Efficient healthcare with large language models: optimizing clinical workflow and enhancing patient care

Abstract

Purpose

This article explores the potential of large language models (LLMs) to automate administrative tasks in healthcare, alleviating the burden on clinicians caused by electronic medical records.

Potential

LLMs offer opportunities in clinical documentation, prior authorization, patient education, and access to care. They can personalize patient scheduling, improve documentation accuracy, streamline insurance prior authorization, increase patient engagement, and address barriers to healthcare access.

Caution

However, integrating LLMs requires careful attention to security and privacy concerns, protecting patient data, and complying with regulations like the Health Insurance Portability and Accountability Act (HIPAA). It is crucial to acknowledge that LLMs should supplement, not replace, the human connection and care provided by healthcare professionals.

Conclusion

By prudently utilizing LLMs alongside human expertise, healthcare organizations can improve patient care and outcomes. Implementation should be approached with caution and consideration to ensure the safe and effective use of LLMs in the clinical setting.

Introduction

Demands on clinicians’ time have grown dramatically since the proliferation of the electronic medical record (EMR). In addition to caring for more patients in less time, healthcare providers are overwhelmed with administrative responsibilities.¹ This trend has led to burnout and decreased the quality of patient care delivered.² Recent advancements in natural language processing and artificial intelligence (AI) and the development of technologies like large language models (LLMs) could make it possible to automate some of this administrative work and free up clinicians to focus on patient care.³

A number of LLMs have recently been released, including some with chat interfaces such as ChatGPT from OpenAI. Most of these models have billions of parameters and have been trained on billions to trillions of data points.⁴ Because these models can interpret natural language with all of its semantic complexities, they can produce responses to text prompts that resemble the way humans converse with one another. As a result, LLMs could become appropriate tools for supporting clinicians in clinical documentation and record-keeping, navigating the prior authorization process, educating patients, and potentially improving access to care (Figure 1).

Figure 1.

Overview of large language models (LLMs) in healthcare workflow.

The success of an LLM depends on the large volume of data it has been trained on, as seen in multiple recent studies that presented high accuracies for GPT models in various clinical applications. These applications include optimizing emergency departments for referrals,⁵ structured reporting,⁶ and context-based chatbots following the ACR Appropriateness Criteria.⁷ The GPT-3.5 and GPT-4 models even demonstrated high performance on radiology board-style examinations, showcasing their ability to decipher medical information.⁸

Moreover, neural networks have already outperformed rule-based systems in many applications due to their flexibility, natural language understanding, and adaptability.^9–11 LLMs can further enhance patient scheduling and engagement with more understanding of natural language, the ability to handle ambiguity and provide contextually relevant responses. Suitable for dynamic environments like healthcare, LLMs can process unstructured data, adapt to different phrasings, and maintain conversation context.

Personalizing patient scheduling

Imaging examinations or medical procedures that require prescreening for safety and appropriateness can pose unexpected challenges as individuals try to navigate care for themselves or their loved ones. However, with the help of LLMs, patients and their caregivers can receive personalized information regarding these procedures based on their medical history. By analyzing a patient's medical records, LLMs can dynamically guide patients to schedule their treatment at the appropriate location, help them understand preparation for an imaging exam or procedure, and answer questions about what to expect.^12–14 For instance, for a patient with a prior allergic reaction to a particular intravenous contrast agent or one with a pacemaker or defibrillator who needs a magnetic resonance imaging examination, LLMs can inform patients of the potential risks and suggest alternative options. LLMs can also alert primary care providers when patients are overdue for follow-up testing by digesting discrete EMR data, encounter notes from specialist visits, and relevant scanned documents.^14–16 Additionally, LLMs can streamline the prescreening process by gathering the necessary information to ensure patient safety and counseling patients before the appointment.

Improving clinical documentation

LLMs have the potential to assist with clinical documentation and record-keeping.^17,18 OpenNotes, a national initiative to share clinicians’ notes with their patients, has been shown to improve record accuracy by giving patients access and edit capabilities.¹⁹ Similarly, LLMs can help manage medical records by flagging potential contradictions or discrepancies, providing smart and dynamic clinical decision support, alerting clinicians to incomplete follow-up recommendations, or flagging actionable test results. This can help ensure that patient data is accurate and up to date, improving the overall quality of care delivered and reducing the likelihood of documentation errors. Furthermore, LLMs can help clinicians document components necessary for appropriate and accurate billing, which can contribute to better patient care and increase revenue.²⁰ However, allowing LLMs to modify patient charts, which are treated as legal documents, also raises concerns. Although LLMs are sophisticated, they can still make mistakes. If left to operate autonomously, they could introduce unexpected and undetected errors into patients’ records, and subsequent medical decisions could be based on inaccurate information. The security of patients’ data is also at stake, especially if the LLMs can access or modify sensitive information. Therefore, careful consideration and implementation of appropriate safeguards would be necessary to ensure the safe and accurate use of LLMs in managing patient medical records.

Facilitating insurance prior authorization

The prior authorization process is particularly onerous and frustrating for clinicians in the United States.²¹ LLMs could assist clinicians by compiling the elements of the patient’s record necessary to submit a complete and thorough request for prior authorization for a particular treatment or procedure.⁸ Insurance companies could similarly use LLMs to automate the review of submitted documentation and highlight aspects contributing to their decisions for approval or rejection, decreasing the need for time-consuming and error-prone manual review and improving the overall clarity of the eventual response. LLMs can determine if a patient's insurance plan covers a procedure or service by analyzing medical records, insurance policies, and other relevant data. LLMs could also be used to reinvent the insurance peer-to-peer review by allowing clinicians on both sides of the consultation to leverage decisions in other related cases and make more informed decisions. This could reduce clinician insurance company, and administrative burdens and minimize errors and delays that directly affect patients’ ability to receive the care they need.

Increasing patient engagement

In addition to aiding clinicians with administrative tasks, LLMs can improve patient engagement in their care. Providing patients with understandable health information can be challenging, as medical terminology can often be complex and difficult to comprehend.¹⁹ Although patient-facing materials are recommended to be written at a fifth-grade reading level, many resources are often found to be more advanced.^22,23 Furthermore, medical documentation is typically intended for other clinicians and requires a much higher readability level.²⁴ If LLMs could automatically convert complex medical information to a target reading grade level of the patient’s choosing, they could empower them to participate in the decision-making around their health.²⁵ Providing patients with personalized education generated by LLMs can also increase engagement and potentially improve health outcomes.²⁶ Patients can access LLMs to learn about their disease and treatment options, including potential side effects, medication schedules, testing needs, and the importance of follow-up appointments. Additionally, LLMs can provide patients with personalized advice on self-care, including exercise, a healthy diet, and techniques to improve mental health. At the same time, it is essential to keep in mind that patients should only be using LLMs for education and not for clinical advice and decision-making. Moreover, these LLMs should be trained on curated medical datasets that prevent LLMs from hallucinating and providing inaccurate information.

Decreasing barriers to access to healthcare

LLMs can also help patients overcome barriers to accessing healthcare. For example, patients who live in remote or underserved areas may not have access to specialist care. With the help of LLMs, patients can receive remote consultations, schedule appointments, and access educational resources from the comfort of their own homes. This can improve the overall quality of care and patient outcomes by increasing access to healthcare services. LLMs could also facilitate the transfer of care when necessary in underserved areas by synthesizing the often repetitive and potentially error-prone patient record for the receiving physician, seamlessly importing it into the receiving facility’s EMR, and highlighting aspects of the workup that are incomplete and need attention. Medical errors often happen during handoffs, and LLMs can decrease the likelihood that such errors reach patients and cause harm.²⁷ Increasing the efficiency of the data entry process would improve the efficiency of the clinical workup, allowing physicians to spend more time with patients and less time on administrative tasks. In addition, by improving the accuracy and quality of clinical documentation, LLMs can potentially enable improved decision-making and patient outcomes.

Security and privacy considerations

The Health Insurance Portability and Accountability Act (HIPAA) compliance prevents users from taking data from the EMR and feeding it unchanged as input to an LLM outside a hospital or health system firewall. However, integrating an LLM with the EMR securely could create a HIPAA-compliant environment in which to use these tools. Recent integration by Microsoft and Epic with Azure’s OpenAI Service in Epic's EMR software seeks to increase productivity, enhance patient care, and improve financial integrity in healthcare systems worldwide.²⁸ The initial solutions automatically draft message responses and bring natural language queries and interactive data analysis to SlicerDicer, Epic's self-service analytics tool. While industry experts have highlighted the need for health systems and hospitals to address intense pressures on costs and margins, generative AI integration promises to increase productivity and efficiency and improve financial sustainability.

At the same time, using LLMs such as GPT-4 in healthcare raises significant concerns about patient privacy and cybersecurity.^3,29 LLMs have the potential to process and generate vast amounts of sensitive healthcare data, which must be protected from unauthorized access or disclosure. Additionally, healthcare organizations that integrate AI tools into their operations must ensure that patient data is handled in compliance with applicable laws and regulations, such as HIPAA in the United States.

Furthermore, deploying AI in healthcare systems creates new opportunities for cyberattackers to exploit vulnerabilities and gain unauthorized access to sensitive patient data and medical systems.³⁰ Cybersecurity attacks can compromise the confidentiality, integrity, and availability of patient data, leading to significant harm to individual patients as well as healthcare organizations. The introduction of AI and now LLMs into the healthcare environment raises significant concerns regarding the security and privacy of patients' data. Cyberattacks, data breaches, and other similar threats have become more prevalent in recent years, making it imperative for healthcare organizations to take all necessary precautions to protect patient data. LLMs have the potential to improve the efficiency and quality of healthcare, but they must operate within the confines of expected patient privacy rules and regulations. The responsibility to ensure the appropriate use of new technologies falls on healthcare providers. Any breach of patient privacy or security could lead to significant harm to patients and damage the reputation of the healthcare provider. It is essential to prioritize the safety and security of patient data and continuously assess and improve the measures to prevent cyberattacks and data breaches when planning to implement such advances in the healthcare setting.

Limitations and future direction

Although LLMs demonstrate proficiency in generating text, their reliability in medical contexts is uncertain because of the lack of exposure to reliable medical datasets.¹⁵ Curated medical data encompasses essential information derived from clinical recommendations, scholarly articles, and patient records. LLMs that have not had this exposure may provide advice that is inaccurate or incredible.³¹ The presence of inconsistent knowledge within the training data might potentially result in the distortion of medical information.

To address these constraints, it is necessary to provide training to LLMs using trustworthy medical sources and incorporate rigorous validation approaches for identifying and minimizing bias. Using domain-specific and diverse datasets³² in the fine-tuning process of LLMs can lead to improved accuracy and more contextual relevance. Lastly, implementing continuous learning and validation processes with human involvement is crucial to maintaining the accuracy of these models in the ever-changing landscape of medical practices.³³

Conclusions

The use of LLMs in the clinical environment has the potential to reduce the amount of time spent by physicians on administrative tasks rather than direct patient care and improve the efficiency of clinical workflow. By automating some administrative tasks and offering easily accessible and tailored patient education, LLMs have the potential to help doctors concentrate more on the patient and less on the EMR. This can ultimately lower the administrative load they face and improve the quality of care they deliver to patients. While LLMs are helpful tools, they have limitations and cannot replace human connection and care. Patients should not be required to depend only on chatbots or LLMs for their healthcare needs but instead should continue to have access to clinicians as well. As with AI elsewhere in healthcare, using LLMs to improve both the clinician and patient experience is possible but must be carried out with caution and consideration.

Author contributions

All authors contributed equally to the writing and development of the manuscript.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Conflicts of interest

The authors declare that there are no competing interests.

Data availability

There are no new data associated with this article.