The Journal of Arthroplasty 33 (2018) 1309e1314

Contents lists available at ScienceDirect

The Journal of Arthroplasty

journal homepage: www.arthroplastyjournal.org

The 2018 Definition of Periprosthetic Hip and Knee Infection:

An Evidence-Based and Validated Criteria
Javad Parvizi, MD a, *, Timothy L. Tan, MD a, Karan Goswami, MD a, Carlos Higuera, MD b,
Craig Della Valle, MD c, Antonia F. Chen, MD, MBA a, Noam Shohat, MD a, d
a
Rothman Institute, Thomas Jefferson University, Philadelphia, PA
b
Cleveland Clinic, Cleveland, OH
c
Rush University Medical Center, Chicago, IL
d
Sackler Faculty of Medicine, Tel Aviv University, Ramat Aviv, Israel

a r t i c l e i n f o a b s t r a c t

Article history: Background: The introduction of the Musculoskeletal Infection Society (MSIS) criteria for periprosthetic
Received 6 February 2018 joint infection (PJI) in 2011 resulted in improvements in diagnostic confidence and research collaboration.
Received in revised form The emergence of new diagnostic tests and the lessons we have learned from the past 7 years using the MSIS
19 February 2018
definition, prompted us to develop an evidence-based and validated updated version of the criteria.
Accepted 20 February 2018
Methods: This multi-institutional study of patients undergoing revision total joint arthroplasty was conducted
Available online 26 February 2018
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 ¼ 684) and
Keywords:
periprosthetic joint infection
aseptic cases underwent one-stage revision for a noninfective indication and did not fail within 2 years (n ¼
diagnosis 820). Serum C-reactive protein (CRP), D-dimer, erythrocyte sedimentation rate were investigated, as well as
hip synovial white blood cell count, polymorphonuclear percentage, leukocyte esterase, alpha-defensin, and sy-
knee novial CRP. Intraoperative findings included frozen section, presence of purulence, and isolation of a pathogen
diagnostic criteria 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.
Results: 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/mL), alpha-defensin (signal-to-cutoff ratio >1), leukocyte
esterase (þþ), 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%.
Conclusion: This study offers an evidence-based definition for diagnosing hip and knee PJI, which has
shown excellent performance on formal external validation.
© 2018 Elsevier Inc. All rights reserved.

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.
* Reprint requests: Javad Parvizi, MD, Rothman Institute, 125 S 9th Street. Suite
1000, Philadelphia, PA 19107.
Diagnosing periprosthetic joint infection (PJI) of the hip and
knee remains a major challenge as there is no test with absolute
accuracy [1,2]. The diagnosis of PJI is based on a combination of
clinical findings, laboratory results from peripheral blood and sy-
novial fluid, microbiological culture, histological evaluation of
periprosthetic tissue, and intraoperative findings.

https://doi.org/10.1016/j.arth.2018.02.078
0883-5403/© 2018 Elsevier Inc. All rights reserved.
1310 J. Parvizi et al. / The Journal of Arthroplasty 33 (2018) 1309e1314
The Musculoskeletal Infection Society (MSIS) and the Infectious
Diseases Society (IDSA) have previously developed criteria to
standardize the definition of PJI [3,4]. A further consensus meeting
in 2013 endorsed the MSIS definition of PJI and modified it slightly
[5]. These definitions have become widely accepted among sur-
geons 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 [6].
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
[7e9], including serum D-dimer [10], synovial leukocyte esterase
(LE) [11e13], synovial alpha-defensin [14,15], synovial C-reactive
protein (CRP) [16,17], and molecular techniques such as next-
generation sequencing [18]. Moreover, publications in the recent
years have shown different weights (sensitivity and specificity) for
the various tests used [9,19] and highlight the value of a high pre-
test probability in the overall diagnosis [14,20,21].
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 under-
going revision total hip arthroplasty (THA) and total knee arthro-
plasty (TKA) arthroplasty from 3 academic centers between January
2001 and July 2016. We excluded patients without serum eryth-
rocyte 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.
Patient Population
Developmental Model
Patients were classified as having a PJI if they met major diag-
nostic criteria of MSIS and International Consensus Meeting (ICM)
[3e5], 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 peri-
prosthetic tissue or synovial fluid samples. Patients coded as
infected who did not meet these definitions and those with meg-
aprosthesis 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 hematoge-
nous 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 develop-
mental model. This validation was performed on a group of PJI and
aseptic patients.
PJI Cases. As there is no “gold standard” for PJI, we chose a repre-
sentative 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 Revi-
sion codes for PJI: 996.6, 996.66, 996.67, 998.5, 100, and 998.59).
Data from the first infection were documented.
Aseptic Cases. A randomly selected holdout sample of 200 aseptic
revisions was excluded from the developmental model for valida-
tion purposes. These patients met the same aforementioned criteria
for aseptic revision and did not fail with infection within 1 year
after surgery.
Data Collection
Patient characteristics, comorbidities, laboratory results (serum,
synovial, and microbial), and intraoperative findings (purulence
and histopathology) were documented (see Appendix A). Labora-
tory 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 [22].
Statistical Analysis
To maximize applicability to clinical practice, diagnostic
markers were evaluated in a stepwise classification model based on
the American Academy of Orthopaedic Surgeons guidelines ac-
counting for simplicity and invasiveness [23]. This stepwise
approach allowed us to take into account the pretest probability for
infection and minimize missing data and the use of imputed data
[24]. Missing data were filled by 10 imputations using multiple
imputation by chained equations [25]. 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 eval-
uate the relative weight and importance of each examined variable.
Random forest is a robust method for ranking the prediction ability
 J. Parvizi et al. / The Journal of Arthroplasty 33 (2018) 1309e1314 1311

Table 1
Characteristics of Patients Who Were Included in the Developmental Model (n ¼ 1504).

Variable Overall (n ¼ 1504) PJI Cohort (n ¼ 684) Aseptic Cohort (n ¼ 820) P Value

Age (y) 65.4 (10.9) 65.9 (11.0) 64.9 (10.8) .079

Gender (male) 718 (47.7%) 366 (53.5%) 352 (42.9%) <.001a
Race (white) 1270 (84.4%) 569 (83.2%) 70 (85.5%) .05
Joint (knee) 841 (55.9%) 409 (59.8%) 432 (52.7%) .009a
Time from the most recent surgery (yr) 6.0 (8.7) 4.3 (9.3) 7.4 (7.9) <.001a
Most recent surgeryerevision procedure 416 (27.7%) 284 (41.5%) 132 (16.1%) <.001a
Body mass index (kg/m2) 31.1 (6.8) 31.4 (7.5) 30.9 (6.1) .192
Charlson Comorbidity Index (mean) 1.80 (1.8) 2.2 (1.7) 1.3 (1.8) <.001a
History of rheumatoid arthritis 99 (6.6%) 62 (9.1%) 37 (4.5%) <.001a
History of malignancy 70 (4.7) 57 (8.3%) s13 (1.6%) <.001a
History of diabetes 261 (17.4%) 152 (22.2%) 109 (13.3%) <.001a

Data are presented as mean (standard deviation) or number (%); kilogram (kg); meter (m); year (yr).
PJI, periprosthetic joint infection.
a
Statistically significant.

of the different variables [26]. The variables within each step were Results
then evaluated in a multivariate logistic regression model. An
integer point scoring system was created on the basis of the final Overall 1504 patients (684 PJI and 820 aseptic revisions) were
beta coefficients; to make the scoring more practical and user included in the developmental model. Of these, 663 (44.1%) un-
friendly, beta coefficients were rounded to the nearest integer. derwent THA and 841 (55.9%) underwent TKA. The average age was
Variables showing a high degree of colinearity (r > 0.7) were 65.4 years (standard deviation 10.9), and 718 (47.7%) were male.
grouped together within a single criterion. Variables in the Patients in the infected group were more likely to be male (P <
regression model which were not statistically significant were kept .001), had most recently undergone a revision (as opposed to a
and scored, as we aimed to provide a scoring system that could be primary procedure, P < .001), had undergone knee as opposed to
used when all markers are not available and not to guide which hip arthroplasty (P ¼ .009), and had a higher Charlson Comorbidity
markers should be used. Index with a history of rheumatoid arthritis, diabetes, and malig-
A preoperative diagnostic score incorporating stages 1 and 2 and nancy (P < .001, Table 1).
an intraoperative diagnostic score incorporating patients with an
ambiguous preoperative diagnostic score and intraoperative find-
ings (stage 3) were then created. Relative scores were assigned Development of the Scoring Model
based on the previously mentioned beta coefficients. Diagnostic
cut-points were determined and validated using the holdout In order of importance, the random forest analysis determined
sample of 200 aseptic revisions and 222 PJI cases as predefined. No that elevated serum CRP (>1 mg/dL), D-dimer (>860 ng/mL), and
imputations were used for missing markers/tests. The diagnostic ESR (>30 mm/h) as the most important variables associated with
performance (false positive, negative, true positives and negatives, PJI in that order (Table 2); the beta coefficients were rounded to 2, 2,
sensitivity and specificity) of the final scoring system was evaluated and 1, respectively. Owing to high collinearity between serum CRP
and compared to the traditional MSIS and ICM definitions. All sta- and D-dimer (r ¼ 0.72), they were grouped together within a single
tistical analyses were performed using R version 3.4.3. criterion. Further results from the random forest analysis of syno-
vial markers pointed out elevated synovial fluid WBC count
(>3000), alpha-defensin (signal-to-cutoff ratio >1), LE (þþ), poly-
Table 2 morphonuclear percentage (>80%), and synovial CRP (6.9 mg/L) as
Simple Importance Based on Random Forest and Beta Coefficients Derived From a the most important variables associated with PJI in that order
Multivariate Regression Analysis of Each Step. (Table 2); the beta coefficients were rounded to 3, 3, 3, 2, and 1,
Step Random Beta Standard P Value Score respectively. Owing to high collinearity between synovial WBC and
Forest Error LE (r ¼ 0.75), they were grouped together within a single criterion.
Step 1 The overall performance showed an AUC of 0.99. The relative scores
Serum CRP >1 mg/dLa 198 2.48 0.28 <.001 2 from both steps were incorporated to a preoperative diagnostic
Serum D-dimer > 860 ng/mLa 134 2.41 0.62 <.001 2 score with patients having a score of greater than or equal to 6
Serum ESR >30 mm/h 112 1.39 0.29 <.001 1
considered infected, while patients with a score of 0 or 1 classified
Step 2
Synovial WBC count >3000 109 2.65 0.80 .001 3 as not infected (Fig. 1).
(cells/mL)a In cases where the preoperative score was between 2 and 5,
Synovial alpha-defensin 79 2.64 1.24 .041 3 intraoperative variables were considered to attempt to classify the
Synovial LE (þþ)a 63 2.56 1.02 .017 3 patient as infected or aseptic. The intraoperative findings listed in
Synovial PMN% >80% 47 1.73 0.92 .121 2
Synovial CRP >6.9 mg/L 22 0.85 1.12 .449 1
descending order of importance were positive histology (3 points),
Step 3 purulence (3 points), and a single positive culture (2 points)
Histologyb 17 3.21 1.02 .002 3 (Table 2). The ability to discriminate PJI from aseptic revisions at
Purulence 12 3.47 1.32 .007 3 this third stage alone was lower, with an AUC of 0.92. However,
Single culture 8 2.25 1.45 .122 2
taking the results of the preoperative scores into account resulted
CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; LE, leukocyte esterase; in an AUC of 0.97. Thus, the relative preoperative scores were
PMN%, polymorphonuclear %; WBC, white blood cell. incorporated into the intraoperative diagnostic score with patients
a
The following demonstrated a high collinearity (r > 0.7) and thus were grouped
into a single criterion in the final model.
having an aggregate score (combined preoperative and intra-
b
Greater than 5 neutrophils per high-power field in 5 high-power fields observed operative) of greater than or equal to 6 classified as infected, scores
from histologic analysis of periprosthetic tissue at 400 magnification. of 4-5 as inconclusive, and scores of 3 or less not infected (Fig. 1).
1312 J. Parvizi et al. / The Journal of Arthroplasty 33 (2018) 1309e1314

Fig. 1. New scoring based definition for periprosthetic joint infection (PJI). Proceed with caution in: adverse local tissue reaction, crystal deposition disease, slow growing organisms.
CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; LE, leukocyte esterase; PMN, polymorphonuclear; WBC, white blood cell. aFor patients with inconclusive minor
criteria, operative criteria can also be used to fulfill definition for PJI. bConsider further molecular diagnostics such as next-generation sequencing.

External Validation
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 un-
infected (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 indi-
vidually (Appendix D).
Discussion
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

Table 3
Performance of the New Definition Compare With the Traditionally Used Musculoskeletal Infection Society (MSIS) and International Consensus Meeting (ICM) Criteria.

Criteria PJI Cohort (n ¼ 222) Aseptic Cohort (n ¼ 200) Sensitivity (95% CI) Specificity (95% CI)

True Positives False Negatives Inconclusive True Negative False Positives Inconclusive

MSIS (2011) 176 (79.3%) 46 (20.7%) - 199 (99.5%) 1 (0.5%) - 79.3% (73.4-84.4) 99.5% (97.3-99.99)
ICM (2013) 193 (86.9%) 29 (13.1%) - 199 (99.5%) 1 (0.5%) - 86.9% (81.8-91.1) 99.5% (97.3-99.99)
New definition (2018) 212 (95.5%) 5 (2.3%) 5 (2.3%) 195 (97.5%) 1 (0.5%) 4 (2.0%) 97.7% (94.7-99.3) 99.5% (97.2-99.99)

CI, confidence interval; PJI, periprosthetic joint infection.

 J. Parvizi et al. / The Journal of Arthroplasty 33 (2018) 1309e1314
Table 4
Proposed Thresholds Based on the 2013 ICM Combined With Current Findings.
Marker Chronic
(>90 d)
Serum CRP (mg/dL) 1.0
Serum D-dimer (ng/mL) 860
Serum ESR (mm/h) 30
Synovial WBC count (cells/mL) 3000
Synovial PMN (%) 80
Synovial CRP (mg/L) 6.9a
Synovial alpha-defensin (signal-to-cutoff ratio) 1.0
CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; ICM, International
Consensus Meeting; PMN, polymorphonuclear; WBC, white blood cell.
a
Further studies are needed to validate a specific threshold.
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 standardiza-
tion 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
[14,18,27].
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 bioinfor-
matics 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 defini-
tion 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 [9,11,12]. Third,
the prior definitions do not account for interplay between indi-
vidual test results or their combined influence upon pretest prob-
ability. The additive effects of positive diagnostic markers has been
demonstrated in multiple studies to substantially affect post-test
probability [14,20,21]. By using a stepwise approach in devel-
oping the current criteria, which was based on the current Amer-
ican 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
Table 5
Patients in Whom the Proposed Criteria May Be Inaccurate.
Red Flag Patients
Adverse local tissue reaction (ALTR)
Crystalline deposition arthropathy
Inflammatory arthropathy flare
Infection with slow growing organismsa
a
Such as Propionibacterium acnes, coagulase negative Staphylococcus, and others.
1313
within the preoperative period more easily. It is also more accurate
for those ambiguous cases with a high pretest probability, for which
Acute we provide a tool for an intraoperative diagnosis. The stepwise
(<90 d) approach also enabled us to avoid false positives, which may have
10 become an issue with increased sensitivity. Finally, external vali-
860a dation confirmed higher sensitivity compared to the established
- MSIS and ICM definitions with no significant increase in false
10,000
positives. The fact that many of these infected patients had only a
90
6.9 limited number of tests/markers available highlights the applica-
1.0 bility of this proposed definition.
Another novel finding of the present definition is the intro-
duction of patients in which a diagnosis is inconclusive. These pa-
tients 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 re-
sults [18].
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 maxi-
mize 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 stan-
dard 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 pro-
posed 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 situa-
tions when a patient is infected and does not meet the diagnostic
criteria and vice versa (Table 5) [28].
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.
References
[1] Ferna
ndez-Sampedro M, Farin ~ as-Alvarez C, Garces-Zarzalejo C, Alonso-
Aguirre MA, Salas-Venero C, Martínez-Martínez L, et al. Accuracy of different
diagnostic tests for early, delayed and late prosthetic joint infection. BMC Infect
Dis 2017;17:592. https://doi.org/10.1186/s12879-017-2693-1.
[2] Ahmad SS, Shaker A, Saffarini M, Chen AF, Hirschmann MT, Kohl S. Accuracy of
diagnostic tests for prosthetic joint infection: a systematic review. Knee Surg
1314 J. Parvizi et al. / The Journal of Arthroplasty 33 (2018) 1309e1314
Sports Traumatol Arthrosc Off J ESSKA 2016;24:3064e74. https://doi.org/
10.1007/s00167-016-4230-y.
[3] Osmon DR, Berbari EF, Berendt AR, Lew D, Zimmerli W, Steckelberg JM, et al.
Executive summary: diagnosis and management of prosthetic joint infection:
clinical practice guidelines by the Infectious Diseases Society of America. Clin
Infect Dis Off Publ Infect Dis Soc Am 2013;56:1e10. https://doi.org/10.1093/
cid/cis966.
[4] Parvizi J, Zmistowski B, Berbari EF, Bauer TW, Springer BD, Della Valle CJ, et al.
New definition for periprosthetic joint infection: from the Workgroup of the
musculoskeletal infection society. Clin Orthop 2011;469:2992e4. https://
doi.org/10.1007/s11999-011-2102-9.
[5] Parvizi J, Gehrke T, Chen AF. Proceedings of the International consensus on
periprosthetic joint infection. Bone Joint J 2013;95-B:1450e2. https://doi.org/
10.1302/0301-620X.95B11.33135.
[6] Koh IJ, Cho W-S, Choi NY, Parvizi J, Kim TK. How accurate are orthopedic
surgeons in diagnosing periprosthetic joint infection after total knee arthro-
plasty?: a multicenter study. Knee 2015;22:180e5. https://doi.org/10.1016/
j.knee.2015.02.004.
[7] Patel R, Alijanipour P, Parvizi J. Advancements in diagnosing periprosthetic
joint infections after total hip and knee arthroplasty. Open Orthop J 2016;10:
654e61. https://doi.org/10.2174/1874325001610010654.
[8] Deirmengian C, Kardos K, Kilmartin P, Cameron A, Schiller K, Parvizi J. Diag-
nosing periprosthetic joint infection: has the era of the biomarker arrived?
Clin Orthop 2014;472:3254e62. https://doi.org/10.1007/s11999-014-3543-8.
[9] Lee YS, Koo K-H, Kim HJ, Tian S, Kim TY, Maltenfort MG, et al. Synovial fluid
biomarkers for the diagnosis of periprosthetic joint infection: a systematic
review and meta-analysis. J Bone Joint Surg Am 2017;99:2077e84. https://
doi.org/10.2106/JBJS.17.00123.
[10] Shahi A, Kheir MM, Tarabichi M, Hosseinzadeh HRS, Tan TL, Parvizi J. Serum D-
dimer test is promising for the diagnosis of periprosthetic joint infection and
timing of reimplantation. J Bone Joint Surg Am 2017;99:1419e27. https://
doi.org/10.2106/JBJS.16.01395.
[11] Wyatt MC, Beswick AD, Kunutsor SK, Wilson MJ, Whitehouse MR, Blom AW.
The alpha-defensin immunoassay and leukocyte esterase colorimetric strip
test for the diagnosis of periprosthetic infection: a systematic review and
meta-analysis. J Bone Joint Surg Am 2016;98:992e1000. https://doi.org/
10.2106/JBJS.15.01142.
[12] Tischler EH, Cavanaugh PK, Parvizi J. Leukocyte esterase strip test: matched
for musculoskeletal infection society criteria. J Bone Joint Surg Am 2014;96:
1917e20. https://doi.org/10.2106/JBJS.M.01591.
[13] Diagnosis of periprosthetic joint infection: the utility of a simple yet unappreciated
enzyme. Available at:https://www.ncbi.nlm.nih.gov/pubmed/?term¼ParviziþJ%
2CþJacovidesþC%2CþAntociþV%2CþGhanemþE%2CþDiagnosisþofþ
periprostheticþjointþinfection%3AþTheþutilityþofþaþsimpleþyetþ
unappreciatedþenzyme. [Accessed 27 September 2017].
[14] Deirmengian C, Kardos K, Kilmartin P, Cameron A, Schiller K, Parvizi J. Com-
bined measurement of synovial fluid a-Defensin and C-reactive protein levels:
highly accurate for diagnosing periprosthetic joint infection. J Bone Joint Surg
Am 2014;96:1439e45. https://doi.org/10.2106/JBJS.M.01316.
€hler C, Kubista B, et al. Qualitative
[15] Sigmund IK, Holinka J, Gamper J, Staats K, Bo
a-defensin test (Synovasure) for the diagnosis of periprosthetic infection in
revision total joint arthroplasty. Bone Joint J 2017;99-B:66e72. https://
doi.org/10.1302/0301-620X.99B1.BJJ-2016-0295.R1.
[16] Tetreault MW, Wetters NG, Moric M, Gross CE, Della Valle CJ. Is synovial C-
reactive protein a useful marker for periprosthetic joint infection? Clin Orthop
2014;472:3997e4003. https://doi.org/10.1007/s11999-014-3828-y.
[17] Omar M, Ettinger M, Reichling M, Petri M, Guenther D, Gehrke T, et al. Sy-
novial C-reactive protein as a marker for chronic periprosthetic infection in
total hip arthroplasty. Bone Joint J 2015;97-B:173e6. https://doi.org/10.1302/
0301-620X.97B2.34550.
[18] Tarabichi M, Shohat N, Goswami K, Alvand A, Silibovsky R, Belden K, et al.
Diagnosis of periprosthetic joint infection: the potential of next-generation
sequencing. J Bone Joint Surg Am 2018;100:147e54. https://doi.org/
10.2106/JBJS.17.00434.
[19] Shahi A, Tan TL, Kheir MM, Tan DD, Parvizi J. Diagnosing periprosthetic joint
infection: and the winner is? J Arthroplasty 2017;32:S232e5. https://doi.org/
10.1016/j.arth.2017.06.005.
[20] Sousa R, Serrano P, Gomes Dias J, Oliveira JC, Oliveira A. Improving the ac-
curacy of synovial fluid analysis in the diagnosis of prosthetic joint infection
with simple and inexpensive biomarkers: C-reactive protein and adenosine
deaminase. Bone Joint J 2017;99-B:351e7. https://doi.org/10.1302/0301-
620X.99B3.BJJ-2016-0684.R1.
[21] Tarabichi M, Fleischman AN, Shahi A, Tian S, Parvizi J. Interpretation of
leukocyte esterase for the detection of periprosthetic joint infection based on
serologic markers. J Arthroplasty 2017;32:S97e100.e1. https://doi.org/
10.1016/j.arth.2017.03.045.
[22] Lerman RI, Yitzhaki S. A note on the calculation and interpretation of the Gini
index. Econ Lett 1984;15:363e8. https://doi.org/10.1016/0165-1765(84)
90126-5.
[23] Parvizi J, Della Valle CJ. AAOS Clinical Practice Guideline: diagnosis and
treatment of periprosthetic joint infections of the hip and knee. J Am Acad
Orthop Surg 2010;18:771e2.
[24] Fox-Wasylyshyn SM, El-Masri MM. Handling missing data in self-report
measures. Res Nurs Health 2005;28:488e95. https://doi.org/10.1002/
nur.20100.
[25] Buuren S van, Groothuis-Oudshoorn CGM. Mice: multivariate imputation by
chained equations in R. J Stat Softw 2011;45. https://research.utwente.nl/en/
publications/mice-multivariate-imputation-by-chained-equations-in-r. [Accessed
9 January 2018].
[26] Liaw A, Wiener M. Classification and Regression by RandomForest, Vol 23; 2001.
[27] Parvizi J, Jacovides C, Antoci V, Ghanem E. Diagnosis of periprosthetic joint
infection: the utility of a simple yet unappreciated enzyme. J Bone Joint Surg
Am 2011;93:2242e8. https://doi.org/10.2106/JBJS.J.01413.
[28] Yi PH, Cross MB, Moric M, Levine BR, Sporer SM, Paprosky WG, et al.
Do serologic and synovial tests help diagnose infection in revision
hip arthroplasty with metal-on-metal bearings or corrosion?
Clin Orthop 2015;473:498e505. https://doi.org/10.1007/s11999-014-
3902-5.
 J. Parvizi et al. / The Journal of Arthroplasty 33 (2018) 1309e1314 1314.e1

Appendix

Appendix A
Data Collection.

Variable Comment

Baseline characteristics
Septic/aseptic revision Based on predefined criteria
Time from last surgery on the joint
Primary/revision (most recent surgery) Was the latest surgery primary or a revision
Gender
Age
Body mass index (kg/m2)
Joint Hip or knee
Race White, black, or others
Charlson Comorbidity Index Overall score
Rheumatoid arthritis
Malignancy
Diabetes
Major criteria
Sinus tract Sinus tract with evidence of communication to the joint or visualization of the prosthesis
2 positive cultures Same pathogen is isolated from at least 2 separate tissues or fluid samples obtained from the affected prosthetic joint
Cultures
Single positive culture
Type of organism and resistance
Serum markers
Serum ESR (mm/h)
Serum CRP (mg/dL)
Serum D-dimer (mg/mL)
Synovial markers
Synovial fluid CRP (mg/L)
Synovial fluid WBC count (cells/mL)
Synovial fluid PMN (%)
Synovial fluid LE Negative, trace, (þ), or (þþ)
Synovial fluid alpha-defensin
Intraoperative
Purulence As described in the operative note
Positive histopathology Greater than 5 neutrophils per high-power field in 5 high-power fields observed from histologic analysis of
periprosthetic tissue at 400 magnification.
Positive molecular findings

CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; LE, leukocyte esterase; PMN, polymorphonuclear; WBC, white blood cell.

Appendix B
Demographics of Patients Which Were Included in the Validation Model (n ¼ 422).

Variable Overall (n ¼ 422) PJI Cohort (n ¼ 222) Aseptic Cohort (n ¼ 200) P Value

Age 64.6 (10.6) 64.3 64.9 .606

Gender (male) 205 (48.6%) 116 (52.3%) 89 (44.5%) .143
Race (white) 364 (86.3%) 191 (86.0%) 173 (86.5%) .782
Joint (knee) 283 (67.1%) 152 (68.5%) 131 (65.5%) .535
Time from last surgery (yr) 5.6 (8.9) 4.2 (10.7) 7.1 (6.1) .001a
Most recent surgery a revision procedure 153 (36.3%) 95 (42.8%) 58 (29.0%) <.001a
Body mass index (kg/m2) 32.5 (7.8) 33.1 (9.1) 31.9 (6.0) .09
Charlson Comorbidity Index (mean) 1.8 (1.9) 2.0 1.7 .102
History of rheumatoid arthritis 31 (7.3%) 18 (8.1%) 13 (6.5%) .216
History of malignancy 17 (4.0%) 12 (5.4%) 5 (2.5%) .13
History of diabetes 100 (23.7%) 61 (27.5%) 39 (19.5%) .02a

Data are presented as mean (standard deviation) or number (%); kilogram (kg); meter (m); year (yr).
PJI, periprosthetic joint infection.
a
Statistically significant.
1314.e2 J. Parvizi et al. / The Journal of Arthroplasty 33 (2018) 1309e1314

Appendix C
Performance of the New Definition Stratified by Hip and Knee Joints.

Joint PJI Cohort (n ¼ 222) Aseptic Cohort (n ¼ 200) Sensitivity (95% CI) Specificity (95% CI)

True Positives False Negatives Inconclusive True Negative False Positives Inconclusive

Knees 148/155 (95.5%) 4/155 (2.6%) 3/155 (1.9%) 128/131 (97.7%) 1/131 (0.8%) 2/131 (1.5%) 97.4% (93.4-99.3) 99.2% (95.8-100)
Hips 64/67 (95.5%) 1/67 (1.5%) 2/67 (3.0%) 67/69 (97.1%) 0/69 (0%) 2/69 (2.9%) 98.5% (91.7-100) 100% (94.6-100)

CI, confidence interval; PJI, periprosthetic joint infection.

Appendix D
Performance of the New Definition for Each of the 3 Different Institutions Individually.

Institution PJI Cohort (n ¼ 222) Aseptic Cohort (n ¼ 200) Sensitivity (95% CI) Specificity (95% CI)

True Positives False Negatives Inconclusive True Negative False Positives Inconclusive

Center A 41/42 (97.6%) 0/42 (0%) 1/42 (2.4%) 45/46 (97.8%) 0/46 (0%) 1/46 (2.2%) 100% (91.4-100) 100% (92.1-100)
Center B 56/60 (93.3%) 2/60 (3.3%) 2/60 (3.3%) 57/59 (96.6%) 1/59 (1.7%) 1/59 (1.7%) 96.6% (88.1-99.6) 98.3% (90.8-100)
Center C 115/120 (95.8%) 3/120 (2.5%) 2/120 (1.7%) 93/95 (97.9%) 0/95 (0%) 2/95 (2.1%) 97.5% (92.8-99.5) 100% (96.1-100)

CI, confidence interval; PJI, periprosthetic joint infection.