Clinical practice has evolved over time, and the doctor’s tools used in the past may no longer be the tools that are regularly utilized today. However, that brings up the question: what tools will healthcare utilize today or in the future? Emily Ray, MD, helps to answer this question in an educational session presented at SABCS 2024.
One of the most utilized tools that has revolutionized healthcare has been electronic health records (EHRs). They provide health systems with an easier way to access prior medical records, as well as see a semi-comprehensive list of medications, test results, and medications a patient may be on. Dr Ray, noting that there are benefits in efficiency and patient history clarity, also noted that EHRs may lead to clinicians being overwhelmed with information. The need for a clinician to synthesize a wealth of information from an EHR may be something that artificial intelligence (AI) assists with, and a study by Lee et al presented by Dr Ray explores that idea. This study sought to develop and validate an AI model to develop clinical summaries of oncology cases. The study utilized 50 breast cancer cases, and each case was processed via AI utilizing a large language model alone. The summaries were then blinded and reviewed by 8 oncology specialists, who ranked the summaries on faithfulness, completeness, and succinctness. This study found that oncology specialists preferred the full AI and AI-assisted summaries over a clinician alone. AI-assisted summaries ranked higher than full AI and clinician-only in faithfulness, completeness, and succinctness, and full AI ranked higher than clinician-only in completeness and succinctness. Though the data is striking, Dr Ray noted some challenges. Large language models often require training on key data elements, and these elements can be different based on specialty. Furthermore, most EHRs lack the human or computing power to be able to incorporate new AI tools.
Another viable use of AI is estimating risks. Existing risk stratification tools utilize regression methods or Cox proportional hazard modeling, and most clinicians are comfortable with these tools. However, AI tools are potentially more powerful, as they can leverage large data sets compared with modeling techniques to help develop risk stratification tools. However, there are challenges with both. First, having risk stratification tools outside of the EHR, whether modeling or AI, requires the clinician to input information from the EHR into the tool, increasing the time needed. AI risk stratification tools particularly, similar to the summary tool mentioned above, often require large amounts of computing power that an EHR may not have. To overcome these challenges, Dr Ray states that EHR systems must figure out how to integrate AI tools, and there must be improvement in the quality of data derived from the EHR. Finally, Dr Ray touched on some other uses of AI, including treatment plan selection and patient communication. Oncology Clinical Pathways (OCPs) are perhaps the most well-known application of AI in treatment decision-making. Although OCPs may increase the consistency of care, Dr Ray notes that going down the wrong pathway may be easy, and that because data between the pathway and EHR systems are not bidirectional, there is frequently a large amount of information logging that needs to be done. AI in patient communication may have some potential, but more investigation is needed. A study by Seale et al showed that AI-generated draft replies resulted in increased message read time for physicians, no change in the timeliness of response, and longer length of response. Dr Ray does state that though these conclusions may sound negative, there is still potential for AI applications in patient communications if models are trained on prior clinician responses. She also acknowledged that some providers may not want to utilize AI because of the lack of provider awareness of their patients’ concerns and health conditions.
AI is impressive in its capabilities; however, the limitations of EHRs are often an obstacle to the integration of AI into healthcare practices. However, as stated by Dr Ray, AI tools in multiple applications may need some improvements. When those improvements are finally realized, we may see a new age of healthcare.
Ray EM. Digital tools for clinical practice. Presented at: San Antonio Breast Cancer Symposium. December 10-13, 2024; San Antonio, TX.
Learn more about our family of publications.
View Our Publications