D-dimer in the diagnosis of periprosthetic joint infection: a systematic review and meta-analysis

Background D-dimer, a coagulation-related indicator, has recently been used as a tool for the diagnosis of periprosthetic joint infection (PJI), but its reliability is uncertain. The purpose of this systematic review and meta-analysis was to explore the accuracy of D-dimer in the diagnosis of PJI after joint arthroplasty. Methods We systematically searched the MEDLINE, EMBASE, and Cochrane databases for relevant literature about D-dimer in the diagnosis of PJI. QUADAS-2 was used to assess the risk of bias and clinical applicability of each included study. We used the bivariate meta-analysis framework to pool the sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and area under the SROC curve (AUC). Univariate meta-regression and subgroup analyses were performed to explore the sources of heterogeneity. Results We included 8 eligible studies. The pooled diagnostic sensitivity and specificity were 0.82 (95% CI, 0.70–0.89) and 0.70 (95% CI, 0.55–0.82), respectively. The pooled PLR, NLR, and DOR were 2.7 (95% CI, 1.7–4.4), 0.26 (95% CI, 0.15–0.46), and 10 (95% CI, 4–25), respectively. The AUC was 0.83 (95% CI, 0.8–0.86). Serum D-dimer might have higher diagnostic accuracy than plasma D-dimer for PJI (pooled sensitivity: 0.88 vs 0.67; pooled specificity: 0.76 vs 0.61). Conclusions D-dimer has limited performance for the diagnosis of PJI.


Introduction
Periprosthetic joint infection (PJI) is a rare and devastating complication that affects 0.7-2.4% of patients after hip or knee arthroplasty [1][2][3]. PJI not only affects the quality of life of infected patients but also increases the risk of death [4].
Because the typical clinical manifestations of patients with PJI may not appear and pain can be caused by other diseases, PJI is difficult to diagnose. The Musculoskeletal Infection Society (MSIS) formulated diagnostic criteria for PJI and tried to reduce the incidence rate of this dreaded complication [5,6]. In 2018, the International Consensus Meeting (ICM) modified the criteria and added D-dimer and alpha-defensin into the new definition of PJI for the knee and hip joint [7] (Table 1).
D-dimer is a specific degradation product of fibrin monomer that is crosslinked by activating factor XIII and then hydrolyzed by fibrinolytic enzyme [8]. It is a specific marker of the fibrinolysis process and mainly reflects the function of fibrinolysis [8]. A study suggested that D-dimer could be used to determine prognosis in systemic sepsis [9]. D-dimer levels continue to rise due to the host's inflammatory response to infection in sepsis [9].
Currently, some studies have examined the diagnostic value of D-dimer for PJI, but diagnostic accuracy varies in different studies. Therefore, the purpose of this systematic review and meta-analysis was to evaluate the diagnostic accuracy of D-dimer for PJI.

Materials and methods
This systematic review and meta-analysis strictly followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [10] ( Fig. 1).  [7] Major criteria

Search strategy
We systematically searched all literature about D-dimer in the diagnosis of PJI in MEDLINE, Embase, and the Cochrane Library (from the inception of each database until November 2019), without language restrictions. The search strategies are shown in Table 2.

Eligibility criteria
We included all studies that reported the accuracy of Ddimer in the diagnosis of PJI after hip or knee arthroplasty and used the MISS or modified MISS criteria. Studies lacking sensitivity and specificity values and those that had duplicated data were excluded.
Two authors independently scanned the titles, abstracts, and full texts sequentially and screened the literature based on the eligibility criteria. The third author settled any disagreements that arose.

Data extraction
Two authors independently classified all studies and extracted data using standardized scales. We extracted all baseline data (author name, publication year, country, average age, sex distribution, BMI, joint type, patient exclusion criteria, diagnostic criteria, etc.) and outcome indicators (sensitivity, specificity, PLR, NLR, DOR, AUC, etc.). The third author resolved any disagreements that arose.

Quality evaluation
The quality of each included study was evaluated using the QUADAS-2 tool [11], which mainly includes four parts: patient selection, indicator testing, reference standard, and flow and timing. The first three parts are also needed to evaluate clinical practicability. According to the answers ("yes," "no," or "uncertain") to the relevant landmark questions included in each part, the risk of bias level was determined as "low," "high," or "uncertain." Two authors independently evaluated the quality, and the third author decided the final result in the event of any divergences.

Statistical analysis
We used the bivariate meta-analysis framework to pool the sensitivity, specificity, PLR, NLR, DOR, and AUC by using the "Midas" command [12]. Compared with the traditional summary ROC curve, the bivariate model is a development and expansion [13]. The joint modeling of sensitivity and specificity is used as the starting point for the analysis, and a random effects model is used [13]. Thus, the diagnostic accuracy may be more reliable with this method [14]. The I 2 statistic was used to estimate the heterogeneity among studies. The value of I 2 is between 0 and 100%. An I 2 value of < 50% indicates low heterogeneity, while an I 2 value of > 50% indicates high heterogeneity.
When there was high heterogeneity, we evaluated the threshold effect through the Spearman correlation coefficient of the logarithm of sensitivity and 1-specificity. When the P value was < 0.05, the threshold effect was considered significant. At the same time, we used univariate meta-regression to find the potential sources of heterogeneity. Then, we conducted a subgroup analysis to further investigate the source of heterogeneity. A test for publication bias (Deeks' funnel plot) was also used to analyze the sources of heterogeneity. When the P value was < 0.05, the tests for publication bias were considered statistically significant [15]. Stata 14.0 software and Meta-DiSc 1.4 were used for data analysis.

Quality assessment
According to the QUADAS-2 tool, we evaluated the quality of all included studies (Table 4 and Fig. 2). The risk of bias in reference standards and flow and timing was low in all studies. Six studies [16][17][18][20][21][22] were at high risk of bias for patient selection because of inappropriate discharge standards and case-control trials. Because retrospective studies and thresholds were not set in advance in 7 studies [16][17][18][19][20][21]23], the bias of the index test was high. All studies scored between 6 and 9 (the total score is 10 points).   (Fig. 3). The AUC was 0.83 (95% CI, 0.8-0.86) (Fig. 4). The Spearman correlation coefficient was − 0.071 (P = 0.867). The heterogeneity might be unrelated to the threshold effects.

Meta-regression
We performed univariate meta-regression to search for the potential sources of heterogeneity (Fig. 5). For sensitivity and specificity, the sample differences and racial differences had the most significant impacts on the heterogeneity of the results (P < 0.05). Based on these results, we performed subgroup analysis to further explore the source of heterogeneity. When I 2 < 50% or P > 0.05, we considered the heterogeneity to be low in the subgroup.

Publication bias
The Deeks' funnel plot asymmetry test of DOR did not show significant asymmetry (P = 0.34), indicating that publication bias might not exist (Fig. 6).

Discussion
The diagnosis of PJI after arthroplasty is a complicated problem for every orthopedist. With early diagnosis, patients can undergo debridement or conservative treatment to treat PJI and avoid 1 or 2 stage revision. Therefore, the quick and accurate diagnosis of PJI is critical. Many potential blood and synovial fluid biomarkers for the diagnosis of PJI have been evaluated, but the clinical gold standard for the diagnosis of the disease has still not been found. Therefore, it is necessary and meaningful to develop a new and accurate diagnostic method for PJI. D-dimer is familiar to medical workers and has not been valued in the past few decades. It has only been used to screen venous thromboembolism [24,25]. Recently, some studies showed that D-dimer was associated with inflammation and might be elevated in infected patients [26,27]. Rodelo et al. found that higher levels of D-dimer were associated with increased 28-day mortality in septic patients [9]. In addition, D-dimer is recommended as a critical diagnostic indicator for infectious diseases such as endocarditis and mycoplasma pneumonia [28,29]. Subsequently, D-dimer levels attracted the attention of plastic surgeons. Shahi et al. [20] reported in his study that serum Ddimer has high diagnostic value for PJI in lower limbs, with a sensitivity and specificity of 89% and 93%, respectively, which preluded the diagnosis of PJI by Ddimer. Parvizi et al. [30] believe that the diagnosis of PJI, such as ankylosing spondylitis, rheumatoid arthritis, and endocarditis, should depend on a combination of various diagnoses, so they added D-dimer and redefined the diagnosis of the PJI standard. The new diagnostic criteria were validated in 222 PJI patients and 200 sterile patients. They found that the sensitivity and specificity of the new diagnostic criteria were 97.7% and 99.5%, respectively, while the sensitivities of the MSIS and ICM diagnostic criteria were only 86.9% and 79.3%, and their specificities were both 99.5%. The ICM passed this diagnostic criterion in 2018, but the pass rate was only 68%. Since 2019, an increasing number of articles about Ddimer in the diagnosis of PJI have been reported, and its diagnostic value is suspected. This is the first systematic review and meta-analysis about the utility of D-dimer for the diagnosis of PJI. We found that D-dimer has limited performance for the diagnosis of PJI, with a pooled sensitivity and specificity of 0.82 and 0.70, respectively, and had a poorer diagnostic value than that of CPR and ESR reported by Carli AV et al. [31]. In this systematic review, the pooled sensitivity and specificity of CRP were 0.85 and 0.81, respectively, and the pooled sensitivity and specificity of ESR were 0.82 and 0.79. The results of the subgroup analysis showed that serum D-dimer might have a higher diagnostic accuracy than plasma D-dimer for PJI (the pooled sensitivity was 0.88 vs 0.67, and the pooled specificity was 0.76 vs 0.61), and D-dimer had better accuracy in subgroups with Caucasian and African American races than in subgroups with East Asian races (the pooled sensitivity was 0.92 vs 0.72, and the pooled specificity was 0.74 vs 0.65).
One possible reason for the variance in the subgroup results was that the samples for the quantification of Ddimer were different: serum vs plasma. Serum is the liquid part of blood after coagulation, while plasma is the liquid part of the blood where coagulation has been prevented. Their density is similar, but their composition is different. The main difference is that there are more fibrinogen and coagulation proteins in plasma [32].
Boisclair et al. [33] reported that there was a very high correlation between plasma and serum D-dimer levels (r = 0.931, P < 0.01), but the diagnostic sensitivity was not consistent. The study reported that the sensitivities of plasma D-dimer for DIC, DVT, and MI  were 100%, 90.4%, and 60%, respectively, but the sensitivities of serum D-dimer were 100%, 94.1%, and 22.2%. The different sensitivities of plasma and serum might be due to the more significant uncertainty in assigning a cut-off for elevated levels of serum Ddimer. The D-dimer assay was operating at its lower detection limit when used to measure non-elevated levels in serum [33]. However, whether different sensitivities between plasma and serum exist in PJI is not supported by relevant literature. Another possible reason was that the level of D-dimer is easily affected by other diseases. Busso et al. [34] reported that the inflammatory synovium secretes a large amount of fibrin in patients with rheumatoid arthritis, and the degradation of this protein subsequently leads to an increase in the level of D-dimer in serum and synovial fluid. In addition, thrombosis [35], malignancies, autoimmune diseases, pregnancy, and heart and brain vascular diseases might affect the determination of Ddimer levels in the blood [36,37]. Li et al. [19] found that the diagnostic accuracy of D-dimer was poor in the subgroups containing these diseases in their study.
In addition, racial differences may affect the diagnostic accuracy of D-dimer for PJI. Shahi and Pannu's studies [20,23] were conducted in the USA, and the population may be predominantly Caucasian and African-American. In the six other studies reported by Chinese scholars, the patients were predominantly of the East Asian race. The studies found that D-dimer levels varied between races, such as between African American and Caucasian patients [38,39]. We suspect that there are also differences in D-dimer levels between the East Asian population and the other races, which will affect the result. However, there are no studies to support this view.
Synovial fluid viscosity tests and several other plasma biomarkers have been reported to diagnose PJI. The synovial fluid viscosity level was significantly lower in patients with PJI than in patients with aseptic failure, with a sensitivity of 0.99 and a specificity of 0.67 [22]. Both plasma fibrinogen and fibrin degradation product (FDP) are coagulation-related indicators. When the threshold for plasma fibrinogen was 4.01 g/L, the sensitivity and specificity values were 0.763 and 0.862 [19], respectively. FDP has low sensitivity and specificity, with values of 65.12% and 60.33%, respectively [18].
Our meta-analysis has some strengths and potential limitations. The cases included all involved hip and knee joints. In addition, all studies used MSIS standards [5] or modified MSIS standards [6]. Therefore, the classification bias was minimized. The most important factor was that all D-dimer tests were taken before surgery, excluding the interference of a sharp increase in serum Ddimer levels after surgery [40].
The limitations of our meta-analysis included variability in race, age range, sex ratio, and sample size. In addition, none of the studies considered whether patients used antibiotics before admission. Shahi et al. [20] reported that premature antibiotic treatment could affect the results of D-dimer in the blood. Another limitation of our study is that MSIS standards or modified MSIS Fig. 6 The Deeks' funnel plot of the pooled DOR standards lack the sensitivity to detect chronic and lowgrade PJI; patients with "positive" D-dimer results might be classified as uninfected [41]. Additionally, most studies did not provide information about the measurement of D-dimer. D-dimer assays can be categorized into three types [42]: ELISA, immunoturbidimetric automated assay, and latex-based immunoassays. ELISA is more sensitive than immunoturbidimetric automated assays and latex-based immunoassays [42]. In addition, some studies excluded patients with tumors, rheumatoid arthritis, autoimmune diseases, a history of smoking, and obesity. However, the proportion of such patients in joint replacement is still high. The exclusion of these patients will interfere with the accuracy of D-dimer in the diagnosis of PJI. Finally, the diagnostic thresholds in some studies were not determined in advance, and the threshold values were not unified in this meta-analysis.

Conclusion
D-dimer, a coagulation-related indicator, is inexpensive and easy to measure but has limited performance for the diagnosis of PJI, and the pooled sensitivity and specificity were poorer than those of traditional inflammatory markers such as CRP and ESR. Based on our findings, we suggest using serum samples for the quantification of D-dimer. Additionally, the diagnostic accuracy may be better in Caucasian and African American patients.