Health system-scale language models are all-purpose prediction engines.

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.

A Review of the Literature on the Transciliary Supraorbital Keyhole Approach.

BACKGROUND: Conventional craniotomy approaches involve substantial soft tissue manipulation that can cause complications. The transciliary supraorbital keyhole approach was developed to avoid these complications. The aim of this review is to evaluate the safety and the effectiveness of the transciliary supraorbital keyhole approach. METHODS: We searched the PubMed/Medline database for full-text publications from 1996 onward containing data on 100 or more cases of aneurysm clipping or tumor resection by the transciliary supraorbital keyhole approach. The primary outcome was the incidence of approach-related complications. The secondary outcomes were the aneurysm occlusion rate and the extent of tumor resection. RESULTS: Eight publications met the eligibility criteria. All publications were of the retrospective case-series or case-cohort type without any independent assessment of outcomes. The risk of bias at the individual study level may thus have influenced any conclusions drawn from the overall study population, which included 2783 patients with 3085 lesions (2508 aneurysms and 577 tumors). Approach-related complications included 3.3% cerebrospinal fluid collection or leak, 4.3% permanent and 1.6% temporary supraorbital hypesthesia, 2.9% permanent and 1% temporary facial nerve palsy, and 1% wound healing disturbance or wound infection. Complete aneurysm clipping was achieved in 97% of cases. Complete tumor resection in 90% of cases. The overall surgical revision rate was 2.5%. The esthetic outcome was typically reported as highly acceptable. CONCLUSIONS: This approach may represent a safe, effective, and less invasive alternative to conventional craniotomies in experienced hands and for a well-selected subset of patients. However, higher-level evidence is needed to confirm this hypothesis.