Große Sprachmodelle zur schnellen Vereinfachung der Eingabe von Qualitätssicherungsdaten: Kommentar

Dear Editor,

We would like to discuss on the publication entitled “Large Language Models for Rapid Simplification of Quality Assurance Data Input: Field Trial with Real Data in the Context of Tumour Documentation in Urology [1]”. The study ran an experiment to extract information about cancer diseases from medical reports using chatGPT 4.0. Specifically, they focused on the date of initial diagnosis and tumor classifications. The usefulness of applying AI to tasks related to tumor documentation was evaluated, and each outcome was examined separately for accuracy. According to the report, the application of generative AI in this industry is both cost-effective and promising, with value generation surpassing operating expenses. It was pointed out that, in order to attain maximum performance, AI can only play a supporting function in current processes and must be properly integrated.

The limited sample size of 25 patients, which could not be representative enough to derive findings that are applicable to other situations, is one of the study's methodology's flaws. Furthermore, even though the study mentioned the accuracy of the AI-generated findings, it was not possible to compare them to human performance or to currently in place manual processes, which made it difficult to determine the AI system's actual usefulness. Furthermore, the study made no mention of any potential biases in the training data used by the AI system or potential restrictions on its capacity to correctly interpret medical data.

The lack of a thorough examination of the restrictions and disadvantages of applying AI in this situation is a study weakness. The study notes sporadic problems with the AI-generated responses' comprehensiveness, correctness, clarity, and relevancy, but it doesn't go into great detail to explain these flaws or provide solutions. Furthermore, there isn't enough talk on the ethical ramifications of AI being used in medical records and the value of human supervision in intricate medical procedures.

Future iterations of this study might compare the AI system's performance against that of human specialists and enlarge the sample size to encompass a wider variety of medical reports and tumor types. Additionally, it would be advantageous to look into the particular difficulties and restrictions associated with utilizing AI to the documentation of cancer diseases as well as possible fixes and enhancements to increase the precision and dependability of data supplied by AI. To further advance this topic, research on how AI might be integrated into current healthcare workflows and guidelines for its ethical and responsible use in medical contexts should be developed.

Publikationsverlauf

Artikel online veröffentlicht:
03. April 2025

• Literatur

• 1 Frank J, Merseburger AS, Landmesser J. et al. Large Language Models for Rapid Simplification of Quality Assurance Data Input: Field Trial with Real Data in the Context of Tumour Documentation in Urology. Aktuelle Urol 2024; 55: 415-423