How AI in nursing could alleviate documentation burden
As nurses face increasing levels of burnout, researchers are exploring how large language models could streamline clinical documentation and care planning.
Nurses play a critical role in delivering high-quality care and improving patient outcomes, but nursing shortages and burnout have significantly strained the workforce.
The American Nurses Association (ANA) posits that to fully realize nurses' potential, healthcare organizations must prioritize maintaining an adequate nursing workforce, creating healthy work environments and backing policies that support nurses.
Further, the ANA notes that the COVID-19 pandemic exacerbated existing challenges facing nurses, such as increased healthcare demand, inadequate workforce support and retirements outpacing rates of new nurses entering the workforce.
As the demands on nurses grow, some have asserted that technologies like AI could help alleviate pain points, including clinical documentation and other administrative tasks.
In a recent study published in the Journal of the American Medical Informatics Association, Fabiana Dos Santos, Ph.D., MSN, RN, post-doctoral research scientist at Columbia University School of Nursing, led a team investigating how a ChatGPT-based prompting framework could help generate care plan suggestions for a lung cancer patient.
In an interview with Healthtech Analytics, Santos detailed how the framework illustrates the potential -- and perils -- of AI chatbots in nursing.
The challenges of nursing care plan documentation
Creating care plans is instrumental in ensuring patients receive the timely, adequate care best suited to their needs. Nurses are crucial to this process, but they face many obstacles during care plan documentation.
"Nurses are at the front line of care and spend most of their time in close contact with patients, contributing valuable clinical assessments to patients' electronic health records (EHRs)," Santos explained. "However, many existing documentation systems are cumbersome, leading to documentation burdens in which nurses spend a large portion of their workday interacting with EHRs, which may result in cognitive burden, stress, frustration, inconsistencies or redundancies, and disruption of direct patient care."
The American Association of Critical-Care Nurses (AACN) asserts that electronic documentation creates significant burdens for nursing staff, taking up an average of 40% of a nurse's shift. The AACN further states that time spent on documentation is inversely correlated with time spent on care, meaning that more time documenting leads to less time with patients.
This documentation burden can lead to increased burnout and cognitive load, alongside decreases in necessary self-care and job satisfaction. In turn, these lead to patient-related issues like increased risk of medical errors and hospital-acquired infections, which lower patient satisfaction.
When combined with nurses' overburdened workloads, inconvenient documentation tools can make the burdens of care planning even more difficult to manage.
"The demands of providing direct patient care and managing multiple administrative tasks simultaneously limit nurses' time to develop individualized care plans. The non-user-friendly interfaces of many EHR systems further exacerbate this challenge, making it difficult to capture all aspects of a patient's condition, including physical, psychological, social, cultural and spiritual dimensions," she continued.
"These negative impacts on [a nurse's] workday underscore the urgency of improving EHR documentation systems to reduce these issues," she noted. "AI tools, if well designed, can improve the process of developing individualized care plans and reduce the burden of EHR-related documentation."
The promises and pitfalls of AI
Care plan development requires nurses to pull from their expertise to address issues like symptom management and comfort care. This is especially true when planning care for patients with complex cases and needs.
Santos emphasized that advanced technologies, such as generative AI (GenAI), can streamline this process by enhancing documentation workflows and assisting with administrative tasks to tackle issues like time constraints, documentation errors or inconsistencies, and redundancy.
AI tools can rapidly process large amounts of data and generate care plans more quickly than traditional methods. This efficiency might help nurses save time on administrative tasks and focus more on direct and holistic care to patients and their families.
Fabiana Dos Santos, Ph.D., MSN, RNPost-doctoral research scientist at Columbia University School of Nursing
"AI tools can rapidly process large amounts of data and generate care plans more quickly than traditional methods. This efficiency might help nurses save time on administrative tasks and focus more on direct and holistic care to patients and their families," she explained.
However, Santos indicated that careful validation of AI models is essential. This can be achieved by focusing on the critical role of nurses' clinical judgment and expertise when evaluating the effectiveness of AI-generated care plans.
"New technologies can help nurses improve documentation, leading to better descriptions of patient conditions, more accurate capture of care processes and, ultimately, improved patient outcomes," she said. "This presents an important opportunity to use novel generative AI solutions to reduce nurses' workload and act as a supportive documentation tool."
This emphasis on AI's role as a support tool for nursing staff is key to addressing the pitfalls of chatbot-generated nursing care plans.
Santos highlighted that while AI chatbots show promise for bolstering efficient nursing documentation, the limitations of these models must be carefully managed by nurses.
"Chatbots require further development to be effectively implemented in supporting nursing care plans," she stated, explaining that AI-generated outputs can contain inaccuracies or errors. "While AI can produce comprehensive care plans, the risk of incorrect or irrelevant information being included necessitates careful review and validation by nurses."
Further, AI tools might lack the nuanced understanding of a patient's unique needs that only a nurse can provide through personal, empathetic interactions, such as the ability to interpret a patient's specific cultural or spiritual needs, she indicated.
Despite this, large language models (LLMs) and other GenAI tools are creating significant hype in the industry. They are expected to be deployed in several healthcare applications, including EHR workflows and nursing efficiency.
This is where Santos' research comes into play.
To conduct the study, the researchers developed and validated a method for structuring ChatGPT prompts -- instructions the LLM uses to generate responses -- that can be used to produce high-quality nursing care plans.
The approach involves providing detailed patient information and questions that should be considered when creating an appropriate care plan to create model prompts. The research team achieved this by refining the Patient's Needs Framework over 10 rounds using 22 diverse hypothetical patient cases, which helped ensure that ChatGPT-generated plans were consistent and aligned with typical nursing care plans.
"Our findings revealed that ChatGPT could prioritize critical aspects of care such as oxygenation, infection prevention, fall risk and emotional support, while also providing thorough explanations for each suggested intervention, making it a valuable tool for nurses," Santos indicated.
The future of chatbots in nursing
While the study focused on care plans for lung cancer, Santos underscored that the research is just the beginning of exploring how LLMs and other tools could augment nurses.
"This study signifies a step toward integrating AI into nursing care plan documentation systems, with the potential to transform how nursing care is delivered and improve patient outcomes," she said.
Santos and her colleagues plan to expand this research and explore other ways AI could positively impact nursing.
"Our immediate plan is to demonstrate the [generalizability] of our framework by applying it to generate care plans for various clinical contexts beyond lung cancer," she explained, noting that doing so will require testing and validating the framework with different patient populations and conditions to ensure its robustness and adaptability.
To that end, the team aims to apply the ChatGPT prompting framework to other clinical scenarios -- including gestational diabetes and renal failure -- to assess its effectiveness across a broad range of conditions and diseases.
Santos also indicated that the prompting framework is compatible with different versions of ChatGPT, such as GPT-3.5 and GPT-4, and the researchers expect it to achieve better outcomes as more advanced versions of the LLM are released.
"While the care plans generated by AI show promise in our study, it is crucial for nurses to evaluate these plans critically in the context of the patient's unique needs," Santos said. "We can expect to see more AI applications in nursing with the potential to enhance the accuracy and precision of care. We must, however, remember that the nurses' expertise, communication, and empathy with their patients remain irreplaceable."
Santos underscored that developing a comprehensive Patient Needs Framework for nursing assessment and targeted questions is crucial in guiding AI-assisted nursing care planning.
Shania Kennedy has been covering news related to health IT and analytics since 2022.
