404 - Validating a Predictive Algorithm for Inappropriate Antibiotic Prescribing

Publication Number: 2392.404

Samantha Ratner, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, PA, United States; Jeffrey S. Gerber, Childrens Hospital of Philadelphia, Philadelphia, PA, United States; Torsten Joerger, Childrens Hospital of Philadelphia, Philadelphia, PA, United States; Ziyi Wang, Children’s Hospital of Philadelphia, Exton, PA, United States

Background: A core principle of antibiotic stewardship is reducing unnecessary antibiotic prescriptions, which necessitates developing and validating methods to evaluate appropriateness. A seminal CDC study created a tier-based algorithm for classifying antibiotic prescriptions by dividing ICD-9 codes into 3 tiers: antibiotics almost always indicated (Tier 1), may be indicated (Tier 2), and not indicated (Tier 3). The authors concluded that approximately 30% of outpatient antibiotic prescriptions in the United States were inappropriate. This algorithm has been updated to ICD-10 codes and used in dozens of influential studies to determine the frequency of inappropriate antibiotic prescribing and measure the effectiveness of stewardship initiatives. However, despite its widespread use, this algorithm has not been validated in pediatric populations.

Objective: This study aimed to assess the performance of the CDC antibiotic tier classification system in pediatric outpatients.

Design/Methods: The CDC antibiotic appropriateness algorithm was applied to encounters in a large pediatric ambulatory care network containing 32 academic and community-based primary care practices, 4 urgent care centers, and 2 emergency departments serving a diverse population of more than 300,000 children with almost 1 million encounters per year from January 1st to December 31st, 2024. Subsequently, 300 randomly selected encounters classified as Tier 3 were manually reviewed using a structured validation protocol to determine antibiotic appropriateness derived from AAP, IDSA, and AAFP guidelines. Percent appropriate was calculated, and 95% confidence intervals (CI) were reported overall and stratified by location. Based on common antibiotic stewardship protocols, we also explored the contribution of watchful waiting prescriptions.

Results: There were 269,480 ambulatory encounters of which 114,590 (42.5%) were associated with an antibiotic prescription. The CDC algorithm classified 10,370 (8.2%) encounters as Tier 3 (antibiotics not indicated), and 78,142 (68%) as Tier 2 (may be indicated). Of the 300 Tier 3 encounters randomly selected for chart review, 171 (57%) [weighted by location: 0.58, 95% CI (0.52-0.64)] were potentially appropriate based on manual chart review. 64 encounters (21.3%) were "watchful waiting" prescriptions, of which 46.9% were appropriate.

Conclusion(s): The established CDC algorithm did not reliably predict antibiotic appropriateness in our pediatric outpatient population, demonstrating a need for pediatric-focused algorithms for accurately quantifying and characterizing antibiotic appropriateness.

Figure 1: Flowchart
9% of antibiotic prescriptions were associated with Tier 3 encounters, 68% were associated with Tier 2​.

Figure 2
57% of antibiotic prescriptions for Tier 3 encounters were appropriate.