Using a Machine Learning Algorithm to Predict the Likelihood of Presence of Dental Caries among Children Aged 2 to 7

Abstract

Background: Dental caries is the most common chronic childhood infectious disease and is a serious public health problem affecting both developing and industrialized countries, yet it is preventable in most cases. This study evaluated the potential of screening for dental caries among children using a machine learning algorithm applied to parent perceptions of their child’s oral health assessed by survey. Methods: The sample consisted of 182 parents/caregivers and their children 2–7 years of age living in Los Angeles County. Random forest (a machine learning algorithm) was used to identify survey items that were predictors of active caries and caries experience. We applied a three-fold cross-validation method. A threshold was determined by maximizing the sum of sensitivity and specificity conditional on the sensitivity of at least 70%. The importance of survey items to classifying active caries and caries experience was measured using mean decreased Gini (MDG) and mean decreased accuracy (MDA) coefficients. Results: Survey items that were strong predictors of active caries included parent’s age (MDG = 0.84; MDA = 1.97), unmet needs (MDG = 0.71; MDA = 2.06) and the child being African American (MDG = 0.38; MDA = 1.92). Survey items that were strong predictors of caries experience included parent’s age (MDG = 2.97; MDA = 4.74), child had an oral health problem in the past 12 months (MDG = 2.20; MDA = 4.04) and child had a tooth that hurt (MDG = 1.65; MDA = 3.84). Conclusion: Our findings demonstrate the potential of screening for active caries and caries experience among children using surveys answered by their parents.

Keywords

• dental caries
• children
• oral health
• disparities
• machine learning algorithms
• random forest