Session Information
Session Type: Poster Session B
Session Time: 10:30AM-12:30PM
Background/Purpose: The development of knee osteoarthritis (OA) is generally characterized by a slow evolution. However, its progression and severity may occur rapidly in some individuals. Conventional methodologies are ineffective in predicting the knee OA progression speed. Stratification of OA patients early during the disease process is key to achieving optimal disease management. Circulating biomarkers have the potential to enable such classification, and microRNAs (miRNAs) show promise. This work aims to develop a miRNA predictive model for identifying knee OA structural progressors using miRNA as inputs combined with integrated machine/deep learning.
Methods: Baseline serum of 456 miRNAs from 152 Osteoarthritis Initiative (OAI) participants were sequenced. Participants were assigned a label for their probability of being knee structural progressors/non-progressors based on our published machine learning algorithm,1 which utilized three baseline features from MRI and two from X-rays. The development of the present prediction model first included a feature dimensionality reduction through the VarClusHi clustering. Next, the identification of the top miRNAs and OA risk factors (age, sex, body mass index, and race) predictive of knee structural progression using a comprehensive evaluation of eight machine-learning models. Prediction optimization was conducted by exploring an array of machine/deep learning models to achieve the highest area under the curve (AUC), accuracy, sensitivity, and specificity. Validation of the final prediction model included the Monte Carlo cross-validation and the evaluation of baseline plasma (30) from a different OAI cohort than for the modelling.
Results: The clustering of the original 456 miRNAs resulted in a reduced dataset of 107 miRNAs, exhibiting a low level of multicollinearity. For the model development, 152 sera were used to select the top features in forecasting knee OA structural progressors. The Elastic Net algorithm showed superior performance (AUC: 0.83; accuracy: 0.83; sensitivity: 0.82; specificity: 0.85) with a predictive model of ten features comprising seven miRNAs and three risk factors. Optimization of the predictive model was performed with an Artificial Neural Network, which exhibited excellent performance with five features, including age and the miRNAs, hsa-miR-556-3p, hsa-miR-3157-5p, hsa-miR-200a-5p, and hsa-miR-141-3p (AUC: 0.94; accuracy: 0.84; sensitivity: 0.89; specificity: 0.75). Validation with the cross-validation and an independent cohort further confirmed the model’s robustness and generalizability (AUC: 0.92, 0.81; accuracy: 0.85, 0.83; sensitivity: 0.84, 0.71; specificity: 0.88, 0.94, respectively).
Conclusion: This work introduces a novel miRNA prognosis model for knee OA patients at risk of structural progression. The developed model, requiring five baseline features, demonstrating excellent performance, and validated with an independent cohort, could have high clinical relevance to guide decision-making and hold promise for a translational application in personalized therapeutic monitoring.
1. Jamshidi A, et al. Ther Adv Musculoskelet Dis 2020;12:1-12
To cite this abstract in AMA style:
Jamshidi A, Espin-Garcia O, Wilson T, Loveless I, Pelletier J, Ali A, Martel-Pelletier J. Harnessing MicroRNA and Machine/Deep Learning for Early Prediction of Knee Osteoarthritis Structural Progression [abstract]. Arthritis Rheumatol. 2024; 76 (suppl 9). https://acrabstracts.org/abstract/harnessing-microrna-and-machine-deep-learning-for-early-prediction-of-knee-osteoarthritis-structural-progression/. Accessed .« Back to ACR Convergence 2024
ACR Meeting Abstracts - https://acrabstracts.org/abstract/harnessing-microrna-and-machine-deep-learning-for-early-prediction-of-knee-osteoarthritis-structural-progression/