Incorporating comorbidities into latent treatment pattern mining for clinical pathways

Area of Science:

• Health Informatics
• Computational Medicine
• Data Mining in Healthcare

Background:

• Clinical pathway (CP) design is complex due to multiple diagnoses and comorbidities.
• Existing data mining techniques often overlook the impact of comorbidities on treatments.
• Understanding comorbidity influence is crucial for comprehensive CP analysis.

Purpose of the Study:

• To extract latent treatment patterns considering both first-diagnosis and comorbidities.
• To unveil associations between diagnoses (including comorbidities) and treatments.
• To quantify the contribution of comorbidities to treatment patterns within clinical pathways.

Main Methods:

• Proposed a generative statistical model extending latent Dirichlet allocation.
• Incorporated an additional layer for diagnosis modeling to capture comorbidity effects.
• Validated the model on a real-world clinical dataset of 12,120 unstable angina patient traces.

Main Results:

• Discovered three distinct latent treatment patterns from the clinical data.
• Identified latent correlations between comorbidities and specific treatments.
• Demonstrated the model's ability to detect comorbidity-focused patterns and treatment adaptations.

Conclusions:

• The proposed model effectively extracts meaningful latent treatment patterns.
• It highlights the significant influence of comorbidities on treatment selection in clinical pathways.
• Offers potential applications in personalized treatment recommendation systems.