Revisiting Drug Recommendation From a Causal Perspective.

Area of Science:

• Medical Informatics
• Computational Chemistry
• Causal Inference

Background:

• Drug recommendation is crucial in healthcare, with molecular graphs offering potential insights.
• Current methods often fail to fully utilize molecular graph data or consider patient health context.
• Existing approaches struggle with complex molecular structures and spurious correlations.

Purpose of the Study:

• To develop a novel drug recommendation method using a causal perspective.
• To address limitations of existing methods, including inflexibility and disregard for patient health conditions.
• To improve the accuracy and reliability of drug recommendations.

Main Methods:

• Proposed a conditional causal representation learner for drug recommendation.
• Separated drug molecular representations into causal and spurious components based on patient health.
• Employed causal intervention to eliminate spurious correlations.

Main Results:

• Achieved new state-of-the-art performance on MIMIC-III and MIMIC-IV datasets.
• Demonstrated significant improvements, including 6.68% Jaccard improvement on MIMIC-III (p-value 0.05).
• Effectively reduced spurious correlations in drug molecular representations.

Conclusions:

• The proposed causal approach significantly enhances drug recommendation accuracy.
• This method offers a more robust and reliable way to prescribe medications by considering patient-specific factors.
• Future work can explore further applications of causal inference in personalized medicine.