Q&A: AI can triage patients in primary care, streamline health care services

Key takeaways:

• A machine learning model can effectively triage primary care patients who have respiratory symptoms.
• One researcher told Healio the innovation indicates the potential for artificial intelligence in primary care.

Using artificial intelligence for patient triaging has the potential to streamline health care services while also tackling issues like antibiotic resistance, according to researchers.

Respiratory symptoms are the most common complaint primary care physicians see, Emil Lárus Sigurðsson, MD, PhD, a PCP and professor in the department of family medicine at the University of Iceland, Steindór Ellertsson, MD, the co-founder of an app for insomnia treatment, and colleagues wrote in Annals of Family Medicine. Symptoms can indicate a severe illness but could also resolve on their own.

Considering this range, triaging patients before in-person consultations could help physician workload and health care costs in offering low-risk patients other means of communication. An artificial intelligence (AI) solution would further expedite the process.

So, the researchers conducted a retrospective study to understand if a machine learning model could triage patients with respiratory symptoms before a visit to a primary care clinic. They found that the technology reduced the number of referrals for chest X-rays (CXRs) by eliminating them in low-risk patients.

Healio spoke with Ellertsson and Sigurðsson to learn more about the machine learning model’s capabilities, the potential of AI in primary care and more.

Healio: Will you describe your study? Why did you look into AI in primary care?

Ellertsson and Sigurðsson: Our study examines the effectiveness of AI models in evaluating patients with respiratory symptoms in a primary care setting, specifically when patients are still at home. Health care systems worldwide are facing challenges due to overutilization, which is increasing burnout among health care workers. This issue is further complicated by the fact that health care professionals are spending less and less time with patients. A large number of patient consultations for respiratory issues are for conditions that could be managed with symptomatic treatment only, without requiring a clinical assessment. Yet, these patients often undergo unnecessary diagnostic tests and receive antibiotics, a practice often inconsistent with clinical guidelines.

Being a primary care physician myself, I am particularly interested in research exploring ways to enhance our service quality and reduce the workload for primary care staff. Using AI could be a promising approach to address these issues by aiding in the initial evaluation and triaging of patients, potentially reducing unnecessary consultations and interventions.

Healio: What were your findings and their importance?

Ellertsson and Sigurðsson: This retrospective study suggests the feasibility of classifying patients based on the severity of their symptoms, utilizing only their symptomatic history prior to their seeking help from the healthcare system. The results illustrate that roughly one-third of patients are designated as low risk. This group has the following characteristics:

• All their CXRs are negative for pneumonia and tumors, although a significant number of these scans reveal incidental findings (incidentalomas), which are often clinically nonsignificant (all incidentalomas were nonsignificant in our study) and lead to increased cost and discomfort for patients.
• They have low C-reactive protein values, which indicates a high proportion of viral disease.
• There's a high incidence of International Classification of Diseases (ICD) codes in this group, where antibiotics or further medical evaluation are not necessary. Despite this, such treatments are frequently administered. Interestingly, no patient in the low-risk group was diagnosed with pneumonia, either through an ICD code or a positive CXR.
• There are low rates of re-evaluation within a 7-day period in both primary care settings and emergency departments.

These observations underscore the potential efficiency of patient triaging through initial symptom-based assessment, with the intention of reducing unnecessary medical visits and treatments.

Healio: What are the clinical implications of your study?

Ellertsson and Sigurðsson: Currently, we lack the ability to determine the severity of a patient's illness before they seek assistance from the health care system. If we were equipped with an AI model which could reliably make this assessment, we could guide patients more effectively on where to seek assistance, whether they should stay at home and receive symptomatic treatment or opt for a telemedical consultation as opposed to an in-person visit. The COVID-19 pandemic substantially amplified the use of telemedicine consultations; however, there are no existing guidelines that dictate who is eligible for such consultations. The study shows we might be able to reduce physical clinical consultations by up to 35%.

Antibiotic resistance is a growing issue worldwide, and the most effective way to mitigate this problem is by reducing unnecessary antibiotic usage. This study demonstrates that the application of AI could decrease antibiotic use by 25%.

Overutilization of diagnostic imaging is another concern, but our study suggests that we could potentially reduce referrals for imaging by 35% without risking missed diagnoses of severe conditions such as pneumonia. Hence, implementing AI in patient triaging could enhance the efficiency and accuracy of health care delivery.

Healio: Is there anything else you’d like to add?

Ellertsson and Sigurðsson: Our research demonstrates the potential of AI in primary care settings. By implementing AI for patient triaging, we can not only streamline health care services but also contribute significantly towards combating global issues such as antibiotic resistance and overutilization of diagnostic resources. It's an exciting step forward in our continuous pursuit of improving health care outcomes while maintaining the efficiency of our health care system. However, we recognize that this is just the beginning, and more extensive research and trials are necessary to refine and optimize the use of AI in health care.

Reference:

• Sigurðsson EL, et al. Ann Fam Med. 2023;doi:10.1370/afm.2970.