Despite increased survival rates of neonates born extremely preterm, brain lesions remain a major problem and are associated with high mortality and morbidity.1 The incidence of periventricular haemorrhagic infarction (PVHI) as a consequence of intraventricular haemorrhage (IVH) has remained high recently. According to the Vermont Oxford Network (https://public.vtoxford.org), 24% of very-low-birthweight infants experienced IVH in the industrialized world during the last decade. IVH is commonly detected in the neonatal intensive care unit using cranial ultrasound (CUS), and though there is a correlation between ultrasound-based IVH grading and neurodevelopmental outcome, large inter-study variations exist.2-5 In light of the large number of infants born preterm at risk of brain injury in combination with increasing survival rates of infants born extremely preterm, there is an urgent need to develop a robust method to appraise lesions in the neonatal brain.

Although certain CUS variables have been linked to poor subsequent development,6-11 it remains challenging to accurately predict outcome. CUS-based categorical features include ventricular area, size of intraventricular echodensity, size of PVHI, shape, topography, bilaterality of PVHI, and division according to venous anatomy. In contrast to CUS, which is heavily operator-dependent, magnetic resonance imaging (MRI) allows for a more accurate and less biased assessment of the neonatal brain. During the last decade, MRI-based scores at term-equivalent age have been established as prognostic markers of outcome in neonates born preterm.12-14 To our knowledge, all published MRI-based scoring systems have been developed for the entire population born preterm, and though severity and location of brain damage are thought to play an important role in later development, scoring systems especially designed for infants with IVH have not yet been established.

The aim of the present study was to create an MRI-based score specific to infants born preterm with IVH, assessing important brain areas as well as potential additional abnormalities commonly associated with this type of brain injury, and thereby, going beyond the scoring of the conventional 4-grade IVH classification.

METHOD

This retrospective, two-center, observational cohort study included neonates born preterm less than 34 weeks’ gestational age with IVH. The study cohort was further restricted to those patients who underwent MRI during their clinical course and who had standardized neurodevelopmental follow-up. Exclusion criteria were major congenital anomalies, cerebral malformations, metabolic disorders, chromosomal abnormalities, as well as death before follow-up. Patients were treated at either the Department of Pediatrics and Adolescent Medicine of the Medical University of Vienna, Austria or the Department of Pediatrics I of the University Hospital Essen, Germany, between October 2000 and January 2016.

CUS examinations were performed repeatedly from birth until term and IVH grades were classified based on maximum lesion extension, as seen on CUS according to Papile et al.15 This is refered to as the conventional IVH classification. In addition, the presence of posthaemorrhagic ventricular dilatation and information on whether neurosurgical intervention was required was recorded.

Magnetic resonance imaging

In Vienna, neonatal brain imaging was performed around term-equivalent age using a 1.5 Tesla MRI scanner (Philips Ingenia, Philips Healthcare, Best, the Netherlands) and an adult head or knee coil in combination with a vacuum air extraction device. In Essen, a 1.5 Tesla MRI (Magnetom Avanto, Siemens Healthcare, Erlangen, Germany) and standard coil was used until February 2011 and thereafter, a 3 Tesla MRI (Skyra, Siemens Healthcare, Erlangen, Germany) in combination with a magnetic resonance compatible incubator. MRI was reviewed and approved for analysis by a paediatric neuroradiologist (GK) and analysed together by three investigators (GK, KG, KKS) who were blinded to perinatal data, clinical course, previous CUS findings, and outcome data. In cases of disagreement between investigators, a consensus was reached by discussion. For analysis, imaging data of multiplanar T2-weighted turbo spin echo sequences was used to allow a wide use of the proposed scoring system, even if only standard sequences are available. Diffusion-weighted images and T1-weighted spin echo sequences were not used for scoring; however, these were available for the reviewing neuroradiologist in order to differentiate between periventricular infarcted regions (persisting defect) and reversible perihaemorrhagic oedema.

The MRI-based scoring system developed within this study was partly based on previously published scores,12-14 but adjusted to include lesions that particularly occur after IVH. It consists of 11 items (Fig. 1) that take into account eight brain areas (four within the grey matter: gyrus precentralis, gyrus postcentralis, hippocampus, and basal ganglia; and four within the white matter: pyramidal tract/posterior limb of the internal capsule, corpus callosum, radiatio optica, and crossroad16) and three potential additional abnormalities (periventricular leukomalacia and/or white matter volume loss, hydrocephalus, cerebellar tissue loss). Figure 1 gives a general overview of the created score while specifics on functional topography of chosen areas and a detailed MRI score description (step-by-step instruction) are shown in Appendix S1 (online supporting information). Exemplary MRI showing scoring of neonates born preterm included in this study is shown in Figures 2 and 3. As described, graded scores were used for each studied area (0–3, increasing score with increasing pathology). After appraisal, a grey matter score, white matter score, and total MRI score (composed of grey matter score, white matter score, and additional points) were calculated.

Score description. aCrossroad: the assessed area is corresponding to the C4 crossroad described in the paper by Judas et al.16 PLIC, posterior limb of the internal capsule; EVD, external ventricular drainage; WM, white matter; PHVD, posthaemorrhagic ventricular dilatation; CSF, cerebrospinal fluid; MRI, magnetic resonance imaging.

Exemplary scoring 1 (all items and magnetic resonance imaging [MRI] scores). All three patients showed a developmental quotient of >3 standard deviations below the norm for age, cerebral palsy (example [E] 1: Gross Motor Function Classification System [GMFCS] level I; E2 and E3: GMFCS level IV) and had functionally impaired vision improved by aids (E1 and E3: strabismus and myopia; E2: severe amblyopia and optic nerve atrophy). Furthermore, E1 and E2 underwent surgery for epilepsy at 12 to 19 months of age as they had epilepsy refractory to antiseizure medication. Gpre, gyrus precentralis; Gpost, gyrus postcentralis; HC, hippocampus; BG, basal ganglia; PyrT, pyramidal tract; PLIC, posterior limb of the internal capsule; CC, corpus callosum; RO, radiatio optica; CR, crossroad; WMI, white matter injury; PHVD, posthaemorrhagic ventricular dilatation; GMS, grey matter score; WMS, white matter score.

Exemplary scoring 2 (selected items). CC, corpus callosum; RO, radiatio optica; CR, crossroad.

Neurodevelopmental outcome

Outcome assessment was performed at the respective follow-up clinic by pediatricians and developmental psychologists using the Bayley Scales of Infant Development (Second or Third Edition) between 2 years and 3 years of age. Published German Bayley norms were used. Outcomes were dichotomized into two groups using conventional standard deviation (SD) banded cut-off points: favourable outcome (no–mild impairment, ≥70) and unfavourable outcome (moderate–severe impairment, <70, equivalent to >2SDs below the norm).17 Furthermore, functional assessment of visual and hearing ability were included. In the presence of cerebral palsy (CP), a Gross Motor Function Classification System (GMFCS) level was assigned. CP was dichotomized into two groups: no CP+ambulant CP (GMFCS level I–II) and non-ambulant CP (GMFCS level III–V).

Data analysis and ethics

Statistical analysis was performed using SPSS, version 20.0 (IBM Corp., Armonk, NY, USA) and SAS, version 9.4 (SAS Institute Inc., Cary, NC, USA). Data are summarized with medians and interquartile ranges (IQRs), means and SDs, and counts and percentages, as appropriate. Univariate associations between IVH grade based on CUS and MRI scores were summarized with Pearson’s correlation coefficient.

Outcome variables were analysed as continuous as well as binary variables in two separate regression models. In the first model, both scores were treated as left censored variables and analysed in a linear model adapted for left truncation using the procedure ‘nlmixed’ in SAS. A random intercept term was added for each patient because of repeated measurements. R2 values were calculated to evaluate the prognostic relevance of independent variables. In the second model, both scores were treated as binary variables with a threshold value of ≤70 points and >70 points and analysed via a logistic regression model including a random intercept term using the ‘genmod’ procedure in SAS. Finally, receiver operating curves and area under the curve were calculated for total MRI score as well as conventional IVH classification.15 P-values of <0.05 were considered statistically significant. No adjustment of p-values for multiple testing was applied.

Scoring duration was evaluated in 25% of patients. Intra- and interobserver reliability were assessed in 10% of all included neonates. Therefore, scoring was repeatedly performed after several months by a single observer (GK). The interobserver reliability was evaluated by repeating scoring in every infant by a second observer (VS) unaware of the first observers’ data. Intraclass correlation coefficients were calculated using the two-way random model for absolute agreement and interpreted according to the Brennan and Silman strength of agreement scale.

The respective local research ethics committee (ethics committee of the Medical University of Vienna: EK 1968/2017; ethics committee of the medical faculty of the University Essen: 17-7877 BO) approved this study.

RESULTS

The study cohort consisted of 103 neonates born preterm with IVH (61 males, 42 females; median gestational age 26wks 6d [IQR 25+1–29+1], median birthweight 856g [IQR 675–1200]) diagnosed using CUS. Detailed descriptive data are summarized in Table 1. Forty-five (44%) neonates with posthaemorrhagic ventricular dilatation were treated by insertion of subcutaneously tunneled external ventricular drainage or Rickham/Ommaya reservoir once the ventricular index crossed the 97th centile plus 4mm. Subsequently, in cases of progressive ventricular enlargement after external ventricular drainage removal, 30 (29%) neonates required the implantation of a permanent ventriculoperitoneal shunt and three (3%) needed a ventriculostomy (requirements for both procedures were weight >2kg and cerebrospinal fluid protein <200mg/dL). Median scoring duration per patient was 1.7 minutes (IQR 1.2–2.5).

Table 1. Descriptive data Characteristic Median (IQR); n [%] Birth year 2011 (2007–2013) Gestational age birth, wks 26+6 (25+1–29+1) Gestational age MRI, wks 39+2 (37+3–40+3) Weight, g 856 (675–1200) Male/female 61 [59]/42 [41] Apgar 1′ 7 (4–8) Apgar 5′ 8 (7–9) Apgar 10′ 9 (8–9) IVH grade based on CUS 3 (2–4) IVH II 34 [33] IVH III 27 [26] IVH IV=PVHIa 42 [41] Bihemispheric IVH based on CUS 76 [74] PHVD 51 [50] No intervention 6 [12] Only EVD/Reservoir 12 [24] EVD/reservoir+shunt 30 [59] EVD/reservoir+ventriculostomy 3 [6] MRI scoring Grey matter score 0 (0–2) White matter score 3 (0–6) Additional points 2 (0–3) Total MRI score 6 (1–10) Neurodevelopmental outcome Cognitive outcomeb 69 (50–85) Motor outcomeb 65 (49–82) GMFCS level No cerebral palsy 32 [32] GMFCS level I 47 [47] GMFCS level II 7 [7] GMFCS level III 4 [4] GMFCS level IV 9 [9] GMFCS level V 1 (1] Vision No impairment or diagnosisc not sufficient to require aids 65 [66] Functionally impaired vision improved by aids 32 [32] Blindness 2 [2] Hearing No impairment or hearing loss not sufficient to require aids 100 [99] Hearing loss improved by aids 1 [1] Hearing loss not improved by aids 0 [0] CUS, cranial ultrasound; PHVD, posthaemorrhagic ventricular dilatation; EVD, external ventricular drainage; MRI, magnetic resonance imaging; GMFCS, Gross Motor Function Classification System. a The diagnoses intraventricular haemorrhage (IVH) grade IV and periventricular haemorrhagic infarction (PVHI) were regarded as synonymous. b Median composite scores. c Includes strabismus and refractive errors not sufficient to require aids. MRI scores increase with severity of haemorrhage in neonates with IVH

To determine whether MRI scores reflect standard IVH grading and, if present, the extent of PVHI, 103 previously-obtained term-equivalent MRIs were reviewed and analysed. MRIs were performed at a median gestational age of 39 weeks 2 days (IQR 37+3–40+3) corresponding to a median postnatal age of 12.9 weeks (IQR 9.4–15.6). Fifty-three (51%) neonates did not show parenchymal injury on MRI, while 50 (49%) showed PVHI in either one (n=34, 33%) or both hemispheres (n=16, 16%). The median total MRI score was 6 (IQR 1–10) with a maximum of 30. The median values for all three subscores were 0 (IQR 0–2) for grey matter, 3 (IQR 0–6) for white matter, and 2 (IQR 0–3) for additional points. A significant positive correlation between CUS-based IVH grade and total MRI score as well as all MRI subscores was observed (R2=0.59/0.54/0.57/0.44 for total MRI score/grey matter score/white matter score/additional points; each p<0.001). Results were divided into no, unilateral, or bilateral PVHI and are shown in Table S1 (online supporting information).

MRI scores predict neurodevelopment

To evaluate the association between individual brain areas and additional abnormalities with later development, detailed analyses were performed. Outcome data were available from 95 patients and are shown in Table 1. As an influencing cofactor, gestational age at birth was significantly correlated with later neurodevelopmental outcome (cognitive, p=0.039; motor, p<0.001) and, therefore, all analyses were adjusted for gestational age.

Predictive strength of total MRI score and all individual subscores for outcome were high when using the following model: outcome = b0 + b1 × total MRI score + b2 × gestational age. The R2 for total MRI score was 0.65 for cognitive outcome and 0.81 for motor outcome. The R2 for all subscores was 0.56/0.68/0.74 (grey matter score/white matter score/additional points) for cognitive and 0.75/0.82/0.86 (grey matter score/white matter score/additional points) for motor outcome. Receiver operating curves demonstrated that total MRI score could differentiate between favourable and unfavourable outcomes, with a high area under the curve of 0.78 (95% confidence interval [CI] 0.68–0.88) for cognitive, 0.84 (95% CI 0.75–0.92) for motor outcome, and 0.94 (95% CI 0.88–0.99) for CP. Compared with the conventional IVH grading based on CUS, area under the curve of total MRI score were significantly higher by 0.16 for cognitive, 0.17 for motor outcome, and 0.15 for CP (Fig. S1 [online supporting information]). Based on these findings, our aim was to propose a tool for clinical implementation. This was achieved by translating the results from our model into easy-to-use tables based on gestational age and total MRI score, allowing probability estimation for later development (Tables 2 and 3, Fig. 4).

Table 2. Charts allowing probability estimation of favourable cognitive outcome Total MRI score GA (wks) % Difference between GA of 23wks and 34wks 23 24 25 26 27 28 29 30 31 32 33 34 0 60 63 67 70 73 76 78 81 83 85 87 88 29 1 55 58 62 65 69 72 75 78 80 82 84 86 32 2 50 53 57 61 64 68 71 74 77 79 82 84 34 3 45 48 52 56 59 63 66 70 73 76 78 81 36 4 40 43 47 51 54 58 62 65 69 72 75 77 38 5 35 38 42 46 49 53 57 60 64 67 71 74 39 6 30 34 37 41 44 48 52 56 59 63 66 69 39 7 26 29 33 36 39 43 47 50 54 58 62 65 39 8 23 25 28 31 35 38 42 45 49 53 57 60 38 9 19 22 24 27 30 34 37 40 44 48 52 55 36 10 16 18 21 23 26 29 32 36 39 43 47 50 34 11 14 16 18 20 22 25 28 31 34 38 42 45 31 12 11 13 15 17 19 22 24 27 30 33 37 40 29 13 10 11 13 14 16 18 21 23 26 29 32 35 26 14 8 9 10 12 14 15 18 20 22 25 28 31 23 15 7 8 9 10 11 13 15 17 19 21 24 27 20 16 5 6 7 8 9 11 12 14 16 18 20 23 18 17 4 5 6 7 8 9 10 12 13 15 17 20 15 18 4 4 5 6 7 8 9 10 11 13 15 17 13 19 3 4 4 5 5 6 7 8 9 11 12 14 11 20 3 3 4 4 5 6 7 8 9 10 12 10–12 21 3 3 4 4 5 6 6 7 9 10 8–10 22 3 3 3 4 5 5 6 7 8 6–8 23 3 3 4 4 5 6 7 5–7 24 3 3 4 4 5 6 4–6