Page 23 - AIH-1-2

Artificial Intelligence in Health LLMs-Healthcare: Application and challenges

learners, meaning once trained on massive datasets, they accuracy, personalize therapy options, and streamline
can be retrained for new domains utilizing a small number patient care in oncology. By analyzing vast amounts of
of domain-specific examples. 3 data, LLMs can provide insights that potentially improve
LLMs have become increasingly prevalent in the treatment outcomes and patient management strategies. In
medical domain, due to their versatility, and expanding the subsequent discussion, we explore the studies dedicated
influence. Their applications in healthcare are to integrating LLMs within oncological care, encapsulating
multifaceted, ranging from processing vast quantities of the innovative efforts to harness LLMs’ capabilities in
medical data and interpreting clinical notes to generating enhancing the diagnostic, treatment, and management
comprehensive, human-readable reports. This broad processes associated with cancer care.
4
spectrum of functionalities shows how LLMs are not In a study conducted by Sorin et al., the capabilities
6
just tools for data processing but are also instrumental in of ChatGPT, an LLM were explored as a decision-
providing innovative solutions across various aspects of support tool for breast tumor boards. The research’s
healthcare. LLMs are increasingly being utilized to tackle primary objective was determining how ChatGPT’s
critical challenges in patient care. This includes providing recommendations align with expert-driven decisions
customized educational content to patients, assisting during breast tumor board meetings. For this purpose,
health-care professionals in making complex diagnostic clinical data from ten patients discussed in a breast tumor
decisions, and easing the administrative burdens often board at their institution were inputted into ChatGPT-3.5.
associated with health-care provision. 4,5 Subsequently, the model’s management recommendations
While LLMs have been applied across a spectrum were compared with the final decisions made by the tumor
of activities in healthcare, including medical question board. Moreover, two senior radiologists independently
answering, examination, pure research-oriented tasks, and evaluated ChatGPT’s responses, grading them on a
administrative duties in hospitals, this review will focus scale from 1 (complete disagreement) to 5 (complete
exclusively on their practical applications in healthcare, agreement) across three categories: summarization of the
such as diagnostics and treatment purposes. We uncover case, the recommendation provided, and the explanation
their deployment in critical areas such as cancer care, for that recommendation. Most patients in the study
dermatology, dental, mental health, and other core medical (80%) had invasive ductal carcinoma, with one case each
specialties listed in Figure 1. This exploration is crucial, as of ductal carcinoma in situ and a phyllodes tumor with
it showcases LLMs’ capacity to innovate and streamline atypia. ChatGPT’s recommendations aligned with the
medical diagnostics, patient care, treatment tasks, and also tumor board’s decisions in seven out of the ten cases,
address the challenges and opportunities in harnessing marking a 70% concordance. On grading, the first reviewer
their full potential in complex medical areas. In this gave mean scores of 3.7, 4.3, and 4.6 for summarization,
review, we conduct an in-depth analysis of the applications recommendation, and explanation, respectively, while the
of LLMs across different medical fields. We focus on second reviewer’s scores were 4.3, 4.0, and 4.3 in the same
the advancements and challenges of integrating these categories. As an initial exploration, the study suggests
sophisticated models into routine health-care practices. We that LLMs like ChatGPT are potentially valuable tools
offer insights into the current state of progress and identify for breast tumor boards. However, as technology rapidly
barriers to their widespread adoption in clinical settings. advances, medical professionals must know its advantages
The paper is structured to cover each medical specialty and and potential limitations.
associated challenges, followed by examining various data In a study by Lukac et al. in January 2023, the
7
types in the medical field. The conclusion summarizes the capabilities of ChatGPT to assist in the decision-making
findings and implications. process for therapy planning in primary breast cancer
2. Cancer care (oncology) cases were investigated. Although the ChatGPT was able
to identify specific risk factors for hereditary breast cancer
Cancer is characterized by the uncontrolled growth of and could discern elderly patients requiring chemotherapy
abnormal cells in the body, a topic encompassed under assessment for cost/benefit evaluation, it generally offered
the big umbrella discipline called oncology – the study non-specific recommendations concerning various
of cancer types and related factors. Adopting LLMs such treatment modalities such as chemotherapy and radiation
as ChatGPT in oncology has become a focal point of therapy. Notably, it made errors in patient-specific
recent research, especially in supporting decision-making therapy suggestions, misidentifying patients with Her2 1+
processes for cancer treatment. These advanced models are and 2+ (FISH negative) as candidates for trastuzumab
being explored for their capability to enhance diagnostic therapy and mislabeling endocrine therapy as “hormonal

Volume 1 Issue 2 (2024) 17 doi: 10.36922/aih.2558