Inferring actual treatment pathways from patient records.

Area of Science:

• Computational biology
• Health informatics
• Machine learning

Background:

• Treatment pathways guide medical care but revising them requires analyzing complex health records.
• Inferring actual patient treatments from administrative health data is challenging due to coding schemes and lack of annotations.

Purpose of the Study:

• Develop a novel method to infer patient treatment pathways from administrative health records.
• Address technical gaps in current treatment pathway inference research.
• Create a robust framework for testing and validating pathway inference methods.

Main Methods:

• Introduce Defrag, a novel neural network (NN)-based method for inferring treatment pathways.
• Utilize a self-supervised learning objective to capture semantic and temporal meaning in healthcare event sequences.
• Develop a comprehensive testing and validation framework for pathway inference evaluation.

Main Results:

• Defrag effectively identifies treatment pathway fragments for breast cancer, lung cancer, and melanoma in public health records.
• Experiments with synthetic data demonstrate Defrag's characteristics and superior performance over non-NN baselines.
• The method significantly outperforms existing pathway inference techniques.

Conclusions:

• Defrag provides an innovative and effective approach for inferring treatment pathways from complex health data.
• While outperforming existing methods, comparing computationally derived pathways to clinical guidelines remains a challenge.
• Open-source code and a testing framework are provided to foster further research in treatment pathway inference.