ABSTRACT
OBJECTIVE: To study the prevalence of Macrophage Activation Syndrome (MAS) in children with Kawasaki disease (KD) and to devise a classification tree for predicting MAS in early KD based on easily available clinical and laboratory information using artificial intelligence (AI) technology. METHODS: A prospective cross-sectional observational study was conducted (March 2020 - October 2021) during which hospitalized children aged 1-18 years with KD were consecutively enrolled. Those with positive RTPCR test or IgM/IgG serology for COVID-19 were excluded. The clinical and laboratory profiles of children with and without MAS were studied. A multivariable logistic regression (LR) model was developed utilizing backward elimination method to determine the relationship between select candidate predictor variables and MAS in patients with KD. A classification tree was created based on these using artificial intelligence algorithms. RESULTS: Sixty-two children were diagnosed with KD during the study period, of these, 42 children with KD were included; 14 (33.3 %) were diagnosed with MAS. The median (IQR) duration of fever (days) was significantly more in MAS than those without MAS [7 (5, 15) vs 5 (5, 9), P < 0.05]. Serum albumin (g/dL) was significantly lower in those with MAS [2.3 (2.2, 2.7) vs 2.8 (2.3, 3.1), P = 0.03]. The classification tree constructed by the AI-based algorithm predicted that in children with KD who had myocardial dysfunction, serum albumin ≤ 2.8 g/dL and fever > 6 days duration at admission had increased likelihood of developing MAS. In children without myocardial dysfunction, alanine transaminase (ALT) levels > 70 U/L and fever > 5 days were equally predictive of MAS. CONCLUSION: Nearly one-third of the children with KD had MAS. Clinicians should consider screening all children with KD for MAS at admission. A classification tree based on the presence of myocardial dysfunction, duration of fever > 6 days, ALT levels and hypoalbuminemia can identify MAS in the course of KD.
