Abstract
BACKGROUND AND AIMS
Preterm infants are prone to neonatal infections such as late-onset sepsis (LOS). The consequences are severe and potentially life-threatening. Unfortunately, often LOS presents with unspecific symptoms, and early screening laboratory tests have limited diagnostic value. Aim was to build a predictive algorithm to aid doctors in earlier detection of LOS in very preterm infants.
METHODS
In a retrospective cohort study, all consecutively admitted preterm infants (GA ≤ 32 weeks) from 2008 until 2019 were included. Infants were classified according to blood culture results, currently the gold standard, in LOS and control patients. Routinely and continuously measured oxygen saturation and heart rate were extracted from electronic medical records to generate features. Care was taken to not include variables indicative of existing LOS suspicion. Timing of blood culture served as proxy for LOS-onset. An equivalent timestamp was generated in GA-matched controls. Two machine learning techniques (Generalized Additive Model and Logistic Regression) were used to build a classification algorithm up to 24 hours before blood culture. Hourly predictions were generated for the total hospitalization period.
RESULTS
389 infants with LOS were GA-matched to 1501 controls, median GA was 28.1 and 30.3 weeks, respectively. The algorithm yielded an AUC of 0.76 (p
CONCLUSIONS
Our algorithm based on routinely collected data can potentially accelerate clinical decisions, even with relatively restricted inputs. Prospective validation is needed to prove benefit in clinical practice.
Preterm infants are prone to neonatal infections such as late-onset sepsis (LOS). The consequences are severe and potentially life-threatening. Unfortunately, often LOS presents with unspecific symptoms, and early screening laboratory tests have limited diagnostic value. Aim was to build a predictive algorithm to aid doctors in earlier detection of LOS in very preterm infants.
METHODS
In a retrospective cohort study, all consecutively admitted preterm infants (GA ≤ 32 weeks) from 2008 until 2019 were included. Infants were classified according to blood culture results, currently the gold standard, in LOS and control patients. Routinely and continuously measured oxygen saturation and heart rate were extracted from electronic medical records to generate features. Care was taken to not include variables indicative of existing LOS suspicion. Timing of blood culture served as proxy for LOS-onset. An equivalent timestamp was generated in GA-matched controls. Two machine learning techniques (Generalized Additive Model and Logistic Regression) were used to build a classification algorithm up to 24 hours before blood culture. Hourly predictions were generated for the total hospitalization period.
RESULTS
389 infants with LOS were GA-matched to 1501 controls, median GA was 28.1 and 30.3 weeks, respectively. The algorithm yielded an AUC of 0.76 (p
CONCLUSIONS
Our algorithm based on routinely collected data can potentially accelerate clinical decisions, even with relatively restricted inputs. Prospective validation is needed to prove benefit in clinical practice.
| Original language | English |
|---|---|
| Publication status | Published - 10 Oct 2022 |
| Event | European Acadamy of Pediatric Societies - Barceolona, Spain Duration: 7 Oct 2022 → 11 Dec 2022 https://eaps2022.kenes.com/ |
Conference
| Conference | European Acadamy of Pediatric Societies |
|---|---|
| Abbreviated title | EAPS |
| Country/Territory | Spain |
| City | Barceolona |
| Period | 7/10/22 → 11/12/22 |
| Internet address |