The use of CRP within a clinical prediction algorithm for the differentiation of septic arthritis and transient synovitis in children

Transcription

1 CHILDREN S ORTHOPAEDICS The use of CRP within a clinical prediction algorithm for the differentiation of septic arthritis and transient synovitis in children R. Singhal, D. C. Perry, F. N. Khan, D. Cohen, H. L. Stevenson, L. A. James, J. S. Sampath, C. E. Bruce From Alder Hey Children s Hospital, Liverpool, United Kingdom R. Singhal, MBBS, MS (Ortho), MRCSEd, Registrar D. C. Perry, MB, ChB (Hons), MRCS (Eng), Registrar, PhD Student F. N. Khan, MBBS, MRCSEd, Specialty Trainee D. Cohen, MBChB, FRCS (Tr & Orth), Registrar H. L. Stevenson, MBCh, FRCS (Tr & Orth), Registrar L. A. James, MBBS, FRCSEd (Tr & Orth), Consultant Orthopaedic Surgeon C. E. Bruce, MBChB, FRCS (Tr & Orth), Consultant Orthopaedic Surgeon Alder Hey Children s Hospital, Department of Orthopaedic Surgery, Eaton Road, West Derby, Liverpool L12 2AP, UK. J. S. Sampath, MBBS, MSc, FRCSEd (Tr & Orth), Consultant Orthopaedic Surgeon SPARSH Hospital, Department of Orthopaedics, The Health City, Bommasandra Industrial Area, Hosur Road, Bangalore , India. Correspondence should be sent to Mr R. Singhal; singhal.rohit75@gmail.com 2011 British Editorial Society of Bone and Joint Surgery doi: / x.93b $2.00 J Bone Joint Surg Br 2011;93-B: Received 3 February 2011; Accepted after revision 15 July 2011 Clinical prediction algorithms are used to differentiate transient synovitis from septic arthritis. These algorithms typically include the erythrocyte sedimentation rate (ESR), although in clinical practice measurement of the C-reactive protein (CRP) has largely replaced the ESR. We evaluated the use of CRP in a predictive algorithm. The records of 311 children with an effusion of the hip, which was confirmed on ultrasound, were reviewed (mean age 5.3 years (0.2 to 15.1)). Of these, 269 resolved without intervention and without long-term sequelae and were considered to have had transient synovitis. The remaining 42 underwent arthrotomy because of suspicion of septic arthritis. Infection was confirmed in 29 (18 had micro-organisms isolated and 11 had a high synovial fluid white cell count). In the remaining 13 no evidence of infection was found and they were also considered to have had transient synovitis. In total 29 hips were categorised as septic arthritis and 282 as transient synovitis. The temperature, weight-bearing status, peripheral white blood cell count and CRP was reviewed in each patient. A CRP > 20 mg/l was the strongest independent risk factor for septic arthritis (odds ratio 81.9, p < 0.001). A multivariable prediction model revealed that only two determinants (weight-bearing status and CRP > 20 mg/l) were independent in differentiating septic arthritis from transient synovitis. Individuals with neither predictor had a < 1% probability of septic arthritis, but those with both had a 74% probability of septic arthritis. A two-variable algorithm can therefore quantify the risk of septic arthritis, and is an excellent negative predictor. Several algorithms have been proposed to assist in differentiating between septic arthritis and transient synovitis of the hip in children. 1-6 This distinction is crucial as the natural history, and hence treatment, of the two diseases differs greatly. Whilst transient synovitis is a benign condition with little potential for long-term serious damage, septic arthritis can lead to a poor outcome, particularly if the treatment is delayed. 7 Kocher, Zurakowski and Kasser 1 initially described several variables to aid in the differentiation of these conditions: temperature > 38.5 C, inability to weight-bear, erythrocyte sedimentation rate (ESR) > 40 mm/hr and peripheral white blood cell count > cells/mm 3. However, in these conditions ESR has largely been replaced by estimation of the C-Reactive Protein (CRP). 8 CRP is an acute phase reactant, which has a more rapid response to disease than ESR. It is influenced less by external factors and its measurement requires a smaller blood volume. In a further validation study by Kocher et al 3 it was acknowledged that CRP had benefits over ESR, yet they were unable to include this parameter owing to their lack of necessary data. Whilst some studies have included CRP in prediction models this has been in combination with ESR. 4,6 In our centre we frequently make decisions in differentiating transient synovitis and septic arthritis based on the use of CRP without consideration of ESR. This study was designed to assess the use of CRP in a clinical prediction algorithm to differentiate transient synovitis and septic arthritis, based on the framework suggested by Kocher et al. 3 Patients and Methods A review was undertaken of the records of all patients under the age of 16 years presenting with an acute, new-onset, atraumatic limp or hip pain between January 2004 and October All patients underwent an ultrasound evaluation of the hip as part of a protocol to investigate the irritable hip. Neonates were excluded. Demographic data, clinical examination, haematological parameters and ultrasound 1556 THE JOURNAL OF BONE AND JOINT SURGERY

2 THE USE OF CRP FOR THE DIFFERENTIATION OF SEPTIC ARTHRITIS AND TRANSIENT SYNOVITIS 1557 Table I. Organisms isolated from the hip aspirate Organism Number of patients Ultrasound scan 735 Staphylococcus aureus 5 Streptococcus pyogenes 4 Streptococcus agalactiae 2 Coagulase negative 3 staphylococcus Neisseria meningitidis 1 Enterococcus species 1 Gram-positive 1 diphtheroids Methicillin-resistant Staphylococcus 1 aureus Total 18 Positive 311 Excluded 55 Negative 369 Arthrotomy 42 Resolved 269 Negative 13 Transient synovitis findings in all patients were recorded. Only patients with a proven effusion were included in the analysis. Patients with a diagnosis other than transient synovitis and septic arthritis were excluded. In all, a total of 735 hips underwent ultrasound examination during the study period. However, several patients were excluded as a clear cause for limp or hip irritability was identified. Those excluded involved two patients with discitis, one with an aneurysmal bone cyst, two with juvenile rheumatoid arthritis, fourteen with Perthes disease, six with a fracture of the femur, one with a fracture of the fibula, one with neurological disease, seven with other rheumatological diseases and 21 with non-hip-related infection, which included ischial osteomyelitis, proximal femoral osteomyelitis, sacroiliac sepsis, fibular osteomyelitis, piriformis pyomyositis, cellulitis, and septic arthritis at the ankle and knee. Accordingly 680 patients remained after exclusions, of whom 311 had an effusion of the hip on ultrasound and therefore formed the study group (217 males and 94 females with a mean age of 5.3 years (0.2 to 15.1)). Each of the following findings were considered to be significant: a temperature taken in the emergency department of > 38.5 C, an inability to bear weight or inability to move the leg in children of non-weight-bearing age, a white blood cell count of more than /mm 3 ( cells/l), and a CRP level of > 2.0 mg/dl (> 20 mg/l). The treating clinician determined the need for intervention on the basis of the clinical findings and results of investigations. If septic arthritis was considered likely an arthrotomy was undertaken. Fluid was taken for culture and antibiotics commenced. If a diagnosis of transient synovitis was considered more likely, patients were managed by simple observation and analgesia. If resolution of symptoms occurred and normal gait was restored, patients were considered to have experienced a confirmed episode of transient synovitis. If, however, symptoms deteriorated these patients also underwent arthrotomy and antibiotic therapy for presumed septic arthritis. Those patients who were treated with arthrotomy and antibiotics were further investigated to determine if there was truly septic arthritis. Septic arthritis was defined on the basis of synovial white cell count and culture. Patients with Organisms 18 Synovial fluid 11 Septic arthritis Fig. 1 Flow diagram of patient group allocation (number indicates the number of patients within each group). either a positive culture of the hip aspirate, or with a microscopic abundance of white cells (++ or more per high power field) within the hip aspirate were deemed to have septic arthritis. Those without such positive findings were deemed to have transient synovitis. Statistical analysis. Stata10 statistical package (StataCorp LP, College Station, Texas) was used for the statistical analysis. Univariate analysis assessed the significance of each of the key variables (temperature, weight-bearing status, CRP and peripheral white blood cell count) using Fisher s exact test. All variables remained significant and were entered into a multivariate model using logistic regression to allow comparison with previous studies. A p-value of < 0.05 was considered to be significant. A probability table was constructed based around the algorithm for each permutation of variables. Stepwise logistic regression was then completed to select the best-fit model based upon the variables, and a further probability table was constructed. Results Of the 311 patients, 42 (13.5%) underwent arthrotomy of the hip for suspected septic arthritis. The remaining 269 (86.5%) resolved without requiring antibiotic therapy or operative intervention and were considered to represent transient synovitis. No patient in the ultrasound negative group required subsequent treatment for septic arthritis. These patients all made a complete spontaneous recovery with no sequelae at three months follow-up. Of the 42 patients treated as septic arthritis, 29 met the criteria for confirmed septic arthritis. Organisms were isolated from the hip aspirate in 18 patients (Table I) and a further 11 were classified as septic arthritis based on the white cell count of the synovial fluid. In 13 patients, VOL. 93-B, No. 11, NOVEMBER 2011

3 1558 R. SINGHAL, D. C. PERRY, F. N. KHAN, D. COHEN, H. L. STEVENSON, L. A. JAMES, J. S. SAMPATH, C. E. BRUCE Table II. Demographics of the patients included in the study Variable Transient synovitis (n = 282) Septic arthritis (n = 29) Mean age (yrs) (range) 5.5 (0.5 to 15.0) 3.2 (0.2 to 15.1) Male gender (n, %) 204 (72.3) 13 (44.8) CRP > 20 mg/l (n, %) 20 (7.1) 25 (86.2) Fever > 38.5 C (n, %) 2 (0.7) 8 (27.6) Refusal to bear weight (n, %) 51 (18.1) 26 (89.7) Peripheral white blood cell count > cells/mm 3 (n, %) 20 (7.1) 14 (48.3) Table III. Univariate analysis of the four-variable model Explanatory variable Odds ratio p-value C-reactive protein 81.9 < Temperature 53.3 < Weight-bearing status 39.3 < Peripheral white blood cell 12.3 < count microscopy and culture of synovial fluid was negative, and were accordingly assigned to the transient synovitis group, giving a total of 282 patients classified as having transient synovitis of the hip (Fig. 1). The results are outlined in Table II. Univariate analysis conducted on these patients demonstrated that all of the variables were significant with p < CRP appeared the best single predictor of septic arthritis (odds ratio (OR) = 81.9 (95% confidence interval (CI) 7.8 to 120.3) and temperature the second best single predictor of septic arthritis (OR = 53.3 (95% CI 3.7 to 58.7)) (Table III). Four-variable predictive model. In order to compare with previous studies, all variables were taken forward to a multivariable model using logistic regression analysis. The Hosmer-Lemeshow statistic 9 demonstrated no significant departure from a good fit model (p = 0.09). Independent associations of peripheral white blood cell count and temperature were eliminated on multivariate analysis. CRP remained the most significant independent predictor of septic arthritis (Table IV). A probability algorithm was created (Table V). When none of the predictors was present the probability of septic arthritis was < 1% and when all the predictors were present the probability of septic arthritis was 87%. Assuming a p-value > 0.5 to indicate septic arthritis, the model correctly classified 94.8% individuals with sensitivity of 75.9%, specificity of 96.8%, positive predictive value of 71.0% and negative predictive value of 97.5%. Two-variable predictive model. Not all of the factors remained significant within the regression model owing to close relationships between the variables; the measurement of CRP and peripheral white blood cell count may be measuring a similar process and therefore the relative strength of CRP removes the importance of the white cell count. Stepwise logistic regression was used to eliminate variables with such relationships (p < 0.2) (Table VI). The Hosmer- Lemeshow statistic demonstrated no significant departure from a good-fit model (p = 0.17). A further probability algorithm was created (Table VII); fitting the model to our dataset (assuming p > 0.5 indicated septic arthritis) demonstrated identical efficacy to the four-variable model. The results demonstrate that CRP is a strong predictor of septic arthritis, and its inclusion within a model eliminates the significance of other variables. Additional precision may be gained from adding more variables into the diagnostic model, but this did not alter the sensitivity or specificity when applied to our dataset. Discussion Prediction algorithms were introduced to simplify clinical medicine, being first widely used to manage myocardial infarction. 10 The purpose of algorithms in clinical medicine is to identify key factors that guide the clinician to the appropriate diagnosis. Kocher et al 1 demonstrated an ability to differentiate between transient synovitis and septic arthritis using a fourvariable model with a predictive probability of septic arthritis of 99% when all four variables were positive. Subsequently prospective internal validation demonstrated that the probability of septic arthritis fell to 93%. 3 External validation has been less successful with Luhmann et al 2 showing only a 59% probability of septic arthritis when all four variables were present. The benefit of CRP in identifying septic arthritis was proposed by Levine et al 11 who measured CRP in 133 patients with an effusion in any joint, of which 39 were classified as septic arthritis. CRP proved a better independent predictor of disease than ESR. Consequently Jung et al 6 created a prediction model to include ESR, CPR, white cell count, temperature and size of the radiological joint space. They identified patients with atraumatic hip pain in whom plain radiographs were normal and classified them into septic arthritis or not septic arthritis groups. They also found that CRP appeared a better predictor of septic arthritis, though the definitions of both transient synovitis and septic arthritis were poor. Caird et al 4 updated the model of Kocher et al 1 to include both CRP and ESR, suggesting a 97.5% predictive value in a five-variable model but that study included very few cases of transient synovitis (n = 14). A recent external validation study of the model of Caird et al 4 showed a predictive probability of just 59.9% with all five variables, 5 although again the number of cases of septic arthritis within the validation study was small (n = 5). No existing algorithms recognise the trend to using the CRP in place of ESR. CRP is a reproducible, direct and quantitative measure of the acute phase reaction, unlike ESR, which is an indirect measure of inflammation that may be influenced by factors such as temperature, age, gender, steroid medication and non-steroidal anti-inflamma- THE JOURNAL OF BONE AND JOINT SURGERY

4 THE USE OF CRP FOR THE DIFFERENTIATION OF SEPTIC ARTHRITIS AND TRANSIENT SYNOVITIS 1559 Table IV. Multivariable model using logistic regression for four-variable model Explanatory variable Adjusted odds ratio p-value 95% confidence interval C-reactive protein 30.6 < to Weight-bearing 14.6 < to 58.7 status Temperature to 17.5 Peripheral white blood cell count to 4.7 Table V. Probability algorithm for four-variable model Fever C-reactive protein Weight -bearing status Peripheral white blood cell count Probability Yes Yes Yes Yes 0.87 Yes Yes Yes No 0.84 Yes Yes No Yes 0.31 Yes Yes No No 0.27 Yes No Yes Yes 0.17 Yes No Yes No 0.15 Yes No No Yes 0.01 Yes No No No 0.01 No Yes Yes Yes 0.71 No Yes Yes No 0.68 No Yes No Yes 0.15 No Yes No No 0.12 No No Yes Yes 0.08 No No Yes No 0.06 No No No Yes 0.01 No No No No 0.01 Table VI. Stepwise logistic regression for two-variable model Explanatory variable Adjusted odds ratio p-value 95% confidence interval C-reactive protein 40.0 < to Weight-bearing status 16.5 < to 64.7 Table VII. Probability algorithm for two-variable model C-reactive protein Weight-bearing Probability Yes Yes 0.74 Yes No 0.15 No Yes 0.06 No No 0.01 tory drugs. 12 The CRP rises early in sepsis with abnormal results evident within six hours of onset, whereas ESR has a more intermediate response to inflammation. 13,14 CRP is being used instead of ESR because of its increasing availability, reduced cost and smaller blood volume requirement, which is of particular importance in children. We have demonstrated that CRP is an important independent risk factor for septic arthritis. Furthermore its addition to the regression model reduced the significance of other variables owing to its relative strength. This facilitates the construction of a more simplified prediction algorithm based upon only two variables (weight-bearing status and CRP > 20 mg/l). These two variables alone offer a 74% predicted probability of septic arthritis. If both weight-bearing status and CRP were negative the predicted probability was < 1%. A number of studies have evaluated algorithms for differentiating between septic arthritis and transient synovitis (Table VIII). One of the most important differences between these studies is the proportion of cases of septic VOL. 93-B, No. 11, NOVEMBER 2011

5 1560 R. SINGHAL, D. C. PERRY, F. N. KHAN, D. COHEN, H. L. STEVENSON, L. A. JAMES, J. S. SAMPATH, C. E. BRUCE Table VIII. Previous studies conducted to differentiate transient synovitis (TS) from septic arthritis (SA) Study Study design Number of TS cases Number of SA cases Ratio TS:SA Source population of patients Kocher et al 1 Retrospective :1 Patients with the presence of fluid on hip aspiration * Jung et al 6 Retrospective :1 All patients presenting with hip pain Kocher et al 3 Prospective :1 Patients with the presence of fluid on hip aspiration * Luhmann et al 2 Retrospective :1 Patients undergoing hip aspiration performed after ultrasound confirmation of effusion in patients in whom SA was a possible diagnosis * Caird et al 4 Prospective :1 Patients undergoing hip aspiration based upon those who were most suspicious for septic arthritis * Sultan and Hughes 5 Retrospective :1 Patients presenting to hospital with an irritable hip Current study Retrospective :1 All patients investigated for irritable hip with the presence of fluid on ultrasound evaluation * criteria warranting hip aspiration unclear no confirmation of effusion sought arthritis compared with those of transient synovitis. This is probably a function of the populations from which septic arthritis and transient synovitis were recruited. Kocher et al 1 used cases positive for effusion on aspiration, but did not indicate how children were selected for aspiration (septic arthritis: transient synovitis ratio = 1.0). Caird et al 4 similarly used patients undergoing aspiration of the hip but acknowledged that there was a high threshold for children to undergo aspiration (septic arthritis: transient synovitis ratio = 2.4). Luhmann et al 2 had more cases of transient synovitis (septic arthritis: transient synovitis ratio = 0.4) but similarly recruited from patients undergoing aspiration of the hip. Sultan and Hughes 5 and Jung et al 6 did not confirm if those with transient synovitis were positive for effusion, and therefore they may not truly represent transient synovitis. We believe that our results are the most inclusive as our study included all cases of ultrasound-confirmed transient synovitis and septic arthritis presenting to our institution. The septic arthritis: transient synovitis ratio of 0.1 in our study appears to represent broadly the case mix within our tertiary level unit. Acknowledging such differences is important as the size of the odds ratios may be different based upon the source population. We believe our prediction rule is therefore applicable to patients from entry into the emergency department, and hence the most universally applicable to date. As with other studies, prospective external validation is ultimately required to confirm these findings. We acknowledge our study differs from that of Kocher et al, 1 Caird et al 4 and Luhmann et al 2 because a large proportion of our patient group were treated without aspiration of the hip. This means that no formal confirmation of an aseptic joint exists in the majority of our transient synovitis group. It is largely through the work of groups such as Kocher et al 1-4 that we now rarely perform invasive tests and are able simply to observe such patients to complete resolution of symptoms. Our study is also limited in its definition of septic arthritis by the way in which laboratory results are reported. Many laboratories in the United Kingdom express white cell counts from joint fluid using an ordinal measure such as +, We recognise the limitations of this when making comparisons with international data, but maintain that we needed to be pragmatic in the collection of historical data. The literature to date arbitrarily designates white cell count per cm 3 as presumed septic arthritis ; 1-4 we believe that this approximates to ++ or greater. We have found an elevated CRP is a strong independent indicator of septic arthritis, and we have demonstrated its utility in simplifying diagnostic predictive models. No benefits in any form have been received or will be received from a commercial party related directly or indirectly to the subject of this article. References 1. Kocher MS, Zurakowski D, Kasser JR. Differentiating between septic arthritis and transient synovitis of the hip in children: an evidence-based clinical prediction algorithm. J Bone Joint Surg [Am] 1999;81-A: Luhmann SJ, Jones A, Schootman M, et al. Differentiation between septic arthritis and transient synovitis of the hip in children with clinical prediction algorithms. J Bone Joint Surg [Am] 2004;86-A: Kocher MS, Mandiga R, Zurakowski D, Barnewolt C, Kasser JR. Validation of a clinical prediction rule for the differentiation between septic arthritis and transient synovitis of the hip in children. J Bone Joint Surg [Am] 2004;86-A: Caird M, Flynn J, Leung Y, et al. Factors distinguishing septic arthritis from transient synovitis of the hip in children: a prospective study. J Bone Joint Surg [Am] 2006;88-A: THE JOURNAL OF BONE AND JOINT SURGERY

6 THE USE OF CRP FOR THE DIFFERENTIATION OF SEPTIC ARTHRITIS AND TRANSIENT SYNOVITIS Sultan J, Hughes PJ. Septic arthritis or transient synovitis of the hip in children: the value of clinical prediction algorithms. J Bone Joint Surg [Br] 2010;92-B: Jung ST, Rowe SM, Moon ES, et al. Significance of laboratory and radiologic findings for differentiating between septic arthritis and transient synovitis of the hip. J Pediatr Orthop 2003;23: Nunn TR, Cheung WY, Rollinson PD. A prospective study of pyogenic sepsis of the hip in childhood. J Bone Joint Surg [Br] 2007;89-B: Husain T, Kim D. C-reactive protein and erythrocyte sedimentation rate in orthopaedics. University of Pennsylvania Orthopaedic Journal 2002;15: Hosmer DW Jr, Lemeshow S. Applied logistic regression. New York: Wiley, 1989: Goldman L, Cook EF, Johnson PA, et al. Prediction of the need for intensive care in patients who come to the emergency departments with acute chest pain. N Engl J Med 1996;334: Levine MJ, McGuire KJ, McGowan KL, Flynn JM. Assessment of the test characteristics of C-reactive protein for septic arthritis in children. J Pediatr Orthop 2003;23: Ng T. Erythrocyte sedimentation rate, plasma viscosity and C-reactive protein in clinical practice. Br J Hosp Med 1997;58: Kallio MJ, Unkila-Kallio L, Aalto K, Peltola H. Serum C-reactive protein, erythrocyte sedimentation rate and white blood cell count in septic arthritis of children. Pediatr Infect Dis J 1997;16: Pepys MB. C-reactive protein fifty years on. Lancet 1981;1: VOL. 93-B, No. 11, NOVEMBER 2011