High-volume Arthroplasty Centers Are Associated With Lower Hospital Costs When Performing Primary THA and TKA: A Database Study of 288,909 Medicare Claims for Procedures Performed in 2019

Abstract

Background

With bundled payments and alternative reimbursement models expanding in scope and scale, reimbursements to hospitals are declining in value. As a result, cost reduction at the hospital level is paramount for the sustainability of profitable inpatient arthroplasty practices. Although multiple prior studies have investigated cost variation in arthroplasty surgery, it is unknown whether contemporary inpatient arthroplasty practices benefit from economies of scale after accounting for hospital characteristics and patient selection factors. Quantifying the independent effects of volume-based cost variation may be important for guiding future value-based health reform.

Questions/purposes

We performed this study to (1) determine whether the cost incurred by hospitals for performing primary inpatient THA and TKA is independently associated with hospital volume and (2) establish whether length of stay and discharge to home are associated with hospital volume.

Methods

The primary data source for this study was the Medicare Provider Analysis and Review Limited Data Set, which includes claims data for 100% of inpatient Medicare hospitalizations. We included patients undergoing primary elective inpatient THA and TKA in 2019. Exclusion criteria included non–Inpatient Prospective Payment System hospitalizations, nonelective admissions, bilateral procedures, and patients with cancer of the pelvis or lower extremities. A total of 500,658 arthroplasties were performed across 2762 hospitals for 492,262 Medicare beneficiaries during the study period; 59% (288,909 of 492,262) of procedures were analyzed after the exclusion criteria were applied. Most exclusions (37% [182,733 of 492,262]) were because of non–Inpatient Prospective Payment System hospitalizations. Among the study group, 87% (251,996 of 288,909) of procedures were in patients who were 65 to 84 years old, 88% (255,415 of 288,909) were performed in patients who were White, and 63% (180,688 of 288,909) were in patients who were women. Elixhauser comorbidities and van Walraven indices were calculated as measures of patient health status. Hospital costs were estimated by multiplying cost-to-charge ratios obtained from the 2019 Impact File by total hospital charges. This methodology enabled us to use the large Medicare Provider Analysis and Review database, which helped decrease the influence of random cost variation through the law of large numbers. Hospital volumes were calculated by stratifying claims by national provider identification number and counting the number of claims per national provider identification number. The data were then grouped into bins of increasing hospital volume to more easily compare larger-volume and smaller-volume centers. The relationship between hospital costs and volume was analyzed using univariable and multivariable generalized linear models. Results are reported as exponential coefficients, which can be interpreted as relative differences in cost. The impact of surgical volume on length of stay and discharge to home was assessed using binary logistic regression, considering the nested structure of the data, and results are reported as odds ratios (OR).

Results

Hospital cost and mean length of stay decreased, while rates of discharge to home increased with increasing hospital volume. After controlling for potential confounding variables such as patient demographics, health status, and geographic location, we found that inpatient arthroplasty costs at hospitals with 10 or fewer, 11 to 100, and 101 to 200 procedures annually were 1.32 (95% confidence interval [CI] 1.30 to 1.34; p < 0.001), 1.17 (95% CI 1.17 to 1.17; p < 0.001), and 1.10 (95% CI 1.10 to 1.10; p < 0.001) times greater than those of hospitals with 201 or more inpatient procedures annually. In addition, patients treated at smaller-volume hospitals had increased odds of experiencing a length of stay longer than 2 days (OR 1.25 to 3.44 [95% CI 1.10 to 4.03]; p < 0.001) and decreased odds of being discharged to home (OR 0.34 to 0.78 [95% CI 0.29 to 0.86]; p < 0.001).

Conclusion

Higher-volume hospitals incur lower costs, shorter lengths of stay, and higher rates of discharge to home than lower-volume hospitals when performing inpatient THA and TKA. These findings suggest that small and medium-sized regional hospitals are disproportionately impacted by declining reimbursement and may necessitate special treatment to remain viable as bundled payment models continue to erode hospital payments. Further research is also warranted to identify the key drivers of this volume-based cost variation, which may facilitate quality improvement initiatives at the hospital and policy levels.

Introduction

Orthopaedic surgery is a leading driver of healthcare expenditures in the United States [25, 42]. Public and private payors are making major efforts to reduce the national financial impact of orthopaedic surgery. Specifically, elective hip and knee arthroplasties have been targeted for cost reduction because of the large and growing volume of procedures. The Centers for Medicare and Medicaid Services has introduced alternative reimbursement systems such as the Bundled Payments for Care Improvement initiative [10] and the Comprehensive Care for Joint Replacement [11] to reward high-quality, lower-cost healthcare. These systems offer a single payment for multiple services received by beneficiaries during an episode of care, which includes the index hospitalization and the postacute care period. Hospitals may be penalized or generate profits depending on quality-adjusted price targets established by the Centers for Medicare and Medicaid. The price targets of such programs, however, are not static and are regularly revised downward. In 2013, for example, the Bundled Payments for Care Improvement target price for lower extremity joint replacement at academic hospitals in New York City was USD 35,000. By 2018, these same hospitals entered the Comprehensive Care for Joint Replacement program with a target price of USD 23,850 [29]. In this environment of rapidly declining hospital payments, improving efficiency and hospital costs is paramount to preserve the sustainability of inpatient arthroplasty services.

Multiple prior studies have examined the determinants of cost variation in primary THA and TKA [7-9, 20, 46, 50, 53]. However, little has been published about the relationship between hospital costs and volume in arthroplasty [19, 22]. Prior research on the subject is either outdated [19] or based on univariable analyses without controlling for potential confounding variables [22]. Moreover, the body of research investigating volume-based cost variation outside orthopaedics is inconclusive [4, 5, 13, 36, 38, 39, 45, 56, 58]. Some authors have observed large hospitals to be more scale efficient [4, 39], while others have found small and medium-sized hospitals to be more productive, suggesting possible diseconomies of scale in healthcare [36, 45]. It is therefore unknown whether contemporary inpatient arthroplasty practices may benefit from increasing economies of scale after accounting for hospital characteristics and patient selection factors. Quantifying the independent effects of volume-based cost variation may be important for guiding and interpreting future value-based health reform.

Therefore, the objectives of this study were to (1) determine whether the cost incurred by hospitals for performing primary inpatient THA and TKA is independently associated with hospital volume and (2) establish whether length of stay (LOS) and discharge to home are associated with hospital volume.

Patients and Methods

Study Design and Setting

This was a large-database study drawn from the Medicare Provider Analysis and Review Limited Data Set, a national, longitudinally maintained database of deidentified clinical and financial records for 100% of Medicare beneficiaries using inpatient hospital services. This database is the most comprehensive repository of claims data available for the inpatient Medicare population. Data for the full 2019 calendar year were included.

We identified patients undergoing primary elective inpatient THAs or TKAs using a method modified from that of Cram et al. [14, 15]. Patients were initially identified using ICD, Tenth Revision, Procedure Coding System codes for primary THA and TKA (Supplemental Table 1; http://links.lww.com/CORR/A975). Exclusion criteria included nonelective hospitalizations, cancer of the pelvis or sacrum identified by ICD-10 code C41.4, and cancer of the long bones of the lower extremities (C40.20, C40.21, and C40.22). We also excluded bilateral and staged procedures occurring during the 30-day billing period.

Hospitalizations that were not reimbursed under the fee-for-service Inpatient Prospective Payment System were excluded using a method published by the Centers for Medicare and Medicaid Services Office of Enterprise Data and Analytics [2]. Hospitals participating in the Inpatient Prospective Payment System include more than three-fourths of acute-care hospitals in the United States [2]. Claims for non–Inpatient Prospective Payment System hospitalizations, such as those reimbursed under Medicare Advantage, are known to be inaccurate and incomplete in the Medicare Provider Analysis and Review dataset [49], and cannot be assessed reliably.

Patients and Hospitals

A total of 500,658 elective primary THAs and TKAs were performed on an inpatient basis for 492,262 Medicare beneficiaries in 2019. A total of 37% (182,733 of 492,262) were excluded because they were non-fee-for-service hospitalizations, which were primarily attributable to patient enrollment in Medicare Advantage. This program accounted for 36% of all Medicare enrollees in 2019 [57], and our data were in line with this national benchmark. An additional 3% (15,096 of 492,262) of patients were excluded because they had nonelective hospitalizations, and 1% (5524 of 492,262) were excluded because they had bilateral procedures or a history of cancer of the pelvis or long bones of the lower extremities. After the exclusion criteria were applied, 288,909 arthroplasties across 2762 hospitals were available for analysis; 70% (201,263) were TKAs and 30% (87,646) were THAs. Among these patients, 87% (251,996 of 288,909) were 65 to 84 years old, 88% (255,415 of 288,909) were White, and 63% (180,688 of 288,909) were women.

Hospitals performing 200 or fewer procedures represented 85% (2345 of 2762) of the hospitals and 45% (130,461 of 288,909) of the arthroplasties included in this study (Fig. 1). They were more likely to be nonteaching hospitals (that is, resident-to-bed ratio of zero) and located in rural communities than high-volume centers were. Smaller-volume centers were also more likely to treat a higher percentage of Black and Hispanic patients (Table 1).

Fig. 1.

This pie chart shows the proportion of hospitals by annual hospital volume. Eighty-five percent of hospitals perform 200 arthroplasties or fewer per year.

Table 1.

Patient and hospital characteristics of elective primary THAs and TKAs performed for Medicare beneficiaries in 2019^a

	≤10 (n = 2305)	11 to 100 (n = 60,778)	101 to 200 (n = 67,378)	201 + (n = 158,448)	p valueb
Gender, % (n)					< 0.001
Men	35 (812)	37 (22,316)	37 (25,153)	38 (59,940)
Women	65 (1493)	63 (38,462)	63 (42,225)	62 (98,508)
Age in years, % (n)					< 0.001
< 65	15 (337)	10 (6087)	8 (5482)	6 (9697)
65 to 74	52 (1189)	53 (32,425)	55 (37,122)	57 (90,570)
75 to 84	28 (655)	31 (18,909)	31 (21,125)	32 (50,001)
85 and older	5 (124)	6 (3357)	5 (3649)	5 (8180)
Race or ethnicity, % (n)					< 0.001
White	76 (1743)	86 (52,481)	88 (59,507)	89 (141,684)
Black	13 (300)	7 (4526)	6 (4095)	5 (8494)
Hispanic	4 (101)	2 (987)	1 (671)	1 (1032)
Other	7 (161)	5 (2784)	5 (3105)	5 (7238)
Elixhauser comorbidities, % (n)
Hypertension	70 (1608)	69 (41,676)	68 (45,662)	67 (106,484)	< 0.001
Obesity	23 (534)	26 (15,532)	26 (17,615)	28 (44,127)	< 0.001
Diabetes mellitus	26 (604)	25 (15,015)	22 (14,990)	20 (31,545)	< 0.001
Chronic lung disease	19 (448)	18 (10,828)	17 (11,439)	16 (24,847)	< 0.001
Heart disease	9 (208)	9 (5598)	9 (6355)	9 (14,270)	< 0.001
Renal failure	9 (201)	9 (5525)	9 (5974)	9 (14,453)	0.24
Arthroplasty type, % (n)					< 0.001
THA	25 (577)	27 (16,651)	30 (20,136)	32 (50,282)
TKA	75 (1728)	73 (44,127)	70 (47,242)	68 (108,166)
Academic status, % (n)c					< 0.001
Nonteaching	68 (1563)	65 (39,220)	53 (35,941)	43 (67,485)
Academic	25 (569)	32 (19,521)	43 (29,016)	52 (82,372)
Major academic	8 (173)	3 (2037)	4 (2421)	5 (8591)
Hospital setting, % (n)					< 0.001
Rural	27 (622)	21 (12,504)	11 (7578)	4 (5709)
Urban	28 (635)	39 (23,409)	48 (32,135)	52 (83,056)
Large urban	44 (1019)	41 (24,679)	41 (27,477)	44 (69,683)
Geographic region, % (n)					< 0.001
Northeast	20 (464)	15 (8885)	18 (12,331)	19 (30,736)
South	39 (896)	36 (22,125)	36 (23,994)	35 (55,854)
Midwest	14 (333)	28 (16,684)	26 (17,448)	29 (46,438)
West	22 (513)	21 (12,789)	20 (13,238)	16 (25,420)
Puerto Rico	3 (70)	0 (109)	0 (179)	0 (0)
van Walraven index, mean ± SD	-1.08 ± 4.29	-1.30 ± 4.16	-1.43 ± 4.08	-1.57 ± 4.04	< 0.001
Resident-to-bed ratio, mean ± SD	0.10 ± 0.22	0.07 ± 0.20	0.08 ± 0.15	0.11 ± 0.19	< 0.001
Wage Index, mean ± SD	1.01 ± 0.26	0.99 ± 0.18	1.01 ± 0.20	1.03 ± 0.19	< 0.001
Hospital cost in USD, median (IQR)	18,845 (11,494)	16,852 (7953)	15,684 (7031)	15,211 (6069)	< 0.001

Lower-volume hospitals were more likely to be nonteaching hospitals located in rural areas, and they were more likely to treat non-White patients.

^aPercentages represent the proportion of total patients per column.

^bp value for Pearson chi-square test, which was used for categorical variables. Either one-way analysis of variance or Kruskall-Wallace test was used, as appropriate, for continuous variables.

^cNonteaching hospitals are defined by a medical resident-to-bed ratio of 0. Academic hospitals are defined by resident-to-bed ratio of 0 to 0.50. Major academic hospitals are defined by resident-to-bed ratios > 0.50.

Variables

Patient age, race or ethnicity, gender, diagnosis codes, total hospital charges, and the national provider identification of the admitting hospital were extracted from the Medicare Provider Analysis and Review dataset. Patient age was not available as a specific number but rather as a categorized range of values. Patient race or ethnicity was obtained by Medicare as a race code located on the first reimbursement claim submitted for each hospitalization. The process of identifying and accurately reporting patient race or ethnicity varied at the hospital level. Total hospital charges were collected to estimate hospital costs using cost-to-charge ratios in a process described later in this section. National provider identification numbers were obtained to calculate hospital procedure volumes. Finally, diagnosis codes were collected to identify Elixhauser comorbidities and calculate van Walraven indices.

The Elixhauser comorbidity index is a validated instrument for analyzing preexisting medical conditions using the diagnosis codes found in administrative databases [16, 41]. The original index included 30 dichotomous variables characterizing preexisting medical conditions that can be distinguished from complications or conditions arising during the hospital stay. In the present study, Elixhauser comorbidities were identified using algorithms that are publicly available on the Healthcare Cost and Utilization Project website [1]. Hypertension, obesity, diabetes mellitus, chronic lung disease, heart disease, and renal failure are reported to illustrate patient health status.

Each patient’s Elixhauser comorbidities were used to calculate his or her van Walraven index, which is a single numeric score summarizing a patient’s disease burden and risk of in-hospital mortality [54]. The index multiplies each Elixhauser comorbidity by a value between -7 and 12, which reflects the strength of the condition’s independent association with in-hospital mortality. Negative weights are the result of statistical artifact and do not denote a decreased risk of death. The van Walraven index has been shown in orthopaedic research to outperform the Charlson comorbidity index and the American Society of Anesthesiologists score in quantifying patient health status and preoperative surgical risk [35, 43, 44, 55].

For each hospital included in our dataset, we obtained the local wage index, urban or rural status, geographic location, and ratio of resident physicians to hospital beds (resident-to-bed ratio) from the Impact File for 2019. The Impact File contains metrics that are used to estimate the financial impact of changes in Medicare policy on hospitals. The Centers for Medicare and Medicaid updates and publishes it annually. It is defined as the ratio of local market wages over the national average hourly wage. Values greater than 1.0 indicate areas of a relatively high cost of labor, while values less than 1.0 indicate a relatively low cost of labor.

Hospital volumes were calculated by stratifying claims by national provider identification number and counting the number of claims per national provider identification number. The data were then categorized into bins of increasing hospital volume and characterized using descriptive statistics. Because costs are unlikely to vary with the addition or subtraction of one or two procedures per year, hospitals were clustered by surgical volume to more easily compare large-volume and smaller-volume centers. Finally, after stratifying the data, we calculated the median hospital costs, mean and median LOS, and the mean percentage of patients discharged home.

Potential Sources of Bias

We focused only on inpatient arthroplasty for Medicare beneficiaries and did not analyze outpatient or ambulatory procedures. In 2019, 100% of Medicare fee-for-service THAs and 64% of TKAs were performed on an inpatient basis [21]. Therefore, only 36% of Medicare fee-for-service TKAs (outpatient TKAs) were excluded from our study. To control for potential selection bias introduced by the omission of outpatient TKAs, “arthroplasty type” (THA or TKA) was included as a variable in the multivariable analysis.

Our exclusion criteria resulted in removal of 42% (211,749 of 500,658) of the inpatient arthroplasties performed in 2019. Most exclusions were because of non-fee-for-service hospitalizations, as described. The number of exclusions per hospital was calculated as a proportion of the hospital volume. The mean percentages and standard deviations of exclusions per hospital were 32% ± 29%, 41% ± 23%, 40% ± 21%, and 40% ± 19% for hospitals performing ≤ 10, 11 to 100, 101 to 200, and 200 or more arthroplasties per year, respectively. A Kruskal-Wallis test showed a difference between the means (H = 30.9, p <0.001). Post hoc pairwise comparisons revealed the smallest hospitals (≤ 10 procedures) were different from the higher-volume hospitals (p < 0.001), but no differences were observed between the higher-volume cohorts. Therefore, exclusions were approximately 10% (or one procedure per year) fewer among the very smallest hospitals (≤ 10 procedures per year) than among larger-volume hospitals, but otherwise statistically equivalent across all groups.

Finally, our data included hospitals participating in bundled payment programs, which may affect hospital-level arthroplasty costs. Voluntary programs such as the Bundled Payments for Care Improvement Initiative are the most likely to attract participation from hospitals with low costs and high procedure volumes. Therefore, we believe that excluding such programs would inappropriately skew our results toward higher-cost programs [3]. Conversely, the impact of mandatory programs such as the Comprehensive Care for Joint Replacement Model on hospital costs is still being determined, but are designed to force cost savings at the hospital level. We do not believe including these hospitals introduces a large systematic bias to our results because a range of high-volume and low-volume arthroplasty centers are included in the Care for Joint Replacement program.

Analysis of Hospital Costs

The primary objective of this study was to determine whether the cost incurred by hospitals for performing THA and TKA is independently associated with hospital volume. We estimated hospital costs occurring during acute hospitalization by multiplying total hospital charges by cost-to-charge ratios obtained from the Impact File for 2019. Estimating costs using these cost-to-charge ratios, which reflect operating and capital costs, is a method published by the Centers for Medicare and Medicaid Research Data Assistance Center and used in medical research [12, 19, 48]. We chose this method because hospitals are subject to multiple identifiable and unidentifiable sources of cost variation [7-9, 20, 46, 50, 53], increasing the difficulty of isolating trends attributable to a single variable. Estimating costs with cost-to-charge ratios enabled us to analyze the large Medicare Provider Analysis and Review dataset and decrease the influence of random cost variation through the law of large numbers.

The median hospital costs (arthroplasty costs) and interquartile ranges (IQR) were calculated and analyzed using univariable and multivariable generalized linear models, assuming a gamma distribution and using a log link function. Generalized linear models were used because of their advantages in handling highly skewed, nonparametric data, which is characteristic of health economics data [6]. Hospital costs were observed to decline with stepwise increments in surgical volume for hospitals performing 10 or fewer procedures (USD 18,845 [IQR USD 11,494]), 11 to 100 procedures (USD 16,852 [IQR USD 7953]), 101 to 200 procedures (USD 15,684 [IQR USD 7031]), 201 to 300 procedures (USD 14,325 [IQR USD 6228]), 301 to 400 procedures (USD 14,499 [IQR USD 6232]), 401 to 500 procedures (USD 14,293 [IQR USD 6402]), and 501 procedures or more (USD 13,856 [IQR USD 5618]) per year. A Spearman rank correlation test confirmed this downward trend to be statistically significant (ρ = -0.177; p < 0.001). A univariable regression analysis demonstrated these costs were 1.34 (95% CI 1.32 to 1.36; p < 0.001), 1.15 (95% CI 1.15 to 1.16; p < 0.001), 1.09 (95% CI 1.08 to 1.09), 1.00 (95% CI 0.99 to 1.00; p = 0.69), 1.01 (95% CI 1.00 to 1.01; p = 0.004), and 0.99 (95% CI 0.98 to 1.00; p < 0.001) times greater than those of hospitals with 501 or more procedures per year, respectively. For hospitals with 201 procedures or more per year, increasing surgical volume was associated with less than or equal to 1% difference in cost compared with centers performing 501 procedures or more per year. As a result, hospitals performing 201 procedures or more annually were consolidated for further analysis (Fig. 2 ). Multivariable modeling was performed to control for hospital characteristics, patient demographics, and health status. Results are presented as exponential coefficients with 95% CIs, which can be interpreted as relative differences in cost.

Fig. 2.

The median cost of elective primary THA and TKA is shown by annual hospital volume. Hospitals performing 200 arthroplasties or fewer per year incurred costs that were 10.1% to 32.2% greater than hospitals performing 201 procedures or more per year.

Analysis of LOS and Discharge to Home

The secondary objective of this study was to determine whether LOS and discharge-to-home are associated with hospital volume. We estimated the median and mean LOS and the percentage of patients discharged to home after stratification of the data by surgical volume. The median LOS (IQR) did not vary with surgical volume. The median LOS was 3 days (IQR 1 day) for hospitals with 10 or fewer procedures and 2 days (IQR 2 days) for hospitals with 11 to 100 procedures, 101 to 200 procedures, and 200 or more procedures per year. The mean LOS, however, declined with increasing surgical volume, reflecting the heavily skewed nature of the data. The mean LOS was 2.72 ± 1.82 days for hospitals with 10 or fewer procedures, 2.29 ± 1.54 days for hospitals with 11 to 200 procedures, 2.17 ± 1.45 days for hospitals with 101 to 200 procedures, and 2.06 ± 1.30 days for hospitals with 201 procedures or more (Fig. 3). A Spearman rank correlation test confirmed this downward trend to be statistically significant (ρ = -0.077; p < 0.001). Because LOS was recorded as a noncontinuous integer value with very little variation around the median, it was transformed into a binary variable identifying patients with LOS greater than 2 days, which was the median LOS. A binary logistic regression analysis accounting for the nested structure of the data was performed on the binary LOS variable.

Fig. 3.

The mean and median LOS by annual hospital volume are shown. Although the median LOS was relatively unchanged, the mean LOS decreased as the annual hospital volume increased, reflecting the highly skewed nature of the data.

The proportion of patients discharged to home increased with increasing surgical volume. The percentage of patients discharged to home was 62.3% (1436 of 2305) for hospitals performing 10 or fewer procedures, 73.4% (44,609 of 60,778) for hospitals performing 11 to 100 procedures, 77.9% (52,505 of 67,378) for hospitals performing 101 to 200 procedures, and 81.8% (129,564 of 158,448) for hospitals performing 201 procedures or more. A Spearman rank correlation test confirmed this upward trend to be statistically significant (ρ = 0.085; p < 0.001). Binary logistic regression accounting for the nested structure of the data was also performed on the discharge-home variable.

Results are presented as ORs with 95% CIs. Statistical significance was established at an a priori level of p ≤ 0.05. All analyses were performed using SPSS Statistics 28 (IBM Corp).

Results

Hospital Costs

The multivariable regression confirmed (p < 0.001) that the annual arthroplasty volume was independently associated with hospital cost after controlling for relevant confounding variables, such as patient age, race or ethnicity, preoperative comorbidities, geographic location, hospital resident-to-bed ratio, and wage index (Table 2). Hospitals with 201 procedures or more annually had the lowest costs. Hospitals with progressively fewer procedures had progressively increasing costs.

Table 2.

Results of multivariable generalized linear model of hospital arthroplasty costs

Variable	Exp(B) (95% CI)	p value
Hospital volume
≤ 10	1.32 (1.30 to 1.34)	< 0.001
11 to 100	1.17 (1.17 to 1.17)	< 0.001
101 to 200	1.10 (1.10 to 1.10)	< 0.001
200 +	Reference
Gender
Men	Reference
Women	0.99 (0.99 to 0.99)	< 0.001
Age in years
< 65	Reference
65 to 74	0.92 (0.91 to 0.92)	< 0.001
78 to 84	0.91 (0.91 to 0.92)	< 0.001
85 and older	0.94 (0.93 to 0.95)	< 0.001
Race or ethnicity
White	Reference
Black	1.02 (1.02 to 1.03)	< 0.001
Hispanic	0.99 (0.98 to 1.00)	0.10
Other	0.98 (0.98 to 0.99)	< 0.001
Arthroplasty type
THA	Reference
TKA	1.02 (1.02 to 1.02)	< 0.001
Hospital setting
Rural	Reference
Urban	0.90 (0.89 to 0.90)	< 0.001
Large urban	0.87 (0.86 to 0.87)	< 0.001
Geographic location
Northeast	Reference
South	1.23 (1.23 to 1.24)	< 0.001
Midwest	1.13 (1.12 to 1.13)	< 0.001
West	1.17 (1.17 to 1.17)	< 0.001
Puerto Rico	0.59 (0.57 to 0.62)	< 0.001
van Walraven index	1.01 (1.01 to 1.01)	< 0.001
Wage index	2.13 (2.11 to 2.15)	< 0.001
Resident-to-bed ratio	1.30 (1.29 to 1.31)	< 0.001

Arthroplasty costs at hospitals with 101 to 200 procedures, 11 to 100 procedures, and 10 or fewer procedures per year were 1.10 (95% CI 1.10 to 1.10; p < 0.001) times greater or 10% higher, 1.17 (95% CI 1.17 to 1.17; p < 0.001) times greater or 17% higher, and 1.32 (95% CI 1.30 to 1.34; p < 0.001) times greater or 32% higher than those of hospitals with 201 or more procedures per year. There was no overlap of CIs. Among all variables included in the model, the only categorical factors associated with less than 10% change in hospital cost, besides annual surgical volume, were hospital setting (urban or rural) and geographic location (Northwest, South, Midwest, and West).

LOS and Discharge to Home

The binary logistic regression analysis confirmed that the odds of a patient experiencing LOS longer than 2 days increased at hospitals with fewer annual procedures. Compared with hospitals with 201 procedures or more per year, the OR of experiencing LOS longer than 2 days was 1.25 (95% CI 1.10 to 1.43; p < 0.001) for hospitals with 101 to 200 procedures, 1.78 (95% CI 1.60 to 1.98; p < 0.001) for hospitals with 11 to 100 procedures, and 3.44 (95% CI 2.94 to 4.03; p < 0.001) for hospitals with 10 or fewer procedures per year.

The odds of discharge to home were also increased at hospitals with greater annual surgical volume. Compared with hospitals with 201 procedures or more per year, the OR of being discharged to home was 0.78 (95% CI 0.71 to 0.86; p < 0.001) for hospitals with 101 to 200 procedures, 0.56 (95% CI 0.52 to 0.61; p < 0.001) for hospitals with 11 to 100 procedures, and 0.34 (95% CI 0.29 to 0.38; p < 0.001) for hospitals with 10 procedures or fewer per year.

Discussion

With bundled payments and alternative reimbursement models expanding in scope and scale, payments to hospitals and surgeons are declining. As a result, cost reduction at the hospital level is paramount for the sustainability of profitable inpatient arthroplasty practices. Although prior studies have examined determinants of cost variation in arthroplasty [7-9, 20, 46, 50, 53], it is unknown whether contemporary arthroplasty practices benefit from economies of scale. Quantifying the independent effects of volume-based cost variation in arthroplasty may be important for guiding and understanding future healthcare policy. Our findings demonstrate that lower-volume hospitals experience greater hospital costs, longer LOS, and lower rates of discharge to home for primary elective THA and TKA than higher-volume hospitals. Stepwise increases in volume also correlated with progressive improvements in each of these metrics. At the hospital level, these findings suggest that smaller-volume hospitals may benefit from increasing surgical volume. At the policy level, these findings suggest that smaller-volume hospitals are disproportionately affected by declining reimbursement and may necessitate special treatment to achieve equitability to larger-volume hospitals as bundled payments and alternative reimbursement models continue to erode hospital payments. Ensuring the sustainability of inpatient arthroplasty practices at small-volume hospitals may be important because many provide critical access to healthcare for many rural and underserved communities.

Limitations

This study has several limitations. For instance, estimating hospital costs using cost-to-charge ratios is an imprecise method of estimating cost. Our goal was not to precisely account for costs, however, but to study the relationship between costs and volume. We believe that using cost-to-charge ratios is sufficient for this purpose. Moreover, hospitals are subject to multiple identifiable and unidentifiable sources of cost variation [7-9, 20, 46, 50, 53], increasing the difficulty of isolating trends attributable to a single variable. By choosing to estimate costs with cost-to-charge ratios, we could use the Medicare Provider Analysis and Review dataset, which is the largest available repository of claims data for inpatient Medicare hospitalizations, and decrease the influence of random variation through the law of large numbers. Further, because of the confidential nature of corporate financial information, we are skeptical about the feasibility of a study comparing costs across multiple institutions of various sizes using verified accounting data. The downside of using the Medicare Provider Analysis and Review database is that it precluded an analysis of non-Medicare and noninpatient surgeries. Because inpatient procedures for Medicare beneficiaries represented a very large portion of the arthroplasties performed in 2019 [21], however, we believe the directional trends reported here are unlikely to be invalidated by the exclusion of these patients.

Another limitation of his study is that our final study group represented only 58% of all procedures meeting our inclusion criteria. Most exclusions were because of participation in Medicare Advantage, whose data are known to be incomplete and inaccurate in the Medicare Provider Analysis and Review dataset [49]. Although we cannot be sure that these exclusions did not bias our findings, we are reassured that the proportions of exclusions were broadly similar across all groups analyzed.

As a retrospective analysis, our results may have been affected by unmeasured confounding variables, such as participation in bundled payment programs, which we could not identify in this dataset. Particularly, our analysis of LOS and discharge to home is susceptible to possible confounding because a multivariable analysis was not performed for those endpoints. Adjustments for measured confounding variables may also have been inadequate. For example, we used the van Walraven index as an adjustment for patient health. Although this metric is validated in orthopaedics for in-hospital mortality [35, 43, 44, 54, 55], it has not been examined for an association with cost and may be inadequate as a case-mix adjustment. To our knowledge, however, no comorbidity index has been examined for associations with cost in total joint arthroplasty. This should be a subject for further study. Our analysis also could not adjust for race and ethnicity in a nuanced way. The Medicare Provider Analysis and Review dataset only provides a single categoric identifier of race or ethnicity that does not reflect the increasingly multicultural and mixed-race environment in which we live. Such reductive categorizations can mask differential outcomes and trends between groups of patients who are not of a single race or ethnicity.

Finally, our data reflected a precoronavirus-19 environment that may not reflect current conditions. We did not include any information on the potential drivers of cost variability, and we did not address revision arthroplasties, nonelective surgeries, or patient outcomes.

Hospital Costs

Our findings demonstrated that for hospitals performing fewer than 201 procedures per year, increasing surgical volume was associated with decreasing hospital costs. These findings are important at the hospital level because they suggest that smaller-volume hospitals, under pressure from bundled payment programs and decreasing reimbursement, may benefit financially from increasing procedure volumes. At the policy level, these findings suggest that smaller-volume hospitals are disproportionately harmed by reductions in payments and may necessitate special treatment to achieve equitability with larger-volume hospitals amid declining hospital reimbursement.

Prior reports investigating economies of scale in healthcare, defined as decreasing unit cost or increasing productivity resulting from increasing operational scale, have reported inconsistent results [4, 13, 19, 34, 36, 38-40, 58, 60]. Some studies have demonstrated empirical evidence of scale effects on hospital cost and efficiency [4, 19, 37, 39, 58, 60], while others have reported evidence of only marginal scale effects [13, 23, 34, 51] and even diseconomies of scale [36-38, 40]. In arthroplasty, prior research broadly corroborated the findings of the present study. Using the 1989 Medicare Provider Analysis and Review dataset, Gutierrez et al. [19] showed that average treatment costs for knee replacement declined continuously up to 100 procedures per year. In a more recent analysis of THAs performed in New York State, Haeberle et al. [22] identified two thresholds of 122 and 310 procedures per year that were statistically associated with cost savings. The latter analysis, however, did not control for potential confounding factors.

It is unknown what factors drive the economies of scale reported in this study. Higher-volume hospitals have been associated with lower rates of medical complications, mortality, and revision surgery than lower-volume centers have been [19, 24, 26-28, 30-33, 47, 59]. Our findings may be mediated through variations in complications and patient outcomes. Rudy et al. [50], however, observed that the operation-related variables of surgery, such as use of regional anesthesia and total operative time, had a greater impact on hospital costs than patient demographics or health status. Therefore, service-level decisions may be contributing to our results. One well-known example of the operational and economic potential of service-level efficiencies is Shouldice Hospital in Canada, which has unusually low costs and high rates of success in primary inguinal hernia repair. Shouldice’s success has largely been attributed to the hospital’s implementation and perfection of streamlined care paths designed specifically for a well-defined population of patients [27]. We noted that the IQR of hospital costs reported in the current study decreased sequentially with surgical volume, which suggests that our findings are in part mediated by a similar type of “practice makes perfect” phenomenon [17-19, 24]. Another explanation for our results is that high-volume centers, which predominantly consist of urban academic hospitals, could leverage their surgical volume and corporate prestige to obtain favorable pricing from vendors of medical devices. With implant prices contributing the greatest direct cost of surgery (more than 50%), the potential benefit of enhanced bargaining power is substantial [9, 24, 52] and likely to be an important driver of our results. Further research is warranted to identify what major factors drive economies of scale in THA and TKA.

LOS and Discharge to Home

Our findings show that increasing surgical volume is associated with decreasing mean LOS and increasing rates of discharge to home for primary elective THA and TKA. These observations support the conclusion that higher-volume centers are associated with lower cost and greater efficiency than smaller-volume hospitals. Additionally, the observation that lower surgical volumes are associated with lower rates of discharge to home suggests that smaller-volume hospitals participating in bundled payment programs may generate lower average profits per procedure, not only during the acute hospitalization but also during the postacute period. Hospitals involved in bundled payment schemes are penalized for patients discharged to skilled nursing or rehabilitation facilities because bundled payments are shared between the postacute care facility and the acute-care hospital. Thus, for hospitals involved in bundled payment programs, the financial benefits of increasing surgical volume may be even greater than what we have reported. These findings also suggest that smaller-volume hospitals may require special accommodations in the broader rollout of bundled payment programs to achieve an equitable outcome to that of larger-volume hospitals. Further research into the mechanisms of these observations is warranted to facilitate hospital and policy-level initiatives targeting these metrics.

Conclusion

Higher-volume hospitals incur lower costs when performing elective primary inpatient THA and TKA. This remains true after controlling for patient demographics, health status, and hospital characteristics. Hospitals performing 200 or fewer arthroplasties per year can expect to incur costs 10% to 32% greater than centers performing 201 procedures or more per year, depending on annual procedure volumes. Higher-volume hospitals are also associated with shorter LOSs and higher rates of discharge to home than lower-volume hospitals. These results suggest that small and medium-sized regional hospitals are disproportionately impacted by ongoing reimbursement cuts, which may be important because many smaller-volume hospitals provide critical access to healthcare for rural and underserved communities. As a result, smaller-volume hospitals may necessitate special treatment to remain viable as bundled payments and alternative reimbursement models continue to erode hospital payments. Further research is also warranted to identify the key drivers of this volume-based cost variation, which may facilitate quality improvement initiatives at the hospital and policy levels.