Microbiome-Augmented Model for Predicting Knee Osteoarthritis Progression Based on Gut Microbiota and Kellgren-Lawrence Classification.

purposeKnee osteoarthritis (OA) is a widespread chronic degenerative condition that may experience slow or rapid deterioration. The gut-joint axis represents a bidirectional relationship in OA onset and progression. This study aimed to establish and validate a prediction model of knee OA disease progression.

methodsThis prospective cohort investigation involved 296 patients diagnosed with knee OA using X-ray and CT scans at Taizhou People's Hospital from January 2020 to January 2022. Fecal samples and general information were collected for gut microbiota analysis. Least absolute shrinkage and selection operator (LASSO) regression and various prediction models, including microbiome-augmented models, were employed for knee OA risk prediction. The models predicting Kellgren-Lawrence classification one year later were evaluated by accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under the curve (AUC).

resultsA total of 270 patients were involved in our study. After random assignment, 214 patients belonged to the training set and 56 patients belonged to the test set. The final intestinal flora included in the analysis included the following 12 species. Shannon index of patients with a Grade I Kellgren-Lawrence Classification after one year was lower than those with a Grade II/III after one year (P=0.018). The best model was the microbiome-augmented model built by Light GBM (LGBM). The AUC of this model in the training set was 0.812 (0.754-0.870), the sensitivity was 0.804 (0.725-0.883), the specificity was 0.744 (0.664-0.823), the PPV was 0.722 (0.638-0.807), the NPV was 0.821 (0.748-0.894), and the accuracy was 0.771 (0.715-0.827). The AUC of this model in the testing set was 0.876 (0.781-0.972), the sensitivity was 0.759 (0.603-0.914), the specificity was 0.917 (0.806-1.000), the PPV was 0.917 (0.806-1.000), the NPV was 0.759 (0.603-0.914), and the accuracy was 0.830 (0.729-0.931).  Conclusion: One year later, the microbiome-augmented model constructed by LGBM for knee OA patients based on general and gut microbiota data using the Kellgren-Lawrence classification demonstrated the highest performance. This approach could aid in identifying patients at risk of rapid disease progression, facilitating early intervention and personalized treatments. Furthermore, it offers a novel perspective on the gut-joint axis's role in OA.