Leveraging Clinical Informatics to Improve Mental Health Care
Thursday, December 2, 2021 – 4:00 – 5:00 pm
Dr. Juliet Beni Edgcomb MD PhD, Fellow Physician at the Division of Child and Adolescent Psychiatry, Semel Institute for Neuroscience & Human Behavior, Department of Psychiatry & Biobehavioral Sciences, UCLA David Geffen School of Medicine.
Health informatics is the science of how we collect, analyze, and use health information to improve health and healthcare – this science brings a nuanced understanding of the promise and pitfalls to clinicians at bedside and researchers seeking to leverage these rich clinical data mines. In what ways is health informatics changing the delivery and quality of mental health care, and transforming mental health services research?
Collection, analysis, and use of electronic health record (EHR) data from naturalistic clinical care settings has the potential to answer a variety of questions such as: What are the characteristics of children presenting to emergency departments for mental health related concerns during the COVID-19 pandemic? Who is at risk for a suicide attempt in the 30 days after medical hospitalization? Is adherence to national quality measures for depression screening associated with reduction in acute mental health service use for suicide? Health informatics, particularly, analysis of electronic health record data, enables discovery of target areas for quality improvement, disparities in mental health service access and delivery, and predictive characteristics of populations at risk for adverse outcomes, such as suicide and psychiatric hospitalizations. At the beginning of the last decade, we saw the start of use medical records to identify patients for genetic studies (e.g., Electronic Medical Records and Genomics Network (eMERGE)). Since then, use of EHR data has grown alongside computational advances in machine learning and natural language processing to become a rising area within mental health research.
There is a gap between the promised panacea of data analytics and the scarce adaptation into clinically useful bedside tools to support quality of mental health care, particularly for child mental health. Many of the papers pushing the field of mental health informatics in terms of computational methods are siloed in informatics journals with limited clinical audiences. Psychiatrists and other mental health providers are often not aware of these methods or the growing role of clinical informatics in their specialty.
Clinical informatics is recognized as a medical specialty with nearly 1700 board certified professionals, and many growing departments of clinical informatics throughout the country. There is great hope that analytics of health data will produce accurate, personalized clinical guidance to advance high quality mental health care (e.g., a stepped care algorithm for suicidal patients). EHR datasets often contain more nuanced and detailed information than other administrative data sources (e.g., an emergency department visit may be associated not only ICD code and payor type, but also with e.g., Columbia Suicide Severity Rating Scale scores, chief complaint, social history, and record of involuntary hold). Also, as the data are naturalistically recorded, we are able to capture more indicators of ‘usual care’ in a broader population than research-specific surveys or other means that require outreach to patients to gather information.
This being said, EHR data suffer from missingness, human error, structural biases, and the ‘open system problem’ (i.e., patients use more than one health care setting and data may not be linked). Greater attention should be paid to how to connect the “bench” of computational methods, predictive analytics, and electronic health record data mining to the “bedside” of point-of-care tools to support high quality and evidenced-based mental health care.
Necessary skills and knowledge:
- an understanding of the EHR infrastructure of the health system(s) provisioning the data
- programming skills to extract the EHR data from the institutional database and curate the provisioned data into an analyzable format
- biostatistical skills to analyze the dataset
- domain expertise to interpret the findings and, when appropriate, iterate on curation and analysis
Assembling a team that includes a biostatistician, clinical expert(s), a programmer, and an informatician is often helpful. For projects involving narrative text data derived from EHRs, inclusion of a natural language processing expert is prudent. For clinicians, some knowledge of programming and data curation is helpful. To work with EHR databases often requires programmers with experience in transforming data into a useable format for research (e.g., a relational database structure).
Hardware and equipment requirements:
- will vary based on the size of the data pull and complexity of analysis. Data reformatting often requires an infrastructure that can support secure servers, hardware space, a research version of the EHR. Once extracted, curation and project-specific analysis of EHR data can often be done with various free and open-source software.
Data: Most healthcare institutions use EHRs and have some form of an EHR database including data on billing codes, prescriptions, laboratory values, and narrative text notes. If your institution partners with a Clinical and Translational Science Institute, this may your first point of contact to request access to EHR data. HIC considerations: This process typically requires IRB approval and appropriate data use agreements prior to release of data. There are often institutional compliance protocols to store, analyze, and share EHR data within your team. As some software packages (e.g., for annotation of clinical note text data) are developed for industry applications, these do not necessarily meet the compliance requirements of HIPAA.
The Accrual to Clinical Trials network: Good for multisite validation; allows researchers to explore and validate feasibility for clinical studies, including those using EHR data, across sites. This may be one resource to explore your patient population in depth as well as find partner sites across the Clinical and Translational Science Association Consortium network.
The National Patient-Centered Clinical Research Network (PCORnet): A distributed network with millions of electronic health records all adhering to a common data model allowing ready capacity for analysis across sites. The PCORnet Front Door is the initial access point.
MIMIC-III: A large, freely-available database comprising deidentified health-related data associated with over 40,000 patients, though this is not mental health specific. May be helpful for testing out computational methods (e.g., deidentification of text records) that can then be applied to other mental health specific EHR datasets.
Tutorials and training opportunities:
- Look to see if your institution has active collaborations and learning opportunities within the Department of Informatics (e.g., https://medicine.yale.edu/ycmi).
- Duke Center for Health Informatics Research Seminar: https://chip.unc.edu/duke-unc-health-informatics-seminar-serie
- The Harvard Medical School Clinical Informatics Lecture Series: https://dbmi.hms.harvard.edu/events/clinical-informatics-lecture-series
The Healthcare Information and Management Systems Society Learning Center:
- The American Medical Informatics Association offers 10x10 Virtual Informatics Education courses (at cost): https://amia.org/education-events/amia-10x10-virtual-courses
- PCORnet also has free data curation learnings shared on GitHub: https://pcornet.org/news/category/data-resource/data-curation
- Mankowitz, S., ed. Clinical Informatics Board Review and Self-Assessment. Springer, 2018.
- Edgcomb, J. B., & Zima, B. (2019). Machine learning, natural language processing, and the electronic health record: innovations in mental health services research. Psychiatric Services, 70(4), 346-349. https://pubmed.ncbi.nlm.nih.gov/30784377
- Liao, K. P., Cai, T., Savova, G. K., Murphy, S. N., Karlson, E. W., Ananthakrishnan, A. N., ... & Kohane, I. (2015). Development of phenotype algorithms using electronic medical records and incorporating natural language processing. BMJ, 350. https://www.bmj.com/content/bmj/350/bmj.h1885.full.pdf
- Tsui, F. R., Shi, L., Ruiz, V., Ryan, N. D., Biernesser, C., Iyengar, S., ... & Brent, D. A. (2021). Natural language processing and machine learning of electronic health records for prediction of first-time suicide attempts. JAMIA open, 4(1). https://academic.oup.com/jamiaopen/article/4/1/ooab011/6174413
- A call to action for training in clinical informatics for psychiatric trainees:- Torous, J., Chan, S., Luo, J., Boland, R., & Hilty, D. (2018). Clinical informatics in psychiatric training: preparing today’s trainees for the already present future. Academic Psychiatry, 42(5), 694-697. https://pubmed.ncbi.nlm.nih.gov/29047074