“Machine Learning” for Knee OA Phenotyping

A new study, “A machine learning approach to knee osteoarthritis phenotyping: data from the FNIH Biomarkers Consortium,” addresses the use of machine learning to define the progression phenotypes that might lead to a new treatment. The study was published in the July issue of Osteoarthritis and Cartilage.

Co-author Amanda Nelson, M.D., associate professor of medicine at the University of North Carolina Thurston Arthritis Research Center and study co-author, explained the strategy behind the study to OTW, “We sought to utilize all available data simultaneously to pick out those features which help to identify patients who did or did not experience progressive symptomatic knee osteoarthritis.”

“Traditional methods require physicians to select predictor variables, while machine learning methods allow physicians to incorporate all available data to determine the most important features, thereby taking advantage of the richness of datasets like the FNIH [Foundation for the National Institutes of Health] Biomarkers cohort, which was the data we used for this analysis.”

The study used all available observations for 597 individuals (59% of whom were women, mean age 62 years and BMI 31 kg/m²) as well as all 73 baseline variables which were available in the FNIH dataset. The investigators noted a clear separation among patients for whom knee pain had progressed and those for whom it had not.

The baseline variables which appeared to have the greatest influence on non-progression at 48 months included WOMAC [Western Ontario and McMaster Universities Arthritis Index] pain, lateral meniscal extrusion, and serum N-terminal pro-peptide of collagen IIA (PIIANP). Those variables which appeared to have the greatest effect on symptomatic knee pain progression included bone marrow lesions, osteophytes, medial meniscal extrusion, and urine C-terminal crosslinked telopeptide type II collagen.

Dr. Nelson summarized the results to OTW, “We were able to confirm the importance of bone marrow lesions, osteophytes, medial meniscal extrusion, and elevated urine CTX-II [C-terminal crosslinked telopeptide type II collagen] to progression of knee OA.”

“We also found that higher baseline pain, lateral meniscal extrusion, and serum PIIANP were associated with lack of progression. The key is that all of these associations were seen in a single analysis of all the data simultaneously, rather than many separate models of a few features at a time, as would be required by traditional methods. This increases the power of the analysis and also allows novel findings (that is, we could find things that we had not hypothesized in advance).”

“Currently, the results are confirmatory in that the FNIH cohort included only known/proven imaging and biochemical biomarkers and was not designed to explore novel or potential biomarkers. The features we found which were associated with progression of knee OA have been previously reported. The strength of this work is in the potential of these novel analytic methods to identify unexpected predictors or patterns in the data that could lead to novel pathways or potential interventions in the future.”