Classification of Opioid Usage Through Semi-Supervised Learning for Total Joint Replacement Patients.

Area of Science:

• Medical Informatics
• Machine Learning
• Public Health

Background:

• Opioid misuse and overdose are significant public health issues in the US, linked to over-prescription for pain management.
• Current opioid prescription practices often use a one-size-fits-all approach, leading to excess medication and potential misuse.
• Opioids are commonly prescribed after total joint replacement surgeries, highlighting a need for personalized dosage prediction.

Purpose of the Study:

• To develop a machine learning model for predicting individual patient opioid usage in the first two weeks post-discharge after joint replacement surgery.
• To address data quality issues caused by missing survey responses regarding patient opioid consumption.
• To improve the accuracy of patient opioid usage classification and inform personalized prescription adjustments.

Main Methods:

• Collected electronic health records, patient prescription history, and consumption survey data.
• Developed a semi-supervised learning model incorporating Bayesian regression to predict missing survey data (opioid amounts).
• Assigned pseudo-labels to patients with predicted opioid usage to enhance classification model performance.

Main Results:

• The proposed semi-supervised learning model demonstrated improved patient classification performance compared to standard methods.
• The model effectively predicted missing opioid consumption data, thereby improving overall data quality.
• Experimental results confirmed the efficacy of the approach in handling incomplete survey data.

Conclusions:

• The developed machine learning model can accurately predict short-term opioid usage in patients undergoing joint replacement surgery.
• Personalized opioid prescription based on predicted usage can help manage opioid prescriptions and pain intervention effectively.
• This approach has the potential to reduce opioid over-prescription, misuse, and associated public health risks.