Take-home points

  • Artificial intelligence can augment the otherwise time-consuming clinical trial recruitment process by sifting through thousands of images and charts to rapidly identify patients who meet inclusion criteria and those who should be excluded.
  • Large language models stand to streamline administrative duties, particularly in the prior authorization and denial workflow, analyzing denied claims and retrieving relevant data from EHRs to aid in appeals.
  • AI voice agents in call centers can reduce office staff burden and offer 24/7 availability to patients for answering common questions, scheduling or canceling appointments and issuing appointment reminders.

Bios

Dr. Liu is the James P. Gills Jr. MD and Heather Gills Rising Professor of Artificial Intelligence in Ophthalmology at Johns Hopkins.
Dr. Popper is a businessperson and lawyer in the diagnostic and medical device space.
Mr. Hayat is a health-care entrepreneur and leader.
Dr. Al-khersan is a vitreoretinal surgeon and clinical trialist at the Retina Consultants of Texas in Houston.

Disclosures: Dr. Liu is an advisor for Optain Health, AKASA, OMNY Health and Amaros AI. He is a consultant for Roche. Mr. Hayat is on Athenahealth’s RCM Advisory Board and Cencora’s Inventory Management Advisory Board. Dr. Al-khersan is a consultant for Abbvie, Adverum, ANI, Annexon, Apellis, Eyepoint, Genentech, Ocular Therapeutix and Regeneron.

Deep learning, the cutting-edge artificial intelligence technique for image analysis and natural language processing, has made strides in medicine and ophthalmology in recent years. Within ophthalmology, the subspecialty of retina has been at the forefront of this AI revolution. For example, deep learning algorithms can predict imminent conversion from dry to wet age-related macular degeneration directly from optical coherence tomography images,1 or automatically segment important clinical endpoints such as geographic atrophy areas.2 Autonomous AI for diabetic retinopathy screening in the primary care setting is also the second most used medical AI application in the United States.3

Deep learning-based image analysis and natural language processing, particularly in the form of large language models, have the potential to transform clinical care. However, the application of these AI tools in hospital and health-care systems’ operational domains is likely to achieve real-world scaling first, due to clear in return-on-investment achieved through operational efficiencies and the relative ease of adoption (with quick evidence generation and without the need for regulatory review).

Against this reality, we’ll explore three practical real-world examples of how AI—particularly deep learning-based systems—could benefit innovators and care providers in the business of retina care: (1) Large-scale AI analysis of retinal images and automated quantification of disease biomarkers can streamline clinical trial recruitment and generate real-world evidence; (2) LLM-based tools might assist with prior authorizations for intravitreal injections and manage insurance claim denials; and (3) AI voice agents can handle voice-related tasks in a retina practice, from call center duties to engaging with insurance payers. These examples illustrate the potential for AI to enhance patient care, improve efficiency and reduce administrative burdens across a wide range of retinal practices.

 

Example 1: Clinical trial recruitment and real-world evidence generation

One promising application of deep learning in retina care is the large-scale analysis of retinal images (such as color fundus photographs and OCT scans) to identify and quantify relevant biomarkers in an efficient and cost-effective manner. This capability can transform clinical trial recruitment in retina care. Traditionally, screening patients for trials is time-consuming, requiring manual review of images and charts to determine whether a patient meets inclusion criteria (e.g., the presence of specific lesions or a threshold of central subfield thickness on OCT). AI can automate this process by rapidly analyzing medical images and extracting the relevant information to determine patient eligibility for trials. By including or excluding patients a priori based on objective imaging criteria, an AI system could streamline patient selection, allowing faster identification and enrollment of suitable participants.

Beyond basic clinical trial recruitment, the same large-scale image analyses could enrich data sets of real-world evidence and serve as a resource to academics, industry innovators and pharmaceutical companies. In ophthalmology, real-world data from tens of thousands of patient visits could be mined by AI to uncover patterns in disease progression and treatment response outside the confines of clinical trials. However, currently, most real-world datasets (with rare exceptions such as the clinicoimaging database spearheaded by Retina Consultants of America) only contain tabular data extracted from electronic health records, which are sub-optimal for our retina subspecialty that is so image-interpretation dependent. Deep learning can quantify imaging biomarker trajectories over time, such as the longitudinal evaluation of changes in intraretinal hyperreflective foci in diabetic macular edema or the evolution of subretinal hyperreflective material in wet AMD in response to anti-VEGF therapies.

These real-world insights enrich our understanding of how therapies perform in routine practice, could aid us in discovering novel clinical endpoints and allow us to perform post-approval safety monitoring to uncover low-incidence safety events that clinical trials may not be powered to detect. With privacy safeguards in place, these real-world datasets, enriched with quantitative image analysis, could represent an alternative revenue steam for retina practices around the country.

Adding AI tools to retina practices can streamline the prior authorization process. Leveraging LLM capabilities, AI can analyze denied claims and aid in appeal drafting, saving office staff the time of sifting through patient records.

Example 2: LLM-based tools for revenue cycle management

Another area where AI can significantly affect the business of retina practice is in revenue cycle management, particularly prior authorizations for intravitreal injections and the management of insurance claim denials. Intravitreal injections are high-cost procedures frequently requiring prior approval from insurers. Failing to obtain the necessary or correct prior authorizations will almost certainly result in a rejected claim. Retina practices must navigate different payers’ policies, which often include step therapy or providing documentation of medical necessity. This process is labor-intensive and time-consuming, and ripe to be disrupted by LLM technologies.

LLM-based tools can act as intelligent co-pilots to streamline prior authorization and denial workflows. These systems leverage the natural language processing prowess of LLMs to understand policy language and clinical documentation. For example, an LLM-driven application could automatically review a patient’s chart and the insurer’s criteria for an intravitreal injection, and then suggest to the human biller the appropriate documents to demonstrate medical necessity. By ensuring the initial submission is thorough and matches payer requirements, such a tool would reduce the chance of denials due to missing information or technical errors.

In cases where a claim is denied, AI can step in to perform denial analysis and appeal drafting. Instead of a human employee sifting through the denial reasons and patient records, an LLM-based system could parse the insurance denial letter, identify the reasons, retrieve relevant data from the electronic health record and propose an appeal letter or corrective action. Similarly, LLM models can efficiently review a large volume of charts to determine if medical coding is consistent with documentation and suggest edits prior to a claim being sent for adjudication. This can also reduce future denials or audit requests.

The rationale for and results of implementing LLM tools for revenue cycle management are clear—clinics can reduce administrative burden, lower denial rates and recover revenue that would otherwise be lost. Instead of battling administrative hurdles, clinicians can focus more on patient care. For patients, fewer delays from prior authorizations could lead to more timely treatments, potentially improving outcomes. Overall, LLM-based prior authorization and denial management systems exemplify AI’s potential to optimize practice operations and provide positive ROIs in both business and clinical domains.

 

Example 3: AI voice agents for call center management and insurance interactions

Voice-related workflows are another domain in which AI can positively impact a retina practice. AI voice agents are advanced conversational programs that interact through speech, and typically contain several components: speech recognition; natural language understanding; dialogue management; and text-to-speech output.

In health care, AI voice agents are already reshaping customer service by automating patient engagement and routine administrative calls. For a retina clinic, an AI voice agent can function as a virtual receptionist and call center assistant, handling inbound and outbound calls 24/7. On the patient-facing side, it can answer common questions (office hours, location, instructions), schedule or cancel appointments, send reminders and route urgent issues to on-call staff. On the back-office side, it can directly call insurance companies or pharmacies to perform tasks such as eligibility checks and obtaining claim status updates.

This automation of voice-related workflows saves considerable time and reduces the administrative load on staff who can be reallocated to higher-value work. From call center management to insurance communications, AI voice agents offer a comprehensive voice interface for the retina practice’s needs, by potentially combining the roles of a skilled receptionist, an appointment scheduler and a benefits coordinator—an all-in-one, always-on system.

 

The takeaway

As these examples demonstrate, AI in the form of deep learning is poised to significantly enhance the clinical and business operations of retina care. Deep learning excels at analyzing retinal images and can automate the identification of patients for clinical trials and the extraction of real-world data, thereby supporting both research and industry efforts. LLM-based tools and AI voice agents can take on burdensome administrative tasks. These breakthrough AI technologies can be leveraged for quicker clinical trial enrollments, smoother approval for sight-saving therapies, more consistent patient outreach and reduced overhead costs—all contributing to a more effective retina practice.

We’re on the threshold of a future in which health care, including retina care, will be provided by a collaboration between human providers and AI. In fact, retina care is a leading specialty in this care transformation. This will require provider retraining and retooling, a renewed focus on the practical and legal elements of data privacy, and cybersecurity systems. We can anticipate evolving regulatory requirements for instruments, software and trial data. In the future, the clinical standard of care may also evolve, where the non-use of AI tools that have been shown to improve quality may represent sub-standard care, with associated liability.

Adopting AI isn’t just about technology for technology’s sake; it directly translates into better patient care and a healthier bottom line for the businesses. As deep learning and AI continue to evolve, retina specialists and clinics that embrace these innovations will be well-positioned to thrive in a data-driven, patient-centered future of eye care. RS

 

REFERENCES

1. Yim J, Chopra R, Spitz T, et al. Predicting conversion to wet age-related macular degeneration using deep learning. Nat Med 2020;26:6:892-899.
2. Spaide T, Jiang J, Patil J, et al. Geographic atrophy segmentation using multimodal deep learning. Transl Vis Sci Technol 2023;12:7:10.
3. Wu K, Wu E, Theodorou B, et al. Characterizing the clinical adoption of medical AI devices through U.S. insurance claims. NEJM AI 2023;1:1. [Epub November 9, 2023]. Available from: https://doi.org/10.1056/aioa2300030.