Clinical predictive factors for septic arthritis of the knee in children: A cohort study

Abstract

Objective

Criteria to differentiate pediatric knee joint effusion remained inexpedient. This study aimed to establish a prediction rule to distinguish infection from other inflammatory arthritis in children presenting with knee pain.

Methods

An ambi-directional cohort was conducted at the university hospital from 1999 to 2021 by including children aged ≤ 15 years with knee pain/swelling, complete physical examination, knee imaging, laboratory, and/or synovial fluid analysis. Contaminated synovial fluid culture and incomplete medical records were excluded. Diagnosis of septic knee (positive joint fluid culture or synovial white blood cell counts > 50,000 cells/mm³) or other inflammatory joints were retrieved. Baseline characteristics, physical examination, and laboratory investigations were analyzed. The probability of septic arthritis was calculated based on numbers of predictors from the final model.

Results

From 48 patients (average age 7.0 ± 4.0 years, and 24 (50%) unilateral involvement), the incidence of septic knee was 9 patients (18.8%), and 39 patients with other inflammatory arthritis. Multivariate logistic regression identified three predictors: unable to bear weight, ballottement, and erythrocyte sedimentation rate (ESR) ≥ 75 mm/h (adjusted odds ratio (OR) 36.2, 19.4, and 31.9, respectively, the model chi-square p < 0.0001, with area under receiver operating characteristic curve = 0.9188). Estimated probability of having septic knee according to 1, 2, and 3 predictors was 27.3% (95% confidence interval (CI) 13.3–45.5%), 61.5% (95% CI 31.6–86.1%), and 100% (95% CI 15.8–100%), respectively. The optimal cutoff value from sum of estimated ORs was ≥ 51.29 (at least 2 predictors) with sensitivity 88.89%, specificity 87.18%.

Conclusions

The clinical predictive factors of septic knee in children may be practically determined by unable to bear weight, ballottement, and ESR at least 75 mm/h. The probability of septic knee increases, especially for positive 2–3 factors. Further large studies would benefit for external validating this prediction rule.

Keywords

Ballottement erythrocyte sedimentation rate inflammation juvenile idiopathic arthritis synovial fluid analysis unable to bear weight

Introduction

Acute septic arthritis is one of the most common emergency surgical conditions in children. Its incidence varies from 5 to 37 per 100,000 individuals.^1–4 Septic arthritis of the knee contributed up to 40–50% of pediatric joint infections.^3–6 Clinical manifestations included fever, irritability, joint effusion, and unable to bear weight. Urgent joint irrigation and drainage are mandatory to preserve the cartilaginous part of immature knees.⁷ Therefore, prompt diagnosis would ensure acceptable therapeutic outcomes and prevent unfavorable prognosis.

Differentiating septic arthritis of the knee from other inflammatory arthritis in children can be very challenging. Both conditions may cause pyrexia, knee pain, swelling, or inability to bear weight on the affected limb.^3,7,8 Additionally, laboratory investigations including erythrocyte sedimentation rate (ESR), c-reactive protein (CRP), and peripheral white blood cell (WBC) counts may elevate in both circumstances.^3,7,9 Knee arthrocentesis is essentially required for synovial cell counts and culture.^9,10 This gold standard is invasive and, sometimes, have to perform under sedation in non-cooperating children putting more risks of anesthetic complications.

Predictive factors have been previously proposed to assess the probability of septic arthritis.^10–12 Kocher criteria (oral temperature > 38.5 Celsius, unable to bear weight, ESR > 40 mm/h, and serum WBC > 12,000/mm³) was originally used for anticipating septic hip.^12,13 Unfortunately, 52% of septic arthritis of knee could be missed even if positive 3 or more criteria.¹⁴ Baldwin et al. introduced criteria for differentiating between septic arthritis and Lyme disease of the knee, at least 3 of 4 criteria (age, history of fever, short arc pain, or CRP) promisingly predicted 84% probability of septic arthritis.¹⁰ With the rare incidence of Lyme disease in Southeast Asia,¹⁵ these predicting tools were definitely inapplicable to differentiate septic knee from other inflammatory arthritis. A recent study reported age under 5 years old and CRP > 2 mg/dl could well differentiate septic arthritis from aseptic knee effusion.¹⁶ However, the authors included presumed septic arthritis (synovial WBC > 25,000/mm³ and infectious/orthopedist final diagnosis) possibly leading to overestimated results.

Therefore, the purpose of this study was to develop a predictive model using clinical parameters to differentiate pediatric septic arthritis from other inflammatory arthritis of the knee. We hypothesized that a set of relevant clinical predictors would practically guide diagnosis and operative decision with a high level of certainty.

Methods

Study design and population

An ambi-directional cohort study (retrospective and prospective data retrieval) was conducted among children aged up to 15 years with knee pain who were treated at Department of Orthopaedics, Faculty of Medicine Ramathibodi Hospital between June 1999 and September 2021. This study was approved by Human Research Ethics Committee, Faculty of Medicine Ramathibodi Hospital, Mahidol University prior to data collection (MURA2020/1527, 29 September 2020). In regard to a prospective phase, written informed consents from all participants’ parents or legal guardians, and informed assent/consent from participants was obtained where appropriate. This study was conducted following Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) statement¹⁷ in accordance with the Helsinki Declaration of 1975 as revised in 2024, and all patients’ details were de-identified.

Eligible patients were consecutively identified through electronic medical record system. Search strategies were children admitted to the study hospital; underwent knee aspiration procedure (the International Classification of Diseases, Ninth Revision; ICD-9 code 81.91)¹⁸ or diagnosed as knee infection (the International Statistical Classification of Diseases and Related Health Problems, Tenth revision; ICD-10 code M00.06, M00.16, M00.26, M00.86, M00.9 M00), transient synovitis of the knee (M67.36), juvenile rheumatoid arthritis of the knee (M08.06), and juvenile knee arthritis (M08.96).¹⁹

The inclusion criteria were children aged 0–15 years with knee pain/swelling, complete physical examination, laboratory results (complete blood count (CBC), ESR, CRP and/or synovial fluid analysis), and imaging of the knee (ultrasonography, radiographs, or magnetic resonance imaging). Contaminated synovial fluid culture and incomplete medical records were excluded. The medical record of each included patient was reviewed to confirm the diagnosis of either septic arthritis or inflammatory arthritis and to extract clinical presentation and laboratory information.

Septic arthritis criteria

Septic arthritis was defined as either the presence of a positive joint fluid culture irrespective of synovial cell count (culture-positive septic arthritis) or synovial WBC count of >50,000 cells/mm³ with a negative joint fluid culture (culture-negative septic arthritis).^14,16,20,21 Inflammatory arthritis (juvenile idiopathic arthritis (JIA), systemic lupus erythematosus (SLE), and other autoimmune disease) was diagnosed by experienced pediatric rheumatologists based on the American College of Rheumatology (ACR) Classification,²¹ and International League of Associations for Rheumatology (ILAR).²² JIA was determined as at least one joint inflammation (as well as oligoarticular, polyarticular, systemic, psoriatic arthritis, enthesitis-related arthritis, and other forms of JIA) more than 6 weeks of duration in children aged less than 16 years old after ruling out other causes.²² SLE was diagnosed in those fulfilling ≥ 1 clinical criterion and ≥ 1 immunologic criterion or biopsy-proven nephritis in association with positive antinuclear antibodies and anti-double-stranded DNA tests.²³ All outcome assessors were unaware of predictors by using clinical presentation other than predictors and objective laboratory measurement.

Data collection

Demographic characteristics, medical comorbidities, physical examination findings, and results of laboratory testing including CBC, ESR, CRP, joint fluid analysis, Gram stain and cultures were retrieved from medical records. In case of multiple peripheral blood and joint fluid analysis during the same admission, the first inspection was collected. Study factors were age, gender, underlying disease, duration of symptoms (day), length of hospital stay (day), affected side (right/left/bilateral), physical examination such as body temperature (Celsius), unable to bear weight, knee pain, warmness, swelling, limited range of motion, and ballottement. Duration of symptoms was a period between the onset of fever or knee pain to the hospital visit. Fever was defined as oral temperature > 38 Celsius on the date of hospitalization. Limited knee range of motion was assessed by active knee flexion less than 120 degrees or unable to perform fully active knee extension.

Besides standardized cutoff values (serum WBC ≥ 15,000 cells/mm^3,24 ESR ≥ 20 mm/h,²⁴ and CRP ≥ 10 mg/L²⁵), and the common cutoff values from septic arthritis of the hip (WBC > 12,000 cells/mm³, ESR > 40 mm/h, and CRP > 20 mg/L),^12,13 the proper cutoff values of serum WBC, ESR, and CRP were identified and applied for possible predictors.

Statistical analysis

Data were analyzed using STATA version 18.0, StataCorp, College Station, Texas, USA. Continuous variables were presented as means and standard deviation or median and range, and categorical variables were presented as frequency and percentages. Continuous data were compared between groups by using unpaired t-test or Mann–Whitney U test for normally or non-normally distributed data, respectively. A chi-square or Fisher's exact test was used to determine significance in the case of binary or categorical data with normal and non-normal distribution, respectively. The optimal cut of pointed for serum WBC, ESR, and CRP were estimated using area under receiver operating characteristic (AUC) at the foremost sensitivity and specificity.

Logistic regression, odds ratio (OR) with 95% confidence interval (CI) were calculated to determine predictors for septic arthritis of the knee. Potential predictors and confounders were initially identified by univariate analysis. Any factors with p < 0.2 were brought into the multivariate analysis. By using forward stepwise logistic regression, the parsimonious model with adjusted ORs (95% CI) was summarized. The model performance was evaluated by Hosmer–Lemeshow goodness of fit test (insignificant p > 0.05 designates the estimated model fitted the data), AUC, and internally validated by bootstrap replication (1000 times).²⁶ The number of predictors and OR from the final model were used as summative scores where appropriate. Diagnostic properties of the number of predictors and the estimated score were analyzed and specified the probability of having septic arthritis of the knee.

Sample size estimation

Level of significance was set at p < 0.05. Sample size for at least one predictive factor was calculated with anticipated R² = 0.15, target expected shrinkage 0.9, the required samples size was 55.²⁶ Sample size for diagnostic accuracy was estimated based on alpha error 0.05, beta error 0.2, accuracy 0.8, and error 0.15, given 56 required cases.

Results

Baseline data

According to search strategies, 1466 cases were identified from the medical records, and 48 patients impeccably met the inclusion and exclusion criteria. Average age was 7.0 ± 4.0 years, 21 (43.8%) males, and 24 (50%) unilateral involvement. Fifteen patients (31.3%) had underlying diseases such as Beta-Thalassemia, Down's syndrome, Crohn's disease, Henoch-Schoenlein purpura, G6PD (glucose-6-phosphate dehydrogenase) deficiency, aplastic anemia, iron deficiency anemia, hypertension, aortic regurgitation, and scleroderma. Twenty-nine out of 48 patients underwent knee radiographs. Of these, 11 patients had further ultrasound (4 patients), and magnetic resonance imaging (7 patients). One patient undertook only knee ultrasound, and 19 patients had non-specific radiographic investigations.

Septic knee

The incidence of septic knee was 9 out of 48 (18.8%) patients, 7 positive/ 2 negative culture. Of 7 culture-positive patients, identified organisms were methicillin-susceptible Staphylococcus aureus, Burkholderia pseudomallei, Streptococcus dysgalactiae, Salmonella, Tuberculosis (2 patients), and Candida albicans. Of 39 (82.8%) patients with inflammatory arthritis, their diagnosis was 28 (66.7%) JIA, 5 (12.8%) SLE, 3 (7.7%) reactive arthritis/synovitis, 2 (5.1%) acute osteomyelitis, 1 (2.6%) acute rheumatic fever, 1 (2.6%) sarcoidosis, and 1 (2.6%) serum sickness.

Septic knee vs inflammatory arthritis

Comparing to inflammatory joints, septic arthritis significantly prolonged length of hospital stays (p = 0.0043), unilateral and left side predominance (p < 0.001), inability to bear weight (p = 0.006), joint warmness (p = 0.018), ballottement (p = 0.020), but lower number of serum WBC (p = 0.0126), Table 1. Average synovial WBC count among septic knee was 64,440 (range 4100–112,500) cells/mm³ with absolute synovial polymorphonuclear leukocytes (PMN) 41,410 (range 2419–100,125) cells/mm³. Six arthrocentesis out of 39 inflammatory arthritis was performed but only 3 of them obtained synovial cell count (WBC 13,826 (range 2500–126,500) cells/mm³, and absolute PMN 9539 (range 1450–113,850) cells/mm³). There was no significant difference of synovial WBC (p = 0.8333), and absolute synovial PMN (p = 0.8333) between groups. All synovial fluid Gram stain showed negative results.

Table 1.

Baseline characteristics.

Variables	Septic arthritis (N = 9)	Inflammatory arthritis (N = 39)	p-value
Age (year), mean (SD)	7.8 (4.1)	6.9 (4.0)	0.5563
Female (%)	5 (55.6)	22 (56.4)	0.623
Underlying disease (%)	5 (55.6)	10 (25.64)	0.115
Length of hospital stay (day), median (range)	12 (1–111)	4 (1–30)	0.0043*
Duration of symptoms (day), median (range)	7 (1–30)	20 (1–365)	0.1059
Fever (%)	5 (55.6)	25 (64.1)	0.711
Body temperature (Celsius), mean (SD)	38.2 (0.9)	37.6 (1.2)	0.1411
Knee pain (%)	9 (100)	26 (66.7)	0.090
Side (%) - Right/Left	1 (11.1)/ 8 (88.9)	7 (18.0)/ 8 (20.5)	< 0.001*
- Bilateral	0	24 (61.5)
Unable to bear weight (%)	7 (77.8)	10 (25.6)	0.006*
Swelling (%)	8 (88.9)	21 (53.9)	0.068
Joint warmness (%)	9 (100)	22 (56.4)	0.018*
Limited range of motion (%)	7 (77.8)	15 (38.5)	0.061
Ballottement (%)	5 (55.6)	6 (15.4)	0.020*
Hemoglobin (g%), mean (SD)	10.4 (1.2)	9.9 (1.8)	0.3647
Serum WBC/mm³, mean (SD)	8451.1 (4290.1)	13,293.5 (6050.2)	0.0126*
ESR (mm/h), mean (SD)	80.8 (24.0)	62.1 (33.7)	0.0691
CRP (mg/L), median (range)	59 (25–165)	37 (1–385)	0.6896
Arthrocentesis (%)	9 (100)	6 (15.4)	< 0.001*

WBC: white blood cell; ESR: erythrocyte sedimentation rate; CRP: c-reactive protein; *significant p < 0.05.

Since all septic arthritis in this study presented with unilateral involvement, joint pain and warmness, serum WBC < 15,000 cells/mm³, ESR ≥ 20 mm/h, and CRP ≥ 10 mg/L, the ORs could not be estimated. At serum WBC > 12,000 cells/mm³, the probability of septic knee reduced with OR 0.39, 95%CI 0.08–1.77, and p = 0.221. ESR > 40 mm/h increased risk of having septic arthritis 2.4 times (OR 2.40, 95% CI 0.26–21.84) but could not reach statistical significance (p = 0.437). The best cutoff values of serum WBC, ESR, and CRP have been explored. Serum WBC ≥ 11,400 mm³ could be differentiated septic knee from inflammatory condition with sensitivity 44.4%, specificity 41.0%, likelihood ratio of positive test (LR+) 0.75, and likelihood ratio of negative test (LR-) 1.35. ESR ≥ 75 mm/h delivered sensitivity 77.8%, specificity 66.7%, LR + 2.33, and LR- 0.33. CRP ≥ 59 mg/L fairly contributed to sensitivity 66.7%, specificity 54.5%, LR + 1.47, and LR- 0.61. These cut points were used for univariate analysis.

Predictors

Potential predictors from univariate logistic regression are presented in Table 2. Independent factors with p < 0.2 to be included in the multivariate analysis were duration of symptoms, body temperature, unable to bear weight, swelling, limited range of motion, ballottement, serum WBC, and ESR ≥ 75 mm/h. The best parsimonious model was identified from the multivariate analysis (the model chi-square p < 0.0001, pseudo R² = 0.5027 and AUC = 0.9188, Figure 1). The final three predictors comprised unable to bear weight (adjusted OR 36.24, 95% CI 2.45–536.30), ballottement (adjusted OR 19.38, 95% CI 1.54–243.59), and ESR ≥ 75 mm/h (adjusted OR 31.91, 95%CI 1.95–522.82), Table 3. Goodness of fit test indicated the data fitted to the model (insignificant p = 0.7717). Bootstrapping for internal validity unveiled unable to bear weight (observed OR 36.24, Bootstrap standard error (BSE) 195.97, p = 0.507), ballottement (observed OR 19.38, BSE 119.14, p = 0.630), and ESR ≥ 75 mm/h (observed OR 31.91, BSE 206.30, p = 0.592).

Figure 1.

The graph between the sensitivity (true positive rate) vs 1-specificity (false negative rate) demonstrated that three predictors had discrimination performance for septic arthritis of the knee with area under receiver operating characteristic (ROC) curve of 0.9188.

Table 2.

Univariate analysis for factors associated with septic arthritis of the knee.

Variables	Odds ratio (95% CI)	p-value
Duration of symptoms (day)	0.95 (0.90–1.02)	0.186
Fever	0.70 (0.16–3.04)	0.634
Body temperature (Celsius)	1.53 (0.81–2.89)	0.189
Unable to bear weight	10.15 (1.80–57.14)	0.009*
Swelling	6.85 (0.78–60.18)	0.082
Limited range of motion	5.60 (1.02–30.62)	0.047*
Ballottement	6.88 (1.42–33.26)	0.017*
Serum WBC/mm³	0.99 (0.99–1.00)	0.029*
Serum WBC ≥ 11400/mm³	0.56 (0.13–2.40)	0.432
ESR (mm/h)	1.02 (0.99–1.05)	0.129
ESR ≥ 75 mm/h	7.00 (1.27–38.57)	0.025*
CRP (mg/L)	0.99 (0.98–1.00)	0.575
CRP ≥ 59 mg/L	1.71 (0.37–7.86)	0.488

WBC: white blood cell; ESR: erythrocyte sedimentation rate; CRP: c-reactive protein; CI: confidence interval; *significant p < 0.05.

Table 3.

Multivariate analysis for predictors of septic arthritis of the knee.

Variables	Odds ratio (95% CI)	Standard error	p-value
Unable to bear weight	36.24 (2.45–536.30)	49.82	0.009*
Ballottement	19.38 (1.54–243.59)	25.03	0.022*
ESR ≥ 75 mm/h	31.91 (1.95–522.82)	45.53	0.015*

CI: confidence interval; ESR: erythrocyte sedimentation rate. *significant p < 0.05.

Predictor scores

Considering number of predictors (Table 4), the presence of 1, 2, and 3 factors estimated probability of having septic knee at 27.3% (95% CI 13.3–45.5%), 61.5% (95% CI 31.6–86.1%), and 100% (95% CI 15.8–100%), respectively. Every septic arthritis patient was positive for at least 1 factor, and 15 out of 39 inflammatory arthritis (38.5%) reported no positive one. Therefore, the probability of having septic arthritis for 0 factor could not be estimated. The OR from each predictor was also used to estimate the score (Table 5). The optimal cutoff value was ≥ 51.29 with sensitivity 88.89%, specificity 87.18%, LR + 6.93, LR- 0.13, and AUC = 0.8803, Table 6 and Figure 2. This cutoff point comprised at least two out of three predictors (Table 5). Its probability of having septic knee was 61.5% (95% CI 31.6–86.1%).

Figure 2.

The graph between the sensitivity (true positive rate) vs 1-specificity (false negative rate) at the cutoff score ≥ 51.29 discriminated septic arthritis of the knee with area under receiver operating characteristic (ROC) curve of 0.8803.

Table 4.

Diagnostic abilities of predictors for pediatric septic knee. Predictors were unable to bear weight, ballottement, and ESR ≥ 75 mm/h.

Predictor	Septic arthritis (N = 9)	Inflammatory arthritis (N = 39)	Sensitivity	Specificity	PPV	NPV	Accuracy	LR+	LR-	AUC	OR
Predictor	Septic arthritis (N = 9)	Inflammatory arthritis (N = 39)	(95% CI)
1	9 (100%)	24 (61.5%)	100 (66.4–100)	38.5 (23.4–55.4)	27.3 (13.3–45.5)	100 (78.2–100)	50.0 (35.2–64.8)	1.6 (1.3–2.1)	0	0.7 (0.6–0.8)	-
2	8 (88.9%)	5 (12.8%)	88.9 (51.8–99.7)	87.2 (72.6–95.7)	61.5 (31.6–86.1)	97.1 (85.1–99.9)	87.5 (74.5–95.3)	6.9 (2.9–16.2)	0.2 (0–1.4)	0.9 (0.8–1.0)	54.4 (6.8-.)
3	2 (22.2%)	0 (0%)	22.2 (2.8–60.0)	100 (91.0–100)	100 (15.8–100)	84.8 (71.1–93.7)	85.4 (72.2–93.9)	-	0.8 (0.6–1.1)	0.6 (0.5–0.8)	-

PPV: positive predictive value; NPV: negative predictive value; LR+: likelihood ratio of positive test; LR-: likelihood ratio of negative test; AUC: area under receiver operating characteristic curve; OR: odds ratio; CI: confidence interval; ESR: erythrocyte sedimentation rate.

Table 5.

Scoring of predictors for differential septic arthritis of the knee from inflammatory knee arthritis.

Unable to bear weight (score 36.24)	Ballottement (score 19.38)	ESR ≥ 75 mm/h (score 31.91)	Number of positive predictors	Total score
-	-	-	0	0
-	+	-	1	19.38
-	-	+	1	31.91
+	-	-	1	36.24
-	+	+	2	51.29
+	+	-	2	55.62
+	-	+	2	68.15
+	+	+	3	87.53

ESR: erythrocyte sedimentation rate; +: positive predictor; -: negative predictor.

Table 6.

Diagnostic performance of cutoff values.

Cutoff values	Sensitivity (%)	Specificity (%)	LR+	LR-
0	100	0	1.00	-
≥ 19.38	100	38.46	1.63	0.00
≥ 31.91	100	46.15	1.86	0.00
≥ 36.24	88.89	69.23	2.89	0.16
≥ 51.29	88.89	87.18	6.93	0.13
≥ 55.62	77.78	92.31	10.11	0.24
≥ 68.15	55.56	94.87	10.83	0.47
≥ 87.53	22.22	100.00	-	0.78

LR: likelihood ratio of positive test; LR-: likelihood ratio of negative test.

Discussion

This study aimed to develop predicting rules for septic knee and other inflammatory arthritis. Clinical manifestation (unable to bear weight), physical examination (ballottement), and laboratory investigation (ESR ≥ 75 mm/h) were able to differentiate these two conditions at reasonable confidence (adjusted OR 36.2, 19.4, and 31.9, respectively). The probability of septic arthritis ranged from 27.3–100% for positive 1–3 factors. The score ≥ 51.29 or positive at least two predictors suggested risk of having septic arthritis (61.5%). Serum WBC and CRP were still unsuitable for differentiating inflammatory arthritis.

Performance of predictors

Age, body temperature, unable to bear weight, limited knee range of motion, serum WBC, ESR, and CRP have been previously endeavored to discriminate infective arthritis from other conditions. Their performances depended on the geographical distribution of comparative diseases. As we have already known, Kocher criteria^12,13 and modified Kocher criteria¹¹ could not properly classify septic knee from either transient synovitis^9,14 or Lyme disease.¹⁰ With regard to transient synovitis, Bisht et al. found temperature > 38.5 Celsius, unable to bear weight, ESR > 40 mm/h, CRP > 2 mg/dL were not good predictors for septic arthritis of the knee.⁹ The authors proposed unable to bear weight and CRP predicted septic knee with probability 18.1%, 77.8%, and 89.7% for positive 0, 1, and 2 factors, respectively. Baldwin et al. developed new criteria to differentiate septic arthritis from Lyme monoarthritis using four independent clinical variables; patient or family-reported history of fever, pain with short arc motion, CRP ≥ 4 mg/L, and age younger than 2 years. The probability of septic arthritis with any one factor present was 18% compared with 100% with all four factors present.¹⁰

The previous study proposed that age < 5 years old, and CRP > 2 mg/dl was potentially categorized septic knee and aseptic effusion with excellent likelihood > 90%.¹⁶ When both predictors are applied to our data, infection is anticipated only 39.4% likelihood. All nine septic arthritis had CRP > 2 mg/dl (20 mg/L) but only one case was under age of 5 years. While 37 out of 39 inflammatory arthritis (94.9%) had positive CRP, and 12 of them (30.8%) were younger than 5 years old. The explanation might be different age groups in the septic groups (this study 7.8 ± 4.1 years vs. the previous study 3.3 years, range 1.5–7.4 years¹⁶), leading to limited generalizability. Moreover, the authors included presumed septic knee (WBC > 25,000/mm³ and infectious/orthopaedist final diagnosis) resulting in very high incidence of septic arthritis (88/122 case, 72.1%)¹⁶ compared to this cohort (18.8%). Their excessively high incidence could lead to overestimated sensitivity and diagnostic ability.

Hanalioglu et al. reported two-step approach to categorize low, moderate, and high-risk septic knee.²⁷ By incorporating major (age ≤ 5 years, serum WBC > 12,000 mm³, CRP > 2 mg/dl) and minor criteria (fever > 38.5 Celsius, non-weight bearing, ESR > 40 mm/h), two major criteria or one major plus two minor criteria indicated high risk for septic arthritis. Presence of either one major or at least one minor criterion was defined as intermediate risk. When their algorithm is applied to our data, all septic cases are six intermediate, and three high risks. On the other hand, 39 inflammatory joints are classified as 2 low risk, 13 intermediate, and 24 high risks. According to their recommendation, every suspected septic knee properly undergo joint aspiration. However, up to 24 inflammatory joints (61.5%) encounter with unnecessary arthrocentesis. In contrast, two predictors of our study captured septic knee at 88.9% vs inflammatory arthritis at 12.8%. One septic case presented only one predictor (ESR ≥ 75 mm/h) in comparison with positive ESR alone in 9 out of 39 (23.1%) inflammatory knees. Based on either two predictors or a single predictor as ESR, all septic knees were detected and 35.9% of inflammatory cases would be over investigated.

In conjunction with clinical and laboratory predictors, imaging modalities might augment diagnostic performance. From this study, diagnostic ability of knee radiographs (29 patients), ultrasound (5 patients), and magnetic resonance (7 patients) did not reach statistical significance. The OR of knee radiographs was 1.52 (95%CI 0.29–8.03) with p = 0.625. All ultrasounds (five patients) were positive for synovitis, and magnetic resonance imaging with negative results (one patient) perfectly predicted inflammatory arthritis, therefore, the OR could not be estimated. Consistent with the other studies,^10,14,27 imaging studies were also used only for diagnostic guidance, not a notable predictor for pediatric septic knee.

Characteristics of pathogens may influence inflammatory marker levels. As Kingella kingae and Lyme disease was undetected in this study, other potential pathogens (TB, bacteria, and Candida albicans) were considered. Only TB significantly differed in ESR (p = 0.0219) and CRP (p = 0.0435) from inflammatory arthritis. Therefore, the pathogens were stratified by TB and non-TB. For TB stratum, the best cutoffs were still at ESR ≥ 75 mm/h (sensitivity 100%, specificity 66.7%, AUC 0.67), and CRP ≥ 59 mg/L (sensitivity 100%, specificity 54.5%, AUC 0.5455). Two predictors (either unable to bear weight, ballottement, or ESR ≥ 75 mm/h) showed sensitivity 50% (95%CI 1.3–98.7), specificity 87.2% (95%CI 72.6–95.7), AUC 0.69 (95%CI 0.19–1.00), and positive predictive value (PPV) 16.7% (95%CI 0.4–64.1). For non-TB stratum, the best cutoff values for non-TB infection were changed to ESR ≥ 84 mm/h, and CRP ≥ 96 mg/L. The best model from multivariate analysis included ballottement (OR 8.02, 95%CI 0.88–73.22, p = 0.065), and ESR ≥ 84 mm/h (OR 11.33, 95%CI 0.99–129.6, p = 0.051) without statistical significance. Diagnostic performance determined the best cutoff by one predictor (either ballottement or ESR ≥ 84 mm/h) with sensitivity 100.0% (95%CI 47.8–100), specificity 64.1% (95%CI 47.2–78.8), AUC 0.82 (0.74–0.90), and PPV 26.3% (95%CI 9.1–51.2).

Periods of inflammatory presentation (1–365 days) overlapped with those of septic arthritis (1–30 days). Duration of symptoms were reported as early as 2–14 days (8 patients, 31%), within 21 days (13 patients, 50%), and within 30 days (18 patients, 69%) among JIA; and 1–30 days among reactive arthritis. The sensitivity analysis of patients who had duration of symptoms within 30 days showed the same different baseline factors. The probability of having septic arthritis slightly increased due to higher incidence of septic arthritis (9 out of 39 cases, 23.1%) when compared with overall analysis (9 out of 48 cases, 18.8%). According to one–three predictors, the probability of septic knee was 33.3%, 66.7%, and 100% for duration of symptoms within 30 days vs. 27.3%, 61.5%, and 100% for overall duration. At the cutoff value of 51.29, the probability of septic knee was 66.7 (95% CI 34.9–90.1%) for duration of symptoms within 30 days vs. 61.5% (95% CI 31.6–86.1%) for overall duration.

Strengths and limitations

Strengths of our study are the inclusion of whole spectrum of septic and inflammatory arthritis; all cases were evaluated and diagnosed by experienced pediatric rheumatologists or pediatric orthopedists; using objective laboratory investigations and arthrocentesis for septic arthritis; and 78% of septic arthritis had positive synovial culture confirming definite diagnosis.

This study had several limitations. The single-center retrospective design, exclusion of both contaminated synovial fluid, and incomplete information may introduce selection and information bias. Moreover, medical record review had the potential loss of data, especially physical examination such as swelling, ballottement, and knee range of motion, leading to a scant number of eligible cases. Novelty of ballottement as a predictor is not generally validated, and ESR cutoff level at ≥ 75 mm/h, above previous reported threshold (< 40 mm/h),^9,27 hinder generalizability. There were only seven culture-positive septic arthritis of the knee. Various pathogens including endemic Burkholderia pseudomallei, Tuberculosis, and Candida albicans could not represent the common pyogenic causes that requires rapid diagnosis. Despite the standard criteria for septic arthritis of the knee were applied,^14,16,20,21 the common pathogen as Kingella kingae may be missed due to mild symptoms and lower level of laboratory findings.²⁸ Polymerase chain reaction is recommended for diagnostic assurance.^28,29 Moreover, Lyme disease that typically presents with afebrile, painful, ambulatory, marked knee swelling, low ESR and CRP^10,16,30,31 has not been found in this study and in Thailand³² leading to impede clinical applicability. Ten (25.6%) out of 39 children, who diagnosed with inflammatory joint diseases, historically used non-steroidal anti-inflammatory drugs. This may minimize the clinical presentation leading to overestimate the diagnostic abilities of clinical predictors. Inadequate sample size (48 cases with the power of study = 0.738) and small number of septic knees in children absolutely affect the validity and precision. The ORs from univariate and multivariate analysis had wide 95% CI. However, all predictors reach statistical significance. The final model fitted well with the data and provided high AUC (0.92). Bootstrapping with 1000 replications fairly ascertained the internal validity. The observed ORs were the same as those in the final model analysis. The 95%CIs were imprecise, and Ps were non-significant. Three predictors required more sample size to achieve excellent internal validity.²⁶ A large cohort or prospective multi-centered studies is essential for both internal and external validation of these predictors. Molecular diagnostics and inclusion of CRP for Kingella kingae would strengthen the model's clinical utility.

Clinical applications

Applying our research findings into clinical practice, septic arthritis could be initially differentiated from inflammatory joint by clinical presentation, physical examination, and simple laboratory investigation in order to preclude an invasive procedure. The algorithm of screening and predicting septic arthritis is proposed (Figure 3). If two or more factors are detected, that patient is more likely to be infected. Additional laboratory and radiographic investigations can then be appropriately used according to the likelihood of provisional diagnosis. In case of positive only ESR, closed follow-up is needed to rule out septic knee.

Figure 3.

A proposed algorithm differentiating between septic arthritis and inflammatory joint.

Conclusion

This study offers a useful predictive algorithm for septic arthritis compared with inflammatory arthritis of the knee. The presence of unable to bear weight, ballottement, and ESR at least 75 mm/h may be practical clinical predictive factors. The more positive factors, the higher the risk of septic knee. External validation of this clinical prediction rule is essential for varying clinical presentation and laboratory cutoff points across population and settings.

Footnotes

Acknowledgements

The authors would like to thank Department of Orthopaedics, Faculty of Medicine Ramathibodi Hospital, Mahidol University for supporting research processes.

ORCID iDs

Kantapon Khoployklang

Thira Woratanarat

Soamarat Vilaiyuk

Chanika Angsanuntsukh

Patarawan Woratanarat

Ethics consideration

This study was approved by Human Research Ethics Committee, Faculty of Medicine Ramathibodi Hospital, Mahidol University (MURA2020/1527) on 29 September 2020. All methods were carried out in accordance with relevant guidelines and regulations.

Consent to participate

According to prospective data collection, written informed consents from all participants’ parents or legal guardians, and informed assent/consent from participants was obtained where appropriate.

Consent to publication

Not applicable.

Authors contributions

KK contributed to conceptualization, methodology, investigation, data curation, data analysis, and writing the original draft. TW contributed to methodology, data analysis, writing, review and editing manuscript, and supervision. SV contributed to conceptualization, methodology, data analysis, and writing, review and editing manuscript. CA contributed to methodology, data analysis, and writing, review and editing manuscript. PW contributed to conceptualization, methodology, validation, formal analysis, data curation, writing, review and editing, visualization, and supervision.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

References

Stoesser

Pocock

Moore

, et al. The epidemiology of pediatric bone and joint infections in Cambodia, 2007–11. J Trop Pediatr 2013; 59: 36–42. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3739458/pdf/fms044.pdf

Pääkkönen

. Septic arthritis in children: diagnosis and treatment. Pediatric Health Med Ther 2017; 8: 65–68. https://www.dovepress.com/getfile.php?fileID=36570

Cohen

Katz

Rahamim

, et al. Septic arthritis in children: updated epidemiologic, microbiologic, clinical and therapeutic correlations. Pediatr Neonatol 2020; 61: 325–330. https://www-sciencedirect-com-443.web.bisu.edu.cn/science/article/pii/S1875957220300243. https://www.pediatr-neonatol.com/article/S1875-9572(20)30024-3/pdf .

Nossent

Raymond

Keen

, et al. Septic arthritis in children: a longitudinal population-based study in western Australia. Rheumatology and Therapy 2021; 8: 877–888. https://doi.org/10.1007/s40744-021-00307-x .

Cho

Burke

Lee

. Septic arthritis. Hosp Med Clin 2014; 3: 494–503. https://www-sciencedirect-com-443.web.bisu.edu.cn/science/article/pii/S2211594314000355

Caksen

Oztürk

Uzüm

, et al. Septic arthritis in childhood. Pediatr Int 2000; 42: 534–540.

Iliadis

Ramachandran

. Paediatric bone and joint infection. EFORT Open Rev 2017; 2: 7–12. https://eor.bioscientifica.com/downloadpdf/journals/eor/2/1/2058-5241.2.160027.pdf

Oberle

Harris

Verbsky

. Polyarticular juvenile idiopathic arthritis - epidemiology and management approaches. Clin Epidemiol 2014; 6: 379–393. https://www.dovepress.com/getfile.php?fileID=22175

Bisht

Burns

Smith

, et al.

The modified Kocher criteria for septic hip: does it apply to the knee?

J Child Orthop 2022; 16: 233–237. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9254023/pdf/10.1177_18632521221106383.pdf

10.

Baldwin

Brusalis

Nduaguba

, et al. Predictive factors for differentiating between septic arthritis and Lyme disease of the knee in children. J Bone Joint Surg Am 2016; 98: 721–728.

11.

Caird

Flynn

Leung

, et al. Factors distinguishing septic arthritis from transient synovitis of the hip in children. A prospective study. J Bone Joint Surg Am 2006; 88: 1251–1257.

12.

Kocher

Zurakowski

Kasser

. 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: 1662–1670.

13.

Kocher

Mandiga

Zurakowski

, et al. 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: 1629–1635.

14.

Obey

Minaie

Schipper

, et al. Pediatric septic arthritis of the knee: predictors of septic hip do not apply. J Pediatr Orthop 2019; 39: e769–ee72. https://www.ingentaconnect.com/content/wk/bpo/2019/00000039/00000010/art00015

15.

Sthitmatee

Jinawan

Jaisan

, et al. Genetic and immunological evidences of Borrelia burgdorferi in dog in Thailand. Southeast Asian J Trop Med Public Health 2016; 47: 71–77.

16.

Thomas

Schiffman

Faino

, et al. Diagnostic criteria for the painful swollen pediatric knee: distinguishing septic arthritis from aseptic effusion in a non-lyme endemic area. Front Surg 2021; 8: 740285. https://www.frontiersin.org/articles/ https://doi.org/10.3389/fsurg.2021.740285

17.

Collins

Reitsma

Altman

, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. Br Med J 2015; 350: g7594.

18.

Slee

. The international classification of diseases: ninth revision (ICD-9). Ann Intern Med 1978; 88: 424–426.

19.

Brämer

. International statistical classification of diseases and related health problems. Tenth Revision. World Health Stat Q 1988; 41: 32–36.

20.

Hujazi

Ambler

Arora

, et al. Role of Newman's classification in predicting outcomes in patients with crystal arthritis. Int Orthop 2012; 36: 1287–1290. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3353081/pdf/264_2011_Article_1458.pdf

21.

Cassidy

Levinson

Bass

, et al. A study of classification criteria for a diagnosis of juvenile rheumatoid arthritis. Arthritis Rheum 1986; 29: 274–281.

22.

Petty

Southwood

Manners

, et al. International league of associations for rheumatology classification of juvenile idiopathic arthritis: second revision, Edmonton, 2001. J Rheumatol 2004; 31: 390–392.

23.

Hochberg

. Updating the American college of rheumatology revised criteria for the classification of systemic lupus erythematosus. Arthritis Rheum 1997; 40: 1725.

24.

Pääkkönen

Kallio

, et al. Sensitivity of erythrocyte sedimentation rate and C-reactive protein in childhood bone and joint infections. Clin Orthop Relat Res 2010; 468: 861–866. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2816763/pdf/11999_2009_Article_936.pdf

25.

Mitchell

Viswanath

Obi

, et al. A prospective study of screening for musculoskeletal pathology in the child with a limp or pseudoparalysis using erythrocyte sedimentation rate, C-reactive protein and MRI. J Child Orthop 2018; 12: 398–405. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6090194/pdf/jco-12-398.pdf

26.

Riley

Ensor

Snell

KIE

, et al. Calculating the sample size required for developing a clinical prediction model. Br Med J 2020; 368: m441.

27.

Hanalioglu

Turker

Kamaci

, et al. Age and C-reactive protein as modifiers of Kocher criteria in pediatric septic knee monoarthritis. Am J Emerg Med 2022; 60: 145–151.

28.

Williams

Cooper

Cundy

. Kingella kingae septic arthritis in children: recognising an elusive pathogen. J Child Orthop 2014; 8: 91–95.

29.

Yagupsky

. Kingella kingae: carriage, transmission, and disease. Clin Microbiol Rev 2015; 28: 54–79.

30.

Grant

Neville

Levas

, et al. Validation of septic knee monoarthritis prediction rule in a Lyme disease endemic area. Pediatr Emerg Care 2022; 38: e881–e8e5.

31.

Yen

Sanborn

Donohue

, et al. Lyme arthritis in the pediatric knee: clinical and magnetic resonance imaging differentiators. JB JS Open Access 2022; 7.

32.

Chaorattanakawee

Tachavarong

Hananantachai

, et al. Seasonal pattern of questing ticks and prevalence of pathogenic Rickettsia and Anaplasmataceae in Khao Yai national park, Thailand. Travel Med Infect Dis 2024; 58: 102696.