Background/Purpose:   The Research in Arthritis in Canadian Children emphasizing Outcomes cohort (ReACCh-Out) showed a 45% chance of attaining inactive disease within a year of diagnosis with conventional JIA treatments.  Children who do not attain inactive disease may miss a window of opportunity to start aggressive treatment.  Here, we used ReACCh-Out data to develop models to assign probability of early remission on medication (ERM) for each child with JIA at diagnosis.

Methods: ReACCh-Out recruited 1497 patients newly diagnosed with JIA in 2005-2010 and followed them for five years or until May 2012.  Children were included in this study if they 1) were enrolled within 90 days of diagnosis, 2) received conventional treatment (no biologics or triple DMARD within 6 months of diagnosis) and 3) had enough follow-up to determine the outcome.  Outcome (ERM): at least six months of inactive disease with the first inactive disease visit occurring within one year of diagnosis.  Documentation of any of the following within that period meant ERM was not attained: an active joint, enthesitis, a physician global assessment of 1 or more in a 10cm scale, systemic JIA manifestations, active uveitis, corticosteroid eye drops, morning stiffness >15 min, ESR >20 mm/h, CRP >5 mg/L.  Patients who discontinued treatment during that period and remained inactive were still counted as having attained ERM.  Eligible subjects were randomly split into a training set to develop candidate models (75% of subjects) and a test set to determine their accuracy (25% of subjects).  Missing data on predictors were imputed using multiple imputation (20 datasets).  Data splitting and model fitting were repeated 10 times for each imputed dataset to assess model stability.

Results: We included 916 children enrolled a median of 2 days after JIA diagnosis, of whom 409 (44.7%) attained ERM.  Among 50-plus assessed variables, 17 were positively or negatively associated with ERM (Table).  A logistic regression model combining these variables had a c-index of 0.62 (95% CI 0.59, 0.66).  Although predicted and observed frequencies of ERM paralleled each other (Figure), the model’s c-index was similar to using JIA category alone (c-index of 0.59; 95% CI 0.56, 0.63).  C-index values >0.70 are considered helpful prediction.

Conclusion: In this cohort, many variables easily available at diagnosis were associated with ERM in children with JIA.  However, a prediction model combining the variables was only marginally better than using JIA category alone.  Novel biomarkers and modelling methods may be needed to improve prediction accuracy.