High-Volume Hospitals Are Associated With Lower Rates of Mortality Following Revision Arthroplasty for Periprosthetic Joint Infection

Abstract

Background

The relationship between hospital case volume and postoperative morbidity following revision arthroplasty for periprosthetic joint infection (PJI) has not been defined. As such, we sought to compare 90-day postoperative mortality rates among patients undergoing antibiotic spacer placement for PJI of the hip or knee between high- vs low-volume hospitals.

Materials and methods

The Premier Healthcare Database was queried from 2016 to 2023 for adult patients undergoing antibiotic spacer placement for PJI. Using the Metropolis–Hastings algorithm, a single hospital volume threshold, below which mortality increased, was identified. Patients were divided by those operated on by high- vs low-volume hospitals. Multivariable logistic regression was conducted to assess odds of 90-day mortality and sepsis.

Results

In a cohort of 55,670 PJIs, postoperative mortality was decreased in patients who were operated on by hospitals performing ≥35 antibiotic spacers annually (high volume). The adjusted odds of mortality (adjusted odds ratio = 0.76; 95% confidence interval = 0.65-0.88, P < .001) and sepsis (adjusted odds ratio = 0.69; 95% confidence interval = 0.66-0.73, P < .001) were significantly lower among patients operating at high-volume hospitals.

Conclusions

Patients receiving PJI care at hospitals performing greater than 35 antibiotic spacers annually died less frequently and had improved outcomes than those presenting to lower volume centers. These findings reinforce emerging evidence in support of centralized PJI care centers to improve patient survival.

Introduction

Periprosthetic joint infection (PJI) remains a highly morbid and costly complication following total joint arthroplasty (TJA). PJI is the leading indication for revision arthroplasty surgery, accounting for 33.1% of revisions after total knee arthroplasty (TKA) and 21.4% of revisions after total hip arthroplasty (THA) [1]. While advancements in surgical techniques and antimicrobial therapies have led to improvements in patient outcomes, mortality rates following revision surgery for PJI remain a significant concern. In addition to patient-specific risks, identifying systemic factors that contribute to postoperative mortality in this population is critical for preoperative risk stratification and patient optimization.

The relationship between hospital TJA case volume and postoperative outcomes after primary, elective TJA is well described in the arthroplasty literature. Previous studies have associated higher hospital TJA volumes with lower all-cause complication rates, shorter lengths of stay, and reduced costs after primary arthroplasty [[2], [3], [4], [5], [6], [7], [8], [9]]. Further studies have suggested the volume–outcome relationship extends to revision arthroplasty, with centers that report high-revision volumes demonstrating decreased risks of mortality and re-revision compared to centers performing low volumes of revisions [[10], [11], [12], [13]]. However, the impact of hospital volume on outcomes following revision arthroplasty to manage PJI remains unclear. Given the medical and surgical complexity associated with the surgical management of PJI, determining an association between hospital volume and postoperative outcomes is essential for this high-risk patient population.

Given the paucity of data on this topic, this study has several aims: (1) identify the relationship between hospital volume and 90-day postoperative mortality in patients undergoing antibiotic spacer placement for PJI of the hip or knee; (2) quantify a threshold hospital volume where the odds of mortality significantly changes; (3) and use this threshold to define low- and high-volume hospitals to compare outcomes in patients presenting to either low- vs high-volume centers. Specifically, we hypothesized that patients undergoing antibiotic spacer placement for PJI at high-volume centers will have significantly lower rates of postoperative complications and odds of 90-day mortality than patients presenting to low-volume centers.

Material and methods

The Premier Healthcare Database (PHD) was retrospectively queried from January 1st, 2016, to December 31st, 2023, for adult patients (≥18 years) who underwent spacer placement for PJI after total hip or knee arthroplasty (THA or TKA). The PHD is a nationally representative database containing patient demographics, hospital characteristics, with diagnosis and procedure billing codes from >1000 hospitals [14]. Additionally, the database contains records of all patient-specific billing information, including implants and medications charged during the hospital admission.

Cohort selection

We identified patients with an admission diagnosis of PJI or other deep infection via International Classification of Disease, Tenth Revision (ICD-10) diagnosis coding. (Supplementary Table S1) Revision procedures and spacer placement were identified via ICD-10 Procedure Coding System and Current Procedural Terminology codes (Supplementary Table S2). Vancomycin, gentamicin, and tobramycin powders used for antibiotic cement were identified using charge data from the day of surgery [15] (Supplementary Table S3). Patients who had antibiotic charges for <2000 mg of vancomycin, <2400 mg of gentamicin, or <2400 mg of tobramycin were excluded to most appropriately capture true antibiotic spacer placement. Direct charges for antibiotic spacers or antibiotic cement were also identified (Supplementary Table S3). Using these charge categories and procedure codes, adult patients undergoing antibiotic spacer placement in the setting of PJI after TJA were identified.

Study endpoints

The study endpoints were identified via ICD-10 clinical modification codes and data provided directly via the PHD. The primary endpoint was 90-day odds of inpatient mortality. Additional secondary endpoints were rates of 90-day sepsis, readmission, and medical complications (acute kidney injury, acute respiratory failure, myocardial infarction, pneumonia, urinary tract infection, and venous thromboembolism).

Modeling case volumes and mortality risk

The annual hospital volumes for antibiotic spacers were calculated over the study period. Hospitals who provided ˂3 years of data were excluded. Given the potential nonlinear relationship between hospital volume and mortality risk, 3-knot restricted cubic splines (RCSs) on logistic regressions were modeled after excluding 4- and 5-knot models to reduce the risk of overfitting while optimizing interpretability. To account for confounding factors, covariates in each RCS model included age, sex, race, marital status, comorbidities, hospital bed size, hospital setting (urban or rural), hospital region (Northwest, Southwest, Northeast, and Southeast), length of stay, the type of involved joint (THA vs TKA) and the interaction term between hospital volume and surgeon volume. Sequential minimization of Akaike’s Information Criterion (AIC) followed by Bayesian Information Criterion (BIC) was used to optimize model selection. The AIC selects a model with optimal coverage of assumptions, thereby increasing the number of potential covariates selected. The BIC penalizes increasing model parameters to a greater extent than AIC, thus simplifying the final model and reducing the risk of overfitting. The area under the receiver operating characteristic curve and the Hosmer–Lemeshow test were then used to validate model selection (Appendix). The final RCS used the median cohort volume as a reference to calculate odds of PJI.

Given the relationship observed on RCS modeling, low-volume hospitals were defined as those with annual volumes below the value where the approximated odds of mortality significantly increased with decreasing volumes. These approximated threshold values were then input into a Markov chain Monte Carlo simulation using the Metropolis–Hastings algorithm to quantitatively identify a single expected threshold value. By iteratively sampling candidate threshold values around the approximated inflection point, this process creates a probability distribution that converges on a single threshold value which best explains the observed relationship between hospital volume and mortality. Values above this final output were defined as “high-volume” and those equal to or below were defined as “low-volume”.

Statistical analyses

Descriptive statistics were used to report patient demographics, hospital characteristics, and comorbidities per the Elixhauser system [16]. A Charlson Comorbidity Index was also calculated [17]. Between-group differences in categorical and continuous variables were identified via chi-square test and independent t-tests, respectively. Multivariable models determining the odds of primary or secondary outcomes while accounting for potential confounding covariates were then created using iterative minimization of AIC and BIC, which included demographic and hospital factors, as well as comorbidities, and length of hospital stay. Statistical significance was defined as P < .05 for two-sided tests. All statistical analyses were performed using Stata (version 17.0; StataCorp LLC, College Station, TX).

Results

A total of 55,670 antibiotic spacers for PJI were identified (THA: 16,002 [28.7%], TKA: 39,668 [71.3%]) (Fig. 1). Of the 578 hospitals identified, the average hospital performed 16 antibiotic spacers (standard deviation: 14; median 11, interquartile range: 7-20) in a given year (Supplemental Fig. 1). These were performed by 394 surgeons who performed an average of 7 antibiotic spacers (standard deviation: 7, median 5, interquartile range: 3-9) in a given year. Based on the aforementioned RCS analysis, high-volume hospitals were defined as those reporting ≥35 antibiotic spacers (9.9% of all hospitals) (Fig. 2). Using these criteria, 18,727 patients (33.6%) and 36,943 patients (66.4%) undergoing antibiotic spacer placement for PJI at high- vs low-volume hospitals were identified, respectively.

Figure 1.

Strengthening the Reporting of Observational studies in Epidemiology diagram of cohort selection.

Figure 2.

Restricted cubic spline modeling the relationship between hospital volume and odds of mortality. Gray dashed lines represent 95% confidence intervals.

Demographics

Patients presenting to high-volume hospitals were marginally younger (66.61 ± 10.75 vs 67.05 ± 10.88 years, P < .001). Patients were admitted for 8 days on average in both cohorts, with high-volume patients costing $2867.13 less on average than their low-volume counterparts ($29,968.34 ± 26.036.32 vs $32,835.47 ± 17,533.70, P < .001). Significant between group differences in race, insurance type, admitting hospital bed size, urban–rural designation, teaching status, and region were observed between groups, however, without differences in gender or marital status (Table 1). There were significant differences in 12 of the 34 comorbidities examined (range: 0.00-2.47%), with the largest differences in rates of history of venous thromboembolism and depression (Table 2). However, no significant difference in Charlson Comorbidity Index was observed between groups (3.6 ± 2.0 vs 3.6 ± 2.1, P = .646).

Table 1.

Demographic and hospital data.

Demographics	High volume N = 18,727		Low volume N = 36,943		P value
	Average	Standard deviation	Average	Standard deviation
Age (years)	66.6	10.8	67.1	10.9	<.001
Length of stay (days)	7.6	7.2	7.7	7.5	.190
Cost ($)	$29,968.34	$26,036.32	$32,835.47	$17,533.70	<.001
	Number	Percent	Number	Percent
Total knee arthroplasty	13,181	70.4%	26,487	71.7%	.001
Women	8862	47.3%	17,660	47.8%	.548
Race					<.001
Asian	83	0.4%	338	0.9%
Black	1844	9.9%	3209	8.7%
Other	1351	7.2%	1665	4.5%
White	15,449	82.5%	31,731	85.9%
Marital status
Married	9106	48.6%	18,904	51.2%	<.001
Other	2244	12.0%	2654	7.2%
Single	7377	39.4%	15,385	41.7%
Insurance type
Medicare	12,437	66.4%	24,435	66.1%	<.001
Medicaid	1146	6.1%	2643	7.2%
Private	4351	23.2%	8409	22.8%
Other	793	4.2%	1456	3.9%
Bed size
<100	0	0.0%	2044	5.5%	<.001
100-199	578	3.1%	6147	16.6%
200-299	2434	13.0%	7063	19.1%
399-399	2047	10.9%	6977	18.9%
400-499	3508	18.7%	4456	12.1%
>500	10,160	54.3%	10,256	27.8%
Urban
Yes	18,414	98.3%	33,132	89.7%	<.001
Teaching status
Yes	13,082	69.9%	17,531	47.5%	<.001
Region
Midwest	3373	18.0%	9310	25.2%	<.001
Northeast	3131	16.7%	5283	14.3%
South	9624	51.4%	16,777	45.4%
West	2599	13.9%	5573	15.1%

Table 2.

Comorbidity data per the Elixhauser system.

Comorbidity	High volume N = 18,727		Low volume N = 36,943		P value
	Average	Standard deviation	Average	Standard deviation
Charlson Comorbidity Index	3.6	2.1	3.6	2.0	.646
	N	%	N	%
Congestive heart failure	2810	15.0%	5426	14.7%	.319
Myocardial infarction	1154	6.2%	2278	6.2%	.985
Arrhythmia	4248	22.7%	7884	21.3%	<.001
Valvular disease	309	1.7%	674	1.8%	.140
Pulmonary hypertension	624	3.3%	1064	2.9%	.003
Chronic pulmonary disease	4362	23.3%	8439	22.8%	.234
Peripheral vascular disease	1227	6.6%	2256	6.1%	.040
Hypertension	10,042	53.6%	19,689	53.3%	.464
Complicated hypertension	4527	24.2%	8850	24.0%	.570
Cerebrovascular accident	1444	7.7%	2930	7.9%	.361
Hemiplegia/paraplegia	115	0.6%	154	0.4%	.002
Other neurological disorders	1563	8.4%	3116	8.4%	.723
Diabetes, uncomplicated	2419	12.9%	5048	13.7%	.015
Diabetes, complicated	3395	18.1%	6753	18.3%	.664
Hypothyroidism	3584	19.1%	7057	19.1%	.919
Renal failure	3508	18.7%	6887	18.6%	.797
Liver disease	1098	5.9%	1898	5.1%	<.001
Chronic peptic ulcer disease	98	0.5%	211	0.6%	.473
Blood loss anemia	306	1.6%	584	1.6%	.636
Deficiency anemia	1265	6.8%	2631	7.1%	.109
Coagulopathy	1230	6.6%	2479	6.7%	.525
Venous thromboembolism	2830	15.1%	4670	12.6%	<.001
Fluid and electrolyte disorders	5266	28.1%	10,941	29.6%	<.001
Rheumatic disease	1693	9.0%	3089	8.4%	.007
HIV/AIDS	47	0.3%	77	0.2%	.314
Lymphoma	120	0.6%	203	0.6%	.180
Solid tumor	322	1.7%	589	1.6%	.272
Metastatic cancer	125	0.7%	227	0.6%	.456
Obesity	6977	37.3%	13,685	37.0%	.623
Weight loss	1124	6.0%	1915	5.2%	<.001
Alcohol abuse	731	3.9%	1405	3.8%	.561
Drug abuse	944	5.0%	1668	4.5%	.006
Psychoses	181	1.0%	354	1.0%	.925
Depression	4353	23.2%	7945	21.5%	<.001

HIV, human immunodeficiency virus; AIDS, acquired immunodeficiency syndrome.

Outcomes

Significantly lower unadjusted mortality rates were observed in the high-volume vs the low-volume cohorts (1.3 vs 2.0%, P < .001). Similarly, lower rates of sepsis (13.6 vs 19.2%, P < .001) and medical complications (25.0 vs 27.4%, P < .001) were observed. Significantly higher readmission rates were observed in the high-volume cohort (13.5 vs 15.0%, P < .001).

After adjusting for confounding factors, patients operated on in high-volume hospitals continued to have significantly lower adjusted odds ratio (aOR) of mortality than the low-volume cohort (aOR = 0.76 [95% confidence interval (CI) = 0.65-0.88], P < .001). Similarly, lower adjusted odds of sepsis (aOR = 0.69 [95% CI = 0.66-0.73], P < .001) and medical complications (aOR = 0.88 [95% CI = 0.84-0.92], P < .001) were observed. In contrast, significantly higher adjusted odds of readmission was observed in the high-volume group (aOR = 1.08 [95% CI = 1.02-1.15], P = .010). These are summarized in Table 3.

Table 3.

Primary and secondary study outcomes.

Complication	High volume N = 18,727		Low volume N = 36,943		P value	Multivariate regression
	N	%	N	%		aOR	95% CI	P value
90-Day mortality	238	1.3%	720	2.0%	<.001	0.76	0.65-0.88	<.001
Sepsis	2549	13.6%	7076	19.2%	<.001	0.69	0.66-0.73	<.001
Readmission	2814	15.0%	4971	13.5%	<.001	1.08	1.02-1.15	.010
Medical complication	4683	25.0%	10,133	27.4%	<.001	0.88	0.84-0.92	<.001

aOR, adjusted odds ratio; CI, confidence interval.

Discussion

In this study of over 55,000 antibiotic spacer placements for PJI, hospital volume was a significant independent risk factor for postoperative mortality, with patients presenting to hospitals performing ˃35 antibiotic spacers annually experiencing lower mortality (1.3% vs 2.0%), reflected by a 24% lower adjusted odds after adjusting for confounding factors. Additionally, these patients faced significantly lower rates of sepsis and other medical complications, potentially reflecting that treatment failure or medical complications may contribute to the elevated mortality rates seen in the low-volume cohort. These findings highlight the potential benefit of designated PJI care centers in improving outcomes and reducing mortality.

Specialized care centers for the treatment of PJI have garnered increased attention. The first widely publicized use of a centralized system for the treatment of orthopaedic infections was described in France where the Health Ministry established a national network dedicated to the treatment of bone and joint infections [18]. Within the United States, individual centers are self-selecting to become regional hubs for designated PJI care with favorable results [[19], [20], [21]]. Referrals to these centers have been associated with improved outcomes, while surgeries performed prior to referral have been shown to decrease the likelihood of successful infection control [19]. The importance of early referral to these centers was explored by Dombrowsky et al. [20] who compared 76 first stage revision arthroplasties for PJI who were referred to a single institution >90 days after diagnosis to 106 who were referred within 90 days, observing significantly lower mortality, failure rates, and costs in those referred early. The authors also noted that the 90-day mortality rate of 4-7% in their overall cohort remained lower than previously reported national rates [20]. A proposed source of the improved survival in the current and prior studies might be related to the higher rates of reimplantation at specialized PJI care centers. In a cohort of 390 patients treated at a specialized PJI care center, Fehring et al. reported 99% reimplantation and 87% final infection-free replantation rates which are notably higher than the 43-62% reimplantation rate reported in prior studies [[22], [23], [24]]. These findings align with the success of specialized treatment centers in other complex surgical disciplines, such as orthopaedic oncology [25]. Collectively, these data provide further support for the establishment of specialized centers for PJI care; however, more work is required to understand what policy-level changes are necessary to facilitate the creation of these centers and how to best implement these in the current insurance landscape. Further, we must emphasize that such centers may fail to address inequities related to access to care and rural health and may in fact exacerbate these disparities.

The relationship between hospital volume and outcomes after arthroplasty has been well-documented [2,[26], [27], [28]]. This trend extends to revision TJA, with both aseptic and septic revision cases showing improved outcomes at higher-volume centers [[9], [10], [11], [12],29]. In a national analysis of a Scottish arthroplasty registry, Yapp et al. observed that hospitals performing fewer than 20 revision TKAs had significantly higher re-revisions, which was significantly greater in patients who were undergoing revision surgery for infection [11]. Comparable results were observed by Samuel et al. in their analysis of revision cases from New York and Florida, where significantly lower in-hospital mortality rates were observed in the highest volume quartiles than the lowest, with the highest mortality rates being observed in the septic revision cases [29]. A similar effect of hospital volume and risk of re-revision and mortality was observed in aseptic revision knee and hip arthroplasty, where a threshold of fewer than 53 cases was a risk factor for inferior outcomes [9,12]. While the threshold of 35 cases described here is unlikely to be broadly generalizable, the observed trend remains consistent with these prior studies and critical to understanding how to optimize outcomes following revision arthroplasty for septic indications.

Costs related to PJI care remain high, representing an over 2-fold greater direct inpatient cost of care than aseptic revision cases [30] with national annual hospital costs expected to rise to almost 2 billion by 2030 [31]. Beyond direct hospitals costs, the indirect societal expenses related to lost productivity and care-related expenditures increases the overall financial impact of a single PJI case to nearly half a million dollars [32]. Our findings suggest that higher-volume centers may help mitigate these costs by improving the efficiency of initial care while also potentially reducing the overall financial impact associated with treatment failure [22]. The influence of hospital volume on reducing the cost of primary joint arthroplasty was previously described by Courtney et al. [2] who reported significantly lower mean hospital charges and Medicare payments. Within the context of septic revision arthroplasty, Frisch et al. [33] determined that high-volume centers were less likely to be high-cost outliers than low-volume centers. Similarly, several groups have described their experience in centers for PJI care where they observed significantly lower length of stay without compromising rates of failure or mortality, thus representing a potential mechanism of cost containment [21,34]. Together, these data suggest the dual benefits of higher-volume centers in delivering both effective and cost-efficient care which will be critical as the national PJI burden continues to rise [35]. However, as hospitals commonly incur greater costs than reimbursement for the episode of care related to PJI care, the present reimbursement models may require adjustment to allow these centers of excellence to be financially sustained.

We must acknowledge several weaknesses of the present study. As in all retrospective reviews of large databases, there is inherent risk of misclassification bias. While an extensive attempt was made to capture first stage spacers only, limitations in ICD-10 and Current Procedural Terminology coding limit our ability to define the exact procedure conducted and what stage of care it involved. Despite these limitations, we believe the use of charge codes, in addition to billing codes, enable the most accurate capture of these cases and minimize the risk of inaccurate classification. Additionally, the Premier Healthcare Database captures only the episodes of care immediately before and after surgery, limiting the assessment of long-term outcomes and mortality rates. As such, we are unable to determine if the differences observed here diverge or converge over time. Another key factor which we were unable to capture was the rates or timing of referred patients which has been shown to influence outcomes after revision arthroplasty.

This study also has several strengths. Principally, the use of a large database also enables the exploration of a rare event (mortality) in an uncommon complication (PJI) while remaining broadly generalizable. There is limited data on the relationship between hospital volume and outcomes in the setting of PJI care, thus these data provide key insights into how to optimize the delivery of care to this population. By leveraging patient-specific billing codes and line-item charges to optimally select patients undergoing antibiotic spacer placement for PJI, then using these data to build a robust model on the relationship between surgeon volume and postoperative mortality, we present compelling evidence in favor of shifting to a more centralized PJI care system.

Conclusions

This study highlights the clinically relevant association between lower hospital volume and increased 90-day mortality following antibiotic spacer placement for PJI. These findings reinforce emerging evidence in support of PJI care centers to improve survival and cost efficiency. However, unless the present reimbursement model for PJI care is modified, financial constraints will continue to limit the number of patients’ health systems and hospitals can sustainably treat. Additionally, we must acknowledge that the consolidation of PJI care may worsen health inequities faced by rural populations and those with limited ability to travel. Continuing to educate providers and empower hospital settings to care for PJI patients is therefore critical to facilitate equitable access to care. Concurrently, policies directed at facilitating measures to optimize PJI care at the system level should be explored further.

Uncited Table

Table 3.

CRediT authorship contribution statement

Julian Wier: Writing – review & editing, Writing – original draft, Methodology, Investigation, Formal analysis, Conceptualization. McKenzie W. Culler: Writing – review & editing, Writing – original draft, Methodology, Investigation, Formal analysis, Conceptualization. Avinash Iyer: Writing – review & editing, Writing – original draft, Methodology, Investigation, Formal analysis. Ryan Palmer: Writing – review & editing, Writing – original draft, Methodology, Investigation, Formal analysis. Sagar Telang: Writing – review & editing, Writing – original draft, Methodology, Investigation, Formal analysis. Jay R. Lieberman: Writing – review & editing, Supervision, Formal analysis, Conceptualization. Nathanael D. Heckmann: Writing – review & editing, Supervision, Methodology, Formal analysis, Conceptualization.

Conflicts of interest

Nathanael Heckmann receives royalties from Corin USA, is a paid consultant for Intellijoint Surgical, MicroPort Orthopaedics, Corin USA, and Zimmer, holds stock in Intellijoint Surgical, and serves as a board member of AAOS, AJRR, AAHKS, the Knee Society, and the Hip Society. Jay R. Lieberman receives royalties from DePuy: A Johnson & Johnson Company and Saunders/Mosby-Elsevier, is a paid consultant for DePuy: A Johnson & Johnson Company, holds stock in BD Surgiphor and Hip Innovations Technologies, and serves as a board member of AAOS, AJRR, AAHKS, the Knee Society, and the Hip Society. The other authors declare that there are no conflicts of interest.

For full disclosure statements, refer to https://doi.org/10.1016/j.artd.2026.102111.

Appendix A. Supplementary data

Supplementary Figure 1.