Validation of an Automated System for Identifying Complications of Serious Pediatric Emergencies

Area of Science:

• Pediatric health outcomes
• Health informatics
• Quality measurement

Background:

• Complications of pediatric illness are adverse outcomes specific to each condition.
• Accurate detection of these complications in administrative data is crucial for quality measurement but remains unclear.
• Administrative data offers potential for widespread quality assessment if complication detection is reliable.

Purpose of the Study:

• To evaluate the accuracy of an automated method for detecting pediatric illness complications using administrative data.
• To compare automated detection with manual health record review for 14 serious pediatric conditions.
• To determine the utility of administrative data for quality measurement of pediatric illness complications.

Main Methods:

• A cross-sectional study analyzed 1534 patient encounters from a pediatric emergency department (2012-2019).
• An automated method using disposition, diagnosis, and procedure codes was developed to identify complications.
• The automated method's performance was assessed against manual record review using positive predictive values (PPVs) and negative predictive values (NPVs).

Main Results:

• The automated method achieved >80% PPVs and NPVs for 8 of 14 conditions, including appendicitis, sepsis, and meningitis.
• Lower PPVs were observed for diabetic ketoacidosis (DKA), empyema, ovarian torsion, and septic arthritis.
• A lower negative predictive value (NPV) was noted for stroke, indicating potential for missed complications.

Conclusions:

• An automated method using administrative data reliably detects complications for several serious pediatric conditions.
• The tool is effective for quality measurement in appendicitis, meningitis, sepsis, and other conditions with high PPVs/NPVs.
• For conditions with lower accuracy, the automated method can serve as a screening tool to identify cases for manual review.