Summary
Amid rising advanced artificial intelligence (AI) technologies, global health systems are embracing the instant capability of natural language processing (NLP) applications. These applications maximise large language models (LLM) algorithms for holistic health care services, to improve patient-centred outcomes. Chatbots powered by LLM have the potential to serve as automated conversational agents that help advance patient-centred care and improve patient outcomes. This shift necessitates a unified definition of value, beginning with the patient-centered outcomes that matter most. This may increase patient trust through sufficient information about AI benefits and increase medical adherence through adequate instruction in doctor-patient diagnosis communication. The objective is to determine how LLM chatbots have been used to achieve patient-reported outcomes in 2024 using Yake and Medspacy.
How can LLMs be Utilised for Patient-Centred Outcomes?
‘Large Language Models (LLMs) are transformer-based neural networks with billions of parameters trained on large text data from various sources’ (Raiaan et al., 2024). They have the potential to revolutionise healthcare delivery by analysing complex medical information and generating context-based answers. However, the lack of standardised rules for reporting methods and results for LLM-oriented research are affected. While LLM offers promising solutions, patient-centered care remains the gold standard of care in the health sector. This care revolves around the patient and the specific health care requirements. According to Reynolds (2009), a higher rate of patient satisfaction, adherence to suggested lifestyle changes and prescribed treatment, better outcomes, and more cost-effective care are associated with patient-centered care. The question now is, how has the latest language model impacted the healthcare sector using a patient-centred care approach?
Data Collection and Analysis
Scopus and Google Scholar databases were used for collecting data using search text, ‘LLM AND chatbot AND in AND patient AND centred AND care’. In 2024, 34 manuscripts were seen in total and among those documents, 27 results from Google Scholar, 4 from Scopus, and 3 results from Google Scholar with 13 abstract documents selected after screening for duplicates. The abstract was preprocessed with Spacy, keywords were extracted with Yake (Campos et al., 2020) as seen in Fig.1 and target rule and text analysis with Medspacy (Eyre et al., 2021) as shown in Fig.2.
Fig 1. Keywords extraction with Yake
Fig 2. Target rules with Medspacy methods
What Can We Infer from Abstracts?
LLMs as one of the conversational application methods to examine its potential in primary care. The collated abstracts emphasise the potential benefits and limitations of using LLMs in primary care by explaining the considerable factors that primary care clinicians should consider when deciding to adopt and integrate such technologies into their clinical practice. Most of the results focused on the application of LLM chatbots in assisting healthcare administrators with administrative tasks and pre-consultation tasks. These applications are explored to educate patients on digital health development and assist them in making better clinical decision making. Instances where patients may have questions that they feel embarrassed to ask or will face delays awaiting responses from their healthcare providers, which may impact their health behaviour. Most of the metrics used focus on safety, accuracy, bias and reproducibility, and suitability using the DISCERN score. Mistral-7b emerged as the top performer among selected lightweight LLMs. As shown in the table, words with patients can be seen as the focus of LLM applications in health care using ChatGPT as instances.
Table 1.0 Abstract composition
Other Insights for Clinicians
LLM is still a work in progress regarding robustness and specificity regarding questions answered by human users – patients, clinicians and health administrators. This is an issue that can be fixed with reinforcement learning from human feedback (RLHF) as medical information during training. The latest development is LLMs can be infused with domain knowledge using Retrieval Augmented Generation (RAG) or Supervised Fine-Tuning (SFT) (Yu et al., 2024). This will ensure that LLM applications give accurate answers and specific responses to questions asked by clinicians or patients.
LLM Limitations in Healthcare
Although the major challenges with the LLM model are the depth and specificity of knowledge provided, many times, wrong information is common with these language models. This information may pose serious and life-threatening implications for patients, which should be considered and addressed in future research.
Conclusion
This write-up delves into how LLM models have been linked to patient-centred care by collating 2024 abstracts on the topic matter and analysis with Yake, Spacy and Medspacy. Additionally, it offered in-depth analysis of LLM capabilities and shortcomings in clinical settings. It therefore concludes that LLM applications hold immense potential to revolutionise healthcare by enhancing patient-centred care, refining clinical decision-making, automating administrative tasks and optimising pre-consultation processes based on patient-reported outcomes.
However, ensuring the safe and effective deployment of LLM models in healthcare is critical to guarantee accurate and reliable results. Likewise, integrating human feedback into the model training process is essential for improving model performance.
Additionally, further research should focus on leveraging domain-specific databases and expert knowledge to enhance LLM capabilities in supporting patient-centred care within medical applications.
