The United States Preventive Services Task Force (USPSTF) Guidelines vs. Electronic Health Record (EHR)-Based...

Area of Science:

• Oncology
• Medical Informatics
• Artificial Intelligence

Background:

• Current prostate cancer screening relies on age-based criteria from the United States Preventive Services Task Force (USPSTF).
• This approach may lead to missed aggressive cancers or over-screening in men outside the typical age range.
• Personalized prediction models incorporating diverse medical histories show promise for improved accuracy.

Purpose of the Study:

• To develop an AI-driven predictive model for prostate cancer using patient medical history.
• To compare the performance of this AI model against traditional age-based screening criteria.

Main Methods:

• Developed a predictive model using electronic health records (EHRs) from the All of Us database.
• Utilized Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression to identify key predictive features from prior health conditions.
• Assessed model performance using McFadden's R² and area under the receiver operating characteristic curve (AUROC).

Main Results:

• The EHR-based AI model achieved a McFadden's R² of 0.36, significantly outperforming the USPSTF age-based model (R² of 0.20).
• AUROC validation confirmed superior sensitivity and specificity of the AI model compared to current screening criteria.
• The model demonstrated enhanced predictive accuracy using non-invasive EHR data.

Conclusions:

• AI-driven prediction models utilizing EHR data offer a more accurate method for prostate cancer risk assessment.
• Personalized screening strategies can improve early detection and reduce invasive procedures associated with unnecessary screening.
• This approach enables focused diagnostic efforts on individuals most at risk, optimizing healthcare resources.