Objectives: Using the ICD-9-CM codes with respect to reimbursement claims data of the Bureau of National Health Insurance, we aimed to demonstrate for Taiwanese emergency pediatrics departments that it was possible to screen out, firstly, the most erratic time points for disease volume and, secondly, the most unexpected abnormalities in diagnoses. Methods: The item response theory based Rasch model with Chi-square-like outfit mean square errors and standardized residual scores was applied to detect the volume of visits at emergency pediatric departments (EPD) by month from 2000 to 2006, and to monitor erratic stability and outbreaks in terms of timing with respect to diagnosis patterns. Results: Sixteen disease groups classified by ICD-9-CM were used to construct a unidimensional latent trait. The month with the most outfit mean square errors was July 2006, in which patients with mental disorders had an abnormally high frequency, with the next highest being May 2002. Furthermore, diseases of the musculoskeletal system and connective tissue exhibited dramatic fluctuations in volume. Furthermore, patient load was sharply decreased during the period of the SARS outbreak in 2003. It was determined that an outfit MNSQ value greater than 2.0 denoted an unexpected abnormity using the Rasch model for diagnosis volume at the EPDs over the past seven years. This group accounted for 8.3% of total patient volume. Conclusions: The Rasch model using a non-parametric approach with the distribution-free characteristics can be easily applied as real-time disease surveillance system to detect changes in hospital patient characteristics as well as form a foundation for further relevant and related research on patient patterns.