Very Preterm Infant Retinal Microanatomy at 36 Weeks Postmenstrual Age and 2-Year Neurodevelopment

Structured Abstract

Importance:

If non-invasive optical coherence tomography (OCT) retinal imaging could predict 2-year neurodevelopment of very preterm (VPT, <32 weeks) infants in neonatal intensive care, it might help to guide management.

Objective:

Evaluate association of retinal microanatomy on OCT in VPT infants at 36-weeks postmenstrual age (PMA) with 2-year neurodevelopmental outcomes.

Design:

Prospective, longitudinal cohort Study of Eye Imaging in Preterm Infants (BabySTEPS) at risk for retinopathy of prematurity (ROP) through their neurodevelopmental visit at age 2 years.

Setting:

Single center NICU and outpatient clinic

Participants:

72 (73.5%) of 98 surviving infants with retinal OCT imaging and neurodevelopmental follow-up, from 118 VPT infants enrolled.

Exposure:

OCT imaging of both eyes concurrent with eye examinations for ROP.

Main Outcomes and Measures:

Multivariate linear regression analysis of associations between OCT measures of retinal layers at 36±2 (SD) weeks PMA and neurodevelopmental outcomes at 2 years, assessed by Bayley Scales of Infant and Toddler Development-III (BSID-III), Child Behavior Checklist (CBCL), and Modified Checklist for Autism in Toddlers-Revised (MCHAT-R), adjusted for infant factors: small for gestational age (SGA) at birth, GA, mother’s highest education level and PMA at imaging.

Results:

In 72 infants (54% male, mean±SD GA 27.6±2.6 weeks and birthweight 946±288 grams) greater retinal nerve fiber layer (RNFL) thickness (regression coefficient (95% confidence interval (CI)) per 10-micron increase) was associated with higher BSID-III Motor (7.50 (4.38, 10.62), P<.001) and Cognitive (3.71 (0.73, 6.69), P=.02) scores, and lower MCHAT-R autism risk (−0.64 (−1.19, −0.09), P=.03). Greater choroidal thickness (per 100 microns) was associated with higher Motor score (4.84 (0.45, 9.23), P=.03).

When considered in addition to infant factors, RNFL thickness improved prediction of BSID-III Motor score (R² from 0.36 to 0.53, difference 0.17 (95% CI: 0.04, 0.26)),and Cognitive score from (R² from 0.29 to 0.35, difference 0.06 (95% CI: 0.00, 0.16)); choroidal thickness improved predictions less than and did not add to that of RNFL.

Conclusions and Relevance:

In this study RNFL thickness in VPT infants at 36 weeks PMA predicted motor and cognitive neurodevelopmental outcomes at age 2 years. These findings, if confirmed independently, could help guide management through 2 years of VPT infants.

Infants born very preterm (VPT, < 32 weeks GA), have high rates of poor neurologic development (50–70%),[1, 2], with rates of impairment, autism [3, 4], learning difficulties [5], special education services [6]), and behavioral problems[7, 8] all higher than for term infants. The immature retina of the VPT infant is at risk for ROP, which has been associated with and likely shares common pathological pathways with neurodevelopment.[9] Systemic inflammation is associated with poorer infant eye-brain neuro- and vascular development,[10–16] head growth suppression with severity of ROP and with serum insulin-like growth factor-1,[17, 18] and low serum levels of brain derived neurotrophic factor with more severe ROP.[19, 20] In a meta-analysis of 38 studies, ROP diagnosis and severity were associated with increased risk for cerebral palsy (CP) and cognitive, behavioral, and neurodevelopmental impairment,[21] and in a study of >79,000 preterm infants, treated ROP (compared with untreated) was associated with intellectual disabilities, psychiatric and behavioral disorders, and motor impairment at age 1–2 years.[22]

Our premise is that in the VPT infant population, the retina, an accessible extension of the diencephalon, reflects common eye-brain developmental pathways disrupted by preterm birth and neonatal intensive care‥ Optical coherence tomography (OCT), which uses reflectivity of infrared illumination to provide a conventional and a cross-sectional near histological view of the retina at multi-micron resolution, is standard in diagnosis and care of adult neurovascular disease [23], accessible for infants through cart-based, clinical (Bioptigen, Envisu SDOIS) and research systems, [24, 25] and low stress for preterm infant imaging when compared with clinical ophthalmoscopic examination. [26] (Figure 1) Thus, OCT imaging provides measures of retinal microanatomic neurovascular development, disease or injury that either reflect disruptions in neurovascular development in the central nervous system, or themselves limit key neurodevelopmental functions, such as visual perception, attention, visual motor skills, and academic and social communication skills.[27]

Figure. Photograph of the Noncontact Retinal Optical Coherence Tomography (OCT) System, OCT Image, Retinal Nerve Fiber Layer Thickness (RNFL) Map for the Temporal Quadrants, and Representative B-Scans From the Volume.

A. Example of the bedside noncontact retinal OCT system in the intensive care nursery. B. Representative macula-centered OCT image with a retinal view, generated from the swept-source OCT system. C. The organizing axis (yellow line) extends from the center of the optic nerve head to the foveal center. The thin pink arc represents both temporal quadrants (from −45° to +45° relative to the organizing axis), and the thick pink arc represents the papillomacular bundle (from −15° to +15° relative to the organizing axis). D. Representative foveal B-scan from macular volume from the swept-source OCT system. E, Representative foveal B-scan from macular volume from the swept-source OCT system with Duke OCT Retinal Analysis Program Marking Code Baby version 2.0 semiautomated segmentation at the internal limiting membrane (pink), outer border of the nerve fiber layer (magenta), inner nuclear layer (inner, aqua; outer, yellow), inner border of the outer nuclear layer (green), ellipsoid zone (blue). retinal pigment epithelium (inner, purple; outer, red), and choroid (orange). F, Representative B-scan from the swept-source OCT system with Duke OCT Retinal Analysis Program Marking Code Baby version 2.0 semiautomated segmentation at the internal limiting membrane (magenta) and the outer border of the nerve fiber layer (blue).

From research OCT imaging in a small cohort of VPT infants who had clinical MRI’s at term equivalent age and 18 – 24 month neurodevelopmental assessments, our group found that ganglion cell axonal layer thinning at the optic nerve head and in the adjacent retina (thinner retinal nerve fiber layer, RNFL) correlated with severity of brain injury on MRI and with neurodevelopment.[28, 29] Presence of cystoid spaces (edema) in the retina of preterm infants was also associated with poorer neurodevelopment.[30] Because MRI and neurodevelopmental measures were completed as part of clinical care, infants at greater risk of brain injury may have been included. A major goal of the BabySTEPS (STudy of Eye imaging in Preterm infantS) prospective cohort studies has been to distinguish elements of retinal microanatomy that predict poor neurodevelopment, and to test the validity of the RNFL and macular edema findings above. In the current study in VPT infants, we obtained OCT imaging with an investigational high speed device at the time of ROP examinations followed by neurodevelopment assessments at 2-years corrected age. We report the relationship between the OCT measures and the neurodevelopmental outcomes for these infants.

Methods

BabySTEPS (NCT02887157), also called Analyzing Retinal Microanatomy in ROP to Improve Care, was approved by the Duke Health System Institutional Review Board and enrolled 118 infants from August 12, 2016, through November 12, 2019, who had parental or legal guardian written consent for participation and who met the American Association of Pediatrics eligibility for ROP screening[31]. Infants with congenital or chromosomal anomalies that profoundly impact brain development were excluded. Participants received compensation during the 2-yr neurodevelopment visit. This study used the STROBE cohort reporting guidelines.[32]

Research procedures for OCT imaging were detailed in prior publications [25, 26]: participants in the neonatal intensive care unit (NICU) with pupils dilated for ROP examination were imaged with the noncontact handheld scanner of the investigational OCT system [25]. (Figure, A) Expert graders, masked to health and ROP, reviewed OCT images captured from both eyes of infants from the session closest to 36 weeks in the predetermined window of 34 through 38 weeks’ premenstrual age (PMA) for predictors of interest. Multiple macular scans were captured, with the best selected for segmentation and grading; this included least tilt and best quality of image [25]. (Figure B-F) OCT predictors included presence of the foveal ellipsoid zone (an indicator of photoreceptor maturity that separates from retinal pigment epithelium after intervening photoreceptor outer segments have grown sufficiently)[33]; presence of macular edema (cystoid spaces) and its severity[34]; and thickness, adjacent to the optic nerve, of the RNFL[29, 35, 36] (Figure, C and F) and at the fovea of inner retina (thickening as a marker of immaturity),[37] inner nuclear layer (the layer most affected by edema), total retina, and choroid. (Figure, B, D, and E).

The primary neurodevelopmental outcome was the Bayley Scales of Infant and Toddler Development-III (BSID-III) Motor, Cognitive and Language scores.[37] Mean scores equal 100 with a standard deviation (SD) of 15, and higher scores associated with better development. Assessments were completed by psychologist examiners certified by the National Institute of Health Neonatal Research Network and masked to OCT imaging data. Secondary outcomes included parent-reported behavioral functioning: the Child Behavior Checklist/Preschool (CBCL)[38] to assess Internalizing Behavior, Externalizing Behavior and Total Problems, (Mean = 50; SD=10) with higher scores indicating more problems, and the Modified Checklist for Autism in Toddlers – Revised (MCHAT-R) to screen for symptoms of autism, (score range 0–20) with higher scores indicating higher risk for autism.[39, 40].

Univariate and multivariate linear regression models were used to evaluate associations between the measures of retinal microanatomy on OCT at 36+/-2 weeks PMA and 2-year neurodevelopmental scores. Multivariate models were adjusted for GA, SGA status at birth, and mother’s highest educational level (influences neurodevelopment [6, 41, 42]), and GA and PMA at imaging (impacts retinal measures[43–45]). Performance of OCT measures in predicting 2-year neurodevelopmental scores was evaluated using R². Study design included sample size of 80 participants to provide 80% power (at type I error rate of 0.05) for detecting associations with R² at least 0.10 between microanatomy measures and neurodevelopmental scores using multivariate model with adjustment of confounders [46]. Statistical analyses were performed in SAS v9.4 (SAS Institute Inc., Cary, NC). All P values were two-sided but not adjusted for multiple analyses.

Results

Of 273 VPT infants screened, 118 were enrolled, 98 were retained until NICU discharge (12 transferred to another hospital and 8 died) and 72 infants (90% of the retention goal of 80) returned for the 2-year neurodevelopmental visit from January 18, 2019, through January 12, 2022, at mean (SD) corrected age 25.2 (1.2) months. The infants in our analyses had a mean GA of 27.6 weeks, a mean birthweight (BW) of 945.5 grams and 18% were SGA. Eleven of the infants were treated for ROP with first treatment from 33 through 44 weeks PMA. Table 1 summarizes participant demographic characteristics, and OCT features and measures. Mean RNFL thickness difference between right and left eyes was 2.59 (6.97) um. The 26 (26.5%) infants lost to follow up had GA of 28.8 weeks, birthweight of 1084 g, and sociodemographic characteristics and OCT findings similar to those who returned. Mean BSID-III composite standard scores for the participant group were more than 1 SD below the BSID-III mean of the normative sample. Mean scores for CBCL and for MCHAT-R autism risk were in the normal range. (Table 2)

Table 1:

Demographics and OCT-based retinal thicknesses and features of the cohort (N=72)

Birth weight (g), Mean (SD)	945.5 (287.8)   497
Small for Gestational Age, N (%)	13 (18.1%)
Gestational age (weeks), Mean (SD)	27.6 (2.6)   498
Female	33 (45.8%)   499
Mother’s Self-identified Race, N (%)
Asian	3 (4.2%)   500
Black	32 (44.4%)   501
More than One Race	2 (2.8%)
White	35 (48.6%)   502
Mother’s Self-identified Ethnicity, N (%)	503
Hispanic or Latino	6 (8.3%)
Not Hispanic or Latino	66 (91.7%)   504
Mother’s Education, N (%)	505
8th grade or less	6 (8.3%)
9th to 12th grade	6 (8.3%)
High School Diploma	8 (11.1%)
Partial college or Associate’s degree	28 (38.9%)
College degree	17 (23.6%)
Graduate degree	7 (9.7%)
PMA at OCT imaging (weeks), Mean (SD)	36.2(1.0)
OCT-based retinal thicknesses* and features
Inner Nuclear Layer at foveal center, Mean (SD)	97.4 (90.0)
Inner Retina** at foveal center, Mean (SD)	51.1(21.7)
Total retina*** at foveal center, Mean (SD)	213.4 (111.8)
Central choroidal thickness, (N=70), Mean (SD)	241.0 (92.5)
Retinal nerve fiber layer thickness, (N=70), Mean (SD)^	61.0 (11.3)
Macular edema^^, N (%)	34 (47.2%)
Mild, moderate, severe, macular edema, N (%)	17 (23.6%)/ 17 (23.6%)/ 9 (12.5%)
Ellipsoid zone at foveal center^^^, N (%)	2/72 (2.8%)

^*Thickness was based on the average of two eyes,

^**Inner retina included retinal nerve fiber layer, ganglion cell layer and inner plexiform layer.

^***Total retina included all retinal layers and retinal pigment epithelium.

^{^}Retinal nerve fiber layer thickness is adjacent to optic nerve in papillomacular bundle.

^{^^}Macular edema was defined based on presence in either eye, and severity was determined based on the more severe eye.

^{^^^}Analysis of photoreceptor development was not performed due to the small number of infants who had the postulated marker of ellipsoid zone at fovea (appears if photoreceptor outer segments lengthen sufficiently [71]).

Table 2:

Distribution of 2-year neurodevelopmental and behavioral scores among study participants (N=72 infants)

Scores	# of infants (%)	Mean (SD)	Median [1st, 3rd quartile]	[range]
BSID-III Composite Scores (corrected for prematurity)
Cognitive	72	83.3 (16.3)	85.0 [70.0, 95.0]	[54.0–120.0]
Mild cognitive delay (70–84)  Mod-severe cognitive delay (<70)	18 (25.0) 12 (16.7)
Language	69*	79.3 (19.4)	79.0 [65.0, 94.0]	[46.0–127.0]
Mild cognitive delay (70–84)	17 (24.6)
Mod-severe cognitive delay (<70)	24 (34.8)
Motor	72	84.3 (19.9)	88.0 [75.3, 97.0]	[46.0–130.0]
Mild motor delay (70–84)  Mod-severe motor delay (<70)	12 (16.7) 15 (20.8)
Behavioral Scores
MCHAT-R	71**	1.8 (2.9)	1.0 [0.0, 2.0]	[0.0–16.0]
Normal (0–2)	56 (78.9)
Suspect (>2, <8)	10 (14.1)
Autism (≥8)	5 (7)
CBCL Internalizing problems	71**	48.0 (10.6)	45.0 [41.0, 55.5]	[29.0–74.0]
Normal <65	67 (94.4)
Abnormal ≥65	4 (5.6)
CBCL Externalizing problems	71**	50.6 (11.3)	51.0 [42.0, 59.5]	[28.0–80.0]
Normal <65	65 (91.5)
Abnormal ≥65	6 (8.5)
CBCL Total problems	71**	50.3 (10.8)	50.0 [44.0, 59.0]	[28.0–73.0]
Normal <65	65 (91.5)
Abnormal ≥65	6 (8.5)

^*Three participants were missing language scores: one due to child primary language other than English and interpreter not available, one due to tracheostomy/ventilator, and another due to behavioral resistance at the time of testing.

^**One parent did not complete behavioral questionnaires.

In the univariate model, higher RNFL thickness was associated with better BSID-III Motor, Cognitive and Language scores, and inversely associated with MCHAT-R score and CBCL Internalizing Problems score (Table 3; Supplemental Figure 1). In the multivariate model, most of the associations appeared to be maintained, but not the Language score. Per 10-micron increase in RNFL thickness, the mean (95% confidence interval (CI)) Motor and Cognitive scores increased by 7.50 (4.38, 10.62, P<.001) and 3.71 (0.73, 6.69, P=.02) points respectively, and for MCHAT-R and CBCL Internalizing Problem scores declined by 0.64 (0.09, 1.19, P=.03) and 2.25 (0.09, 4.41, P=.04) points respectively. RNFL thickness moderately predicted BSID-III Motor (R²=0.32 (95% CI: 0.15, 0.48), P <.001) and very weakly predicted Cognitive (R²=0.17 (0.04, 0.34), P <.001) scores (Supplemental Table 1). When considered in addition to infant factors, RNFL thickness improved the prediction of BSID-III Motor score (R² increased from 0.36 to 0.53, a difference of 0.17 (95% CI: 0.04, 0.26)) and Cognitive score (R² increased from 0.29 to 0.35, a difference of 0.06 (95% CI: 0.00, 0.16)) (Table 4).

Table 3:

Associations between OCT measures at 36 weeks PMA and neurodevelopment measures at 2- years

Retinal Structure from OCT	Retinal Nerve Fiber Layer (RNFL) thickness				Choroidal thickness				Inner retinal thickness
ND Outcome at 2-years	Univariate Model		Full Multivariate Model*		Univariate Model		Full Multivariate Model*		Univariate Model		Full Multivariate Model*
	Regression coefficient per 10 um increase (95% CI)	P value	Regression coefficient per 10 um increase (95% CI)	P value	Regression coefficient per 100 um increase (95% CI)	P value	Regression coefficient per 100 um increase (95% CI)	P value	Regression coefficient per 10 um increase (95% CI)	P value	Regression coefficient per 10 um increase (95% CI)	P value
BSID-III Composite Scores
Cognitive	5.95 (2.81, 9.09)	<.001	3.71 (0.73, 6.69)	.02	6.82 (3.02, 10.62)	<.001	3.74 (−0.14, 7.62)	.06	−2.43 (−4.10, −0.76)	.004	−1.20 (−3.04, 0.64)	.21
Language	5.79 (1.79, 9.79)	.01	2.11 (−1.73, 5.95)	.28	7.00 (2.34, 11.66)	.004	3.10 (1.60, 7.80)	.20	−2.94 (−4.94, −0.94)	.01	−0.68 (−2.91, 1.55)	.55
Motor	10.03 (6.60, 13.46)	<.001	7.50 (4.38, 10.62)	<.001	8.56 (4.05, 13.07)	<.001	4.84 (0.45, 9.23)	.03	−3.10 (−5.06, −1.14)	.003	−0.87 (−3.05, 1.31)	.43
Behavioral Scores
MCHAT-R	−0.87 (−1.44, −0.30)	.004	−0.64 (−1.19, - 0.09)	.03	−0.76 (−1.27, −0.25)	.04	−0.28 (−0.99, 0.43)	.44	0.36 (0.07, 0.65)	.02	0.00 (−0.35, 0.35)	.98
CBCL Internalizing Problem	−2.99 (−5.07, −0.91)	.006	−2.25 (−4.41, −0.09)	.04	−1.50 (−4.13, 1.13)	.27	0.14 (−2.72, 3.00)	.93	−0.36 (−1.50, 0.78)	.53	−1.36 (−2.69, −0.03)	.05
CBCL Externalizing Problem	0.59 (−18.42, 19.60)	.63	0.94 (−1.45, 3.33)	.44	−0.30 (−3.12, 2.52)	.83	0.20 (−2.58, 2.98)	.89	0.37 (−0.85, 1.59)	.54	−0.95 (−1.09, −0.81)	.20

^*Adjusted by gestational age, small for gestational age, postmenstrual age at OCT imaging, and mother’s highest education level, OCT: optical coherence tomography, RNFL: retinal nerve fiber layer, MCHAT-R: Modified Checklist for Autism in Toddlers – Revised, CBCL: Child Behavioral Checklist

Table 4:

Prediction performance (R²) using recognized infant factors without and with inclusion of various retinal structure thicknesses at 36 weeks post menstrual age for predicting 2-year neurodevelopmental and behavior scores

	Infant factors	Retinal structures from OCT added to the analysis
	GA, small for GA, postmenstrual age at OCT imaging, and mother’s highest education level	With RNFL thickness added	With Choroidal Thickness added	With RNFL+ Choroidal Thickness added
	R2 (95% confidence interval*)	R2(95% confidence interval*)	R2(95% confidence interval*)	R2(95% confidence interval*)
Neurodevelopmental Outcomes: Bayley Scales of Infant and Toddler Development-III Composite Scores
Cognitive	0.29 (0.18 – 0.52)	0.35 (0.21 – 0.59)	0.32 (0.20 – 0.55)	0.35 (0.21 – 0.60)
Language	0.31 (0.18 – 0.52)	0.34 (0.20 – 0.57)	0.33 (0.19 – 0.55)	0.34 (0.21 – 0.58)
Motor	0.36 (0.24 – 0.58)	0.53 (0.38 – 0.74)	0.39 (0.26 – 0.62)	0.51 (0.37 – 0.73)
Behavioral Outcomes
MCHAT-R Score	0.25 (0.17 – 0.46)	0.32 (0.21 – 0.58)	0.26 (0.18 – 0.49)	0.31 (0.21 – 0.60)
CBCL Internalizing Problem Score	0.11 (0.04 – 0.34)	0.18 (0.09 – 0.44)	0.11 (0.05 – 0.37)	0.15 (0.09 – 0.44)
CBCL Externalizing Problem Score	0.13 (0.07 – 0.37)	0.14 (0.08 – 0.42)	0.14 (0.08 – 0.41)	0.17 (0.11 – 0.47)

GA: gestational age, OCT: optical coherence tomography, RNFL: retinal nerve fiber layer, MCHAT-R: Modified Checklist for Autism in Toddlers – Revised, CBCL: Child Behavioral Checklist

^*Based on 1000 bootstraps.

In the univariate model, higher choroidal thickness at 36 weeks PMA was associated with better BSID-III Cognitive, Language and Motor scores, and inversely associated with MCHAT-R score (Supplemental Figure 2). In the multivariate model, higher choroid thickness remained associated with better BSID-III Motor scores and trended towards association with Cognitive scores (P=.06). (Table 4) In the multivariate model, for every 100-micron increase in choroidal thickness, mean (95% CI) Motor scores increased by 4.84 (0.45, 9.23, P=.03) points and Cognitive scores increased by 3.74 (−0.14, 7.62, P=.06) points. Choroidal thickness very weakly predicted BSID-III Motor scores (R²=0.16 (95% CI: 0.03, 0.35), P <.001) and Cognitive scores (R²=0.15 (0.03, 0.33), P <.001). When considered in addition to infant factors, choroidal thickness improved the prediction of BSID-III Motor score (from R²=0.36 (0.24, 0.58) without, to R²=0.39 (0.26, 0.62) with) and Cognitive score (from R²=0.29 (0.18, 0.52) without, to R²=0.32 (0.20, 0.55) with) (Table 4). Choroidal and RNFL thicknesses together moderately predicted Motor scores (R²=0.34 (0.18, 0.50), P<.001) and weakly predicted Cognitive scores (R²=0.22 (0.08, 0.41), P<.001) (Supplemental Table 1). When considered with demographic factors, choroidal thickness and RNFL thickness together did not show better prediction for BSID-III outcomes than RNFL alone. (Table 4) In a post hoc analysis, choroidal thickness and RNFL thickness were weakly correlated (R=0.37 (95% CI: 0.14, 0.56), P=.002) (Supplemental Figure 4).

In univariate analyses, thicker foveal inner retina was associated with lower BSID-III Cognitive, Language, and Motor scores and with higher MCHAT-R autism risk score (Supplemental Figure 3); however, none of the associations suggested in the univariate analyses were noted within multivariate analyses (Table 4). In univariate analyses, total retinal thickness, macular edema, or macular edema severity were not associated with subsequent neurodevelopmental or behavioral measures (Supplemental Table 2).

Discussion

This prospective longitudinal, single site, cohort study using a research OCT device in the NICU showed that measures of retinal microanatomy—specifically RNFL and choroidal thickness—of VPT infants at 36 weeks PMA aided in the prediction of neurodevelopmental outcomes at 2-years corrected age. Greater RNFL thickness predicted better motor and cognitive neurodevelopmental outcomes at 2 years, improving prediction performance by 17% beyond known infant factors for motor development and by 9% for cognitive supporting our hypothesis and validating our earlier findings.[29] Greater RNFL thickness was also associated with lower autism risk score and lower internalizing behavior problems score, measures that were not previously tested. Choroidal thickness was associated with better 2-year motor outcomes. Neurodevelopment was not associated with other retinal findings such as macular edema. Thicker inner retina at the fovea is a marker of immaturity, and it was associated with lower BSID-III scores in univariate analysis. but not when corrected for GA.[36].

The findings of this study were supportive of the RNFL thickness association with neurodevelopment identified by Rothman et al in 2015 in 33 VPT infants who were sicker than and from the same hospital as this cohort. There, RNFL thickness weakly correlated with BSID-III Cognitive and Motor scores, and in the multivariate model, both RNFL and birth weight contributed to Cognitive Bayley scores (R²=0.29 (95% CI: 0.18, 0.52), P=.006).[29]. We demonstrated that a 10-micron increase in RNFL associated with 7.5 and 3.71 point increase in BSID-III Motor and Cognitive scores respectively, 2.25 point decrease in CBCL internalizing behavior problems, and 0.64 point lower MCHAT-R raw score (raw score >2 points was associated with increased risk of autism). We believe the 10-micron increase in RNFL is clinically relevant, as the difference between the 1^st and 3^rd quartiles in this population was 15 microns. Our findings point to thickness of RNFL, which is comprised primarily of ganglion cell axons which extend to the lateral geniculate nucleus, as an early marker of parallel neurodevelopment.

A potential value of this study is assessment of retinal structures early in life, before sequelae of ROP treatment and eye and brain growth deform or cause atrophy of retinal tissue[47]. From BabySTEPS, thickness of VPT infant RNFL at 36-weeks PMA is associated with infant health before and after birth. Birthweight, reflecting development in utero, was an independent predictor of RNFL thickness, with a 5.2 micron increase (95%CI: 3.3, 7.0; P<.001) per 250 gm increase in weight [34]; after considering confounding variables, RNFL was 5.6 microns thinner (95% CI: −10.8, −0.3; P =.04) in infants who had prior necrotizing enterocolitis or sepsis.[48] RNFL thickness and choroidal thickness increase with VPT infant age in a biphasic manner, with a decrease in slope at approximately 37–38 weeks PMA[43, 44]. We do not yet know whether earlier measures or the trajectory of growth of RNFL and/or choroid may provide better or earlier predictors of neurodevelopment and plan further study.

The relationship between RNFL thickness and neurodevelopment may be through visual function impacted by ganglion cell injury. In 61 infants (50 also in this cohort) better grating visual acuity at 9-months corrected age correlated with RNFL thickness at 36 weeks PMA independent of GA, birth weight, ROP stage and need for ROP treatment.[49] In school-aged children with comparable spherical equivalent refractive error, RNFL has been found to be overall thinner in those born preterm compared with born at term age [50, 51], to increase with birthweight [51, 52], to have thinning associated with previous stage 3–4 ROP,[51] and to correlate with visual acuity.[50] These studies compared retinal microanatomy with ocular history and visual outcomes, but not neurodevelopment. In children with a history of ROP, worse than expected neurodevelopmental outcomes have been found in those with unfavorable visual status.[53, 54] In a cross-sectional study of 11-year-old children born VPT, peripapillary RNFL thickness was associated with motor development and with better visual acuity, and the volume of the ganglion cell nuclei was associated with cognitive development [55]. We will try to study this cohort to determine the visual, retinal anatomic, and neurodevelopmental outcomes at school age.

We identified a relationship between thicker choroid and higher BSID-III Cognitive and Motor scores. The clinical relevance appears moderate (improvement of 4.84 points Motor and 3.74 points Cognitive for 100-micron increase in choroidal thickness). Although the added impact on neurodevelopment appears to be minimal when compared with that of RNFL, the vascular choroid, provides a distinct independent anatomic measure that relates to neurodevelopmental outcomes. Prior reports from BabySTEPS have shown that thinner choroid was also associated with plus disease of ROP, bronchopulmonary dysplasia, and poorer weight gain early in life.[33, 56] We believe choroidal thickness reflects different pathophysiological processes impacting VPT infant development and may be a relevant marker for therapies affecting these pathways.

Macular cystoid spaces, present in 30 to 60% of preterm infant eyes, have been a provocative finding since their discovery on OCT imaging in 2011[57, 58]. A study of 53 preterm infants that included most infants in the 2015 RNFL study[29], found an association between macular edema and neurodevelopment on BSID-III in preterm infants [30]. We did not find such an association in our univariate analysis. Prior studies have shown distinct morphology of cystoid spaces in infants depending on whether hypoxic ischemic retinopathy (associated with hypoxic ischemic encephalopathy),[59] or other disease was present[60] and on GA or age at imaging of VPT infants.[61] Cystoid spaces at 32–34 weeks PMA have a different prognosis than when present closer to term age.[62] Thus, in future analyses we will consider the timeline of edema relative to neurodevelopment.

Based on findings from this study, use of OCT data may be of benefit to neonatologists working to improve therapeutic strategies to optimize developmental care.[63] Bedside OCT imaging with non-visible infrared light has been shown to induce significantly lower stress in preterm infants compared with a visible light eye examination.[26] While brain MRI is a powerful tool to predict neurodevelopmental outcomes,[64, 65] the cost, location, magnet restrictions, and need for lack of motion limit its use for very sick infants, those prior to term age, and for repeated imaging. Retinal OCT measures can be repeatedly obtained at the bedside to monitor development in at-risk children and inform neurodevelopmental prediction models.[41] OCT analyses relative to brain MRI imaging and neurodevelopmental outcomes is underway in a subset of this cohort. Use of OCT images may afford rapid and early enrollment in inpatient therapies and community early intervention programs, which may be crucial to maximizing the plasticity mechanisms of the brain and improving developmental outcomes in these children.[66] RNFL thickness, as a longitudinal objective measure reflective of neurodevelopment, could be monitored regularly during therapeutic interventions affecting neurodevelopment.

Limitations:

The OCT system used in this study was a research device that is not yet commercially available, which may limit the immediate clinical relevance of these findings. While OCT imaging lacks the stability of chinrest imaging, we have demonstrated good reproducibility of handheld OCT macular and RNFL measurements.[67, 68] This was a single site study; independent validation across multiple sites would help support the generalizability of the findings. Twenty-six infants were lost to follow up by year 2, which may limit the generalizability of our findings; this cohort may have been at greater risk for neurodevelopmental impairment than those who did not return. Neurodevelopmental impairment in the retained group is consistent with the substantial literature of poorer neurodevelopment in preterm born children [3,4].

Conclusions:

In this study, RNFL thickness in VPT infants at 36 weeks PMA predicted motor and cognitive neurodevelopmental outcomes at age 2 years. School age follow up of these children is underway to determine if neonatal retinal microanatomy, such as RNFL and choroid thickness, continue to be associated with adverse neurodevelopmental outcomes, particularly given the persistent neurodevelopmental issues that have been identified in VPT children at school age.[5, 69, 70] Indeed, if subsequent research shows that these findings persist, OCT, which is a noninvasive and low-cost assessment of retinal health, may help predict which VPT infants are most at risk for adverse neurodevelopmental outcomes, allowing for earlier and more intensive developmental and medical intervention designed to help ameliorate these problems.

Key Points.

Question:

Does retinal microanatomy from optical coherence tomography (OCT) imaging of very preterm (VPT, < 32 weeks gestation age [GA]) infants’ eyes predict later neurodevelopmental and behavioral outcomes?

Findings:

In a cohort study of 72 VPT infants, thicker retinal nerve fiber layer (RNFL) measured on bedside OCT at 36 weeks PMA predicted higher Bayley Scales III Composite Motor and Cognitive scores, and lower autism risk score and internalizing behavior problems score at 2 years, after correcting for small for GA, and maternal education.

Meaning:

Retinal OCT imaging before term age may aid in identifying 2-year neurodevelopmental outcomes in VPT infants.