Rule-based Semantics and Big Data Based Methods for Effective Clinical Decision Support (CDS): A Pediatric Severe Sepsis Case Study using ICU Data

Period of Performance: 08/01/2017 - 01/31/2018

$147K

Phase 1 SBIR

Recipient Firm

Computer Technology Associates, Inc.
2033 San Elijo Avenue, Suite 330
Ridgecrest, CA 93555
Principal Investigator

Abstract

The recognition, diagnosis, and management of sepsis remain among the greatest challenges in pediatric critical care medicine. The diagnosis of pediatric severe sepsis is particularly challenging since it is time-critical and frequently must rely on clinician interpretation of an equivocal, non-specific, age-dependent constellation of clinical signs and symptoms that occur in association with an infection or other inciting events. Currently available EMR data screening tools designed to identify pediatric severe sepsis using consensus-based guidelines criteria have poor sensitivity/specificity and/or positive predictive value that can result in poor perceived usability. Recent studies show that, in general, sepsis CDS alert fatigue is common, either in the form of false positives or false negatives (e.g. alerting focused on the most severely ill with unequivocal signs of sepsis). As a result such tools are, at best, modestly clinically effective in improving early detection of unrecognized pediatric sepsis and, in infants, can result in antibiotic administration to large numbers of uninfected newborns. We believe significant improvement in CDS usability can only be achieved with a significant improvement in sensitivity/specificity and positive predictive value (PPV) operating at a clinician-specified level of risk. This will require alerting algorithms that leverage the combined analytic power of computerized semantic models (rules, natural language processing) embodied in consensus pediatric sepsis guidelines, pediatric critical care clinician expertise, and retrospective analytics over existing large repositories of clinical encounter data using machine learning algorithms. Late last year we were awarded a SBIR Phase 1 NIH research grant focused on the use of advanced ontological models combined with ?big data? predictive analytics/machine learning (ML) techniques over ICU data as a foundation for an adult sepsis CDS. We demonstrated an AUC of 98% and PPV in excess of 85% for our sepsis detection tool for a patient sample of 15,811. We will leverage these results to establish exceptionally sensitive and specific tool that monitors PICU data to accurately predict pediatric patients at risk for impending acute clinical deterioration due to severe sepsis/septic shock. Our goal is to commercially deploy highly useable technology that achieves high levels of clinician acceptance and demonstrably influences timely treatment and pediatric patient outcomes. Our product development concept will employ human factors engineering to achieve highly synchronicity with clinical workflows, combined with a clinician-centered design of the interfaces between our CDS and the institutional pediatric sepsis protocols and EMR data sources. Our product vision is an early sepsis detection CDS with actionable accuracy and usability, compatible with any modern EMR in use at a client hospital, that is effective in reducing pediatric sepsis mortality in both critical and non-critical care settings.