BPLT+: a Bayesian-based personalized recommendation model for health care.

Area of Science:

• Medical Informatics
• Machine Learning
• Health Data Science

Background:

• Personalized medicine requires accurate prediction of patient needs.
• Recommending relevant laboratory tests is crucial for effective patient care.
• Existing methods may lack robustness or personalization in test recommendations.

Purpose of the Study:

• To propose and evaluate an Advanced Bayesian-based Personalized Laboratory Tests recommendation (BPLT(+)) model.
• To estimate the probability of a patient taking a specific laboratory test.
• To enhance the effectiveness and robustness of laboratory test recommendations.

Main Methods:

• Developed a Bayesian-based weighting function to assess the likelihood of a laboratory test being taken by a patient.
• Integrated modified smoothing techniques for improved model performance.
• Proposed a novel evaluation framework involving training, validation sets, and a custom metric (CorrectRateM).

Main Results:

• The BPLT(+) model demonstrated effectiveness in recommending laboratory tests.
• Experiments on real medical data validated the model's predictive capabilities.
• The proposed evaluation framework provided objective assessment of the model's performance.

Conclusions:

• The Advanced Bayesian-based Personalized Laboratory Tests recommendation (BPLT(+)) model is effective for personalized healthcare.
• The model accurately predicts patient laboratory test uptake.
• The study highlights the potential of Bayesian methods and smoothing techniques in medical test recommendations.