Artificial intelligence in oncology: ensuring safe and effective integration of language models in clinical practice

Abstract

In this Personal View, we address the latest advancements in automatic text analysis with artificial intelligence (AI) in medicine, with a focus on its implications in aiding treatment decisions in medical oncology. Acknowledging that a majority of hospital medical content is embedded in narrative format, natural language processing has become one of the most dynamic research fields for developing clinical decision support tools. In addition, large language models have recently reached unprecedented performance, notably when answering medical questions. Emerging applications include prognosis estimation, treatment recommendations, multidisciplinary tumor board recommendations and matching patients to recruiting clinical trials. Altogether, we advocate for a forward-looking approach in which the community efficiently initiates global prospective clinical evaluations of promising AI-based decision support systems. Such assessments will be essential to validate and evaluate potential biases, ensuring these innovations can be effectively and safely translated into practical tools for oncological practice. We are at a pivotal moment, where continued advancements in patient care must be pursued with scientific rigor.

Keywords: Artificial intelligence; Clinical trials; Healthcare; Language models; Natural language processing; Oncology.

Fig. 1

Repartition of the number of machine learning methods applied to oncology-related tasks (left panel), as per PubMed query (assessed in February 2024). The right panel specifies the repartition of methods utilizing deep learning. MeSH terms that were used for each query are specified in bold with the related fraction. The color legend indicates the absolute count of results per query.

Fig. 2

Trial Matching example by prompting the Large Language Model GPT4-turbo from https://chat.lmsys.org/. The medical report and synthetic patient profile was extracted from SemEval 2024, and selection criteria of the trial extracted from clinicaltrials.gov.

Fig. 3

Examples of AI-stratified prospective clinical trial designs with randomisation. Scenario#1: randomization to assign patients to use or not the predictions of the tool, inspired from. This applies to situations such as patients monitoring (e.g. using automatic altert systems versus standard procedures), orientation to a specific interventions or another (e.g. orientation to targeted therapies based on the presence of a biomarker), or evaluate a new prognosis tool to decide for an intervention or not (e.g. a surgery, a new treatment line, a clinical trial). This designs alows to derive clear conclusion of the superiority of the prediction compared to the standard of care. Scenario #2: randomization obtain the information now or in a pre-defined periode of time, i.e. intra-patient comparison, also known as N-of-1 trials. In practice, when AI models are used as search engines for medical treatments or clinical trials, it seems unethical to spare patients from the information. It can apply to several clinical decision support systems, to the evaluation of the efficacy of a treatement (using overall reponse rate as primary endpoint), to the impact on the work load of careguivers, the quality of life of patients, among other criterias. Patients that are randomized in the standard of care arm, can have acess to the results of the system after an observation period, here 3 months. The comparison of the endpoint measures between the two period of time allows concluding on the impact of the intervention.