Application of machine learning techniques for warfarin dosage prediction: a case study on the MIMIC-III dataset

Area of Science:

• Medical Informatics
• Computational Biology
• Pharmacogenomics

Background:

• Warfarin dosing is complex due to narrow therapeutic range and patient variability.
• Accurate prediction of international normalized ratio (INR) is crucial for safe and effective anticoagulation.
• Missing data in electronic health records poses a significant challenge for warfarin management.

Purpose of the Study:

• To enhance the accuracy of INR prediction using machine learning on the MIMIC-III dataset.
• To address the challenge of missing data in predicting patient response to warfarin.
• To develop a more personalized and precise warfarin dosing strategy.

Main Methods:

• Applied dimensionality reduction techniques: Principal Component Analysis (PCA) and t-distributed Stochastic Neighbor Embedding (t-SNE).
• Utilized advanced data imputation methods: Denoising Autoencoders (DAE) and Generative Adversarial Networks (GAN).
• Integrated these machine learning approaches for improved INR prediction.

Main Results:

• Achieved significant improvements in predictive accuracy for INR.
• Substantially reduced prediction errors compared to traditional warfarin dosing approaches.
• Demonstrated the efficacy of ML models in handling missing data for anticoagulation therapy.

Conclusions:

• Machine learning models offer a powerful tool for personalized warfarin dosing.
• The proposed methods can significantly reduce the risk of adverse drug events.
• This approach has potential clinical applications for enhancing anticoagulation therapy and other complex treatments with missing data.