Which Serum Biomarkers Predicts Patient-Reported Outcomes Following Treatment for Carpometacarpal Arthritis? A Biomarker Study Using a Machine Learning Approach
Mauro Maniglio, MD.1, Moaath Saggaf, MD2, Mohit Kapoor, MD2, Anusha Ratneswaran, MD2, Nupur Purohit, MD2, Daniel Antflek, BSc3 and Heather L Baltzer, MSc, MD, FRCSC3, (1)Toronto Western Hospital, Toronto, ON, Switzerland, (2)Toronto Western Hospital, Toronto, ON, Canada, (3)University of Toronto, Toronto Western Hospital, Toronto, ON, Canada
First carpometacarpal osteoarthritis (CMC 1 OA) is one of the most frequent joints affected by OA in the hand. Different potential biomarkers are targets of research and advances in OA in general, but few studies focus on hand OA. This study aimed to identify systemic biomarkers in patients with CMC 1 OA at baseline that predict patient-reported outcomes one year after treatment. We hypothesized that baseline systemic biomarkers can predict patient-reported outcomes.
Prospectively collected blood samples and clinical data from 143 patients treated for CMC 1 arthritis with a conservative therapy, fat grafting or trapeziectomy were collected at baseline with follow-up at three months, six months and one year. Supervised machine learning methods using Lasso regularization were applied to the dataset to identify associations among 10 systemic biomarkers known to contribute to cartilage turnover, bone remodelling, pain transmission or lipid metabolism. The model was internally validated using 10-fold cross-validation and included Quick Disabilities of the Arm, Shoulder and Hand (QuickDASH), pain visual analogue scale (VAS) and the Trapeziometacarpal Arthrosis Symptoms and Disability (TASD) questionnaire outcome measures. Generalized estimating equation models were built for each outcome to assess the association between identified baseline biomarkers and 1-year outcomes accounting for age, sex, body mass index, time and treatment.
Mean age of the 143 patients was 61± 8 years (mean±SD), and 99 patients (69 %) were female. The majority of patients were treated conservatively (n=68, 47%), followed by trapeziectomy (n=54, 38%) and fat grafting (n=21, 15%). The supervised machine learning model identified associations between several outcomes and N-propeptide of collagen IIA (PIIANP), visfatin, adiponectin and leptin (Table 1). In the adjusted analyses (Table 2), baseline PIIANP was associated with improvements in VAS (beta = -3.09, 95% CI: -6.07 to -0.11, P=0.04), QuickDASH (beta =-3.99 95% CI: -5.98 to -1.99, P<0.0001) and TASD (beta = -2.42, 95% CI: -4.51 to -0.33, P=0.02) outcome measures longitudinally. Visfatin was associated with worsening in the VAS outcome measure (beta = 3.04, 95% CI: 0.14 to 5.93, P=0.04).
Baseline PIIANP is associated with an improvement in the clinical outcome while baseline visfatin is associated with worsening of CMC 1 OA outcomes up one year following treatment. These two biomarkers or a combination of both may be useful to predict patient specific outcome after treatment of CMC 1 OA.
Back to 2024 Abstracts