Machine Learning and Free-Text Analysis of Notes Improves Patient Identification

MedicalResearch.com Interview with:

Saul Blecker, MD, MHS Department of Population Health New York University Langone School of Medicine, New York, NY 10016

Dr. Saul Blecker,

Saul Blecker, MD, MHS
Department of Population Health
New York University Langone School of Medicine,
New York, NY 10016

Saul.Blecker@nyumc.org

MedicalResearch.com: What is the background for this study? What are the main findings?

Response: The identification of conditions or diseases in the electronic health record (EHR) is critical in clinical practice, for quality improvement, and for clinical interventions. Today, a disease such as heart failure is typically identified in real-time using a “problem list”, i.e., a list of conditions for each patient that is maintained by his or her providers, or using simple rules drawn from structured data. In this study, we examined the comparative benefit of using more sophisticated approaches for identifying hospitalized patients with heart failure.

MedicalResearch.com: What is the background for this study? What are the main findings?

Response: We found that the problem list missed about half of all heart failure patients. Additionally, a simple rule using structured data captured about 80% of heart failure cases but also captured nearly as many cases of patients who did not have heart failure, resulting in many false positives. We found that the best performing approaches used machine learning with unstructured data such as provider notes and radiology reports. These approaches captured 80% of heart failure patients with a precision (positive predictive value) of 90%.

MedicalResearch.com: What should readers take away from your report?

Response: Real-time identification of hospitalized patients is best achieved through analysis of free text notes and reports. Though there is no doubt that implementing such algorithms will help with clinical care and quality improvement efforts, they may require significant information technology resources. At our institution, we are working on implementing these algorithms which we feel will benefit our transitional care team and for EHR based clinical decision support to improve quality of heart failure care. It is important to note that, as EHR vendors begin to natively support such machine learning algorithms, there may be less tradeoff between the benefits versus the cost of implementation.

MedicalResearch.com: What recommendations do you have for future research as a result of this study?

Response: The most exciting finding is that machine learning and free text analysis (advanced analytics) has the potential to improve patient identification. Through this approach there is great opportunity to improve care and outcomes for patients. Future work should examine whether these findings can be translated to other hospital systems, other diseases and other patient settings to improve care across populations.

MedicalResearch.com: Thank you for your contribution to the MedicalResearch.com community.

Citation:

Blecker S, Katz SD, Horwitz LI, Kuperman G, Park H, Gold A, Sontag D. Comparison of Approaches for Heart Failure Case Identification From Electronic Health Record Data. JAMA Cardiol. Published online October 05, 2016. doi:10.1001/jamacardio.2016.3236

Note: Content is Not intended as medical advice. Please consult your health care provider regarding your specific medical condition and questions.

More Medical Research Interviews on MedicalResearch.com

No Comments

Post A Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.