Publications on the application of artificial intelligence (AI) to many situations, including those in clinical medicine, created in 2023–2024 are reviewed here. Because of the short time frame covered, here, it is not possible to conduct exhaustive analysis as would be the case in meta-analyses or systematic reviews. Consequently, this literature review presents an examination of narrative AI’s application in relation to contemporary topics related to clinical medicine. The landscape of the findings reviewed here span 254 papers published in 2024 topically reporting on AI in medicine, of which 83 articles are considered in the present review because they contain evidence-based findings. In particular, the types of cases considered deal with AI accuracy in initial differential diagnoses, cancer treatment recommendations, board-style exams, and performance in various clinical tasks, including clinical imaging. Importantly, summaries of the validation techniques used to evaluate AI findings are presented. This review focuses on AIs that have a clinical relevancy evidenced by application and evaluation in clinical publications. This relevancy speaks to both what has been promised and what has been delivered by various AI systems. Readers will be able to understand when generative AI may be expressing views without having the necessary information (ultracrepidarianism) or is responding as if the generative AI had expert knowledge when it does not. A lack of awareness that AIs may deliver inadequate or confabulated information can result in incorrect medical decisions and inappropriate clinical applications (Dunning–Kruger effect). As a result, in certain cases, a generative AI system might underperform and provide results which greatly overestimate any medical or clinical validity.

Purpose: The purpose of this study was to assess the impact of implementing a Nurse Navigator (NN) to improve the rate and timeliness of molecular tumor testing. Methods: This is an evaluation of the impact of education sessions, consensus building, and NN implementation for molecular tumor testing in patients with epithelial ovarian cancer. The NNs’ responsibilities included attending tumor boards and ensuring Next Generation Sequencing (NGS) is ordered, reviewed, and coordinated for appropriate patients. Results: NNs significantly improved NGS testing rates from 35.29% to 77.27%, p = 0.002. Ordering a targeted panel test (TPT) was the most common reason for not ordering NGS in the pre-NN cohort (13/22, 59%). The total turnaround time for testing was reduced after the introduction of NNs from 145.2 days to 42.8 days, p < 0.0001. The post-NN group had a significantly higher rate of actionable mutations identified for the recurrent setting [67.6% versus 20.8% (p = 0.0005)] and a trend towards a higher rate of actionable mutations identified in the frontline setting [41.2% versus 33.3% (p = 0.41)]. Conclusion: NNs significantly improved somatic tumor testing rates and timeliness for patients with ovarian cancer. Discontinuing TPT in favor of NGS revealed a higher rate of actionable tumor mutations that would have been missed with TPT alone.

When women receive a diagnosis of a gynecologic malignancy, they can have questions about their diagnosis or treatment that can result in voice queries to virtual assistants for more information. Recent advancement in artificial intelligence (AI) has transformed the landscape of medical information accessibility. The Google virtual assistant (VA) outperformed Siri, Alexa and Cortana in voice queries presented prior to the explosive implementation of AI in early 2023. The efforts presented here focus on determining if advances in AI in the last 12 months have improved the accuracy of Google VA responses related to gynecologic oncology. Previous questions were utilized to form a common basis for queries prior to 2023 and responses in 2024. Correct answers were obtained from the UpToDate medical resource. Responses related to gynecologic oncology were obtained using Google VA, as well as the generative AI chatbots Google Bard/Gemini and Microsoft Bing-Copilot. The AI narrative responses varied in length and positioning of answers within the response. Google Bard/Gemini achieved an 87.5% accuracy rate, while Microsoft Bing-Copilot reached 83.3%. In contrast, the Google VA’s accuracy in audible responses improved from 18% prior to 2023 to 63% in 2024. While the accuracy of the Google VA has improved in the last year, it underperformed Google Bard/Gemini and Microsoft Bing-Copilot so there is considerable room for further improved accuracy.

PURPOSE: Tumor next-generation sequencing (NGS) testing identifies possible germline pathogenic variants (PGPVs), creating a dilemma for appropriate recognition, triage, and management. The objective of this study was to determine the clinical utility of an institutional molecular tumor board (MTB) in assessing tumor NGS reports for PGPVs. METHODS: Our institutional MTB reviews all NGS reports to provide treatment and further testing recommendations, including genetic counseling referral and consideration of genetic testing (GC/GT). We studied the patients reviewed by the MTB who were recommended for GC/GT to determine the frequency of referral to a GC, germline test completion, rate of pathogenic germline variants (PGVs), factors related to PGVs, and germline conversion rate (GCR). RESULTS: Of the 2,355 patients reviewed by the MTB during the study period, 609 (25.9%) had a recommendation for GC/GT. Of the 609 with a GC/GT recommendation, only 181 (29.7%) were referred for GC/GT by their treating physicians, and only 107 (17.6%) completed GT. Of the 107 patients completing GT, 29 (26%) had a confirmed PGV. The only factors significantly associated with PGVs were testing due to a PGPV and higher mean variant allele fraction on the tumor NGS. Only 40 patients with a GC/GT recommendation (14.3%) due to a PGPV completed GT; however, the GCR was 42.5% (n = 17/40). CONCLUSION: The MTB review of PGPV is clinically valuable, identifying PGPV in 12% of patients undergoing tumor NGS and a GCR of 42.5%. Rates of GC/GT completion were relatively low due to under-referral by treating physicians. Given the high GCR, the authors encourage institutional algorithms to help increase GC/GT rates for patients found to have PGPV following tumor NGS testing.

• The novel STK11 adnexal tumor may not be associated with Peutz-Jeghers syndrome. • Diagnosis of the STK11 adnexal tumor requires a high index of suspicion given its immunohistochemical and pathologic similarity to mesothelioma. • Further research is necessary to identify and develop evidence-based treatment strategies for these unique tumors.

We have published on the accuracy of the Google Virtual Assistant, Alexa, Siri, Cortana, Gemini and Copilot. Emerging from this published work was a focus on the ac-curacy of AI that could be determined through validations. In our work published in 2023, the accuracy of responses to a panel of 24 queries related to gynecologic oncology was low, with Google Virtual Assistant (VA) providing the most correct audible replies (18.1%), followed by Alexa (6.5%), Siri (5.5%), and Cortana (2.3%). In the months following our publication, there was explosive excitement about several generative AIs that continue to transform the landscape of information accessibility by presenting the search results in impressively engaging narratives. This type of presentation has been enabled by combining machine learning algorithms with Natural Language Processing (NLP). In 2024, we published our exploration of the generative AIs Gemini and Copilot as well as the Google Assistant in relation to how accurately they responded to the panel of 24 queries that we used in the 2023 publication. Google Gemini achieved an 87.5% accuracy rate, while the accuracy of Microsoft Copilot was 83.3%. In contrast, the Google VA’s accuracy in audible responses improved from 18% in the 2023 report to 63% in 2024. Because of our investigation in this area, we have examined the accuracy of results obtained through different AI models in this review. The landscape of the findings reviewed here surveyed 252 papers published in 2024, topically reporting on AI in medicine of which 83 articles are considered in the present review because they contain evidenced-based findings. In particular, the types of cases considered deal with AI accuracy in initial differential diagnoses, cancer treatment recommendations, board-style exams and performance in various clinical tasks. Importantly, summaries of the validation techniques used to evaluate AI findings are presented. This review focuses on those AIs that have a clinical relevancy evidenced by application and evaluation in clinical publications. This relevancy speaks to both what has been promised and what has been delivered by various AI systems. Readers will be able to understand when generative AI may be expressing views without having the necessary information (ultracrepidarianism) or is responding as if the generative AI had expert knowledge when it does not. Without an awareness that AIs may deliver inadequate or confabulated information, incorrect medical decisions and inappropriate clinical applications can result (Dunning-Kruger effect). As a result, in certain cases a generative AI might underperform and provide results which greatly overestimate any medical or clinical validity.