J Antimicrob Chemother 2011; 66: 1590 1593 doi:10.

1093/jac/dkr182 Advance Access publication 11 May 2011

Serial measurement of the C-reactive protein is a poor predictor of treatment outcome in prosthetic joint infection
Philip Bejon 1,2*, Ivor Byren 1, Bridget L. Atkins 1,3, Matthew Scarborough 1, Andrew Woodhouse 1, Peter McLardy-Smith 1, Roger Gundle 1 and Anthony R. Berendt 1
Bone Infection Unit, Nufeld Orthopaedic Centre, Windmill Road, Oxford OX3 7LD, UK; 2Nufeld Department of Medicine, Oxford University, John Radcliffe Hospital, Headley Way, Oxford OX3 9DU, UK; 3Department of Microbiology, John Radcliffe Hospital, Headley Way, Oxford OX3 9DU, UK
*Corresponding author. Bone Infection Unit, Nufeld Orthopaedic Centre, Windmill Road, Oxford OX3 7LD, UK. Tel: +44-1865-741841; Fax: +44-1865-738027; E-mail: pbejon@well.ox.ac.uk
1

Downloaded from http://jac.oxfordjournals.org/ by guest on February 29, 2012

Received 11 February 2011; returned 29 March 2011; revised 12 April 2011; accepted 13 April 2011 Objectives: Prosthetic joint infection is usually treated using surgery and antibiotics. The response to the treatment regimen is often evaluated using serial monitoring of plasma C-reactive protein (CRP) concentrations. In order to examine how useful this monitoring is, we calculated the sensitivity and specicity of CRP concentrations for predicting treatment failure. Patients and methods: We examined 3732 CRP measurements from 260 patients who were treated by either two-stage revision or debridement and retention. We tested the association between CRP concentration and outcome using logistic regression models, and assessed sensitivity and specicity by using receiver operator curves. Results: The areas under receiver operator curves for CRP concentrations predicting outcome ranged from 0.55 to 0.65. Conclusions: CRP concentrations did not accurately predict treatment failure. Serial monitoring may not be of benet. Keywords: debridement and retention, biomarker, two-stage revision, sensitivity, specicity

Introduction
Arthroplasty is a highly cost-effective treatment for arthritis, but is complicated by a 1% incidence of prosthetic joint infection (PJI).1 Management of PJI most commonly includes surgery and prolonged antibiotic treatment. Unsuccessful treatment may lead to further surgery and ultimately to amputation.2 PJI may recur despite an apparent cure, and denitive diagnosis requires invasive sampling.3 Hence, it is attractive to use surrogate markers such as the C-reactive protein (CRP) in an attempt to predict outcome. Comprehensive meta-analysis supports using CRP to diagnose PJI at presentation,4 but smaller studies do not support using CRP to predict cure at re-implantation during two-stage revision surgery for PJI.5 7 We analysed serial CRP measurements from two cohorts of patients with PJI including long-term clinical outcomes,8,9 in order to determine the clinical utility of CRP measurement during treatment of PJI.

Methods
As described previously, two clinical datasets were retrospectively gathered from patients with PJI managed in the Nufeld Orthopaedic Centre, Oxford, (i) by debridement, antibiotic and implant retention (i.e. DAIR) between 1 January 1998 and 30 April 2003 or (ii) by two-stage revision between 1 January 1999 and 30 April 2003.8,9 There was no research-related contact with patients, and informed consent was not required (as advised by our institutional review board). All activity was in accordance with the Declaration of Helsinki and national and institutional standards.

Case denition
Infection was dened as a clinical syndrome (persistent inammation in the tissues around the implant, wound discharge or implant loosening) with any of the following: bacterial growth of an indistinguishable organism from two or more deep peri-prosthetic tissue samples; a neutrophilic inltrate on histology of peri-prosthetic tissue; or a persistent sinus tract.3,8,9

# The Author 2011. Published by Oxford University Press on behalf of the British Society for Antimicrobial Chemotherapy. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com

1590

Serial CRP in prosthetic joint infection

JAC
was 10 (IQR 5 15) after DAIR and 7 (IQR 3 13) after two-stage revision. Examples of the CRP concentrations of individual patients over time are shown in Figure S1 (available as Supplementary data at JAC Online).

Denition of treatment failure

Treatment failure was dened as (i) a draining sinus, (ii) further revision surgery (irrespective of the indication), or (iii) amputation of the affected limb. Undertaking further debridement with implant retention was a secondary outcome.

DAIR Antibiotic management

As described previously,8,9 intravenous antibiotic therapy was given for 6 weeks following surgery, irrespective of the CRP measurements. Patients undergoing DAIR were given prolonged oral antibiotics (mean duration 1.5 years).

Surgical management
Sinuses were excised and the material discarded. Intra-operative samples for culture and histology were taken as previously described.3 The integrity of the cement bone or prosthesis bone interfaces was tested. If the prosthesis was sound, DAIR was undertaken, with exchange of modular prosthesis components and removal of loose bone graft. If the prosthesis was loose, it was removed and the bone ends, medullary cavities and joint cavity debrided. Re-implantation was undertaken at a median of 120 days later.

CRP measurements
CRP measurements were taken weekly for the rst 6 weeks, then at scheduled outpatient visits (typically three times during the rst year then twice yearly), using a Biostat kit and an Aeroset analyser (Abbott, Maidenhead, UK). The limits of measurement are between 8 and 285 mg/L. CRP concentrations below or above the limits of measurement were substituted by values of 4 and 300 mg/L, respectively. The laboratory was accredited by the UK Clinical Pathology Accreditation scheme.

Analysis
STATA version 10 (Stata Corp., TX, USA) was used to t multiple fractional polynomial regression models for CRP over time and clinical outcome, selected by the Royston and Altman algorithm; to plot receiver operator curves (ROCs); and to calculate sensitivity, specicity and likelihood ratios. We divided the CRP data into (i) the period of initial decline and (ii) the subsequent plateau. This resulted in four different datasets: (i) during the rst 180 days after DAIR; (ii) following the rst 180 days, using time to treatment failure or to the end of follow-up; (iii) during the rst 120 days after the rst stage of a two-stage revision (excluding CRPs taken after re-implantation); and (iv) after re-implantation, excluding the rst 28 days post-re-implantation. CRP measurements taken during hospital admissions for inter-current illness were excluded.

The proles of falling CRP concentrations during the rst 180 days after DAIR are shown in Figure 1(a). After adjusting for time, the ratio of CRP readings for treatment failure was 1.57 [95% condence interval (CI) 1.07 2.0, P 0.02]. In other words, CRP readings were, on average, 57% higher in patients experiencing treatment failure. Since time since surgery was more strongly associated with CRP concentration than treatment outcome, we were unable to produce ROCs using absolute CRP concentrations. Instead, we produced ROCs using the relative CRP level after adjusting for time. The area under the ROC (AUROC) was 0.6, indicating that CRP was a poor test (Figure S2, available as Supplementary data at JAC Online). For example, if a CRP concentration of .1.4 times the concentration expected (based on the decline seen during non-failing treatment) was considered to indicate a positive result, then the specicity was 80% for predicting failure and a sensitivity was 22%, with positive and negative likelihood ratios of 1.1 and 1.03, respectively. We were unable to select any CRP threshold that gave reasonable sensitivity and specicity (Figure S2). CRP did not predict the secondary outcomes of early failure (i.e. for treatment failure identied during the rst 180 days, n 6, the CRP ratio was 0.84, 95% CI 0.6 1.14, P 0.3) or additional debridement with implant salvage (1.1, 95% CI 0.97 1.24, P 0.12, n 24). After 180 post-operative days, there was no signicant change in CRP over time in patients not experiencing treatment failure (the ratio of current to previous months CRP was 1.035, 95% CI 0.99 1.07), but a signicant increase over time in subjects who did experience failure (a ratio of 1.15/month, 95% CI 1.06 1.25), P 0.006 for the interaction between time and treatment failure (Figure 1b). When CRP was used as a test to predict failure within 1 year, the AUROC was 0.65, indicating marginal usefulness. For instance, a CRP of .40 mg/L gave a specicity of 80% and a sensitivity of 50%, with positive and negative likelihood ratios of 2.5 and 1.6, respectively.

Downloaded from http://jac.oxfordjournals.org/ by guest on February 29, 2012

Two-stage revisions
CRP fell during the rst 120 days after surgery. The CRP prole was not signicantly different in patients experiencing treatment failure (ratio 0.7, 95% CI 0.5 1.02, P 0.064) (Figure 1c). CRP did not predict additional debridement between stages (ratio1.23, 95% CI 0.8 1.85, P 0.3), but did predict delayed re-implantation (ratio 1.03 for each month of delay, 95% CI 1.01 1.05, P 0.003). CRP at the time of re-implantation was not associated with treatment failure (ratio 0.84, 95% CI 0.65 1.09, P 0.24), or with positive microbiology at re-implantation (ratio 0.9, 95% CI 0.73 1.1, P 0.34, n 21). After re-implantation, there was no signicant change in CRP over time among patients whose treatment was successful (ratio0.99/month, 95% CI 0.98 1.03), but an increase in

Results
There were 2326 measurements of CRP from 109 patients undergoing DAIR (comprising 51 hip replacements, 50 knee replacements and 8 other joints; 84 of which were primary and 25 revised joint replacements) and 1406 separate measurements from 151 patients undergoing two-stage revision (comprising 71 hip, 76 knee and 4 elbow revisions; 62 of which were primary replacements and 89 of which were previously revised). The median lengths of follow-up with CRP measurements were 1.5 years following DAIR [interquartile range (IQR) 1.1 2.2 years] and 2 years following two-stage revision (IQR 4 months to 5 years). The median number of CRP measurements

1591

Bejon et al.

(a) 315

(b) 315

100 CRP

100

32

32

10

10

3 0 (c) 50 100 150 Days since DAIR 200 (d)

3 600 400 200 Days to failure/end of follow-up 0

315 100

315 100 32 10 3

Downloaded from http://jac.oxfordjournals.org/ by guest on February 29, 2012

CRP

32 10 3 0 Individual CRP level: Best-t line: 50 Days since rst stage treatment failure treatment failure 100

3000

2000

1000

Days to failure/end of follow-up treatment success treatment success -

Figure 1. Prole of plasma CRP concentrations over time in patients treated with DAIR according to eventual success or failure of treatment. Best-t lines are t by fractional polynomials with 95% CIs shown by the shaded area for (a) CRP concentrations by days since DAIR, (b) CRP concentrations by days before failure, excluding 180 days post-DAIR, (c) CRP concentrations by days since rst-stage revision, excluding concentrations after second stage and (d) CRP concentrations by days before failure, excluding 28 days post-re-implantation.

patients who experienced treatment failure (ratio 1.02/month, 95% CI 0.971.07, P 0.038 for the interaction between time and treatment failure) (Figure 1d). However, CRP was not a useful test to predict failure within 1 year (AUROC 0.55). We also examined the absolute neutrophil counts. The neutrophil count fell over time after DAIR (22.6109 cells/L/week, P,0.0005) and after two-stage revision (21.2109 cells/L/week, P,0.0005), but was not associated with treatment failure after DAIR (0.23109 cells/L higher, 95% CI 21.3 to 1.8, P0.8) or after two-stage revision (0.38109 cells/L higher, 95% CI 20.4 to 0.8, P0.6), and so was not analysed further.

Discussion
Treatment failure after DAIR was associated with a CRP that was slow to normalize post-operatively, or a high CRP during long-term follow-up. Treatment failure after two-stage revision was associated with a high CRP during long-term follow-up, but not with the rate of post-operative normalization. However, CRP could not be recommended as a diagnostic test based on the sensitivity and specicity values indicated by ROCs. This does not reect limited power of the study, but the wide scatter of individual

readings in both outcome groups, as found in previous studies.5 7 We did not detect statistically signicant trends over time in the white cell count, neutrophil count or platelet count (data not shown), and did not measure procalcitonin or interleukin-6. Measuring a single CRP may cost 15 US dollars (i.e. $56 000 for the 3732 tests analysed here or $215/patient), but inappropriate management decisions (for instance, invasive sampling or delayed re-implantation) may be much more costly. Observational data are prone to bias. For instance, high CRP during two-stage revision was associated with delayed re-implantation. Since CRP was not, in fact, associated with treatment failure or additional surgical debridement, this indicates that implantation was delayed simply by the clinicians response to the high CRP. However, irrespective of potential biases, the very wide scatter of CRP readings irrespective of outcome indicates the limitation of the test. Serial CRP measurements are cheap, biologically plausible, predict the response to treatment for endocarditis10 and so are readily included in the care bundles for infectious diseases. During treatment for PJI, we found that CRP monitoring was a poor test of cure. In order to avoid triggering needless

1592

Serial CRP in prosthetic joint infection

JAC
2 Zimmerli W, Ochsner PE. Management of infection associated with prosthetic joints. Infection 2003; 31: 99 108. 3 Atkins BL, Athanasou N, Deeks JJ et al. Prospective evaluation of criteria for microbiological diagnosis of prosthetic-joint infection at revision arthroplasty. The OSIRIS Collaborative Study Group. J Clin Microbiol 1998; 36: 29329. 4 Berbari E, Mabry T, Tsaras G et al. Inammatory blood laboratory levels as markers of prosthetic joint infection: a systematic review and meta-analysis. J Bone Joint Surg Am 2010; 92: 2102 9. 5 Ghanem E, Azzam K, Seeley M et al. Staged revision for knee arthroplasty infection: what is the role of serologic tests before reimplantation? Clin Orthop Relat Res 2009; 467: 1699 705. 6 Kusuma SK, Ward J, Jacofsky M et al. What is the role of serological testing between stages of two-stage reconstruction of the infected prosthetic knee? Clin Orthop Relat Res 2010; 469: 1002 8. 7 Shukla SK, Ward JP, Jacofsky MC et al. Perioperative testing for persistent sepsis following resection arthroplasty of the hip for periprosthetic infection. J Arthroplasty 2010; 25: 8791. 8 Bejon P, Berendt A, Atkins BL et al. Two-stage revision for prosthetic joint infection: predictors of outcome and the role of reimplantation microbiology. J Antimicrob Chemother 2010; 65: 56975. 9 Byren I, Bejon P, Atkins BL et al. One hundred and twelve infected arthroplasties treated with DAIR (debridement, antibiotics and implant retention): antibiotic duration and outcome. J Antimicrob Chemother 2009; 63: 1264 71. 10 Verhagen DW, Hermanides J, Korevaar JC et al. Prognostic value of serial C-reactive protein measurements in left-sided native valve endocarditis. Arch Intern Med 2008; 168: 3027.

interventions on the one hand, or delaying clinically indicated interventions on the other, we recommend against routinely monitoring CRP during treatment of PJI.

Funding
The study was carried out as part of a routine service audit. P. B. is supported by the NIHR Biomedical Research Centre in Oxford.

Transparency declarations
I. B. has received honoraria for serving on advisory boards (Pzer) and lecture fees (Pzer and Nordic Pharma). P. M.-S. is advisor to Wright Medical Technologies and receives royalties from the Corin Group. A. R. B. has received honoraria for serving on advisory boards (Pzer and Macrochem), for serving on speakers bureaus (Merck) and for producing sponsored non-promotional educational materials (Merck). All other authors: none to declare.

Downloaded from http://jac.oxfordjournals.org/ by guest on February 29, 2012

Supplementary data
Figure S1 and S2 are available as Supplementary data at JAC Online (http://jac.oxfordjournals.org/).

References
1 HPA. Fourth Report of the Mandatory Surveillance of Surgical Site Infection in Orthopaedic Surgery. April 2004 to March 2008. http://www. hpa.org.uk/webc/HPAwebFile/HPAweb_C/1227774003450 (12 April 2011, date last accessed).

1593