Background Newly diagnosed childhood immune thrombocytopenia (ITP), an acquired autoimmune bleeding disease, has a good prognosis: 60-70% of patients recover spontaneously 3 months after diagnosis (transient ITP), whereas 10-20% remain thrombocytopenic beyond 12 months (chronic ITP). A key clinical challenge is the early identification of a patient's disease course to counsel families, inform treatment decisions, and guide additional diagnostics, e.g. screening for systemic autoimmune diseases, immunodeficiencies or genetic thrombocytopenia. Several clinical predictors have been proposed (Heitink-Pollé et al. Blood 2014;142(22)), but it is unclear how they can be integrated to predict disease outcomes.
Objective To develop and validate a clinical prediction model for transient vs. prolonged disease courses in children with newly diagnosed ITP, using clinical characteristics at diagnosis.
Study design Model development and validation in a multinational prospective observational cohort; external validation in a multicenter randomized controlled trial.
Methods Using modern statistical methods, we extended a score by Edslev et al. (BJH 2007; 138) into an updated, multivariate prediction score, using individual patient data from newly diagnosed childhood ITP patients included in the observational, prospective Nordic Pediatric Hematology-Oncology ITP study (NOPHO; N=377; data shared by original investigators). Transient ITP was defined as complete recovery by platelet count 3 months after diagnosis (NOPHO, ≥150x109/L; TIKI, ≥100x109/L). The model was developed by penalized regression (Ridge) with ten-fold cross-validation in patients included during the first half of the NOPHO study period (derivation cohort, N=233) and subsequently validated in the second half (validation cohort, N=144). External validation was performed on children with newly diagnosed ITP included in the Dutch randomized controlled trial Treatment With or Without IVIg for Kids With ITP (TIKI; N=200; N=100 randomized to IVIg and N=100 carefully observed; Heitink-Pollé et al.Blood 2018; 132(9)). Inclusion criteria of both studies included a diagnosis platelet count ≤20x109/L and age below 16 years.
Results Case-mix analyses showed that TIKI and NOPHO cohorts had comparable baseline characteristics, considering age, gender, preceding infections and bleeding. The rate of transient ITP was 67% (NOPHO) and 73% (TIKI). Seven predictors were included in the model: age (years; penalized odds ratio [OR] for transient ITP, 0.97), male gender (OR, 1.07), presence of mucosal bleeding (OR, 1.27), preceding infection (OR, 1.27) or vaccination (OR, 0.99), insidious disease onset (> 14 days; OR, 0.41) and diagnosis platelet count (x109/L; OR, 0.99). We evaluated the clinical prediction model in two independent groups: patients with transient ITP were discriminated with a receiver-operating characteristic AUC of 0.72 (95% CI, 0.61 - 0.84) for the NOPHO validation cohort and 0.71 (95% CI, 0.62 - 0.80) for the TIKI cohort. Additional analyses in the TIKI cohort revealed a similar classification accuracy for patients randomized to IVIg and observation only. At a posterior probability ≥65%, the positive predictive value for transient ITP was 0.85 and negative predictive value was 0.41. For long-term follow-up, when patients were grouped in low, intermediate and high score terciles (obtained in the NOPHO derivation cohort), the rate of persistent ITP after six months was 60/36/14 % (low/medium/high scores; NOPHO) and 50/27/10 % (TIKI). Furthermore, the rate of chronic ITP at twelve months follow-up was 27/16/6 % for the same score terciles (TIKI). Finally, the model score correlated with cessation of mucosal bleeding as well as any bleeding during five clinical follow-up visits during twelve months after diagnosis.
Conclusion The updated clinical prediction model for transient ITP, i.e. recovery from ITP three months after diagnosis, showed adequate performance in two independent validation cohorts. Next to short-term recovery, long-term recovery six and twelve months after diagnosis and bleeding symptoms were predicted. This prediction model may allow the targeting of intensive monitoring or additional diagnostic efforts to children with low scores. On the other hand, in children with high scores, the rate for persistent and chronic ITP is low and they could be monitored `hands-off`, if this is intended.
Disclosures
No relevant conflicts of interest to declare.
Minimum sample size for external validation of a clinical prediction model with a binary outcome
