Physicians make critical time-constrained decisions every day. Clinical predictive models can help physicians and administrators make decisions by forecasting clinical and operational events. Existing structured data-based clinical predictive models have limited use in everyday practice owing to complexity in data processing, as well as model development and deployment1–3. Here we show that unstructured clinical notes from the electronic health record can enable the training of clinical language models, which can be used as all-purpose clinical predictive engines with low-resistance development and deployment. Our approach leverages recent advances in natural language processing4,5 to train a large language model for medical language (NYUTron) and subsequently fine-tune it across a wide range of clinical and operational predictive tasks. We evaluated our approach within our health system for five such tasks: 30-day all-cause readmission prediction, in-hospital mortality prediction, comorbidity index prediction, length of stay prediction, and insurance denial prediction. We show that NYUTron has an area under the curve (AUC) of 78.7–94.9%, with an improvement of 5.36–14.7% in the AUC compared with traditional models. We additionally demonstrate the benefits of pretraining with clinical text, the potential for increasing generalizability to different sites through fine-tuning and the full deployment of our system in a prospective, single-arm trial. These results show the potential for using clinical language models in medicine to read alongside physicians and provide guidance at the point of care.

Safely maximizing extent of resection has become the central goal in glioma surgery. Especially in eloquent cortex, the goal of maximal resection is balanced with neurological risk. As new technologies emerge in the field of neurosurgery, the standards for maximal safe resection have been elevated. Fluorescence-guided surgery, intraoperative magnetic resonance imaging, and microscopic imaging methods are among the most well-validated tools available to enhance the level of accuracy and safety in glioma surgery. Each technology uses a different characteristic of glioma tissue to identify and differentiate tumor tissue from normal brain and is most effective in the context of anatomic, connectomic, and neurophysiologic context. While each tool is able to enhance resection, multiple modalities are often used in conjunction to achieve maximal safe resection. This paper reviews the mechanism and utility of the major adjuncts available for use in glioma surgery, especially in tumors within eloquent areas, and puts forth the foundation for a unified approach to how leverage currently available technology to ensure maximal safe resection.

No abstract

Gliomas are the most common primary intrinsic tumor in the brain and are classified as low- or high-grade according to the World Health Organization (WHO). Patients with high-grade gliomas (HGG) who undergo surgical resection with adjuvant therapy have a mean overall survival of 15 months and 100% recurrence. The renin-angiotensin system (RAS), the primary regulator of cardiovascular circulation, exhibits local action and works as a paracrine system. In the context of this local regulation, the expression of RAS peptides and receptors has been detected in different kinds of tumors, including gliomas. The dysregulation of RAS components plays a significant role in the proliferation, angiogenesis, and invasion of these tumors, and therefore in their outcomes. The study and potential application of RAS peptides and receptors as biomarkers in gliomas could bring advantages against the limitations of current tumoral markers and should be considered in the future. The targeting of RAS components by RAS blockers has shown potential of being protective against cancer and improving immunotherapy. In gliomas, RAS blockers have shown a broad spectrum for beneficial effects and are being considered for use in treatment protocols. This review aims to summarize the background behind how RAS plays a role in gliomagenesis and explore the evidence that could lead to their use as biomarkers and treatment adjuvants.

OBJECTIVEThe choice of treatment modality for optic pathway gliomas (OPGs) is controversial. Chemotherapy is widely regarded as first-line therapy; however, subtotal resections have been reported for decompression or salvage therapy as first- and second-line treatment. The goal of this study was to further investigate the role and efficacy of resection for OPGs.METHODSA retrospective chart review was performed on 83 children who underwent surgical treatment for OPGs between 1986 and 2014. Pathology was reviewed by a neuropathologist. Clinical outcomes, including progression-free survival (PFS), overall survival (OS), and complications, were analyzed.RESULTSThe 5- and 10-year PFS rates were 55% and 46%, respectively. The 5- and 10-year OS rates were 87% and 78%, respectively. The median extent of resection was 80% (range 30%–98%). Age less than 2 years at surgery and pilomyxoid features of the tumor were found to be associated with significantly lower 5-year OS. No difference was seen in PFS or OS of children treated with surgery as a first-line treatment compared with children with surgery as a second- or third-line treatment. Severe complications included new disabling visual deficit in 5%, focal neurological deficit in 8%, and infection in 2%. New hormone deficiency occurred in 22% of the children.CONCLUSIONSApproximately half of all children experience a long-term benefit from resection both as primary treatment and as a second-line therapy after failure of primary treatment. Primary surgery does not appear to have a significant benefit for children younger than 2 years or tumors with pilomyxoid features. Given the risks associated with surgery, an interdisciplinary approach is needed to tailor the treatment plan to the individual characteristics of each child.