Johns Hopkins University researchers have developed an algorithm that correctly predicted septic shock in 85% of patients a median of 28.2 hours before onset, according to study results.
Using routinely collected measurements and laboratory results from the Multiparameter Intelligent Monitoring in Intensive Care-II (MIMIC-II) Clinical Database, we developed and validated TREWScore for septic shock, Suchi Saria, PhD, assistant professor of computer science and health policy at Johns Hopkins University Whiting School of Engineering, and colleagues wrote. At comparable specificities, TREWScore achieved a higher sensitivity than did either a routine screening protocol frequently used to initiate treatment for severe sepsis and septic shock or Modified Early Warning Score, a general severity score that has been used to identify sepsis.
In addition to predicting the majority of septic shock cases, the algorithm maintained false-positive rates that were comparable to other current screening methods.
But the critical advance our study makes is to detect these patients early enough that clinicians have time to intervene, Saria said in a press release.
TREWScore considered 54 potential measurements from electronic health records of the MIMIC-II Clinical Database which includes all 16,234 patients admitted to ICUs between 2001 and 2007 at Beth Israel Deaconess Medical Center in Boston before automatically selecting and then learning weights for 27 measurements most indicative of septic shock, the researchers wrote. The model accounted for patients who received treatment before developing septic shock and excluded patients who received treatment without developing septic shock.
TREWScore identified patients at a median of 28.2 [interquartile range (IQR), 10.6 to 94.2] hours before onset, and achieved a sensitivity of 0.85 and a specificity of 0.67, with a false-positive rate of 33%, according to study results. The model identified 68.8% of patients before they experienced any sepsis-related organ dysfunction a median of 7.43 hours before septic shock.
When compared to an example routine screening tool, TREWScore showed a 58.6% increase in the number of patients identified before any sepsis-related organ failure. [This] indicates that data-driven early warning scores can be powerful tools for adverse-event prediction, Saria and colleagues wrote. When they are coupled with evidence-based therapies and performance improvement initiatives, there is substantial potential to improve patient outcomes and help make real the vision of learning health care systems. by Will Offit
Disclosure: The researchers report no relevant financial disclosures.