Figure S4.
The model uses lymphocyte counts, periodic fever, and chest or abdominal pain as input variables. The flowchart starts with a decision point based on lymphocyte counts being greater than or equal to 896. If yes, it leads to a node indicating a 0.30 predicted probability and 100 percent of patients. If no, it proceeds to another decision point evaluating periodic fever. If periodic fever is present, it further branches to a node with a 0.22 predicted probability and 82 percent of patients. This node then splits based on the presence of chest or abdominal pain. If chest or abdominal pain is present, it leads to a node with a 0.08 predicted probability and 50 percent of patients. If no chest or abdominal pain is present, it leads to a node with a 0.17 predicted probability and 24 percent of patients. If periodic fever is not present, it leads directly to a node with a 0.44 predicted probability and 32 percent of patients. The final node, reached if no periodic fever is present, indicates a 0.67 predicted probability and 18 percent of patients.
Decision tree model for clinical stratification. A CART was constructed using lymphocyte counts, periodic fever, and chest or abdominal pain as input variables. The values within each node represent the predicted probability of harboring a genetic diagnosis (top) and the proportion of patients classified into that node (bottom). Splitting criteria are shown along the branches. This exploratory model provides a clinically interpretable framework, illustrating how combinations of laboratory and clinical features can stratify the likelihood of underlying monogenic IEI in adults with RMD.