International Journal of Infectious Diseases
Volume 13, Issue 6 , Pages e444-e449, November 2009

The use of receiver operating characteristics analysis in determining erythrocyte sedimentation rate and C-reactive protein levels in diagnosing periprosthetic infection prior to revision total hip arthroplasty

The Rothman Institute of Orthopedics, Thomas Jefferson University, 925 Chestnut Street, Philadelphia, PA 19107, USA

Received 25 June 2008; received in revised form 8 February 2009; accepted 27 February 2009. published online 28 May 2009.

Corresponding Editor: William Cameron, Ottawa, Canada

Article Outline

Summary 

Background

Periprosthetic infection (PPI) is a difficult complication in total joint arthroplasty, and while erythrocyte sedimentation rate (ESR) and C-reactive protein (CRP) are acute phase reactants thought to be of high predictive value for diagnosing infection, no clear cut-off values have been defined. The current study aimed to determine the cut-off values for ESR and CRP that improve clinical differentiation between aseptic failure and PPI in total hip arthroplasty (THA).

Methods

Four hundred and seventy-nine patients who underwent revision THA for either aseptic mechanical failure or PPI during the period of 2000 to 2005 were included in the study. Specific exclusion criteria were applied to eliminate inflammatory or other confounding conditions. All patients underwent preoperative testing of ESR and CRP. Receiver operating characteristic (ROC) curves were constructed to determine maximum sensitivity and specificity.

Results

Patients with PPI had significantly higher ESR and CRP values compared to patients undergoing revision for aseptic etiologies. An ESR threshold of 30mm/h gave a sensitivity of 94.3% and a CRP threshold of 10mg/l gave a sensitivity of 91.1%. Combining both ESR and CRP cut-offs for a positive diagnosis increased the sensitivity to 97.6%. However, when calculated by ROC analysis, the predictive cut-offs equated to 31mm/h for ESR and 20.5mg/l for CRP.

Conclusions

The gold standard for diagnosing PPI remains bacterial culture, but sensitivity is negatively affected by prior antibiotic exposure, strongly adherent bacteria, slow growing persisters, and biofilms. ESR and CRP are reflective of systemic changes in infection and pose an attractive, less invasive alternative with reasonable sensitivity and specificity. The current study is the first to identify ideal cut-off values for ESR and CRP in THA patients, providing an optimum balance between sensitivity and specificity based on ROC curves.

Keywords: Periprosthetic infection (PPI), Erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), ROC analysis, Total hip arthroplasty complications, Infection diagnosis

 

Back to Article Outline

Introduction 

Joint replacement has been dubbed as a great success story, with increasingly more patients taking advantage of the procedure.1 The majority of patients are highly satisfied, yet a small minority experience complications.2, 3 Infection, diagnosed in 1–5% of patients, is the most disconcerting problem due to the associated high morbidity and possible mortality.4 The presentation is often chronic or late, difficult to assess and properly diagnose, and the treatment is often limited by the irresolute diagnosis.

Accurate diagnosis is the first step in the successful treatment of periprosthetic infection (PPI). Differentiation of PPI from aseptic loosening or osteolysis is often difficult and requires interpretation of assorted laboratory and imaging data.5, 6, 7 The gold standard is the isolation of an infectious organism from periprosthetic tissue or biofilm colonizing the implant.8 Such a diagnosis, however, requires surgical debridement or even implant retrieval and exchange, leading to further disability, subsequent morbidity, and increased risk for reinfection. Therefore, an accurate preoperative diagnosis can minimize further complications and drastically improve patient results and satisfaction.9, 10

Currently, the diagnosis of PPI is based on a complex algorithm that incorporates all available information, including clinical history, radiological parameters, serological values, microbiology, and pathology. Erythrocyte sedimentation rates (ESR) and C-reactive protein (CRP) values are popular acute phase reactants used in these diagnostic algorithms, especially given their reasonably high sensitivity and acceptable specificity for infection.6, 8 However, the literature shows much diagnostic variability with different threshold values used in each paper based on institutional experience and individual success.11, 12 Such variability further depends on patient selection, confounding factors like inflammation and autoimmune conditions, study biases, institutional volume and experience, and the clinical laboratory standardization. No standardized values for ESR and CRP have been proposed for hip arthroplasty.

The goal of our study was to determine a set of cut-off values for ESR and CRP that would provide the most optimal sensitivity and specificity for detecting PPI. Clearly defined thresholds would improve the surgeon's capacity to differentiate aseptic mechanical failure from PPI, and therefore improve management. To accomplish this goal, we reviewed the ESR and CRP numbers in a large cohort of infected and non-infected patients who underwent total hip arthroplasty (THA) revision at our high volume center. The data were subsequently analyzed statistically using receiver operating characteristics (ROC) curves. The values obtained in our analysis could then be used with currently acceptable diagnostic algorithms and potentially improve patient care and overall surgical success.

Back to Article Outline

Materials and methods 

Study design 

This was a cohort study with retrospectively collected data examining the ESR and CRP diagnostic tests based on previously developed diagnostic criteria for periprosthetic infection in THA. Institutional review board approval was obtained prior to initiation of our comparative study, and all patients gave their consent to participate. We performed a review of the ongoing total joint arthroplasty database at our institution. We identified 499 patients who underwent revision THA for either aseptic mechanical failure or PPI during the period of 2000 to 2005. A detailed review of medical records was performed to extract demographic information and medical history. All patients in the study underwent preoperative testing using ESR and CRP by the hospital clinical laboratory according to their accepted standards and controls. The CRP level was measured using turbidimetric techniques (Beckman Coulter, Brea, CA, USA), while the ESR was measured using an automated analyzer (Mini-Ves, Plymouth, MN, USA). The time interval from performing phlebotomy to measuring ESR levels was within one to two hours in the vast majority of cases. Patients with comorbid conditions known to elevate ESR and CRP, including inflammatory disease, chronic renal failure, hepatitis, and active malignancy or infection in other regions of the body were not considered for the study due to the confounding effects.13, 14, 15 Two or more intra-operative specimens were obtained and sent for aerobic and anaerobic cultures during the revision procedures. Preoperative aspiration of the hip joint with cell counts and cell differential were also performed when the acute phase reactants or clinical suspicion suggested infection.

Infection criteria 

Patients were diagnosed with PPI if they fulfilled one of the following criteria: (1) an abscess or sinus tract was found to be communicating with the joint space, (2) positive preoperative aspiration culture on solid media, (3) two or more positive intra-operative cultures or one positive culture on solid media in conjunction with the presence of other indicators of infection including gross intracapsular purulence or an elevated cell count and differential of the aspirate fluid.

Study population 

Although our initial cohort consisted of 499 total hip revisions, 20 patients who underwent treatment procedures for PPI were excluded from the infected subgroup because they failed to fulfill the infection criteria listed above. The cultures of 20 patients who underwent revision surgery for aseptic mechanical failure were classified as false positives. Therefore, our final study population consisted of 479 patients (53% female, 47% male) with an average age of 66 years (range 23–93 years) at time of revision surgery. Infected patients constituted 26.5% of the population. In the infected patient group, 28% underwent irrigation and debridement with retention of components, 48% required two-resection arthroplasty to control the underlying infection, while 13% were treated with one-stage reimplantation. The remaining 11% of the infected cohort consisted of patients who were diagnosed with PPI after undergoing revision surgery for mechanical failure.

Statistical methods 

The mean and 95% confidence interval (95% CI) of the serological parameters including ESR and CRP of the infected and non-infected groups were determined, and the difference between means was compared using an independent two-sample Student's t-test. A p-value of <0.05 (two-sided) was considered to be statistically significant. All statistical analyses were performed using SPSS version 13 (SPSS Inc., Chicago, IL, USA) and SAS statistical software (SAS Institute Inc., Cary, NC, USA).

ROC curves, which depict the relationship between true-positive (sensitivity) and false-positive (1 – specificity) cases, were constructed for ESR (Figure 1A) and CRP (Figure 1B). The area under the curve (AUC), which depicts the accuracy of the test, was calculated for each of the above variables. An AUC of 1 demonstrates an ideal test with a 100% sensitivity and specificity, while an AUC of less than 0.5 indicates that the diagnostic test is less useful. The ROC curve constructed correlates the true positive and false positive rates for a series of data points. We used a parametric method based on bivariate normal distribution that implements a maximum likelihood estimator to fit a smooth curve to the data points. This estimate assumes one normal distribution for cases with the disease and one normal distribution for cases without, or that the data have been monotonically transformed to normal.16, 17, 18, 19 This binormal form has been found empirically to provide satisfactory ROC fits to data generated in a very broad variety of situations. The diagnostic cut-off values of ESR and CRP were chosen as those values that corresponded to the points on the ROC curves nearest the upper left hand corner of the graph for optimal balance between sensitivity and specificity in diagnosing PPI. We performed a paired t-test analysis (significance accepted at p<0.05) of the AUC for the ESR and CRP to determine which diagnostic test is best suited for diagnosing PPI.

  • View full-size image.
  • Figure 1. 

    Receiver operating-characteristic curves for (a) erythrocyte sedimentation rate (ESR) and (b) C-reactive protein (CRP). The area under the curve for ESR was 0.943 (95% CI 0.917–0.968) and for CRP was 0.941 (95% CI 0.912–0.970).

The ESR and CRP cut-off values of 30mm/h and 10mg/l, respectively, are frequently used in the literature to distinguish between infected and non-infected total joint arthroplasty. The sensitivity, specificity, predictive values, and likelihood ratios were calculated for the arbitrarily chosen and ROC-determined cut-off values noted above. A positive likelihood ratio represents the odds that a positive result will be observed in a patient with an infection compared with the odds that the same result will be observed in a patient without an infection. On the other hand, a negative likelihood ratio represents the odds that a negative result will be observed in an infected patient compared with the odds that the same result will be observed in a non-infected patient.

As a separate evaluation, multiple combinations were created and their diagnostic value was examined. One strategy entailed selecting for either ESR or CRP to define infection, wherein if either laboratory value reached the ROC cut-off value the patient would be classified as infected, therefore improving the sensitivity of tests and their ability to detect infection.20 On the other hand, a second strategy was employed that required both tests to be positive to reach a diagnosis of PPI, with the understanding that the cutoff values were obtained independently. In this strategy, both tests were performed even if one of the tests was negative. Intuitively this improves the specificity of the combination and allows the surgeon to better confirm PPI.20 The estimated sensitivity, specificity, predictive values, and likelihood ratios were calculated for each combination based on the original clinical information, and the 95% CI were reported.

Back to Article Outline

Results 

Infectious criteria and results were evaluated based on acceptable literature standards.12, 21 As stated, most patients had preoperative joint aspiration along with ESR and CRP measurements. The most common organisms retrieved from the preoperative aspirate fluid and intra-operative cultures were: methicillin-resistant Staphylococcus aureus (MRSA; 24.4%), methicillin-sensitive Staphylococcus aureus (MSSA; 21.3%), methicillin-resistant Staphylococcus epidermidis (MRSE; 11.8%), methicillin-sensitive Staphylococcus epidermidis (MSSE; 9.4%), Streptococcus species (8.7%), Enterococcus faecalis (6.3%), Pseudomonas aeruginosa (3.1%), multiple organisms (3.1%), Proteus mirabilis (2.4%), Escherichia coli (1.6%), Serratia marcescens (1.6%), Corynebacterium striatum (1.6%), vancomycin-resistant Enterococcus faecium (VRE) (1.6%), or no growth (3.1%). Loosening of the acetabulum or femur, instability, polywear, malpositioned components, periprosthetic fracture, and component fracture were the reasons for performing aseptic revision surgery. The most common primary diagnosis for index arthroplasty was degenerative osteoarthritis for both infected and non-infected patients.

The mean ESR measurement in patients with PPI was significantly higher than in patients who underwent revision for aseptic etiologies (77mm/h vs. 29mm/h; p=0.0001). Similarly, the mean CRP was significantly higher among infected patients than non-infected patients (14.9mg/l vs. 9.48mg/l; p<0.0001). When using a threshold of 30mm/h for ESR, the sensitivity was 94.3% (95% CI 89–98%) (Table 1), while a CRP threshold of 10mg/l yielded a sensitivity of 91.1% (95% CI 85–95%). However, the specificity, positive predictive value, and positive likelihood ratios of each test were of low clinical value as compared to their sensitivity, negative predictive value, and negative likelihood ratios (Table 1). When the two tests were combined using the arbitrary cut-offs listed above in such a way that both were required to be positive to determine PPI, the specificity and negative likelihood ratio improved at the expense of the sensitivity and positive likelihood ratio. Nonetheless, there remained a 25% chance that patients with both an elevated ESR and CRP were non-infected and a false positive diagnosis would be committed. Combining the two laboratory tests, with the patient being considered infected if at least one of the two tests was positive, increased the sensitivity to 97.6% (95% CI 93–99%) and negative predictive value to 98.4% (95% CI 95–99%) and decreased the odds of PPI being present with a negative test result (negative likelihood ratio) when compared to each individual test.

Table 1. Diagnostic test characteristics for ESR and CRP using arbitrarily chosen cut-off values present in the literaturea
TestSensitivity %Specificity %PPV %NPV %LR(+)LR(−)
ESR ≥30 mm/h94.3 (89–98)70.2 (65–75)55.5 (48–62)96.9 (94–99)3.4 (0.7–7.9)0.09 (0.03–0.27)
CRP ≥10 mg/l91.1 (85–95)76.6 (72–81)60.5 (53–68)95.6 (92–98)4.1 (2.3–7.2)0.13 (0.05–0.38)
ESR or CRP97.6 (93–99)58.7 (53–64)48.2 (42–54)98.4 (95–99)2.5 (1.5–4.4)0.04 (0.01–0.16)
ESR and CRP87.8 (81–93)88.1 (84–92)74.5 (66–81)94.8 (92–97)7.5 (2.4–18.3)0.20 (0.09–0.38)

ESR, erythrocyte sedimentation rate; CRP, C-reactive protein; PPV, positive predictive value; NPV, negative predictive value; LR(+), positive likelihood ratio; LR(−), negative likelihood ratio.

aA periprosthetic infection prevalence of 26.5% was used to calculate the prevalence values.

ROC curves were constructed for both ESR and CRP in order to define a cut-off value with an optimal balance between sensitivity and specificity to predict infection in THA (Figure 1). The areas under the curve (AUC) for the ESR and CRP were 0.943 (95% CI 0.917–0.968) and 0.941 (95% CI 0.912–0.970), respectively; there was no significant difference in AUC between the two tests (p=0.98). The calculated thresholds for ESR and CRP were 31mm/h and 20.5mg/l, respectively. An ESR cut-off of 31mm/h and CRP cut-off of 20.5mg/l yielded a negative predictive value of 94.3% (95% CI 91–98%) and 94.2% (95% CI 91–98%), respectively, in ruling out PPI (Table 2). The use of such cut-off values also improved the positive predictive value of both ESR (71.5%; 95% CI 64–79%) and CRP (80.5%; 95% CI 72–89%). Since both the positive and negative likelihood ratios for ESR and CRP improved using the ROC thresholds as compared to the arbitrary thresholds frequently implemented in the literature, the odds of committing a false positive or false negative error in diagnosis decreased. As expected,22 when using the criteria of either an ESR >31mm/h or CRP >20.5mg/l to detect infection, the sensitivity and negative likelihood ratio improved as compared to the results obtained for each individual test when used separately (Table 2). The combined tests have a very high success at ruling out infection, leaving only 2.7% of cases misdiagnosed as false negatives.

Table 2. Diagnostic test characteristics for ESR and CRP using cut-off values determined from ROC curvesa
TestSensitivity %Specificity %PPV %NPV %LR(+)LR(−)
ESR ≥31mm/h94.5 (90–98)72.2 (65–79)71.5 (64–79)94.3 (91–98)3.5 (0.7–8.0)0.07 (0.02–0.28)
CRP ≥20.5mg/l94.3 (89–98)81.0 (72–89)80.5 (72–89)94.2 (91–98)4.9 (2.6–8.7)0.07 (0.01–0.26)
ESR or CRP96.1 (91–99)59.2 (54–65)50.0 (44–56)97.3 (94–99)2.4 (1.4–4.3)0.06 (0.02–0.30)
ESR and CRP74.8 (66–82)89.0 (85–92)74.2 (66–82)89.3 (85–93)6.8 (1.9–16.2)0.28 (0.15–0.38)

ESR, erythrocyte sedimentation rate; CRP, C-reactive protein; ROC, receiver operating characteristics; PPV, positive predictive value; NPV, negative predictive value; LR(+), positive likelihood ratio; LR(−), negative likelihood ratio.

aA periprosthetic infection prevalence of 26.5% was used to calculate the prevalence values.

Back to Article Outline

Discussion 

Periprosthetic infection after THA is a difficult complication to manage and treat.5 Although early detection may improve outcomes, there is no single diagnostic tool that can diagnose PPI with 100% accuracy.4 The gold standard for diagnosing infection remains culture of bacteria from the implant surface during revision surgery.8 Such a method, however, decides post-factum, after the revision surgery, and therefore serves only as a confirmatory measure. ESR, CRP, and joint aspiration are commonly used preoperative tests for early detection of infection and discrimination of aseptic failure.6, 21 While aspiration has its advantages, biofilm-based infections create small pockets that are strongly adherent to the biomaterial, therefore limiting their distribution in joint fluid.23, 24 On the other hand, ESR and CRP are reflective of systemic changes and they are available for quantification from serum. These acute phase reactants pose an attractive, less invasive alternative to aspiration and other surgical intervention.

When used appropriately, ESR and CRP can be highly effective at predicting periprosthetic infection. No universal normal ranges exist for either ESR or CRP, as both measurements vary with gender, age, and multiple clinical conditions. However, current findings suggest that both ESR and CRP have reasonable sensitivity and acceptable specificity in a selective group of patients, and they provide adequate predictive values. Furthermore, these results agree with previous values reported in the literature. Feldman et al. showed that ESR values over 50mm/h lead to a sensitivity of at least 79% and a specificity of 78%.25 In another patient population, an ESR cut-off value of 30mm/h in a prospective study by Spangehl et al.12 produced 82% sensitivity and 85% specificity. Similarly, the CRP was an even better predictor of infection with CRP levels greater than 1mg/dl being 96% sensitive and 92% specific for PPI.12 The classification scheme of infection in the above investigation was not independent of the variables under study, including the ESR and CRP, which may have subsequently biased the results and conclusions.

A review of the literature confirms that this is the first study of its kind that has attempted to statistically define ideal cut-off values for ESR and CRP in THA patients. Specifically, we suggest that ROC analysis is a better tool for clinical decision-making. Using the ROC, our current THA data agree in principle with knee replacement results published by Greidanus et al.26 However, their conclusions suggest ESR and CRP cut-off values of 22.5mm/h and 1.35mg/dl, respectively. Such values are significantly lower than those obtained in the current study or previously reported in the literature. This discrepancy may be related to the joint operated on, with critical differences between the typical approach to periprosthetic infection in the knee and hip. Different laboratories may produce slightly different values that may require local standardization at the hospital level. Furthermore, the results of ROC curves have to be interpreted in view of different sample sizes and distributions;20 our sample population was nearly 2.5-fold the number of total knee arthroplasty cases revised in the cohort above.26

Nonetheless, our study has some inherent limitations that require consideration. This work was done in a high volume center with a high referral base with infection rates that differ from the general orthopedic population. The infection rate may be significantly different in our hands compared to the rate experienced in a primary treatment center. In view of that, ROC analysis does not account for such prevalence and only tests the hypothesis for this patient population. Likelihood ratios and predictive values may be more appropriate, however most clinicians feel more comfortable with specificity and sensitivity as measurements of the test quality. Other statistical analyses, including Bayesian inference, may be superior to ROC under these settings. Furthermore, the retrospective nature of the study introduces additional biases and confounding factors, even though the analysis would be difficult to perform prospectively without affecting patient care.9 Furthermore, periprosthetic infection can often develop even years after surgery, producing false negatives, but not relevant for immediate prognosis.27 Finally, the diagnostic gold standard remains debatable and elusive. The selected criteria for infection based on microbiology may not always diagnose an infection due to limited sensitivity. Several reports have pointed to the missed infections in orthopedic surgery; using PCR technology, pathogenic organisms were found in as many as 72% of revision cases.28, 29 Strict patient selection is required as both serum markers change drastically with other inflammatory conditions, including rheumatoid arthritis, psoriasis, and gout. Despite the diagnostic value of ESR and CRP, there is no one ideal test for patients with THA, and the combination of other comorbidities or confounding factors makes the clinical decision an arduous task.

These limitations, however, do not detract from the value of this proposed work that attempts to standardize diagnostic measures in clinical practice. Many physicians implement arbitrarily chosen ESR and CRP cut-off values in an attempt to maximize either the sensitivity or the specificity depending on individual subjective risk assessment.11, 12, 25 Of note, an increase in either sensitivity or specificity leads to detrimental changes in the other reactant. As such, ROC analysis affords us the privilege of obtaining cut-off values with an optimal balance for both sensitivity and specificity. Hence, the proposed thresholds of 31mm/h for ESR and 20.5mg/l for CRP, when standardized for the individual laboratory, may allow improved diagnosis and better management decisions without putting the patient at risk for unnecessary procedures, antibiotic treatments, or even life-long disability. One caveat must be kept in mind when implementing the above thresholds in clinical practice. The definition of optimal relates only to sensitivity and specificity but does not take into consideration the costs of misdiagnosing consequent to the false positive and negative rates. We therefore recommend that further research be undertaken to investigate the role of cost-effectiveness of each diagnostic tool in order to portray a more holistic picture during the diagnostic work-up.

We have previously suggested a diagnostic algorithm that provides a roadmap for combining laboratory examination, imaging, microbiology, and pathology in order to amplify the utility of each test.4, 8 Those algorithms, however, are meaningless if the appropriate questions at each decision fork are not answered correctly. To maximize diagnosis and subsequent management, specific values or thresholds should be determined for each test. We suggest that statistical methods like ROC analysis can assist in defining cut-off values for ESR, CRP, or other critical markers involved in the decision process. A clinical approach derived in evidence-based medicine will subsequently lead to improved practice guidelines, better management strategies, and will ultimately benefit the patient.

Conflict of interest: No conflict of interest to declare.

Back to Article Outline

References 

  1. Fujita K, Makimoto K, Higo T, Shigematsu M, Hotokebuchi T. Changes in the WOMAC. EuroQol and Japanese lifestyle measurements among patients undergoing total hip arthroplasty. Osteoarthritis Cartilage. 2008;
  2. Jandric S, Manojlovic S. Quality of life of men and women with osteoarthritis of the hip and arthroplasty: assessment by WOMAC questionnaire. Am J Phys Med Rehabil. 2009;88:328–335
  3. Pulido L, Parvizi J, Macgibeny M, Sharkey PF, Purtill JJ, Rothman RH, et al. In hospital complications after total joint arthroplasty. J Arthroplasty. 2008;23(6 Suppl 1):139–145
  4. Parvizi J, Ghanem E, Sharkey P, Aggarwal A, Burnett RS, Barrack RL. Diagnosis of infected total knee: findings of a multicenter database. Clin Orthop Relat Res. 2008;466:2628–2633
  5. Hanssen AD, Rand JA. Evaluation and treatment of infection at the site of a total hip or knee arthroplasty. Instr Course Lect. 1999;48:111–122
  6. Parvizi J, Ghanem E, Menashe S, Barrack RL, Bauer TW. Periprosthetic infection: what are the diagnostic challenges?. J Bone Joint Surg Am. 2006;88(Suppl 4):138–147
  7. Toms AD, Davidson D, Masri BA, Duncan CP. The management of peri-prosthetic infection in total joint arthroplasty. J Bone Joint Surg Br. 2006;88:149–155
  8. Bauer TW, Parvizi J, Kobayashi N, Krebs V. Diagnosis of periprosthetic infection. J Bone Joint Surg Am. 2006;88:869–882
  9. Peersman G, Laskin R, Davis J, Peterson M. Infection in total knee replacement: a retrospective review of 6489 total knee replacements. Clin Orthop Relat Res. 2001;(392):15–23
  10. Zhan C, Kaczmarek R, Loyo-Berrios N, Sangl J, Bright RA. Incidence and short-term outcomes of primary and revision hip replacement in the United States. J Bone Joint Surg Am. 2007;89:526–533
  11. Bare J, MacDonald SJ, Bourne RB. Preoperative evaluations in revision total knee arthroplasty. Clin Orthop Relat Res. 2006;446:40–44
  12. Spangehl MJ, Masri BA, O’Connell JX, Duncan CP. Prospective analysis of preoperative and intraoperative investigations for the diagnosis of infection at the sites of two hundred and two revision total hip arthroplasties. J Bone Joint Surg Am. 1999;81:672–683
  13. Olshaker JS, Jerrard DA. The erythrocyte sedimentation rate. J Emerg Med. 1997;15:869–874
  14. Pepys MB. C-reactive protein fifty years on. Lancet. 1981;1:653–657
  15. Zlonis M. The mystique of the erythrocyte sedimentation rate. A reappraisal of one of the oldest laboratory tests still in use. Clin Lab Med. 1993;13:787–800
  16. Hajian-Tilaki KO, Hanley JA, Joseph L, Collet JP. A comparison of parametric and nonparametric approaches to ROC analysis of quantitative diagnostic tests. Med Decis Making. 1997;17:94–102
  17. Hanley JA. The robustness of the ‘binormal’ assumptions used in fitting ROC curves. Med Decis Making. 1988;8:197–203
  18. Hanley JA. The use of the ‘binormal’ model for parametric ROC analysis of quantitative diagnostic tests. Stat Med. 1996 30;15:1575–1585
  19. Swets JA. Form of empirical ROCs in discrimination and diagnostic tasks: implications for theory and measurement of performance. Psychol Bull. 1986;99:181–198
  20. Griner PF, Mayewski RJ, Mushlin AI, Greenland P. Selection and interpretation of diagnostic tests and procedures. Principles and applications. Ann Intern Med. 1981;94(4 Pt 2):557–592
  21. Bernard L, Lubbeke A, Stern R, Bru JP, Feron JM, Peyramond D, et al. Value of preoperative investigations in diagnosing prosthetic joint infection: retrospective cohort study and literature review. Scand J Infect Dis. 2004;36:410–416
  22. Pepe MS, Janes H. Insights into latent class analysis of diagnostic test performance. Biostatistics. 2007;8:474–484
  23. Kobayashi N, Bauer TW, Tuohy MJ, Fujishiro T, Procop GW. Brief ultrasonication improves detection of biofilm-formative bacteria around a metal implant. Clin Orthop Relat Res. 2007;457:210–213
  24. Vinh DC, Embil JM. Device-related infections: a review. J Long Term Eff Med Implants. 2005;15:467–488
  25. Feldman DS, Lonner JH, Desai P, Zuckerman JD. The role of intraoperative frozen sections in revision total joint arthroplasty. J Bone Joint Surg Am. 1995;77:1807–1813
  26. Greidanus NV, Masri BA, Garbuz DS, Wilson SD, McAlinden MG, Xu M, et al. Use of erythrocyte sedimentation rate and C-reactive protein level to diagnose infection before revision total knee arthroplasty. A prospective evaluation. J Bone Joint Surg Am. 2007;89:1409–1416
  27. Tsukayama DT, Estrada R, Gustilo RB. Infection after total hip arthroplasty. A study of the treatment of one hundred and six infections. J Bone Joint Surg Am. 1996;78:512–523
  28. Kordelle J, Hossain H, Stahl U, Schleicher I, Haas H. Usefulness of 16S rDNA polymerase chain reaction (PCR) in the intraoperative detection of infection in revision of failed arthroplasties. Z Orthop Ihre Grenzgeb. 2004;142:571–576
  29. Tunney MM, Patrick S, Curran MD, Ramage G, Hanna D, Nixon JR, et al. Detection of prosthetic hip infection at revision arthroplasty by immunofluorescence microscopy and PCR amplification of the bacterial 16S rRNA gene. J Clin Microbiol. 1999;37:3281–3290

PII: S1201-9712(09)00131-3

doi:10.1016/j.ijid.2009.02.017

International Journal of Infectious Diseases
Volume 13, Issue 6 , Pages e444-e449, November 2009

Article Tools

* Image Usage & Resolution