Yanwei JIN 靳颜蔚

I am a second-year MS student in Biostatistics at the University of Minnesota School of Public Health. My research focuses on E-health, disease modeling, and machine learning applications in healthcare.

I work as a Research Assistant with Dr. Feng Xie (UMN Surgery) on emergency sepsis prediction models using MIMIC-IV-ED and UMN datasets, and with Dr. Xiao Zang (UMN Public Health) on drug overdose mortality in older Black populations.

Previously, I received my Bachelor degree from Peking University in 2022 and worked at the Hong Kong Centre for Cerebro-cardiovascular Health Engineering (COCHE) on cuff-less blood pressure estimation using deep learning advised by Prof. Yuan-Ting Zhang.

Publications

My research focuses on E-health, disease modeling, and the application of machine learning and statistical methods to improve healthcare outcomes.

We developed the Emergency Sepsis Risk Prediction (ESRP) score using the AutoScore framework based on electronic health records from three health systems. The ESRP score demonstrates superior performance compared to qSOFA, NEWS, MEWS, and REMS for early sepsis identification in emergency departments across multiple healthcare systems.

We propose multimodal McBP-Net, a hybrid CNN-LSTM architecture for continuous beat-to-beat blood pressure estimation using PPG, ECG, IPG, and skin temperature signals. Validated on 23 subjects during cold pressor test, McBP-Net achieves mean absolute errors of 4.19 and 2.98 mmHg for systolic and diastolic BP. Published in IEEE JBHI (Volume 29, Issue 8).

We introduce MedThink-Bench, a benchmark for rigorous assessment of LLMs’ medical reasoning with 500 questions across ten domains, each annotated with expert step-by-step rationales. We propose LLM-w-Ref, a novel evaluation framework using LLM-as-a-Judge to assess reasoning with expert-level fidelity. Benchmarking twelve state-of-the-art LLMs showed smaller models (e.g., MedGemma-27B) can surpass larger proprietary ones (e.g., OpenAI-o3).

Model Training Method, Physiological Indicator Detection Method, Apparatus, and Electronic Device

Investigates the effects of cold pressor test on heart rate and HRV in 22 subjects. Cold water exposure significantly increased HR (p<0.001) and decreased HRV, reflecting autonomic balance regulation.

A novel deep RNN model with optimal PPG and ECG feature set for continuous blood pressure estimation. Evaluated on 660 subjects, it achieved MAE of 2.514 and 1.383 mmHg for SBP and DBP, attaining grade A according to British Hypertension Society standards.

A systematic review and meta-analysis of 21 studies with 42,855 participants evaluating public risk perception during COVID-19. Results show 51.1% had high-level risk perception, with appropriate coping behaviors including mask wearing (53.2%) and hand hygiene (63.9%).

A qualitative phenomenological study exploring risk perception evolution among medical staff during public health emergencies. Semi-structured interviews with 16 staff identified three stages: Vigilance (high alertness and active protection), Observation (standardized behaviors and emotional fluctuation), and Maintenance (normalized adaptation with stable response).

Modified version of template from here