Vital, a leading artificial intelligence- (AI) driven digital health company, announced a significant milestone with the publication of a groundbreaking peer-reviewed scientific publication in the Journal of Medical Internet Research AI (JMIR AI). The paper focuses on sepsis prediction, marking a crucial step in Vital’s ongoing commitment to delivering solutions that drive better patient outcomes while providing a positive patient experience.
Sepsis is a life-threatening condition that must be treated promptly and is often difficult to detect, even by experienced physicians, in its earliest stages. The peer-reviewed paper, “Sepsis Prediction at Emergency Department Triage Using Natural Language Processing: Retrospective Cohort Study,” highlights the power of artificial intelligence (AI) in predicting sepsis at the earliest stages.
“For years, Vital has helped improve patient experience scores, follow-up adherence, and patient understanding,” said Vital’s CEO Aaron Patzer. “And we’re improving outcomes as well. Early inpatient outcomes appear to be shaving 0.5-days off the average inpatient stay, which is non-trivial. Our latest research on sepsis prediction reinforces this track record of patient impact.”
The study reveals that sepsis can accurately be predicted at initial emergency department (ED) presentation, using nursing triage notes and clinical information available within the patient’s medical record at the time of triage. This breakthrough indicates that machine learning (ML) and natural language processing (NLP) can facilitate timely and reliable alerting for intervention, and holds significant potential for saving lives.
Findings include:
- Timely sepsis prediction: The research demonstrates that ML algorithms can accurately predict sepsis at ED presentation, enabling healthcare providers to intervene promptly and enhance patient care. The manuscript features a dual-model sepsis prediction system. The time-of-triage and comprehensive models, respectively, predicted sepsis each with an area under the curve (AUC) of 0.94 and up to 0.97. Sepsis was accurately predicted in 76% of septic patients where sepsis screening was not initiated at triage and 98% of cases where sepsis screening was initiated at triage.
- Utilizing free-text data: The paper emphasizes the importance of free-text data (e.g., free-form notes from a triage nurse during initial patient assessment that are entered into the patient’s electronic medical record) in improving the performance of predictive modeling at the time of triage and throughout the ED course. This opens new avenues for using unstructured data to drive better patient outcomes.
- Prediction explainability: The manuscript illustrates how model-based sepsis predictions can be explained by highlighting key words and/or phrases that are suggestive or non-suggestive of sepsis in nursing triage documentation and clinical information.
To coincide with this publication, Vital is announcing future plans to deliver solutions that use predictive analytics to notify physicians and staff about quality and safety issues. Vital’s chief medical officer and practicing emergency department physician, Dr. Justin Schrager shared, “Vital is currently beta testing our sepsis model within our client community. The results continue to encourage and validate the need for us to double down on predictive analytics, solidifying our position as leaders in AI who are committed to innovation.”
“While I’m proud of Vital’s usage metrics, the Google ratings we’ve helped clients achieve, and hospitals’ ROI, I’m certainly looking forward to one day putting a big meter on the website: number of healthy years created,” Patzer added.
SOURCE: BusinessWire