Skip to main content
2026 AAPD Annual Meeting Main banner
Cheng Crowns
Information & Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Presentation Icons
Ticket Required
Ticket Required
Sold Out
Sold Out
Recorded Session
Recorded Session
Rate ePoster

Caries

143 - AI-Predicted Versus Clinician-Selected Treatment Decisions for Caries Management in Primary Dentition

Friday, May 22, 2026
2:40 PM - 2:55 PM PST
Has Audio

    Presenting Author(s)

  • Travis R. Wagner, DDS (he/him/his)

    Resident
    The Ohio State University, Columbus, OH
    Columbus, Ohio, United States

    • Co-Author(s)

  • Daniel Claman, DDS

    Chief of Dentistry
    Nationwide Children's Hospital
    The Ohio State University/Nationwide Children's Hospital
    Columbus, Ohio, United States

  • JP

    Jin R. Peng, MD, MS. PhD

    The Abigail Wexner Research Institute at Nationwide Children’s Hospital

  • Andréa Zandoná, DDS, MS, PhD (she/her/hers)

    Executive Associate Dean and Professor
    The Ohio State University College of Dentistry
    The Ohio State University College of Dentistry
    Columbus, Ohio, United States

    • Research Mentor(s)

  • Daniel Claman, DDS

    Chief of Dentistry
    Nationwide Children's Hospital
    The Ohio State University/Nationwide Children's Hospital
    Columbus, Ohio, United States

    • Program Director(s)

  • Kimberly J. Hammersmith, DDS, MPH, MS

    Associate Professor
    The Ohio State University/ Nationwide Children's Hospital
    The Ohio State University and Nationwide Children's Hospital
    Columbus, Ohio, United States

Abstract:


Purpose: The purpose of this study was to develop and evaluate an artificial intelligence (AI) model that integrates demographic, social, medical, behavioral, and tooth-level clinical factors to predict treatment decisions for carious primary teeth and to compare AI-predicted treatment decisions to clinician-selected treatment decisions.



Methods: This retrospective study included 8,403 pediatric patients contributing 29,639 primary teeth with caries. Predictor variables include patient demographics, social determinants of health, dental and medical history, behavioral factors, and tooth-level characteristics such as tooth type and caries extent. Tooth-level treatments were categorized into five mutually exclusive outcome groups based on CDT procedure codes: preventative care, direct restoration, full coverage restoration, pulp therapy, and extraction. Multinomial logistic regression (LASSO) and gradient boosted (XGBoost) AI models were trained using predictor variables and treatment categories, then generated independent treatment decisions for carious lesions utilizing the predictor variables alone. Model performance was evaluated using accuracy, area under the receiver operating characteristic curve, and confusion matrices. Agreement between AI-predicted and clinician-completed treatments was assessed using weighted kappa statistics and class-specific performance metrics.


Results: Final results are currently pending. The median patient age at visit was 6 years (IQR: 5–8), with a mean age of 6.3 years. The cohort was 52% male, 38% White, and 36% Black, with 27% of patients reporting a non-English preferred language. At the tooth level, full coverage restorations accounted for 40% of treatments, followed by extraction (24%), preventative care (20%), pulp therapy (6%), direct restorations (5%), and no treatment (5%). Overall, the XGBoost model produced 64% accuracy in predictions (95% CI: 62.5-64.5%, p<0.001), with a weighted kappa score of 0.46 indicating moderate agreement with clinician-selected treatment plans.

Identify Supporting Agency and Grant Number: