Analysis of Brain Injury Biomarker Neurofilament Light and Neurodevelopmental Outcomes and Retinopathy of Prematurity Among Preterm Infants

Key Points Question How do serum levels of the brain injury biomarkers neurofilament light (NfL) and glial fibrillary acidic protein (GFAP) change after preterm birth, and are these biomarkers associated with neonatal morbidities? Findings In this cohort study of 221 infants born preterm, infants experienced a postnatal increase in NfL that continued for at least 1 month after birth. High serum levels of NfL were associated with the retinal neurovascular disease retinopathy of prematurity (ROP) and, in an exploratory analysis, associated with an unfavorable neurodevelopmental outcome at 2 years of age. Meaning These findings suggest that serum levels of the neuronal injury biomarker NfL might be used as an early clinical biomarker of ROP development and neurodevelopmental outcome; however, this should be investigated further.


Analytical measurements
Samples for measurements were collected for included infants according to cohort-specific protocols. The samples were frozen and stored below -70 o C in adequately monitored freezers until the measurements. When the samples have been used for earlier measurements, they have been carefully thawed on ice for a minimal time before re-frozen.
Samples were assayed for NfL and GFAP according to the kit protocol using a 1:4 onboard dilution. Samples were assayed in singlicate due to the limited volumes available; samples with insufficient volumes were excluded. Internal controls, pooled samples of serum and plasma, and calibrators were assayed in duplicate to estimate the inter-assay and intra-assay variability. Internal controls at high and low levels were assayed before and after study samples in 24 different runs. The NfL assay had inter-assay repeatability of 8.1% at a low level (average 11 ng/L) and 8.3% at a higher level (average 49.4 ng/L), and a mean (range) intra-assay coefficient of variability (CV) of 6.6% (0.5-19.5%). The GFAP assay had interassay repeatability of 11.7% at a low level (97.9 ng/L) and 9.7% at a higher level (281.5 ng/L), with a mean (range) intra-assay CV of 5.7% (0.1-22.3%).
Interpolated values were used for results outside the calibration limits. For GFAP, we analyzed 1 026 available samples from 202 of the 221 included infants. Ten samples had measurements outside the calibration limit (>4 000 ng/L); for 2 of these samples, no interpolated values were obtained. They are interpreted in this study as 10 000 ng/L, which was above the highest interpolated value. For NfL, we analyzed 1 273 samples representing 209 infants; 1 sample had a value above the quantification limit (2 268 ng/L), and 5 samples had values below the quantification limit (<2.2 ng/L).

Variable definitions
Longitudinal data on NfL and GFAP were categorized into subgroups based on the postnatal day (PND) of sampling described in the manuscript. 48 infants had more than one NfL value per period, and 34 infants had more than one GFAP value. We included a total of 1 225 samples for NfL and 992 for GFAP in the final analysis.

Statistics
Orthogonal Projections to Latent Structures (OPLS) modeling using SIMCA software version 15.0 (Umetrics AB, Umeå, Sweden), was used to investigate birth characteristics and morbidities associated with serum NfL AUC weeks 2-4 and GFAP AUC weeks 2-4. The AUCs for NfL and GFAP were log-transformed to meet a normal distribution. Permutation tests (n=999) were applied to validate the OPLS models. Unit variance scaling was used in all models. Variable importance for the projection (VIP) scores for the model with NfL AUC weeks 2-4 (n=155) are listed in eTable 2. The model included 1 predictive but no orthogonal component (R 2 X=18.6% [goodness of fit in X], R 2 Y =49.3%, and Q 2 =46.2% [goodness of prediction], ANOVA P<.001, eFigure 3). When underlying variables explaining GFAP AUC weeks 2-4 were explored, it yielded a non-significant model (not shown).