Braydon Dymm

Dymm

Prompting is All You Need: How to Make LLMs More Helpful for Clinical Decision Support

IMPORTANCE: Large language models (LLMs) offer potential decision support, but their accuracy varies. Prompt engineering can generally enhance LLM behavior in a clinical context, yet best practices have yet to be formally explored in realistic clinical contexts for neurology. OBJECTIVE: To evaluate the impact of structured prompting versus naive prompting on the performance of four LLMs (two closed-source: OpenAI GPT-4o, OpenAI o3; three open-source: Meta Llama-4-Scout-17B-16E-Instruct, Llama-3.3-70B-Instruct-Turbo, and the reasoning model r1-1776) for thrombolytic clinical decision support (CDS) in acute stroke. DESIGN: Models responded to three novel ischemic stroke vignettes using either a naive question ("Should this patient be offered thrombolytics?") or a five-step structured prompt (CARDS) guiding information extraction, timing analysis, contraindication checking, decision process explanation, and risk-benefit discussion. Outputs were assessed across seven domains: guideline adherence, unsafe recommendations, risk recognition, guideline grading accuracy, inclusion of conversational explanation, clarity, and overall helpfulness. RESULTS: Structured prompts significantly enhanced performance across most domains, with varying effects between model families. For closed-source models (GPT-4o, o3), prompts structured in the CARDS style improved guideline adherence from 83.3% to 100%, eliminated unsafe recommendations (16.7% to 0%), and increased specific guideline grading accuracy from 0% to 100%. Similarly, the open-source reasoning model r1-1776 achieved these top-tier outcomes (100% adherence, 0% unsafe, 100% grading, 100% conversation) when structured prompts were applied, with grading and conversation improving from 0%. In contrast, other open-source models (Llama-4-Scout, Llama-3.3-70B) showed more modest gains: risk recognition improved (83.3% to 100%) and guideline grading accuracy increased (0% to 66.7%), while guideline adherence (66.7%) and unsafe recommendations (33.3%) persisted. Overall, structured prompting yielded the largest improvements in guideline grading accuracy and conversational reasoning across multiple models. CONCLUSION AND RELEVANCE: Structured prompting substantially enhances LLM performance for acute stroke thrombolysis CDS. Notably, some models, including the proprietary GPT-4o and o3, and the open-source reasoning model r1-1776, achieved excellent safety and adherence with structured prompts. For clinical deployment of any LLM, structured prompts are crucial, and vigilant human oversight remains essential.

DymmSeminars in Neurology

Artificial Intelligence in Neurology and Stroke Education: Current Applications and Future Directions

Dymm

IMPROVE Time to Anti-Coagulation Reversal for Hemorrhagic Strokes

Abstract Background Oral anticoagulation (OAC) is a risk factor for intracerebral hemorrhage (ICH) which is an important source of disability and mortality. OAC-associated ICH (OAC-ICH) patients have worse outcomes as compared to ICH patients not on OAC, likely because of the associated larger stroke volumes, higher propensity to intraventricular hemorrhage, and a higher risk of rebleeding. Although current guidelines recommend that OAC should be reversed quickly, many health care systems have not developed a process for optimizing that aspect of care. Methods Through the IMPROVE Stroke Care Consortium, a group of nine Hub hospitals and their 57 regional community hospitals, a systems of care improvement project was implemented. Performance reviews identified best practices which were disseminated throughout all centers. We compared the median door-to-reversal (DTR) time before and after an institutional campaign to speed the process with a target time of 60 minutes. Results Over two years of the study, there were 6,699 ischemic strokes, 152 subarachnoid hemorrhages, and 889 intracerebral hemorrhages. During that time, 36 hemorrhagic stroke patients received reversal agents emergently. The overall baseline median DTR time was 123 minutes (IQR 99, 361 minutes). By the end of the program, the median DTR time had trended down to 84 minutes (IQR 58.5, 151 minutes) which is a 31.7% reduction of DTR from baseline, though times remained somewhat variable (p = 0.08). Conclusions Utilizing an integrated stroke systems of care approach, we were able to reduce DTR times for patients presenting with acute ICH and concurrent use of anticoagulants despite lack of definitive guidelines around targets for reversal times or operational guidance on protocols and agents.

Longitudinal Qualitative Study of Career Decision-making of First-Year Medical Students

OBJECTIVE: The growing shortage of neurologists is in part due to suboptimal recruitment. Little is known about students' decision making regarding a career in neurology, particularly early in training. Using a longitudinal qualitative approach, we aimed to understand factors that influence first-year medical students' decisions about neurology. METHODS: We conducted 1-on-1 semistructured interviews with 15 first-year medical students at 1 institution before and after the preclinical neurology course (2018-2019). In the first interview, we asked about career intentions, factors likely to influence specialty choice, and perceptions of neurology. In the second interview, we asked about changes in students' views over the year. Using thematic analysis, we generated codes and clustered coded data into themes. RESULTS: The 2 most prominent factors influencing career choice in general were lifestyle and personal interest. No students expressed concerns about lifestyle in neurology. Most students were neutral about neurology or had a positive personal interest, which typically increased after the neurology course. Students frequently worried about content difficulty and the curative potential of neurology. CONCLUSIONS: Interventions should include early education about the factors important to students in determining specialty choice, including lifestyle, and address potentially negative perceptions of neurology. Increasing time allotment to the preclinical neurology course may combat perception of the content as difficult.

Suprascapular neuropathy: A review of 87 cases

SSN is commonly associated with axillary neuropathy. Trauma remains the most common cause of SSN. Electrodiagnostic findings aid in the initial diagnosis and may indicate the need for close clinical follow-up based on the severity of the axonal injury.