- and high-complexity
123
208 Table 1 Baseline characteristics
a
Some data missing Risk Adjustment for Congenital Heart Surgery c Interquartile range b
Pediatr Cardiol (2015) 36:205–213
2006 N = 12,913 (%) Died 315 (2.4) Age category [1 year 5,537 (42.9) 1 month to 1 year 5,190 (40.2) \1 month 2,186 (16.9) Female 5,850 (45.3) Insurancea Private 6,061 (46.9) Medicaid 5,615 (43.5) Other 1,231 (9.5) Race/ethnicity Non-Hispanic white 5,219 (40.4) Non-Hispanic black 1,107 (8.6) Hispanic ethnicity 2,232 (17.3) Other race/ethnicity 1,193 (9.2) Missing data 3,162 (24.5) RACHS categoryb 1 1,659 (12.9) 2 4,434 (34.3) 3 4,855 (37.6) 4 1,506 (11.7) 5 19 (0.1) 6 440 (3.4) Premature birth 320 (2.5) Non-cardiac structural 581 (4.5) defects present Hospital type General hospital 7,058 (54.7) Pediatric hospital 5,855 (45.3) Emergency admission 4,003 (31.0) Unadjusted cost (2009 USD) 40,483 (25,923–71,564) Median (IQRc) Length of stay (days) 7 (4,14) Median (IQRc)
procedures, respectively). Complexity-specific volume largely paralleled total hospital volume (correlation coefficient: RACHS 1 and 2: 0.966; RACHS 3 and 4: 0.986; RACHS 5 and 6: 0.863, p \ 0.001 for all comparisons). Across volume categories, there were global differences in complication rates (p \ 0.001) and median unadjusted hospitalization costs (p \ 0.001), while no difference was noted in standardized mortality (SMR range: 0–22 %; p = 0.778). The median facility charge for inpatient hospitalization requiring pediatric cardiac surgery was $106,844 (2009 US Dollars), while the median unadjusted cost was $39,627 USD. Median unadjusted hospitalization costs were lower at low-volume hospitals without adjustment for other covariates (Table 2). In the univariate analysis of patient-level variables, hospital mortality, male sex,
123
2009 N = 12,079 (%)
Total N = 24,992 (%)
224 (1.9)
539 (2.2)
6,355 (52.6) 3,641 (30.1) 2,083(17.2) 5,620 (46.5)
11,892 (47.6) 8,831 (35.3) 4,269 (17.1) 11,470 (45.9)
5,707 (47.3) 5,339 (44.2) 1,023 (8.5)
11,768 (47.1) 10,954 (43.8) 2,254 (9.0)
5,588 1,157 2,126 1,231 1,977
10,807 (43.2) 2,264 (9.1) 4,358 (17.4) 2,424 (9.7) 5,139 (20.6)
(46.2) (9.6) (17.6) (10.2) (16.4)
1,497 (12.4) 4,193 (34.7) 4,667 (38.7) 1,343 (11.1) 12 (0.1) 367 (3.0) 297 (2.5) 699 (5.8)
3,156 (12.7) 8,627 (34.5) 9,522 (38.1) 2,849 (11.4) 31 (0.1) 807 (3.2) 617 (2.5) 1,280 (5.1)
6,849 (56.7) 5,230 (43.3) 2,698 (22.3)
13,907 (55.7) 11,085 (44.3) 6,701 (26.8)
38,685 (25,153–68,770)
39,627 (25,570–70,083)
6 (4,13)
6 (4,14)
premature birth, younger age, non-white race/ethnicity, Medicaid insurance, non-cardiac structural defects, emergent admission, increasing number of complications, and greater surgical complexity were associated with increased hospitalization costs (Fig. 1, p \ 0.05). In terms of hospital characteristics, admissions to a pediatric, teaching, or urban hospital were associated with increased hospitalization costs (p \ 0.05). Hospital SMR and complication rate were correlated with individual hospitalization costs (r = -0.0396 and 0.0188, respectively; p \ 0.05). Multivariate Analysis Baseline logistic regression models examining the association between hospital volume and mortality and
Pediatr Cardiol (2015) 36:205–213
209
Table 2 Hospital characteristics by volume categories
Hospital volume category Low
Medium
High
\200
200–400
[400
2006
219
29
10
2009
302
22
9
2006
4,583
5,341
3,286
2009
5,341
3,801
2,937
26 (14–48)e
82 (70–94)
138 (110–158)
e
83 (66–99) 7 (4–10)
170 (125–209) 14 (9–20)
1.7 (0–4.1)e
1.6 (0.96–2.8)
1.5 (0.93–2.3)
63 (39–93)e
53 (44–87)
75 (65–104)
43,140 (34,685–54,172)
42,178 (36,908–48,248)
Cases/year Number of hospitals
Number of admissions
Surgical volume Median (IQR) RACHSa 1 and 2 a
a
Risk Adjustment for Congenital Heart Surgery
b c
Standardized mortality rate Interquartile range
d
Number of complications/100 surgeries
e
Excluding hospitals with surgical volume B10 cases/year
RACHS 3 and 4 RACHSa 5 and 6
25 (12–35) 0 (0–3)e
SMRb (%) Median (IQRc) Complication rate Median (IQRc)
d
Unadjusted cost (2009 USD) Median (IQRc)
30,908e (24,418–40,287)
occurrence of one or more complications were constructed, adjusting for surgical complexity, age category, insurance, race/ethnicity, prematurity, non-cardiac structural defects, hospital type, emergency admission, hospital urban status, and teaching status (Table 3). In the hospital mortality model, low surgical volume (\200 cases/year) was associated with increased odds of mortality compared to hospitals with high surgical volume ([400 cases/year) (odds ratio (OR) 1.83, 95 % confidence interval (CI) 1.20–2.79). The odds of mortality were not statistically different between medium- and high-volume hospitals. In contrast to its relationship with mortality, surgical volume was not associated with the occurrence of a complication. The relationship between hospitalization costs, surgical volume, and hospital-level quality indicators, adjusting for patient- and hospital-level covariates, are presented in Fig. 2. Without quality indicators (no indicators) and with all other covariates held at the national average, high-volume hospitals were associated with the lowest adjusted hospitalization costs among all hospitals ($37,775; p \ 0.01), and medium-sized hospitals trended toward lower costs ($41,085) than low-volume hospitals ($43,270). In the models adjusting for SMR and hospital complication rates, high-volume hospitals continued to have the lowest adjusted costs of all hospitals. However, in the SMR-adjusted model, this difference was no longer significant. In all of the quality-adjusted models, medium volume hospitals had the highest costs of all volume groups. The predicted costs associated with the other
covariates in the regression model without quality indicators are depicted in Fig. 1. Of all patient-level variables, hospital mortality was associated with the greatest increase in cost ($39,405/hospital admission). In separate models restricted to non-survivors and patients experiencing a complication, high-volume hospitals were not associated with reductions in cost (Fig. 2). Additionally, in the regression models restricted to survivors and patients having no complication, high-volume hospitals had adjusted higher costs compared to mediumand low-volume hospitals (p \ 0.05).
Discussion In this study, hospitals performing [400 surgical cases/ year had the lowest hospitalization costs for children undergoing congenital heart surgery. These savings appear to be due to improvements in quality afforded by highvolume hospitals. By reducing hospital mortality, the risk factor associated with the greatest increase in cost, highvolume hospitals were able to offer improvements in quality with concomitant reductions in cost. While low-volume hospitals may have lower unadjusted costs in the univariate analysis, this may be confounded by the patient mix and severity of lesions dealt with at hospitals with different surgical volumes. However, when these other covariates are held constant, high-volume hospitals demonstrate lower hospitalization costs when
123
210
Pediatr Cardiol (2015) 36:205–213
Characteristic
Univariate
Multivariate
Survived to Hospital Discharge Died in Hospital Age >1year Age 1month−1year Age
