Phenotype Modeling and Outcome Mapping for Pain Management Decision Support

Chronic pain is a major medical challenge: treatment with medications may lead to depression, hopelessness, and drug dependence. The condition creates a large financial drain on social services, insurance providers, and family resources. Lacking independent tests for pain intensity, diagnosis depends on discussion between the patient and the health professional, and can be quite subjective. Effectiveness of treatment is variable from patient to patient and is frequently an expensive process of trial and error. Improving the understanding between how a patient presents, their phenotype, and their genetics, is one path to reduce the subjective nature of chronic pain treatment. Utilizing health information technology to create chronic-pain phenotypes has the potential to vastly improve evidence-based protocols for pain management. Included in these phenotypes are the physical, emotional, and social status of a patient; the influence of these on the intensity of pain; the impact of pain on them; and the response of the patient to treatment.

The overall goal of this project was to demonstrate proof-of-principle that chronic-pain phenotypes could be developed from electronic medical record (EMR) data and linked to patient treatments and outcomes. Phenotypes were developed by applying natural language processing (NLP) principles to progress notes in the EMR, routine patient questionnaires, and practice management data. This study was conducted at Michigan Pain Consultants, a large community specialist practice.

The specific aims of this study were as follows:

• Identify, extract, and organize data to support phenotype-intervention-outcome model construction. 
• Iteratively construct and evaluate pain phenotype-intervention-outcome models until optimized to available data.

While the investigators made great progress with the first aim during the project period, the modeling for the second aim continues to remain a challenge and was ongoing at the end of the grant funding. This project furthered the investigators’ goal of creating phenotypes for individuals with chronic pain. In addition, the team now has a demonstrated NLP approach to apply once the majority of the progress note concepts are extracted and analyzed. The results will inform completion of the modeling and help the investigators toward their ultimate goal of creating clinical decision support tools for, and reducing the subjective nature of, pain management in the primary care setting.