OpenRounds

LLMs Recommend Far More Care Than Doctors

Sometimes a serendipitous study is just what the doctor ordered, especially when you need some empirical evidence to support your case on AI alignment. Ask, and you shall receive. A new NEJM AI study ...

Precision targeting using machine learning to optimize medication therapy management delivery in value-based care: A quasi-experimental study using real-world evidence.

Published in Journal of managed care & specialty pharmacy. Abstract: Medication therapy management (MTM) is a core component of managed care pharmacy that improves medication safety, adherence, and clinical outcomes in high-risk populations. However, traditional eligibility criteria lack precision, limiting scalability and impact. To provide real-world evidence of how machine learning (ML) integration into pharmacy workflows can improve MTM delivery by identifying patients most likely to engage. We conducted a retrospective cohort study using a de-identified pharmacy dataset. MTM-eligible patients in 2025 were assigned to centralized or community pharmacists via an ML-based precision-targeting model, whereas a 2024 cohort was assigned using traditional eligibility-based assignment. The propensity score matching model incorporated demographics, medication patterns, social determinants of health, engagement history, and plan-level/store-level characteristics. The primary outcome was the call favorability rate (acceptance of a pharmacist consultation). After propensity score matching, 183,888 matched patient pairs were analyzed. The ML cohort showed a 21.2% call favorability rate vs 12.6% in the historical cohort, an 8.6% absolute increase and 68% relative improvement (risk ratio = 1.68). Results were robust to moderate unmeasured confounding (E value = 2.75). Results were statistically significant ( P < 0.001) and remained stable with checks for bias. ML-based precision targeting was associated with higher MTM call favorability compared with tra...

Clinical AI governance for scalable innovation in value based healthcare systems

Label-free Femtosecond Laser Imaging Combined With the Fast Lung Artificial Intelligence Model for Rapid Intraoperative Diagnosis of Lung Surgical Specimens: A Multicentre, Prospective, Parallel-workflow, Non-inferiority Study. [RECRUITING]

This study aims to evaluate whether femtosecond laser imaging combined with FastLung AI model can provide intraoperative diagnostic performance that is non-inferior to standard frozen section diagnosi...

Fine-tuned Medical Vision-Language Model (MedGemma 1.5 4B) with and without Structured Radiologist Descriptors versus EU-TIRADS for Malignancy Risk Stratification of Thyroid Nodules at Ultrasound: A Tunisian Multicenter Diagnostic Accuracy Study

Purpose. To evaluate a medical vision-language foundation model fine-tuned on thyroid ultrasound images, with or without the radiologist's structured descriptors, against expert-rated EU-TIRADS for the classification of thyroid nodules as benign or malignant. Materials and Methods. This retrospective study included nodules that underwent ultrasound-guided fine-needle aspiration (FNA) between January 2019 and August 2025 in two Tunisian centers, with consecutive recruitment, together with a selection of images from an international public dataset. Of 1,580 nodules, 1,456 (272 malignant; 1,280 patients) with an interpretable cytological or histological reference standard were allocated at patient level to training (n = 942), validation (n = 258) and a test set sealed before training (n = 256; 51 malignant). MedGemma 1.5 4B was fine-tuned with low-rank adaptation under four input configurations: the image with nodule biometry, with or without age and sex (A, B); A with chain-of-thought supervision (C); and A plus the eleven qualitative descriptors of the radiologist's structured report (D). The EU-TIRADS category was never a model input. Areas under the receiver operating characteristic curve (AUROCs) were compared with DeLong's test and a prespecified non-inferiority margin of 0.05; EU-TIRADS >= 4 was the prespecified comparator, with >= 5 as a sensitivity analysis. Results. Config D reached an AUROC of 0.953 (95% CI 0.909-0.976), versus 0.948 (0.894-0.972) for EU-TIRADS (difference +0.004, 95% CI -0.016 to +0.025), which met the non-inferiority criterion; Configs A-C, without the qualitative descriptors, reached 0.883-0.892 (Holm-adjusted p = 0.045 versus EU-TIRADS). At >= 4, EU-TIRADS had a sensitivity of 98.0% and produced 94 false positives among 205 benign nodules; at a threshold fixed on validation data, Config D reclassified 72 of them as negative while missing t...

A hospital AI deal done in under two months | Dr. Ilya Burkov, Nebius & Thomas Knox, VitVio

Dr. Ilya Burkov, Global Head of Scientific AI and Healthcare at Nebius, and Thomas Knox, Co-founder and CEO of VitVio, join Christina Farr to talk about why AI companies are crowding into healthcare and what it takes to actually get adopted. [Excerpt; full description in article]

Before You Buy an AI Tool for Patient Questions, Count the Questions

The following is a guest article by Keith Schmidt, MD Every practice that sends patients home with pre-procedure instructions is already running a patient education program. It just is not staffed, me...

Microsoft’s Mustafa Suleyman says its AI outperforms doctors on hard cases as he chases ‘medical superintelligence’

Microsoft’s new AI diagnostic system beat generalist doctors on hard cases, according to a study in Science. But most of the improvement seems to come from the AI models underneath, not from Microsoft...

Generative artificial intelligence in medical oncology: A clinically focused narrative review of large language, vision-language, and agentic artificial intelligence systems.

Published in Cancer. Abstract: Rapid advances in cancer genetics and genomics, pathophysiology, diagnostics, and therapeutics have substantially increased the complexity of clinical oncology. Integrating large and diverse data streams into timely, patient-specific decisions provides a compelling rationale for incorporating artificial intelligence (AI) into cancer care. Generative artificial intelligence (GenAI) encompasses several classes of models capable of generating new content. This clinically focused review concentrates on GenAI applications based primarily on large language models (LLMs) and vision-language models (VLMs), as well as emerging agentic systems that use these models to perform coordinated, multistep tasks. LLMs demonstrate substantial medical knowledge, strong performance on oncology-related assessments, and emerging capabilities in clinical reasoning. Guideline- and retrieval-augmented systems can improve concordance with expert recommendations and interpretation of complex treatment algorithms. LLMs and VLMs have been applied to documentation, patient communication, information extraction and summarization, clinical decision support, and multimodal data synthesis. However, performance varies substantially across tasks and clinical domains, with important limitations including incomplete or outdated information, training-data bias, automation bias, and confabulation. Agentic AI represents an emerging extension of these technologies in which LLMs or VLMs are incorporated into systems ca...