Continuous Glucose Monitoring Linked to an Artificial Intelligence Risk Index: Early Footprints of Intraventricular...

To develop and validate a new risk score for intraventricular hemorrhage (IVH) in preterm neonates based on continuous glucose monitoring (CGM).

We retrospectively analyzed CGM traces obtained from 50 very preterm neonates, grouped into two sub-cohorts started on CGM within 12 and 48 h of birth, respectively. A CGM linked to an Artificial Intelligence Risk (CLAIR) index was developed to quantify glucose variability during the first 72 h of life in neonates with and without IVH. Brain-US was performed at least twice a day for the first 5 days of birth. An integrated remote monitoring platform was developed to capture major clinical events in real time and gather data for the risk index. The new score performance was further compared with other measures of glucose variability (coefficient of variation [CV] and standard deviation [SD]) and with a clinical risk index for babies II (CRIB-II) as a predictor of IVH event. The two cohorts were analyzed separately for internal validation of the method.

The primary cohort consisted of 26 neonates (gestational age 30 [28, 31] weeks; BW1275 g[1090, 1750]). Controls (n = 23) exhibited higher CLAIR index than cases (P = 0.004). A cut-off of 0.69 for the new CLAIR index allowed a 100% sensitivity and an 83% specificity for IVH prediction. The CLAIR index was the sole significant predictor for IVH (P = 0.003) when compared with clinical variables, CV, SD, and CRIB-II. In a subgroup analysis in very low-birth-weight infants, the CLAIR index was the sole variable significantly associated with IVH (P = 0.009). Analysis on the secondary cohort (five cases and 16 controls) confirmed a higher CLAIR index in the controls (P = 0.008), in the absence of a difference for CV, SD, and CRIB-II between the two groups.

CGM, combined with the AI-algorithm, provides a high-sensitivity index for risk detection of IVH that reflects the glycemic impairment preceding IVH.