Development and validation of diagnostic and prognostic prediction tools for dental caries in young children: A protocol

Introduction

Dental caries is the most common oral disease worldwide, affecting up to 90% of children globally. It can lead to pain, infection, and impaired quality of life. Early prevention is a key strategy for reducing the prevalence of dental caries in young children. Valid and reliable diagnostic or prognostic tools that enable accurate individualised prediction of current or future dental caries are essential for facilitating personalised caries prevention and early intervention. However, no efficacious tools currently exist in early childhood—the optimal period for disease prevention. We aim to develop and validate diagnostic and prognostic prediction tools for dental caries in young children, utilising a combination of environmental, physical, behavioural and biological early life data.

Methods and analysis

Data sources include two prospective studies, with a total sample size of approximately 600 children. These cohorts have collected detailed demographic, antenatal, perinatal and postnatal data from medical records and parent-completed questionnaires and biological samples including a dental plaque swab. Candidate predictor variables will include sociodemographic characteristics, health history, behavioural, and microbiological characteristics. The outcome variable will be the presence, incidence, or severity of dental caries diagnosed using the International Caries Detection and Assessment System (ICDAS). Statistical and machine learning approaches will be utilised for selection of predictor variables and model development. Internal validation will be conducted using resampling methods (i.e. bootstrapping) and nested cross validation. Model performance will be evaluated using standard performance metrics such as accuracy, discrimination, and calibration. Where feasible, external validation will be performed in an independent cohort. Model development and reporting will be guided by the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD) statement and the Prediction model Risk Of Bias Assessment Tool (PROBAST) guidelines.

Discussion

This protocol describes data collection procedures, outcome and predictor variable configuration, and planned omics-based diagnostic and prognostic prediction analyses. The study employs a discovery-driven approach for the development and validation phases, allowing findings from select steps to inform subsequent stages.

Ethics and dissemination

This study has ethical and governance approval from The Royal Children’s Hospital Melbourne Human Research Ethics Committee (HREC/111803/RCHM-2024).

Strengths and limitations of this study

• Analysis of two rare, prospectively collected datasets from early childhood, comprising longitudinal and cross-sectional early-life questionnaire data, oral microbiome profiles, and clinical dental assessments.

• Use of a gold-standard dental caries measurement based on ICDAS, providing a validated, standardised, and detailed evaluation of caries status.

• Rigorous study design supported by a comprehensive and pre-specified data analysis plan.

• Variability in early-life questionnaire data collection across cohorts may introduce heterogeneity, potentially affecting the comparability and transportability of the prediction models.

• The lack of external validation in an independent dataset beyond the research setting may limit the generalisability of the findings to broader paediatric populations.