If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
The introduction of the Musculoskeletal Infection Society (MSIS) criteria for periprosthetic joint infection (PJI) in 2011 resulted in improvements in diagnostic confidence and research collaboration. The emergence of new diagnostic tests and the lessons we have learned from the past 7 years using the MSIS definition, prompted us to develop an evidence-based and validated updated version of the criteria.
This multi-institutional study of patients undergoing revision total joint arthroplasty was conducted at 3 academic centers. For the development of the new diagnostic criteria, PJI and aseptic patient cohorts were stringently defined: PJI cases were defined using only major criteria from the MSIS definition (n = 684) and aseptic cases underwent one-stage revision for a noninfective indication and did not fail within 2 years (n = 820). Serum C-reactive protein (CRP), D-dimer, erythrocyte sedimentation rate were investigated, as well as synovial white blood cell count, polymorphonuclear percentage, leukocyte esterase, alpha-defensin, and synovial CRP. Intraoperative findings included frozen section, presence of purulence, and isolation of a pathogen by culture. A stepwise approach using random forest analysis and multivariate regression was used to generate relative weights for each diagnostic marker. Preoperative and intraoperative definitions were created based on beta coefficients. The new definition was then validated on an external cohort of 222 patients with PJI who subsequently failed with reinfection and 200 aseptic patients. The performance of the new criteria was compared to the established MSIS and the prior International Consensus Meeting definitions.
Two positive cultures or the presence of a sinus tract were considered as major criteria and diagnostic of PJI. The calculated weights of an elevated serum CRP (>1 mg/dL), D-dimer (>860 ng/mL), and erythrocyte sedimentation rate (>30 mm/h) were 2, 2, and 1 points, respectively. Furthermore, elevated synovial fluid white blood cell count (>3000 cells/μL), alpha-defensin (signal-to-cutoff ratio >1), leukocyte esterase (++), polymorphonuclear percentage (>80%), and synovial CRP (>6.9 mg/L) received 3, 3, 3, 2, and 1 points, respectively. Patients with an aggregate score of greater than or equal to 6 were considered infected, while a score between 2 and 5 required the inclusion of intraoperative findings for confirming or refuting the diagnosis. Intraoperative findings of positive histology, purulence, and single positive culture were assigned 3, 3, and 2 points, respectively. Combined with the preoperative score, a total of greater than or equal to 6 was considered infected, a score between 4 and 5 was inconclusive, and a score of 3 or less was not infected. The new criteria demonstrated a higher sensitivity of 97.7% compared to the MSIS (79.3%) and International Consensus Meeting definition (86.9%), with a similar specificity of 99.5%.
This study offers an evidence-based definition for diagnosing hip and knee PJI, which has shown excellent performance on formal external validation.
]. The diagnosis of PJI is based on a combination of clinical findings, laboratory results from peripheral blood and synovial fluid, microbiological culture, histological evaluation of periprosthetic tissue, and intraoperative findings.
The Musculoskeletal Infection Society (MSIS) and the Infectious Diseases Society (IDSA) have previously developed criteria to standardize the definition of PJI [
]. These definitions have become widely accepted among surgeons worldwide and have dramatically improved diagnostic confidence and facilitated treatment. Moreover, their use in research allowed for consistency in definition between studies and enhanced the potential for collaboration and the overall quality of literature. However, existing guidelines were largely generated by expert opinions and have not been validated. Furthermore, while relatively specific, there is concern about the sensitivity of the current definitions [
Although definite evidence or major criteria for infection are identical between the different definitions, the supportive evidence or minor criteria differ and are less agreed upon. In the recent years, numerous markers have been evaluated and become available [
These advancements in the field of PJI diagnosis call for the modification of current diagnostic criteria to an evidence-based one that is inclusive of the recent tests and considers the relative weights of the different tests. The purpose of this multi-institutional study was, thus, to: (1) generate an evidence-based, weight-adjusted scoring system for the definition of PJI of hip and knee, (2) validate it on an external cohort, and (3) compare its performance against currently available definitions.
Materials and Methods
After the institutional review board approval, we conducted a retrospective review of the medical records of all patients undergoing revision total hip arthroplasty (THA) and total knee arthroplasty (TKA) arthroplasty from 3 academic centers between January 2001 and July 2016. We excluded patients without serum erythrocyte sedimentation rate (ESR) or serum CRP, as well as those without a joint aspiration or an attempt at aspiration. Patients in whom the aspiration or serum testing was performed more than 8 weeks before surgery were also excluded.
Patients were classified as having a PJI if they met major diagnostic criteria of MSIS and International Consensus Meeting (ICM) [
], namely the presence of a sinus tract (with evidence of communication to the joint or visualization of the prosthesis) or 2 positive cultures isolating the same pathogen from the periprosthetic tissue or synovial fluid samples. Patients coded as infected who did not meet these definitions and those with megaprosthesis or missing surgical data were excluded from the study. In addition, we excluded acute PJI cases, defined as occurring less than 3 months from the index arthroplasty, and acute hematogenous PJI, defined as acute symptoms occurring for less than 6 weeks but more than 3 months from index surgery. Aseptic revisions were defined as cases undergoing single-stage revision for a diagnosis other than infection (loosening, wear, instability, malalignment, adverse local tissue reactions, or other aseptic causes) who did not fail with infection, nor had any further reoperation on the same joint.
Validation of the New Criteria
We performed external validation on separate patients from the same 3 institutions, who were not included in the initial developmental model. This validation was performed on a group of PJI and aseptic patients.
As there is no “gold standard” for PJI, we chose a representative sample of infected cases that was independent from any intrinsic bias from the commonly used definitions for infection. This group composed of patients who were treated as PJI cases (undergoing 2-stage revision THA and TKA) and failed with a reinfection within 1 year. All these patients were coded as infected (based on the International Classification of Diseases, Ninth Revision codes for PJI: 996.6, 996.66, 996.67, 998.5, 100, and 998.59). Data from the first infection were documented.
A randomly selected holdout sample of 200 aseptic revisions was excluded from the developmental model for validation purposes. These patients met the same aforementioned criteria for aseptic revision and did not fail with infection within 1 year after surgery.
Patient characteristics, comorbidities, laboratory results (serum, synovial, and microbial), and intraoperative findings (purulence and histopathology) were documented (see Appendix A). Laboratory values and histopathology results were dichotomized as elevated or not based on the ICM cutoffs.5 For markers not present in the ICM definition (serum D-dimer, synovial CRP), a cutoff was determined using the Gini index [
To maximize applicability to clinical practice, diagnostic markers were evaluated in a stepwise classification model based on the American Academy of Orthopaedic Surgeons guidelines accounting for simplicity and invasiveness [
]. The probability for infection was evaluated independently for each step; progression from one step to another was determined to minimize false positive and negatives. The discriminatory capability of each step was then assessed by receiver-operating characteristic curve analysis. Area under the curve (AUC) scores were typically considered acceptable if the AUC exceeded 0.7, with an AUC of 0.5 representing a poor test (toss of a coin) and an AUC of 1.0 signifying a perfect test. The first step included serum markers (CRP, D-dimer, and ESR). Subjects with an extremely low probability for PJI would not proceed to step 2. Step 2 included testing of synovial markers which requires more invasive testing (synovial fluid white blood cell [WBC] count, polymorphonuclear percentage, CRP, LE, and alpha-defensin). Subjects with an extremely low and high probability for infection would not proceed to step 3. The evaluation of intraoperative findings (histology, purulence, and single positive culture) was performed in step 3. Single culture was evaluated in step 3 (and not in step 2) as patients reaching this point already have a high index of suspicion (pretest probability) for infection, thus minimizing the chance for false positive cultures and increasing its overall performance.
For each step, a random forest analysis was performed to evaluate the relative weight and importance of each examined variable. Random forest is a robust method for ranking the prediction ability of the different variables [
]. The variables within each step were then evaluated in a multivariate logistic regression model. An integer point scoring system was created on the basis of the final beta coefficients; to make the scoring more practical and user friendly, beta coefficients were rounded to the nearest integer. Variables showing a high degree of colinearity (r > 0.7) were grouped together within a single criterion. Variables in the regression model which were not statistically significant were kept and scored, as we aimed to provide a scoring system that could be used when all markers are not available and not to guide which markers should be used.
A preoperative diagnostic score incorporating stages 1 and 2 and an intraoperative diagnostic score incorporating patients with an ambiguous preoperative diagnostic score and intraoperative findings (stage 3) were then created. Relative scores were assigned based on the previously mentioned beta coefficients. Diagnostic cut-points were determined and validated using the holdout sample of 200 aseptic revisions and 222 PJI cases as predefined. No imputations were used for missing markers/tests. The diagnostic performance (false positive, negative, true positives and negatives, sensitivity and specificity) of the final scoring system was evaluated and compared to the traditional MSIS and ICM definitions. All statistical analyses were performed using R version 3.4.3.
Overall 1504 patients (684 PJI and 820 aseptic revisions) were included in the developmental model. Of these, 663 (44.1%) underwent THA and 841 (55.9%) underwent TKA. The average age was 65.4 years (standard deviation 10.9), and 718 (47.7%) were male. Patients in the infected group were more likely to be male (P < .001), had most recently undergone a revision (as opposed to a primary procedure, P < .001), had undergone knee as opposed to hip arthroplasty (P = .009), and had a higher Charlson Comorbidity Index with a history of rheumatoid arthritis, diabetes, and malignancy (P < .001, Table 1).
Table 1Characteristics of Patients Who Were Included in the Developmental Model (n = 1504).
In order of importance, the random forest analysis determined that elevated serum CRP (>1 mg/dL), D-dimer (>860 ng/mL), and ESR (>30 mm/h) as the most important variables associated with PJI in that order (Table 2); the beta coefficients were rounded to 2, 2, and 1, respectively. Owing to high collinearity between serum CRP and D-dimer (r = 0.72), they were grouped together within a single criterion. Further results from the random forest analysis of synovial markers pointed out elevated synovial fluid WBC count (>3000), alpha-defensin (signal-to-cutoff ratio >1), LE (++), polymorphonuclear percentage (>80%), and synovial CRP (6.9 mg/L) as the most important variables associated with PJI in that order (Table 2); the beta coefficients were rounded to 3, 3, 3, 2, and 1, respectively. Owing to high collinearity between synovial WBC and LE (r = 0.75), they were grouped together within a single criterion. The overall performance showed an AUC of 0.99. The relative scores from both steps were incorporated to a preoperative diagnostic score with patients having a score of greater than or equal to 6 considered infected, while patients with a score of 0 or 1 classified as not infected (Fig. 1).
Table 2Simple Importance Based on Random Forest and Beta Coefficients Derived From a Multivariate Regression Analysis of Each Step.
In cases where the preoperative score was between 2 and 5, intraoperative variables were considered to attempt to classify the patient as infected or aseptic. The intraoperative findings listed in descending order of importance were positive histology (3 points), purulence (3 points), and a single positive culture (2 points) (Table 2). The ability to discriminate PJI from aseptic revisions at this third stage alone was lower, with an AUC of 0.92. However, taking the results of the preoperative scores into account resulted in an AUC of 0.97. Thus, the relative preoperative scores were incorporated into the intraoperative diagnostic score with patients having an aggregate score (combined preoperative and intraoperative) of greater than or equal to 6 classified as infected, scores of 4-5 as inconclusive, and scores of 3 or less not infected (Fig. 1).
The validation cohort included 222 infected and 200 aseptic revisions. Patient characteristics are presented in Appendix B. Of the PJI cohort, 212/222 (95.5%) were correctly diagnosed (true positives) as infected, 5/222 (2.3%) were falsely diagnosed as uninfected (false negatives), and in 5/222 (2.3%), no definite diagnosis could be made. In the aseptic cohort, 195/200 (97.5%) were correctly diagnosed as not infected (true negatives), 1/200 (0.5%) were falsely diagnosed as infected (false positives), and in 4/200 (2.0%), no definite diagnosis could be made.
Relative Performance Compared With Current Definitions
The overall sensitivity and specificity of the scoring system was 97.7% (95% confidence interval [CI] 94.7%-99.3%) and 99.5% (95% CI 97.3%-99.99%), respectively. Our scoring system demonstrated a higher sensitivity compared to the ICM (86.9%, 95% CI 81.8%-91.1%) and MSIS (79.3%, 95% CI 73.4%-84.4%) definitions, with similar specificity (Table 3). The sensitivity and specificity remained high in a subgroup analysis examining hips and knees separately (Appendix C), as well as for each of the 3 different intuitions individually (Appendix D).
Table 3Performance of the New Definition Compare With the Traditionally Used Musculoskeletal Infection Society (MSIS) and International Consensus Meeting (ICM) Criteria.
In the absence of a test with absolute accuracy, the diagnosis of a clinical condition needs to rely on a combination of criteria. This scenario is common in medicine, with the diagnosis of numerous conditions such as ankylosing spondylitis, rheumatoid arthritis, endocarditis, and sepsis all relying on diagnostic criteria. PJI is also one such clinical entity, the diagnosis of which has relied on the presence of criteria. Up until recently and before the introduction of the diagnostic criteria by MSIS and IDSA, the definition of PJI was fairly subjective and at the discretion of the treating clinician. Furthermore, research investigations also suffered from lack of standardized criteria. The efforts invested by the MSIS and the IDSA in introducing the diagnostic criteria for PJI brought standardization and uniformity into the field that has served many patients. Over the last few years, extensive research efforts were invested in diagnosis of PJI, and a few novel tests have been introduced [
The objective of this study was to conduct a comprehensive review of the literature to identify tests that may have a role in diagnosis of PJI and devise diagnostic criteria for PJI that may improve the accuracy of the prior diagnostic criteria further. A group of established biostatisticians with expertise in bioinformatics and statistics from the University of Tel Aviv were recruited to help us in this initiative. The intention was to validate the newly developed diagnostic criteria on a holdout cohort of patients from 3 academic institutions. In addition, by using appropriate statistical methods, we were able to assign a relative and quantitative weight for established and newly introduced tests. We were able to confirm that the newly introduced diagnostic criteria will have better performance, in terms of accuracy, compared to previous MSIS and ICM definitions.
The new diagnostic criteria were introduced to address a few of the limitations of the prior definitions. First, the prior definitions do not consider chronicity or invasiveness of the diagnostic tests. For instance, 3 of the 6 minor criteria within the current MSIS definition are typically intraoperative findings (purulence, single culture growth, and positive histology). This can make the preoperative diagnosis of infection extremely difficult. Second, novel tests that allow us to reach a preoperative diagnosis more easily and with high diagnostic performance are not accounted for [
]. Third, the prior definitions do not account for interplay between individual test results or their combined influence upon pretest probability. The additive effects of positive diagnostic markers has been demonstrated in multiple studies to substantially affect post-test probability [
]. By using a stepwise approach in developing the current criteria, which was based on the current American Academy of Orthopaedic Surgeons guidelines, we were able to assess the relative weights of these tests and take into account their pretest probability. Our score can be used to diagnose patients within the preoperative period more easily. It is also more accurate for those ambiguous cases with a high pretest probability, for which we provide a tool for an intraoperative diagnosis. The stepwise approach also enabled us to avoid false positives, which may have become an issue with increased sensitivity. Finally, external validation confirmed higher sensitivity compared to the established MSIS and ICM definitions with no significant increase in false positives. The fact that many of these infected patients had only a limited number of tests/markers available highlights the applicability of this proposed definition.
Another novel finding of the present definition is the introduction of patients in which a diagnosis is inconclusive. These patients are often encountered in clinical practice and represent a real diagnostic challenge. Pointing out this unique group or “gray area” of patients promotes awareness in both clinical practice and the need for further research focused at this cohort. Interestingly, all patients with an inconclusive diagnosis had negative cultures and may thus benefit from molecular diagnostic testing such as next-generation sequencing which has recently shown promising results [
This study is not without limitations. The retrospective design and incorporation of data from multiple centers may have resulted in variation in both data collection and reporting. While we implemented a stepwise stratification approach, which allowed us to minimize missing data, some data were imputed which may have limited the power of certain variables. Nevertheless, the scoring system was validated on an external model and remained very sensitive and specific. Another point to make is that we deliberately included certain variables in the final model which were not statistically significant. The reason for this was to maximize usability for physicians by incorporating markers used in everyday practice. It is, therefore, important to emphasize that the present scoring system is not designed or intended to be used as a guide for which tests should be ordered; rather, it should be used as a tool to diagnose patients when a panel of tests are already available. Furthermore, it is important to stress that the major diagnostic criteria was used as the “gold standard” for PJI in the development of this model. This group may have a more overt infection that may have affected the value of each score in the final model. Nevertheless, we tried to minimize bias by using the standard ICM threshold values for each individual test (Table 4). Another limitation relates to the use of chronic infections for model development and validation. While we believe these criteria should apply also for acute infections, both the scoring system and proposed thresholds require further validation. Finally, while we show an excellent performance, clinical judgment should still prevail and guide physicians in management of patients. There may be situations when a patient is infected and does not meet the diagnostic criteria and vice versa (Table 5) [
In conclusion, we present the first validated, evidence-based criteria for diagnosing PJI after hip and knee arthroplasty. We hope this new definition will guide clinicians and further improve the quality of research surrounding this devastating condition. Despite extensive investigations, diagnosis of the problem in some patients remains uncertain. These patients may benefit from the use of novel techniques such as next-generation sequencing.
One or more of the authors of this paper have disclosed potential or pertinent conflicts of interest, which may include receipt of payment, either direct or indirect, institutional support, or association with an entity in the biomedical field which may be perceived to have potential conflict of interest with this work. For full disclosure statements refer to https://doi.org/10.1016/j.arth.2018.02.078.
We read with interest the article entitled “The 2018 Definition of Periprosthetic Hip and Knee Infection: An Evidence-Based and Validated Criteria” by Parvizi et al . Our group includes pathologists with subspecialty expertise in hemostasis and thrombosis testing and orthopedic surgeons; we are especially interested in the proposed use of D-dimer in modified scoring criteria for diagnosis of periprosthetic joint infection.