Structural and Pain Progressors vs Nonprogressors in Knee Osteoarthritis

Patient receiving therapy for knee osteoarthritis.
Researchers used machine learning to predict structural and/or pain progression in patients with knee osteoarthritis.

The use of a structure (S) predicted progression score and pain (P) predicted-progression score, as provided by machine learning (ML) models, allowed for differentiation between actual progressors and nonprogressors over a 2-year follow-up period in patients with knee osteoarthritis (OA), according to study results published in Rheumatology (Oxford).

The IMI-APPROACH study ( identifier: NCT03883568) was conducted in the Netherlands. Researchers assessed the actual 2-year progression in the IMI-APPROACH study with respect to predicted disease progression scores. For each participant, an index knee was chosen based on American College of Rheumatology (ACR) clinical criteria for knee OA. If both of a participant’s knees met the ACR criteria, the most painful knee, according to the participant, was designated as the index knee. If both of the knees were deemed equally painful, the right knee was selected as the index knee.

Minimum Joint Space Width (minJSW) was used to measure actual structural progression. The Knee Injury and Osteoarthritis Outcome Score (KOOS) pain questionnaire was used to evaluate actual pain progression. Progression was defined as actual change (Δ) after 2 years, as well as progression over 2 years based on a per-patient fitted regression line that used 0-, 0.5-, 1-, and 2-year values.

A total of 297 patients were included in the study. The mean age was 66.5±7.1 years at baseline; the women-to-men ratio was 230:67. Participants’ mean body mass index was 28.1±5.3 kg/m2. The mean minJSW was 2.5 (95% CI, 2.4-2.7); the mean KOOS pain score was 66.4 (95% CI, 64.2-68.5).

Of the 297 original participants, 249 completed the study. The S and P predicted-progression scores did not differ significantly between those participants who completed the study and those who withdrew from the study. The main reason for study withdrawal was an individual’s unwillingness to visit a hospital during the COVID-19 pandemic.

Initially, actual structural progressors were assigned higher S predicted-progression scores compared with structural nonprogressors. Actual pain progressors were given higher P predicted-progression scores compared with pain nonprogressors.

The area under the receiver operating characteristic curve (AUC-ROC) for the S predicted-progression score to identify actual structural progressors was poor (ie, 0.612 and 0.599 for ΔminJSW and regression minJSW, respectively). This demonstrated that the S predicted-progression score was not as likely to distinguish actual structural progressors from structural nonprogressors.

Conversely, the AUC-ROC for the P predicted-progression score to identify actual pain progressors was good (ie, 0.817 and 0.830 for ΔKOOS pain and regression KOOS pain, respectively).Therefore, the P predicted-progression score was better able to differentiate between actual pain progressors and pain nonprogressors than the S predicted-progression score for distinguishing actual structural progressors from structural nonprogressors.

A major limitation of the study is the fact that the participants were chosen from multiple different European OA cohorts. Different data collection points were used, which led to different types of historical data for each cohort. In future trials, the study authors suggested the use of uniform data collection protocol when selecting participants from multiple cohorts.

The study authors concluded that, “More uniformly acquired data are needed to adjust the models to improve the accuracy of the S and P predicted-progression scores so that, in future trials, the use of ML models might improve patient selection by increasing the number of actual structural and/or pain progressors and with that reduce the trial sample size.”

Disclosure: Some of the study authors have declared affiliations with biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of authors’ disclosures. 


van Helvoort EM, Jansen MP, Marijnissen ACA, et al. Predicted and actual 2-year structural and pain progression in the IMI-APPROACH knee osteoarthritis cohort. Rheumatology (Oxford). Published online May 16, 2022. doi:10.1093/rheumatology/keac292