On April 19th, 2024, the Medical College of Wisconsin’s Clinical and Translational Science Institute of Southeast Wisconsin (CTSI) and Data Science Institute (DSI) held their first annual Southeast Wisconsin Data Science (SEAWINDS) Collaborative: All Thing Data Science Research Symposium. Their vision was to create a synergistic research environment in Southeast Wisconsin and build a network to advance clinical and translational research.
The symposium was composed of presentations from research leaders at MCW, MSOE, UW-Milwaukee, UW-Madison, and Northwestern Mutual. Many of the speakers considered themselves to be outliers in a room full of people interested in medical research. Their expertise lay in data science application within philosophy, ethics, math, financial marketing, and engineering. However, that was one of the aims of the SEAWINDS; to showcase the importance of interdisciplinary collaboration necessary to foster innovation and research, especially in medicine.
AN.AI Lab holds similar values, demonstrated with the research we shared at SEAWINDS 2024. One of the goals in our lab is to develop tools to better leverage electronic health data and improve clinical and surgical care through digital health applications. SEAWINDS was a great opportunity to showcase the exciting work happening at AN.AI Lab. I am also grateful to have received the Best Early Career Research Poster award from CTSI and DSI for my research.
The research I presented at SEAWINDS was titled, “Integrating a Machine Learning Model into Web-Based App to Predict After CRS +/- HIPEC.” Cytoreductive surgery along with hyperthermic intraperitoneal chemotherapy is a treatment for peritoneal carcinomatosis. This high-risk surgery is associated with high rates of readmission. The purpose of my study was to find the best machine learning (ML) model for predicting readmission and create an application to deploy the model for patients undergoing CRS +/- HIPEC at MCW. To do so, a cohort of MCW patients who underwent CRS +/- HIPEC were used to develop the best performing ML model. The best performing model had an accuracy of 83.1% and AUC of .853. The predicted probability of the model was used to classify patients as low, moderate, and high risk. I then developed an R shiny application that would allow clinicians to provide inputs to predict risk of readmission prior to a patient's discharge, ultimately to create a personalized follow up care plan.
Another project from our lab led by our resident researcher, Adhit Ramamurthi, is titled, "Utilizing GPT- 4 and Unstructured Note Data for Case length Prediction." Accurate case length is essential for optimizing operating room logistics. This study assesses the feasibility and accuracy of GPT- 4 for case length prediction to potentially integrate into our electronic health records (EHR). The study's cohort is comprised of patients undergoing inpatient elective surgery longer than 30 minutes in duration. For each patient, all notes were inputted into the LLM for summarization and again for case length prediction with summarized notes. Varying model temperatures were used at each step. GPT-4 was able to successfully predict case lengths similar to the EHR generated estimates.
Another study led by 2nd year medical student, Carson Gehl, titled, "Wearable Device/Smart Wat activity tracker as a Tool for Surgical Risk Stratification", aims to leverage the All of Us Research Program to investigate associations between preoperative step counts, measured by a Fitbit, and postoperative complications. Using the All of Us Researcher Workbench, 27,150 patients were found to have undergone surgical procedures and 475 patients had preoperative Fitbit data. After adjusting for age, race, sex, comorbid disease, and body mass index, patients with less than a 7,500-baseline step count were at increased odds for postoperative complications. This study supports the use of wearable devices and step count to help to measure preoperative fitness.
It’s always inspiring to see the research we do here at AN.AI Lab and I am thrilled to be a part of it as someone early in my career in clinical and AI research. It’s evident that our research aims to advance patient health, clinical, and surgical practices not only within MCW and Southeast Wisconsin, but also on a broader scale.