Toward Identifying a Multivariate Correlation of Septic Arthritis With a Machine Learning Approach: Time to Reset the...

To understand the complexity of disease pathology through the prism of septic arthritis, especially the reliability of popular and, yet, arbitrary thresholds like synovial leucocyte counts of ≥ 100,000/μL suggestive of it, with the help of statistical analysis and logistic regression.

An anonymized patient dataset comprising 360 swollen joint episodes was collated along with a range of patient attributes, including age, gender, comorbidity (e.g., diabetes, gout, pseudogout, immunosuppression), prior administration of antibiotics and washout of the affected joint, isolation of crystals from synovial aspirate, blood/synovial fluid culture growth, and synovial aspirate cell count. The dataset was subjected to statistical analysis (e.g., sensitivity, specificity, predictive and likelihood ratios) and logistic regression modeling, with results compared to the synovial leucocyte count thresholds of ≥ 100,000/μL and ≥ 50,000/μL.

The logistic regression model (sensitivity 50%, specificity 97.04%) outperformed the models based on arbitrary thresholds like a synovial leucocyte count of ≥ 100,000/μL (sensitivity 48.21%, specificity 88.16%) or ≥ 50,000/μL (sensitivity 64.29%, specificity 69.74%) in predicting septic arthritis. Independent variables like age, presence of gout, and autoimmune arthritis as comorbidities, hip joint involvement, synovial aspirate leucocyte count, and crystals in aspirated fluid demonstrated a significant (p < 0.05) correlation to septic arthritis.

Septic arthritis presents a multivariate correlation that deserves a holistic oversight rather than singling out individual factors. Data mining platforms like logistic regression can investigate the complex interplay among these individual variables while making a diagnosis not only in septic arthritis but also in other diseases with multisystem involvement, infective or non-infective alike.