<scp>OpenAI</scp> 's ‘Deep Research’ for the Generation of Comprehensive Referenced Medical Text: Uses and Cautions

摘要

Surgical research has become an area of great productivity in Australia and New Zealand, with many pre-eminent and prospective surgeons electing to engage in research projects alongside their clinical duties [1]. These research endeavours contribute greatly to the betterment of the surgical sciences and the dissemination of evidence-based medicine [2]. Research advances the surgical vocation by driving the innovation, refinement, and optimisation of surgical procedures and guidelines that dictate contemporary standards of practice [3]. One evolving area of surgical research pertains to the application of artificial intelligence (AI) to streamline, enhance, and accelerate various aspects of the surgical craft. In February 2025, OpenAI (the developers of ChatGPT) released a new platform called Deep Research (DR) [4]. The AI platform was designed to rapidly access and analyse vast datasets of literature, synthesise their findings, and then generate a written output in the desired format, language, and complexity of the user. DR presents an opportunity to bridge the gap between the ever-expanding reservoir of medical knowledge and the surgeon's ability to comprehend and apply it. However, the system poses certain risks that, if left unaddressed, have the potential to compromise the integrity of surgical scholarship and undermine the very foundations of evidence-based medicine and clinical decision-making. In this way, we believe that a balanced approach to this new platform is required. DR, which can be found under the ‘Explore GPTs’ panel of ChatGPT, is advertised as an ‘AI-powered research assistant’. It allows a user to input a research topic and ask several clarifying questions, after which it will generate a written output several pages in length. The platform is already attracting significant attention in the scientific community due to its ability to produce referenced text [5]. For instance, we used DR to generate written pieces on two potentially controversial evidence-based medicine topics; antithrombotic therapy for carotid dissection, and surgical and transcatheter aortic valve replacement for complex asymptomatic aortic stenosis (Supporting Information S1 and S2). The output remains non-deterministic, with identical prompts producing differing results. The quality of these outputs is shown in circumstances beyond controversial evidence-based topics. Potential utility could be envisioned in the summarisation of institutional websites to create patient handouts, and synthesising information on questions for topics appropriate for trainee education (Supporting Information S3). Such comprehensive referenced outputs also have potential utility in the preparation of grant applications and other related research documents, such as ethics applications and project proposals. Thus, the prospective benefit to the surgical community is significant. However, while DR presents several advantages to the broader surgical community, its application in research has many potential limitations, which may complicate surgical education, decision-making, and subsequent patient care. A large proportion of evidence-based surgery revolves around the ability to critically appraise and synthesise information [6]. However, generative AI systems present users with information that has been pre-synthesised, already featuring the elimination of unnecessary data and having made preformed conclusions. In this way, there is a legitimate concern that future generations of researchers may become overly reliant on algorithmic outputs, leading to a decline in their ability to independently assess the validity, methodology, and applicability of primary studies. This can be considered analogous to the concerns that intraoperative robotic-assisted surgery would result in the erosion of core surgical skills [7]. Another major concern is the possibility of algorithmic bias in AI systems, especially given their proprietary and often opaque nature. This is especi