Combating Misinformation in the Digital Age: A Machine Learning Approach to Protect Community Water Fluoridation and Promote Oral Health Equity.

Objectives

Community water fluoridation (CWF) is an effective public health intervention for preventing dental caries. However, the widespread dissemination of misinformation on social media platforms, such as Twitter (now “X”), threatens public acceptance and may exacerbate oral health inequities. This study aimed to develop, evaluate, and deploy machine learning (ML) and deep learning (DL) models to identify misinformation about CWF on Twitter and assess implications for public health communication and surveillance.

Methods

We collected 19,960 English-language tweets about CWF posted between 2014 and 2024 using keyword-based queries. Tweets originated globally; however, only US-based geotagged tweets were used for sociodemographic analysis because reliable demographic and oral health surveillance data (e.g., US Census, BRFSS, NHANES) were available for linkage. Veracity was determined using authoritative public health criteria from the CDC, WHO, and ADA, with a subset of tweets manually annotated as factually correct or misinformation. Six machine learning and deep learning models were trained and evaluated. Additional analyses included sentiment scoring, thematic content coding, and geospatial-demographic comparisons.

Results

The Support Vector Classifier achieved the highest accuracy (91.6%). A hybrid BERT + XGBoost model (89.9% accuracy) was selected for deployment due to its strong performance and interpretability. Overall, 78.8% of tweets were classified as misinformation, with dominant themes including fluoride toxicity, distrust of government, and individual autonomy. Misinformation tweets were shorter, more engaging, and concentrated in socioeconomically disadvantaged areas with a high prevalence of misinformation tweets, where poverty rates and untreated dental caries were also greater. Sentiment analysis showed pro-CWF tweets were, on average, more positive.

Conclusions

ML and DL models can effectively detect CWF-related misinformation on social media. Integrated with equity-focused communication strategies, these tools may help sustain public trust in CWF and reduce misinformation-related oral health inequities.