Risk assessment for canine periodontal disease using a hybrid causal Bayesian network.

Area of Science:

• Veterinary Medicine
• Epidemiology
• Artificial Intelligence

Background:

• Canine periodontal disease is common but underdiagnosed, creating a gap in preventive veterinary care.
• Disease risk is influenced by genetics, age, breed, and modifiable factors like dental hygiene.
• Current evidence-based interventions for canine periodontal disease are underutilized.

Purpose of the Study:

• To develop and validate a hybrid Bayesian network for canine periodontal disease risk assessment.
• To integrate diverse data sources for a comprehensive understanding of disease probability.
• To provide a tool supporting clinical decision-making in veterinary preventive medicine.

Main Methods:

• Constructed a directed acyclic graph (DAG) to map causal relationships of risk factors.
• Developed a Bayesian network integrating electronic health records, owner questionnaires, and expert knowledge.
• Validated the network using four independent datasets, assessing performance metrics like ROC AUC, sensitivity, and specificity.

Main Results:

• The network successfully differentiated high-risk from low-risk breeds and identified associations with age, size, head shape, and hygiene.
• Key clinical indicators significantly increased periodontal disease probability (e.g., gingivitis to 47.0%).
• Validation datasets showed robust performance with ROC AUC values up to 0.962.

Conclusions:

• The hybrid Bayesian network effectively quantifies canine periodontal disease risk by integrating complex interactions.
• The model's bidirectional inference capability supports both probabilistic and causal reasoning for clinical applications.
• This approach demonstrates the utility of Bayesian networks for complex veterinary conditions, improving diagnostic accuracy and preventive strategies.