Biomarkers for Pediatric Bacterial Musculoskeletal Infections in Lyme Disease–Endemic Regions

We conducted a prospective cohort study of children undergoing evaluation for Lyme disease in 1 of 8 pediatric EDs each located in Lyme disease–endemic areas: Nemours Children’s Hospital (Wilmington, Delaware), Boston Children’s Hospital (Boston, Massachusetts), Children’s Minnesota (St. Paul and Minneapolis, Minnesota), Children’s Hospital of Philadelphia (Philadelphia, Pennsylvania), Children’s Hospital of Pittsburgh (Pittsburgh, Pennsylvania), Children’s Hospital of Wisconsin (Milwaukee, Wisconsin), and Hasbro Children’s Hospital (Providence, Rhode Island). Details of study enrollment have been previously described.³⁷  The study protocol was approved by the institutional review board at each participating site with permission for data sharing.

For this study, we selected children from the parent study who were aged 1 to 21 years who presented to the ED between June 2015 and January 2022 with monoarthritis. We defined monoarthritis as the presence of swelling in a single joint on examination in the absence of pain or swelling in any other joint as determined by the treating clinician. Given the challenge of evaluating hip joint swelling based on examination alone, we also included children with hip pain without documented swelling.

At the time of enrollment, treating clinicians completed standardized data collection forms that included patient demographics, presenting history, and physical examination findings. Research biosamples were collected after clinical blood samples were obtained and frozen at –80 °C after initial processing. The collected research biosamples were batch shipped to the Pediatric Lyme disease biobank (Boston Children’s Hospital, Boston, Massachusetts). Clinical biomarkers (ie, WBC count, ANC, CRP, ESR) were obtained at the discretion of the treating provider and performed at each individual site. Approximately 1 month after enrollment, study staff reviewed medical records to abstract laboratory, microbiology, and radiology test results that were performed during this initial ED encounter. A structured telephone follow-up using a standardized script was performed 1 month after enrollment to assess for potentially missed diagnoses. All data were entered in the REDCap data collection tools (Vanderbilt University) hosted by Harvard University.³⁸ 

Using available biobanked samples, we performed a PCT assay on all study patients enrolled between June 2015 and December 2019. To increase the number of PCT samples from children with an MSKI, we also performed a research PCT assay on children diagnosed with an MSKI enrolled between January 2020 and January 2022. All PCT analyses were performed in batches by the clinical laboratory at Boston Children’s Hospital using the Food and Drug Administration–approved Elecsys BRAHMS PCT electrochemiluminescent immunoassay (Roche Diagnostics) on a Cobas e601analyzer.³⁶ 

We evaluated all study patients for Lyme disease using standard 2-tier Lyme disease serology. The Branda laboratory (Massachusetts General Hospital, Boston, Massachusetts) performed a research first-tier enzyme immunoassay (EIA): from 2015-2019 Immunetics C6 (Boston, Massachusetts) and from 2020–2022 Diasorin VlSE (Seattle, Washington). For children with a positive or equivocal first-tier EIA, we performed a second-tier immunoblot (ARUP laboratories; Salt Lake City, Utah) interpreted using standardized criteria.³⁹  Positive Lyme serology was defined as a positive or equivocal first-tier EIA with a positive immunoglobin G immunoblot.

Our primary outcome was an MSKI that included septic arthritis, osteomyelitis, and/or pyomyositis. We defined a case of septic arthritis as the growth of pathogenic bacteria in synovial fluid culture, or if synovial cultures did not growth pathogenic bacteria, a positive blood culture with a synovial fluid pleocytosis (>50 000 WBC/high-powered field).⁹  Because Kingella kingae does not always grow well in routine bacterial culture, we included a positive polymerase chain reaction test of synovial fluid in our septic arthritis case definition.⁴⁰  We classified bacterial pathogens not associated with human infection in immunocompetent hosts as contaminants a priori (eg, non-aureus staphylococci, Viridans streptococci). We defined a case of osteomyelitis and pyomyositis based on results of bone scan or MRI as interpreted by the attending radiologist provider. We did not require identification of a bacterial pathogen as blood cultures can be falsely negative and few children had a diagnostic bone or muscle biopsy performed.⁴¹  For children who did not have an MSKI, we defined Lyme arthritis with a positive 2-tier Lyme disease serology, with a positive Lyme serology defined using the Centers for Disease Control and Prevention interpretation of a positive supplemental immunoblot: a positive immunoglobin G (>5 of 10 bands) or a positive immunoglobin M (>2 of 3 bands) in children <30 days of symptom duration.^42,43  All other children were classified as having other inflammatory arthritis (eg, transient synovitis, reactive arthritis, juvenile idiopathic arthritis).

Our sample size was predetermined by our cohort size. We performed a priori power analysis to determine our ability to detect a difference in accuracy between the research PCT and the serum WBC count. We found that for an estimated biomarker correlation of ≥0.2, we had at least 80% power to detect a greater than 0.14 absolute difference in biomarker accuracy.

We described our patient population using medians and their associated interquartile ranges for continuous variables and counts with proportions for nominal and ordinal variables. We compared demographic and clinical characteristics between children with and without an MSKI using χ² tests and Wilcoxon rank-sum tests for categorical and continuous variables, respectively. We constructed receiver operating characteristics (ROC) curves for each of the biomarkers to determine their test performance as measured by the area under the ROC curve (AUC). We compared the AUCs for ANC, CRP, ESR, and PCT to the AUC for WBC count.⁴⁴  We chose WBC count as the reference standard because this test is frequently obtained and included in widely used MSKI clinical decision rules.⁹  We then evaluated the performance of each biomarker in children with 1 of the 2 most commonly affected joints (knee and hip monoarthritis).

Next, using the Liu method, which maximizes the product of sensitivity and specificity,^36,45  we selected the optimal threshold for each individual biomarker from the ROC curve analysis to maximize test accuracy with obtained values rounded to improve ease of clinical use.^36,45  Using the selected cut-points, we compared each biomarker’s ability to identify MSKIs using standard test characteristics (sensitivity, specificity, positive and negative likelihood ratios) using the derived cut-points from the ROC curve analysis as well as commonly used reference points from the literature.^9,19,21,30  To simulate biomarker accuracy in regions where Lyme disease is not endemic, we assessed biomarker accuracy after excluding children with Lyme arthritis.

We performed all statistical analysis using SAS version 9.4 software, 2012 SAS Institute Inc. (Cary, North Carolina).

Of the 1958 children undergoing evaluation for Lyme disease with joint pain enrolled in the parent Pedi Lyme Net study, we included 1423 (72.7%) children with monoarthritis (Fig 1). The median patient age was 7 years (interquartile range, 4-11 years) and 907 (63.7%) were male. The most commonly inflamed joint was the knee (n = 753; 52.9%), followed by the hip (n = 530; 37.2%). Other joints involved included: ankle (n = 77; 5.4%), elbow (n = 38; 2.7%), wrist (n = 14; 0.9%), shoulder (n = 4; 0.3%), finger (n = 3; 0.2%), toe (n = 3; 0.2%), and sternoclavicular (n = 1; 0.1%) joints.

Among the study patients, 82 (5.8%) had an MSKI (25 septic arthritis alone, 44 osteomyelitis alone, 4 pyomyositis alone, 5 septic arthritis and osteomyelitis, 3 osteomyelitis and pyomyositis), 405 (28.5%) had Lyme arthritis and 936 (65.8%) had other inflammatory arthritis. Bacterial pathogens were identified in 50/82 (61%) MSKI cases. Identified pathogens included: 43 (86%) Staphylococcus aureus, 3 (6%) Streptococcus pyogenes, 3 (6%) Kingella kingae, and 1 (2%) Pasteurella multocida. Compared with children without an MSKI, children with an MSKI were older and more likely to have ankle monoarthritis (Table 1).

WBC count was the most commonly performed biomarker for children undergoing evaluation for monoarthritis in our cohort (Table 2). Overall, children with acute MSKIs had higher WBC count, ANC, CRP, ESR, and PCT than those with other types of arthritis (Table 2), although the biomarkers overlapped considerably between children with and without an MSKI (Fig 2).

We next compared the ROC curves for each of the biomarkers (Fig 3). When compared with WBC count, CRP, ESR, and PCT had higher AUC and ANC had a similar AUC for MSKIs. The empirically derived cut-points were lower than the previously published cut-points,^9,21,30  with the exception of CRP, which had a similar cut-point (Table 3).¹⁹  Compared with the previously published cut-points, the empirically derived cut-points resulted in improved sensitivity, at the expense of specificity, except for in the case of CRP, where there was a minimal difference. The performance of each individual biomarker in knee and hip monoarthritis compared with all joints is summarized in Supplemental Table 4.

Last, we examined the performance of the biomarkers in the 1016 children after excluding those with positive Lyme serologies. The biomarkers performed similarly for the identification of children with an MSKI in this cohort of children with Lyme excluded: WBC count (AUC, 0.62; 95% confidence interval [CI], 0.54–0.69), ANC (AUC, 0.67; 95% CI, 0.60–0.74), CRP (AUC, 0.88; 95% CI, 0.85–0.92), ESR (AUC, 0.82; 95% CI, 0.78–0.87), and PCT (AUC, 0.82; 95% CI, 0.77–0.88).

In our multicenter prospective cohort of children undergoing evaluation for monoarthritis in Lyme disease–endemic areas, we evaluated the ability of biomarkers to distinguish a child with an MSKI from other types of arthritis including Lyme arthritis. When compared with WBC count, CRP, ESR, and PCT were each more accurate⁴⁶  and could be used to inform clinical decision-making for children with potential bacterial bone or joint infections.

Previous studies have evaluated commonly available biomarkers for the diagnosis of an MSKI. WBC count, CRP, and ESR are commonly elevated in patients with both septic arthritis^16,17,47  and osteomyelitis⁴⁸  compared with children without these infections. In a retrospective cohort of 133 children undergoing diagnostic arthrocentesis for septic arthritis, CRP had better diagnostic accuracy than ESR.¹⁷  Our work builds on a previous study comparing Lyme and septic arthritis by its prospective design, inclusion of PCT, boarder geographic inclusion, large sample size, and inclusion of a broader range of pyogenic bone and joint infections.¹¹  Additionally, children from Lyme disease–endemic regions in the Northeast, Mid-Atlantic, and Upper Midwest were enrolled. Given the overlap in clinical and laboratory features between children with an MSKI and Lyme arthritis, all children had 2-tier Lyme disease serology obtained.^11,14 

Our evaluation of PCT as a biomarker for an MSKI expands on previous investigations.^{23,30,32,34,36,49}  In a meta-analysis of studies of mostly adult patients,³²  the pooled sensitivity of PCT for any MSKI was 54% (95% CI, 41–66) and specificity 95% (95% CI, 87–98). Nine of the 10 studies, including the 2 included pediatric studies, used a PCT cut-point of 0.50 ng/mL, and the other study used a cut-point of 0.25 ng/mL.^22,50  A recent study, completed after the publication of this meta-analysis, enrolled 155 children with a suspected MSKI who presented to a single-center located in a Lyme disease–endemic region.³⁶  PCT accurately discriminated children with an MSKI from other types of arthritis (AUC, 0.71; 95% CI, 0.62–0.81). In a recent study of 258 children with a suspected MSKI evaluated at a single center located in a non-Lyme disease–endemic region,³⁴  a PCT above 0.1 ng/mL alone identified 85% of MSKIs, outperforming WBC count, CRP, and ANC. Our study’s selected PCT cut-point (≥0.15 ng/mL) was similar to the previously derived cut-points.^34,36  Taken together, we suggest that PCT may be an accurate biomarker to identify children at lowest risk of an MSKIs using lower cut-points than for other conditions. Importantly, PCT must be interpreted in conjunction with clinical findings as well as other biomarkers in the evaluation of a child with a potential MSKI.³⁵ 

A clinician evaluating a child with a possible MSKI will frequently take a stepwise approach. Blood biomarkers can be used to identify children at low-risk for an MSKI, which may safely avoid invasive testing (eg, arthrocentesis), cross-sectional imaging (eg, MRI), or parenteral antibiotics while awaiting bacterial cultures and Lyme test results.^21,42  Invasive procedures and cross-sectional imaging frequently require procedural sedation for successful completion in young children,⁵¹  which can delay care and cause additional medical risks.⁵²  In a retrospective cohort of children with Lyme arthritis, more than one-half had arthrocentesis and one-fifth had an operative joint washout.²¹  The use of biomarkers has the potential to help reduce the need for these invasive interventions in low-risk patients; however, our data suggest that no single biomarker has the accuracy to be used in isolation.

Multivariable clinical prediction rules more accurately identify children at low and high risk of an MSKI. The Kocher criteria⁹  identify children with hip arthritis at low risk of septic arthritis with the absence of the following 4 high-risk predictors: history of fever >38.5 °C, inability to bear weight, WBC count ≥12 000 cells/µL, and ESR ≥40 mm/h.^13,15  The Kocher criteria were subsequently modified to include CRP ≥2 mg/dL.¹⁹  A single-center study found that both the original and the “modified” Kocher criteria can also identify risk of septic arthritis of other joints.⁵³  However, the Kocher criteria have not been rigorously evaluated in Lyme disease–endemic areas. Our recently published multivariable clinical prediction rule developed in a Lyme disease–endemic area identified children at low risk of an MSKI with these high-risk predictors³⁶ : PCT ≥0.50 ng/mL, CRP ≥0.6 mg/dL, and a positive first-tier Lyme EIA test.³⁵  This previously published multivariable clinical prediction rule did not include an evaluation of the accuracy of each individual biomarker. Although the rule had high sensitivity (100%, 95% CI 91%, 100%) in identifying MSKIs, broad external validation is required before clinical implementation.

Our study must be interpreted in the context of its limitations. First, only children undergoing evaluation for Lyme disease were enrolled and we may not have captured all of the children with an MSKI. However, given the clinical overlap between Lyme and septic arthritis,⁵⁴  most children presenting with monoarthritis in Lyme disease–endemic regions are evaluated for Lyme disease.^54,55  Second, biomarkers were sent at the discretion of treating providers and the rates of obtaining these biomarkers differed between those with and without an MSKI. Third, we did not obtain a research PCT on all study patients. This was partly because of availability of research samples; furthermore, between January 2020 and January 2022, only children with an MSKI underwent PCT testing and were included in our analysis. This decision was made to optimally power our study for MSKI. Because we do not report test characteristics that are dependent on disease prevalence (ie, predictive values), our selected population should not impact our results. Fourth, the approach to evaluation for MSKI was not standardized, risking potential patient misclassification.⁵⁶  In particular, not all patients had a blood culture or Kingella PCR tests performed, and other children were pretreated with an antibiotic before obtaining a bacterial culture, risking potential misclassification. To reduce this risk, we performed 1-month clinical follow-up to assess for missed MSKI diagnoses and found none. Fifth, because our study population included children with monoarthritis from Lyme disease–endemic regions, our findings may not be applicable in regions where Lyme disease is uncommon or with multiple joint involvement. However, in the subgroup of children without Lyme arthritis, the 5 biomarkers performed similarly for the diagnosis of MSKI. Sixth, we used the treating clinical team’s diagnosis for osteomyelitis based on clinical picture and MRI interpretation. We preferred this broader case definition to evaluate our candidate diagnostic biomarkers to avoid misclassifying true cases of MSKI. Lastly, despite our multicenter multiyear cohort, MSKIs were uncommon, limiting the precision of our estimates of diagnostic accuracy.

Commonly available biomarkers including WBC count, ANC, CRP, ESR, and PCT can assist clinical decision-making by identifying children at risk for having an MSKI, even in Lyme disease–endemic regions. However, because each biomarker has only moderate diagnostic accuracy, multivariable clinical prediction tools or novel biomarkers are needed to provide a more accurate and timely identification of children with an MSKI.

Members of Pedi Lyme Net include previous site primary investigators, Aris C. Garro, MD (Rhode Island Hospital and Warren Alpert Medical School of Brown University emeritus), and Jonathan E. Bennett, MD (Nemours Children’s Hospital and Sidney Kimmel Medical College of Thomas Jefferson University).

ANC

absolute neutrophil count

AUC

area under the curve

CI

confidence interval

CRP

C-reactive protein

ED

emergency department

EIA

enzyme immunoassay

ESR

erythrocyte sedimentation rate

MRI

magnetic resonance imaging

MSKI

musculoskeletal infection

PCT

procalcitonin

ROC

receiver operating characteristic

WBC

white blood cell