Target association rule mining to explore novel paediatric illness patterns in emergency settings

Area of Science:

• Pediatric Critical Care Medicine
• Data Mining in Healthcare
• Clinical Informatics

Background:

• Hospitalized children in pediatric emergency departments (EDs) present complex clinical scenarios.
• Identifying patterns in clinical and laboratory attributes is crucial for understanding pediatric critical illness.

Purpose of the Study:

• To apply association rule mining (ARM) to discover novel patterns among clinical and laboratory attributes of sick children admitted to the pediatric ED.
• To analyze the association of these attributes with disease patterns and treatment outcomes.

Main Methods:

• An observational study enrolled 158 children aged 1 month to 12 years.
• Hotspot data mining, specifically ARM, was employed to analyze clinical data, laboratory investigations, and predefined outcome parameters.

Main Results:

• Thirty association rules were identified, linking various clinical and laboratory parameters to patient discharge.
• Key predictors for discharge included duration of hospitalization, lactate levels, platelet count, serum potassium, SBP, PaO2, and Glasgow Coma Scale scores.

Conclusions:

• Association rule mining is an effective technique for uncovering meaningful patterns in pediatric critical illness using clinical data.
• ARM can aid in analyzing the relationship between clinical attributes, disease characteristics, and the success of interventions.