Skip to content Skip to navigation

Philadelphia Hospitals’ Informatics Teams Target Sepsis Response

May 6, 2013
| Reprints
Clinical decision support warning, sepsis registry aim to improve outcomes

Sepsis, a dangerous condition caused by the immune system’s response to a serious infection, is one of the toughest challenges hospitals face. It causes 17 percent of all hospital deaths and is the leading cause of death in non-cardiac intensive-care units. Additionally, the incidence of sepsis is increasing in the United States. But healthcare informatics researchers are working on ways to quicken response times and improve outcomes.

I saw two presentations on such efforts at the recent Mid-Atlantic Healthcare Informatics Symposium in Philadelphia. Katherine Clark, a student in the University of Pennsylvania School of Medicine, described Penn Medicine’s addition of an early warning system (EWS) about severe sepsis to its clinical decision support tool.  The EWS uses real-time laboratory and vital sign data to identify patients with sepsis at high risk for clinical deterioration and to notify providers in a timely and efficient manner.

Clark noted that the effectiveness of any clinical decision support tool depends on the perception of providers concerning its usefulness. So Penn Medicine also surveyed users about the new system to help fine-tune its use. The alert criteria for nurses and physicians were determined by a multidisciplinary team of healthcare professionals and derived from the International Sepsis Definitions Conference criteria.

After being given the warning, providers were asked to complete brief surveys immediately following their assessment of the patient. For a period covering 247 alerts, 127 surveys were collected from physicians and 105 from nurses. The survey asked a few questions trying to assess whether the alerts improved patient care. Although only a small minority of respondents reported that the alert prompted them to recognize critical illness or clinical instability sooner, about half felt the alerts altered patient management, Clark said. “In addition, about half of the respondents reported that the alerts were helpful and that they improved patient care.”

Based on these responses, Penn Medicine plans to continue fine-tuning the system in an attempt to reduce any nuisance alerts and improve alert accuracy.

Clark’s talk was soon followed by a presentation from Robert Grundmeier, M.D., director of clinical informatics in the Children's Hospital of Philadelphia’s (CHOP’s) Center for Biomedical Informatics.

Grundmeier explained that comprehensive clinical data available in electronic medical records could improve understanding of important health problems such as sepsis, where the data are challenging to understand, the disease evolves very rapidly, and the therapy has many different components.

CHOP has built an electronic pediatric sepsis registry as an alternative to manual chart abstraction for use in the emergency department and pediatric intensive care unit. The registry pulls together 244 electronically extracted variables collected for up to 28 days after onset of sepsis.        

The research team sought to compare the accuracy of the electronically extracted items with a manual chart review. They categorized discrepancies between manual abstraction and automated data extraction regarding sepsis recognition time, antibiotics, intravenous fluids, lab tests and co-morbidities.

CHOP researchers identified 140 discrepancies within the cohort of 40 children in their validation sample. For example, electronic data included some laboratory tests inappropriately. The majority of the discrepancies could be resolved by changing the automated extraction methods, Grundmeier said, although at least seven discrepancies in antibiotic administration times were due to errors in manual abstraction.

“The electronic sepsis registry required considerable upfront effort and careful manual validation,” he said. But CHOP believes that the investment in infrastructure has led to accurate and timely data collection and will be valuable to future quality improvement and research endeavors. Efforts spent on programming and data validation will be tracked in the second year of the registry’s operation to estimate costs of ongoing maintenance.This was one of the most challenging projects we have worked on, Grundmeier said, “because it involves working with such complicated clinical data sets.” He added that workflow is often not standardized for pediatric sepsis, so the quality of electronically available data at other institutions may differ.