Development and internal validation of diagnostic prediction models using machine-learning algorithms in dogs with...

Area of Science:

• Veterinary Medicine
• Machine Learning
• Canine Health

Background:

• Canine hypothyroidism is frequently misdiagnosed, leading to potential overtreatment or delayed diagnosis.
• Diagnostic prediction models can improve clinical decision-making and reduce unnecessary testing.

Purpose of the Study:

• To develop and internally validate machine learning models for diagnosing hypothyroidism in dogs.
• To assess the accuracy of models using clinical signs and clinicopathological variables.

Main Methods:

• A cross-sectional study analyzed data from 82 hypothyroid and 233 euthyroid dogs.
• Four models were created using combinations of clinical signs and lab values (e.g., cholesterol, tT4, cTSH).
• Six machine learning algorithms were applied, with internal validation via 10-fold cross-validation and AUROC analysis.

Main Results:

• Models incorporating thyroid function tests (tT4, cTSH) achieved higher accuracy (AUROC up to 0.99).
• The best models demonstrated high positive (up to 0.97) and negative (up to 0.99) predictive values.
• Naive Bayes, logistic regression, and random forest algorithms showed strong performance depending on the model.

Conclusions:

• Machine learning models show high accuracy for predicting canine hypothyroidism based on internal validation.
• Further external validation is necessary to confirm the clinical utility of these diagnostic tools.