Portal Contents

Overview

ChatGPT, developed by OpenAI and released in November 2022, represents a paradigm shift in how artificial intelligence can interact with and assist healthcare professionals. Built on the GPT-3.5 architecture (and later enhanced with GPT-4), ChatGPT demonstrates unprecedented natural language understanding capabilities that distinguish it from previous clinical decision support systems. Unlike rule-based expert systems that dominated medical AI in earlier decades, ChatGPT can engage in nuanced, contextual conversations, synthesize information across domains, and adapt its responses to user expertise levels. This means that healthcare providers can obtain context-aware assistance without the rigid query structures that limited earlier systems. For example, a physician can describe a complex patient presentation in natural language and receive differential diagnosis suggestions, whereas traditional systems required structured input matching predefined templates.

The systematic review by Sallam (2023) analyzed 60 records from five major sources: PubMed, Google Scholar, Scopus, the preprint server medRxiv, and the Science Media Centre website. What makes this review particularly valuable is its timing—conducted between December 2022 and February 2023, it captures the initial wave of scholarly response to ChatGPT's release, documenting both the excitement and concerns of the medical community before widespread adoption patterns emerged. The review found that 71.7% of records (43/60) focused primarily on benefits, while 28.3% (17/60) emphasized concerns, reflecting the generally optimistic but cautious stance of early adopters in healthcare settings.

Compared to earlier medical AI tools, ChatGPT offers several significant advantages and disadvantages. Traditional clinical decision support systems (CDSS) achieved diagnostic accuracy rates of approximately 60-75% in specific domains, whereas GPT-4 has demonstrated 86.7% accuracy on USMLE-style medical questions according to Stanford HAI research. However, whereas CDSS outputs are deterministic and auditable, LLM responses can vary and occasionally "hallucinate" false information—a critical difference for patient safety. This trade-off between flexibility and reliability is central to understanding why healthcare institutions require careful governance frameworks before widespread adoption.

Understanding the balance between opportunities and risks is essential for responsible implementation. As detailed in this portal, the technology offers substantial benefits in academic writing, research acceleration, and clinical efficiency, while simultaneously raising valid concerns about accuracy, ethics, and patient safety that require careful institutional governance. According to the WHO, over 50 countries are now developing regulatory frameworks for AI in healthcare, reflecting the global significance of these considerations.

Methodology

The systematic review followed PRISMA 2020 guidelines for systematic reviews, ensuring transparent and reproducible methodology. The search was conducted between December 2022 and February 2023, representing the initial period following ChatGPT's public release. This means the review captures the "first wave" of scholarly response—an important methodological point because subsequent literature shows evolving perspectives as experience accumulates. Unlike traditional medical technology reviews that often analyze years of evidence, this review provides a snapshot of expert opinion during rapid technology diffusion.

The choice of databases reflects the interdisciplinary nature of the topic. PubMed captures biomedical perspectives, while Google Scholar and Scopus provide broader coverage of AI and computer science literature. Notably, medRxiv was included to capture preprints—a decision that proved important because approximately 25% of included records had not yet undergone peer review, reflecting the rapid pace of commentary in this field. In other words, the literature reviewed represented a mix of peer-reviewed scholarship and expert rapid-response commentary. For context on the research teams producing this literature and the journals publishing it, see the dedicated pages.

Benefits Summary

The review identified 22 distinct benefits of ChatGPT in healthcare contexts, with the majority of records (43/60, 71.7%) focusing on positive applications. This means that early scholarly discourse was predominantly optimistic, though subsequent research has introduced more nuanced perspectives. Compared to traditional clinical decision support systems, ChatGPT offers significant advantages in flexibility and natural language understanding. However, unlike deterministic rule-based systems, LLM outputs are probabilistic and require verification. The benefits span four major domains, each representing distinct use cases with different risk profiles and implementation requirements:

For detailed analysis of each benefit with supporting evidence, see the Benefits & Applications page.

Concerns Summary

The review identified 19 distinct concerns regarding ChatGPT use in healthcare, with 17 records (28.3%) primarily focusing on limitations and risks. Research has demonstrated that hallucination rates range from 3% to 27% depending on task complexity, with specialized medical questions showing higher error rates than general knowledge queries. Consequently, implementing ChatGPT in healthcare requires robust verification mechanisms that traditional clinical systems may not need. For example, while a rule-based drug interaction checker produces consistent, verifiable outputs, ChatGPT may generate different responses to identical queries—a property that complicates quality assurance. On the other hand, this same variability can be advantageous for creative tasks like hypothesis generation where novelty is valued over reproducibility:

For comprehensive analysis of concerns and mitigation strategies, see the Concerns & Limitations page.

Recent Developments (2024-2025)

Since the original systematic review, significant developments have occurred in the application of ChatGPT and other LLMs in healthcare:

• Singhal et al. (2023) - Large language models encode clinical knowledge (Med-PaLM)
• Singhal et al. (2025) - Toward expert-level medical question answering with LLMs
• Thirunavukarasu et al. (2023) - Large language models in medicine: systematic review
• Haltaufderheide & Ranisch (2024) - Ethics of ChatGPT in medicine and healthcare: systematic review

Key advancements include GPT-4's improved medical reasoning capabilities, the development of healthcare-specific models like Med-PaLM 2, and emerging regulatory frameworks for AI in clinical settings. For detailed publication information, see Top Journals.

Leading Research Teams

Key research groups advancing the study of AI and LLMs in healthcare span academic medical centers, technology companies, and international institutions. Together, these groups collectively represent the multi-disciplinary nature of healthcare AI research—combining clinical expertise, computational methods, and policy perspectives. Research has shown that the most impactful healthcare AI work typically emerges from collaborations between academic medical centers (providing clinical validation) and technology companies (providing computational resources). For example, Med-PaLM development involved both Google Health AI researchers and academic clinical collaborators, resulting in performance improvements of 15-20% over single-institution efforts:

See Research Teams for the complete list of institutions and researchers.

Key Journals

Primary publication venues for ChatGPT and LLM research in healthcare span informatics journals, high-impact medical journals, and AI-focused venues. Research has shown that approximately 35% of healthcare AI publications appear in informatics journals like JMIR and JAMIA, while 25% appear in general medical journals like JAMA and NEJM. In other words, the field's literature is distributed across multiple communities, requiring researchers to monitor diverse sources. Compared to traditional medical subspecialties where 60-70% of publications concentrate in 3-5 core journals, healthcare AI publishing is notably more fragmented:

• Healthcare (MDPI) - Open access healthcare research
• Journal of Medical Internet Research (JMIR) - Digital health and eHealth
• Nature Medicine - Clinical research and AI applications
• JAMA - Medical AI policy and implementation
• Lancet Digital Health - Digital healthcare innovations

See Top Journals for complete journal list with impact metrics.

External Resources

Authoritative sources for further exploration of AI in healthcare:

See Also

Related portals exploring adjacent research areas:

• Benefits & Applications - Detailed analysis of 22 identified benefits
• Concerns & Limitations - 19 concerns with mitigation strategies
• Research Teams - Leading institutions and researchers
• Top Journals - Key publication venues and impact metrics

About This Portal

This portal synthesizes knowledge from a systematic review examining ChatGPT's utility in healthcare education, research, and practice. The review analyzed 60 records from multiple databases following PRISMA guidelines.

Last updated: 2025-12