Gram-negative aetiology in hip and knee prosthetic joint infections: risk factors and implications for empirical antibiotic therapy—a retrospective cohort study Authors: Sara Tedeschi, Marta Fernandez-Sampedro, Eleonora Zamparini, Nicolò Rossi, Giuseppe Ferraro, Alberto Pecorelli, Giacomo Fornaro, Maria Pia Neri, Cesare Faldini, Stefano Zaffagnini, Massimiliano De Paolis, Alessandra Maso, Simone Ambretti, Maddalena Giannella, Pierluigi Viale Published: 2026 ABSTRACT AbstractBackgroundEmpirical broad-spectrum antibiotic therapy is currently recommended for patients undergoing surgical treatment of prosthetic joint infections (PJIs), while awaiting intraoperative culture results. However, widespread use of Gram-negative coverage may lead to more adverse reactions, resistance selections and costs.ObjectivesThis study aimed to identify the factors associated with having a PJI caused by Gram-negative bacteria (GNB).MethodsThis was an observational retrospective study of patients with hip and knee PJIs undergoing infection surgery between 2015 and 2019. PJI was defined according to 2018 Musculoskeletal Infection Society-International Consensus Meeting criteria. To assess risk factors for GNB-PJI, univariate and multivariate logistic regression analysis was carried out; statistical significance was considered for P values <0.05.ResultsA total of 281 patients were included, 137/281 (49%) were males, median age was 68 years (IQR 55-75), and the hip was involved in 147/281 (52%) cases. The most frequent surgical approach was a two-stage revision, carried out in 171/281 (61%), and the overall treatment failure rate was 19% (54/281). A Gram-negative aetiology was found in 51/281 (18%) cases; in this group the most frequent aetiological agents were Enterobacterales (n = 36). After surgery, 192/281 (68%) patients received empirical broad-spectrum antibiotic therapy with GNB coverage. In the logistic regression analysis, hip PJI was an independent risk factor for GNB-PJI (OR 2.88, 95% CI 1.44–5.77, P = 0.003) while delayed infection was associated with lower risk (OR 0.45, 95% CI 0.21–0.94, P = 0.03).ConclusionsIn our cohort, hip PJI was associated with a higher likelihood of GNB infection, whereas delayed-onset PJI was associated with a significantly lower likelihood. Introduction Prosthetic joint infection (PJI) is probably the most severe complication of joint replacement and occurs in 1%–2% of hip and knee primary arthroplasty.1 Gram-positive bacteria account for most PJIs,2,3 while Gram-negative bacteria (GNB) have been described as aetiological agents of PJIs mainly in early post-surgical PJI cases or in older patients, and have been associated with variable outcomes after treatment.3–7 A significant increase in the overall incidence of GNB-related PJIs (GNB-PJIs) has been described in recent years, reaching up to 30% in a Spanish cohort.8 However, to date, information about epidemiology and outcome of GNB-PJIs derives mainly from quite small, non-multinational cohorts.9 Currently, in the absence of a preoperative aetiological diagnosis, broad-spectrum antimicrobial therapy is recommended for patients undergoing surgical treatment of PJIs, pending the results of intraoperative cultures.10 The appropriate choice of this empirical antibiotic regimen is crucial, especially in the case of retention or one-stage exchange of the prosthesis, as initial therapeutic inaccuracy negatively affects outcome.11–14 Therefore, including an anti-GN agent, usually a broad-spectrum β-lactam, is recommended for patients undergoing infection surgery for PJI. However, this strategy may lead to widespread use of GNB coverage, increasing resistance selection, the risk of adverse events and healthcare costs. We hypothesized that the preoperative identification of patients at increased risk for GNB-PJI may enable a more tailored empirical antibiotic strategy, reducing the indiscriminate use of broad-spectrum agents against GNB. Patients and methods Study design and population We performed an observational, retrospective, multicentre cohort study including patients with hip and knee PJIs treated at the Infectious Diseases Unit of the IRCCS Azienda Ospedaliero-Universitaria di Bologna and at the Marques del Valdecilla University Hospital in Santander, Spain. Eligible patients were those who underwent surgical management of PJI [e.g. debridement, antibiotics and implant retention (DAIR), one-stage exchange, two-stage exchange, resection arthroplasty] from January 2015 to December 2019 with at least three intraoperative peri-prosthetic tissue samples obtained for microbiological culture. Patients were excluded if they had undergone previous infection surgery for the same PJI or were lost to follow-up within 12 months after index surgical procedure. Antimicrobial susceptibility was assessed according to the EUCAST recommendations; the MIC was provided for each tested antimicrobial.15 Definitions PJI was defined according to 2018 MSIS (Musculoskeletal Infection Society)-ICM (International Consensus Meeting) criteria.16 A PJI was considered a GNB-PJI in the presence of culture of at least two intraoperative peri-prosthetic tissue samples or one tissue sample and sonication fluid yielding GNB with >50 cfu/mL. PJIs were classified according to timing and characteristics of infection onset as: (i) early postoperative in case of signs/symptoms onset <4 weeks from prosthesis placement; (ii) delayed in case of signs/symptoms onset between 4 weeks and 24 months from prosthesis placement; and (iii) acute late in case of sudden, acute (lasting <3 weeks) onset of signs/symptoms of infection after >12 weeks from prosthesis placement, in a patient who was fully recovered after the primary arthroplasty.17–19 Treatment failure was defined as the need for further surgery to control infection (DAIR or prosthesis removal after a DAIR procedure or after a revision, and amputation), the need for antibiotic suppression, or infection related-death within 12 months after index surgery. Variables The following variables were retrospectively collected using a pre-established case record form (CRF), based on hospital and laboratory records as the data source: patients’ demographics; comorbidities and baseline characteristics; year of presentation; site and timing of PJI (hip or knee; early postoperative, delayed, acute late); symptoms of infection; surgery (date, type and number of surgical procedures); microbiology (culture results with pathogen identification and main mechanisms of antibiotic resistance, if applicable); antibiotic therapy (drug, dosage, duration); and outcome. Statistics Descriptive statistics were obtained for all the variables considered. Categorical variables were presented as absolute numbers and their frequencies; continuous variables were presented as mean ± SD or median and IQR, according to their distribution. Differences between patients’ groups were tested with chi-square tests or Fisher’s exact test when appropriate for categorical variables, and Student’s t test or Mann–Whitney U test for normally and non-normally distributed continuous variables, respectively. To assess risk factors for GNB-PJI, univariate and multivariate logistic regression analysis was performed. Statistical significance was considered for P values <0.05. Analysis was carried out with SPSS 26. Ethics The study was conducted according to the principles of the Helsinki Declaration and was approved by the Ethics Committee of the coordinating centre (Comitato Etico Area Vasta Emilia Centro, study code 779/2021/Oss/AOUBo). Results Overall, 281 patients were enrolled. Characteristics of the study population are summarized in Table 1. Briefly, 49% of patients were males, median age was 68 years (IQR 55–75), median Charlson comorbidity index was 3 (IQR 2–5) and 52% had a hip PJI. The majority (69%) of patients had a delayed PJI, and the most frequent presenting symptom was pain, which was reported by 65% of patients. Table 1.Characteristics of study populationAll patients(n = 281)GNB-PJI(n = 51)Other aetiology(n = 230) P Age, median (IQR), y68 (55–75)69 (54–77)67 (55–75)0.573Males, n (%)137 (49)28 (55)109 (47)0.332Hip prosthesis, n (%)147 (52)37 (72.5)110 (48)0.001CCI, median (IQR)3 (2–5)4 (2–5)3 (2–5)0.795BMI, median (IQR), kg/m227 (24–31)26 (24–30)27 (24–31)0.491Non-smokers, n (%)154 (55)28 (55)126 (55)0.890Active smokers, n (%)66 (23.5)13 (25.5)53 (23)Previous smokers, n (%)61 (22)10 (20)51 (22)COPD, n (%)72 (26)7 (14)65 (28)0.031Dementia, n (%)8 (3)3 (6)5 (2)0.150Diabetes, n (%)46 (16)6 (12)40 (17)0.354eGFR <60 mL/min, n (%)19 (7)2 (4)17 (7)0.372Indication for prosthesis, n (%)0.816 Osteoarthritis181 (64)31 (61)150 (65) Tumour35 (13)6 (12)29 (12) Fracture38 (13.5)9 (18)29 (12) Other27 (10)5 (10)22 (10)Mega or custom-made prosthesis, n (%)37 (13)8 (16)29 (13)0.521Previous revision, n (%)93 (33)21 (41)72 (31)0.182Timing of infection, n (%)<0.001 Early54 (19)18 (35)36 (16) Delayed195 (69)26 (51)169 (73.5) Acute late32 (11)7 (14)25 (11)Clinical presentation, n (%) Rubor116 (41)23 (20)93 (40)0.541 Calor119 (43)25 (50)94 (41)0.246 Tumor123 (44)23 (45)101 (43.5)0.833 Dolor183 (65)32 (63)151 (66)0.393 Wound dehiscence44 (16)14 (32)30 (13)0.010 Skin necrosis6 (2)2 (4)4 (2)0.329C-reactive protein, median (IQR), mg/dL2.26 (0.69–6.16)3.70 (1.25–9.14)1.94 (0.64–5.34)0.009Blood leucocyte count, median (IQR), ×106/L7790 (6310–9700)7850 (6280–9700)7725 (6300–9740)0.963Preoperative arthrocentesis, n (%)93 (34)19 (37)74 (32)0.639Mobilization, n (%)76 (29)7 (19)69 (30)0.161Surgery, n (%)0.082 DAIR51 (18)18 (35)33 (14) One-stage revision11 (4)011 (5) Two-stage revision171 (61)26 (51)145 (63) Partial revision29 (10)3 (6)26 (10) Arthrodesis2 (0.7)1 (2)1 (0.4) Resection arthroplasty16 (6)3 (6)13 (6) Amputation1 (0.4)01 (0.4)Treatment failure, n (%)54 (19)13 (25.5)41 (18)0.209CCI, Charlson comorbidity index; DAIR, debridement, antibiotics and implant retention; eGFR, estimated glomerular filtration rate; GNB-PJI, Gram-negative bacilli prosthetic joint infection. Characteristics of study population CCI, Charlson comorbidity index; DAIR, debridement, antibiotics and implant retention; eGFR, estimated glomerular filtration rate; GNB-PJI, Gram-negative bacilli prosthetic joint infection. The most frequent surgical approach was a two-stage revision (69%), followed by DAIR (18%). An attempt at aetiological diagnosis before surgery was carried out in 93 (33%) patients who underwent a synovial fluid culture, leading to a preoperative aetiological diagnosis in 79/281 (28%) cases. Results of intraoperative cultures were concordant with those of synovial fluid culture (both negative or both positive for the same microorganisms) in 63/93 (68%) cases; synovial fluid was positive and intraoperative negative in 13/93 (14%) cases; these two microbiological tests were positive for different microorganisms in 11/93 (12%) cases, while in only one patient was the synovial fluid culture negative and the intraoperative sample positive. All patients were treated with antibiotic therapy, which was started immediately after surgery; the most frequent empirical regimen was a combination of a glycopeptide (vancomycin or teicoplanin) or lipopeptide (daptomycin) plus piperacillin/tazobactam, administered in 119 patients. Overall, 68% (192/281) of patients received broad-spectrum postoperative antibiotic therapy with GNB coverage; a GNB aetiology was presumed before surgery in eight cases according to the results of preoperative synovial fluid culture. Fifty-one patients (18%) were diagnosed with a GNB-PJI; polymicrobial infections were found in 25 (11%), and culture-negative infections in 51 (18%). The most frequent microorganisms were CoNS (n = 108), followed by Staphylococcus aureus (n = 45), while among GNB-PJI the most frequent aetiological agents were Enterobacterales (n = 36) and Pseudomonas aeruginosa (n = 14) (Table 2). Among Gram-negative isolates, 27.5%, 13.7% and 5.9% of strains were resistant to fluoroquinolones, third-generation cephalosporins and carbapenems, respectively. Table 2.Microbiology of Gram-negative PJIMicroorganism n (%)Enterobacterales Escherichia coli14 (27.5%) Klebsiella spp.6 (11.8) Enterobacter aerogenes5 (9.8) Proteus mirabilis4 (7.8) Serratia spp.3 (5.9) Morganella spp.2 (3.9) Pantoea agglomerans2 (3.9) Pseudomonas aeruginosa 14 (27.5%) Sphingobacterium multivorum 1 (1.9%) Acinetobacter lwoffii 1 (1.9%) Pasteurella multocida 1 (1.9%) Bacteroides spp.1 (1.9%) Microbiology of Gram-negative PJI Overall, 54 patients (19%) experienced treatment failure within 12 months from infection surgery; treatment failure was numerically higher amongst those with GNB-PJI but not statistically significant (P = 0.209). Univariate and multivariate logistic regression analysis selected hip PJI (OR 2.88, 95% CI 1.44–5.77, P = 0.003) as an independent risk factor for GNB-PJI, while delayed infection (OR 0.45, 95% CI 0.21–0.94, P = 0.03) was associated with a lower risk of GNB-PJI (Table 3). Table 3.Univariate and multivariate logistic regression analysis of the risk factors for a prosthetic joint infection caused by Gram-negative bacteriaFactorOR (95% CI) P valueaOR (95% CI) P a Age, y1 (0.98–1.02)0.89Male gender0.74 (0.40–1.36)0.33BMI0.98 (0.93–1.05)0.69Charlson index1 (0.89–1.13)0.85Hip prosthesis2.88 (1.48–5.62)0.0022.88 (1.44–5.77) 0.003Indication for prosthesis0.81 Osteoarthritis1 Fracture1 (0.33–2.61) Tumour1.5 (0.64–3.48) Other1.1 (0.38–3.12)Delayed infection0.37 (0.2–0.69)0.0020.45 (0.21–0.94) 0.03Serum C-reactive protein1.04 (1.02–1.08)0.041.03 (0.98–1.07)0.20Wound dehiscence2.53 (1.22–5.21)0.011.25 (0.53–2.95)0.61aOR, adjusted odds ratio. aBold values denote statistical significance. Univariate and multivariate logistic regression analysis of the risk factors for a prosthetic joint infection caused by Gram-negative bacteria aOR, adjusted odds ratio. aBold values denote statistical significance. Discussion In this study, we investigated potential risk factors for having a PJI caused by GNB, trying to find a strategy to streamline the use of anti-GN agents in the empirical treatment of patients with PJIs undergoing infection surgery. We found that the likelihood of GNB infection was significantly higher for hip PJI and significantly lower for delayed infections. GNB account for 10%–20% of PJI cases and, despite representing a minority of all PJIs, these infections are considered particularly challenging due to the high virulence and the resistance pattern of the causative microorganisms. Indeed, according to a recent literature review, treatment failure rates are high, with 3 out of 10 patients treated for GNB-PJIs failing initial surgical treatment within 2 years.9 Successful treatment of PJI requires a combination of surgery and antibiotic treatment. The aetiology of infection is often unknown before surgery, and empirical antibiotic therapy is needed pending the results of intraoperative cultures. This empirical postoperative antibiotic treatment, routinely consisting of an anti-MRSA agent plus a broad-spectrum β-lactam, is usually continued until results of intraoperative cultures are available and, given that in case of PJIs an incubation time of 10–14 days is the main consensus,20,21 this strategy may entail up to 2 weeks of unnecessary broad-spectrum antibiotic treatment in some patients with PJIs, particularly outside specialized referral centres and where access to dedicated microbiological and infectious diseases expertise, as well as timely interpretation of microbiological results, may be more limited. Considering the already high incidence of PJIs, and its expected increase in the future,1 the impact of this antibiotic exposure is not negligible, in terms of adverse events, antimicrobial resistance spread and even healthcare costs. Indeed, each day of β-lactams has been associated with an increased risk of Clostridioides difficile infection (CDI) of over 2%, and antibiotic exposure is one of the main risk factors for developing an infection caused by MDR microorganisms.22,23 Therefore, it is necessary to find strategies to refine the use of antibiotics in the treatment of PJIs. Some attempts have been made to address this issue, including an early reassessment of empirical antibiotic treatment 5 days after surgery—a 5 day incubation period could be applicable for acute infections, but it could lead to insufficient yield in chronic PJIs.24 However, in a specialized tertiary referral centre in the UK, a standardized approach to the antibiotic treatment of bone and joint infections including PJIs is used, consisting of a short course of IV vancomycin plus meropenem started immediately after microbiological sampling and quickly rationalized to more specific therapy as soon as culture results are available (usually beginning at 48 h after surgery), without waiting for final culture results.25 Other groups have proposed different empirical antibiotic regimens according to the timing of PJI presentation, the pathogenetic mechanism of the infection, and the presence/absence of systemic involvement. Empirical anti-GN coverage was always proposed for patients with sepsis and for those with early postoperative infections.3,26 According to current recommendations, most patients in our cohort received broad-spectrum postoperative antibiotic therapy including an anti-GN agent, and piperacillin/tazobactam was the most frequently prescribed in empirical regimens. This choice was justified considering the proportion of strains resistant to fluoroquinolones and third-generation cephalosporins, and the overall prevalence of P. aeruginosa as the aetiological agent in our cohort. However, local epidemiology and individual risk factors are important in determining the risk of antibiotic-resistant GNB and/or P. aeruginosa infections and should always be considered in the choice of empirical anti-GN coverage. Of note, an attempt to obtain an aetiological preoperative diagnosis through preoperative synovial fluid aspiration was infrequently carried out (93/281 patients). Trying to obtain a preoperative aetiological diagnosis could be useful to prescribe a targeted post-surgical antibiotic treatment. Indeed, some studies have analysed the concordance in results between preoperative aspiration and intraoperative synovial fluid cultures. According to a recent systematic review, including 1677 patients from seven retrospective studies, mainly at high risk of bias, the concordance between preoperative synovial fluid culture and intraoperative tissue cultures in patients with hip and knee PJI undergoing revision varied between 52% and 78%.27 The 68% concordance rate we found in our study is consistent with these data. Higher quality studies are needed to obtain a more accurate estimate of this concordance rate before recommending the widespread adoption of a targeted postoperative antibiotic treatment based on preoperative synovial fluid culture results. Our study has limitations that should be acknowledged. First, its retrospective nature could have introduced selection bias, and the inclusion of only two centres runs the risk that results are influenced by local epidemiology, limiting the generalizability of our findings. Second, our definitions of PJI timing (early, delayed and acute late) may be controversial. However, they are based on largely accepted classification systems and due to the retrospective nature of the study it was not possible to assess reliably the pathogenetic mechanism of late infections (new haematogenous or exacerbation of a chronic infection). To conclude, our findings confirm that GNB are involved in a minority of PJIs but cannot be ignored. In a clinical setting where preoperative aetiological diagnosis is infrequent, we identified two factors independently associated with a GN aetiology of PJIs: (i) hip PJI was associated with a higher likelihood of GNB infection, and (ii) delayed-onset PJI was associated with a significantly lower likelihood. The latter finding is consistent with previous studies reporting a lower incidence of GNB in chronic PJI.3,7,26,28 However, although delayed-onset infection may identify a subgroup at lower risk of GNB aetiology, our findings do not demonstrate that this risk is sufficiently low to support withholding empirical Gram-negative coverage on the basis of infection timing alone. Further studies in larger cohorts are needed to determine whether a subgroup of patients at sufficiently low risk of GNB infection can be reliably identified, thereby allowing empirical antibiotic therapy to be safely narrowed. Acknowledgements The results of the present study have been presented at the 43rd Annual Meeting of the European Bone and Joint Infection Society, 11–13 September 2025, Bologna, Italy. Contributor Information Sara Tedeschi, Department of Medical and Surgical Sciences, Alma Mater Studiorum University of Bologna, Bologna, Italy; Infectious Diseases Unit, Department for Integrated Infectious Risk Management, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Via Massarenti 11, 40138 Bologna, Italy. Marta Fernandez-Sampedro, Infectious Diseases Service, Hospital Universitario Marqués de Valdecilla-IDIVAL, Cantabria, Spain; CIBER de Enfermedades Infecciosas, Instituto de Salud Carlos III (CIBERINFEC, ISCIII), Madrid, Spain; Spanish Society of Infectious Diseases and Clinical Microbiology (GEIO-SEIMC), Madrid, Spain. Eleonora Zamparini, Infectious Diseases Unit, Department for Integrated Infectious Risk Management, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Via Massarenti 11, 40138 Bologna, Italy. Nicolò Rossi, Infectious Diseases Unit, Guglielmo da Saliceto Hospital, Piacenza, Italy. Giuseppe Ferraro, Section of Infectious Diseases, Department of Translational Medicine, University of Ferrara, Ferrara, Italy. Alberto Pecorelli, Department of Medical and Surgical Sciences, Alma Mater Studiorum University of Bologna, Bologna, Italy. Giacomo Fornaro, Infectious Diseases Unit, Department for Integrated Infectious Risk Management, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Via Massarenti 11, 40138 Bologna, Italy. Maria Pia Neri, II Clinic of Orthopaedics and Traumatology, IRCCS Istituto Ortopedico Rizzoli, Bologna, Italy. Cesare Faldini, 1st Orthopaedics and Traumatology Clinic, IRCCS Istituto Ortopedico Rizzoli, Bologna, Italy; Department of Biomedical and Neuromotor Science-DIBINEM, University of Bologna, Bologna, Italy. Stefano Zaffagnini, II Clinic of Orthopaedics and Traumatology, IRCCS Istituto Ortopedico Rizzoli, Bologna, Italy; Department of Biomedical and Neuromotor Science-DIBINEM, University of Bologna, Bologna, Italy. Massimiliano De Paolis, Orthopedics and Traumatology Unit, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Bologna, Italy. Alessandra Maso, Laboratory of Microbiology and GMP Quality Control, IRCCS Istituto Ortopedico Rizzoli, Bologna, Italy. Simone Ambretti, Department of Medical and Surgical Sciences, Alma Mater Studiorum University of Bologna, Bologna, Italy; Microbiology Unit, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Bologna, Italy. Maddalena Giannella, Department of Medical and Surgical Sciences, Alma Mater Studiorum University of Bologna, Bologna, Italy; Infectious Diseases Unit, Department for Integrated Infectious Risk Management, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Via Massarenti 11, 40138 Bologna, Italy. Pierluigi Viale, Department of Medical and Surgical Sciences, Alma Mater Studiorum University of Bologna, Bologna, Italy; Infectious Diseases Unit, Department for Integrated Infectious Risk Management, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Via Massarenti 11, 40138 Bologna, Italy. Funding This study was carried out as part of our routine work Transparency declarations None to declare. References