PREDICTION OF LATE ONSET SEPSIS IN VERY LOW BIRTH WEIGHT INFANTS BY A SOFTWARE APPLICATION--ARE WE THERE YET?].

Early prediction of late onset sepsis is imperative in order to improve survival and reduce long-term complications. Since clinical deterioration is detrimental, empiric antibiotic treatment is initiated once sepsis is suspected. Symptoms that may indicate invasive infection are non-specific. Previous risk scores offered to improve clinical decision-making but provided low predictive values. To evaluate the quantitative early alert of software application compared to clinical judgment by the treating physician, and the "gold standard" of positive blood and/or positive cerebrospinal fluid. Weight, heart and respiratory rates, episodes of bradycardia and desaturation, and temperature were collected for each neonate and loaded daily into the system for a period of 30 days by a registered nurse. The medical team and the registered nurse were blind to the system alerts. Analysis of the correlation between the software alerts, the clinical suspicion of sepsis and bacteremia was conducted. Forty-five very low birth weight consecutively born infants who did not have early onset sepsis and survived, were evaluated, of whom 17 infants had culture proven bloodstream infection. The software positive predictive value was 6%, 23%, 31%, at 12, 24, 48, hours respectively for alerts approximately to positive cultures. The positive predictive value of clinical suspicion of LOS was 28% but increased from 25% with low levels of clinical suspicion to 34% with high levels of clinical suspicion. The software application did not improve sepsis prediction. However, further trials may develop a more accurate algorithm that will alert the physician to be more attentive to infants in special cases.

Citation

Zivanit Ergaz, Shmuel Benenson, Noa Ofek-Shlomai, Sinan Abu-Leil, Benjamin Bar-Oz. PREDICTION OF LATE ONSET SEPSIS IN VERY LOW BIRTH WEIGHT INFANTS BY A SOFTWARE APPLICATION--ARE WE THERE YET?]. Harefuah. 2016 Jan;155(1):15-9, 68

Mesh Tags

Substances

PMID: 27012068

search → result