Objectives: To predict the occurrence of inactive disease in JIA in the first 2 years of disease.
Methods: An inception cohort of 152 treatment-naïve JIA patients with disease duration <6 months was analysed. Potential predictors were baseline clinical variables, joint US, gut microbiota composition and a panel of inflammation-related compounds in blood plasma. Various algorithms were employed to predict inactive disease according to Wallace criteria at 6-month intervals in the first 2 years. Performance of the models was evaluated using the split-cohort technique. The cohort was analysed in its entirety, and separate models were developed for oligoarticular patients, polyarticular RF negative patients and ANA positive patients.
Results: All models analysing the cohort as a whole showed poor performance in test data [area under the curve (AUC): <0.65]. The subgroup models performed better. Inactive disease was predicted by lower baseline juvenile arthritis DAS (JADAS)-71 and lower relative abundance of the operational taxonomic unit Mogibacteriaceae for oligoarticular patients (AUC in test data: 0.69); shorter duration of morning stiffness, higher haemoglobin and lower CXCL-9 levels at baseline for polyarticular RF negative patients (AUC in test data: 0.69); and shorter duration of morning stiffness and higher baseline haemoglobin for ANA positive patients (AUC in test data: 0.72).
Conclusion: Inactive disease could not be predicted with satisfactory accuracy in the whole cohort, likely due to disease heterogeneity. Interesting predictors were found in more homogeneous subgroups. These need to be validated in future studies.