Clinical outcomes for asynchronous teledermatology
Dennis H. Oh, MD PhD
San Francisco VA Medical Center, San Francisco, CA San Francisco, CA
June 2022 -
Background: Store-and-forward teledermatology is a significant part of Department of Veterans Affairs’ telehealth portfolio. While considerable evidence supports teledermatology’s potential to provide timely access to expert dermatologic care, its effectiveness in achieving clinical outcomes that are equivalent to usual in- person care has not been as well documented due to the lack of objective outcome measures for many skin diseases. Clinicians typically document skin diseases using non-standardized qualitative language. Manual review to extract meaningful outcomes data from relatively unstructured text is typically prohibitive. Significance/Impact: Natural language processing (NLP) offers a previously unexplored approach to objectively and systematically identify relevant text in the electronic medical record to gauge patients’ clinical responses following either in-person dermatology and asynchronous teledermatology consultation. This project will leverage NLP to follow clinical courses of important skin conditions in the medical record and to compare the outcomes and effectiveness of teledermatology relative to usual office-based dermatology consultation. It will also serve as a test for other outcome measures such as access times that are often assumed to be proxies for quality of care for Veterans. The results may help influence VA telehealth strategy and policies to enhance access of patients to high quality skin care and to improve patient safety. Innovation: This project represents a novel application of NLP methods to understand how key clinicians document skin conditions and to provide a large-scale, systematic and rigorous assessment of teledermatology’s effectiveness in caring for Veterans with a variety of skin diseases. The project will also result in NLP systems which may be translatable to create practical operational quality management tools for monitoring the quality of follow-up care of both dermatology and teledermatology patients in VA. Specific Aims: Aim 1 will survey expert and non-expert clinicians to learn how each group evaluates and documents clinical change in five common skin diagnostic categories. We will test novel annotation methods, and identify differences between clinician groups in annotated survey responses. Aim 2 will use our annotated data sets to train and validate NLP models to extract concepts and relationships for our five diagnostic categories from actual VA clinical notes. This information will be used to create a document classifier capable of assigning a clinical change status to follow-up notes. Aim 3 will integrate output from our NLP tools to assign an overall clinical outcome to dermatology and teledermatology referrals. Other important clinical events and activities available as structured data will be correlated with NLP outcomes to further interpret their significance. Commonly used access outcome measures will also be compared as a test of their validity. Methodology: Aim 1 will survey dermatologists and primary care providers to annotate and compare their responses. Aims 2 and 3 will create trained and validated NLP tools to assign condition and outcome status to actual clinical notes. Aim 3 will use our tools to compare clinical outcomes following teledermatology and dermatology consultation and will will utilize the VA Corporate Data Warehouse to obtain structured data on other key clinical events and access measures. Implementation/Next Steps: The NLP models that result from this project may be extendable beyond routine in-person and teledermatology care to generally track clinical course outcomes related to other forms of telehealth such as dermatology e-consults and video telehealth. In addition, the models may be adaptable to create a practical dashboard tool to allow providers and quality management staff to monitor the effectiveness and quality of teledermatology delivered to Veteran patients.