Longitudinal Changes in Choroidal Thickness Across the Eye in Young Children With Premyopia

Abstract

Background

The choroid is thought to be implicated in eye growth regulation, yet longitudinal choroidal data in young children are limited.

Methods

Choroidal thickness (ChT) was measured in 92 children with functional emmetropia (baseline age, 7.4 ± 0.8 years; cycloplegic refractive error, +1.01 ± 0.55 diopters) enrolled in the 3-year PICNIC (Preventing Myopia: Investigating Contributing Factors to Nearsightedness in Children) study. Wide-field spectral domain OCT radial scans (12-line, enhanced depth imaging) were segmented using a neural network and sampled at annular eccentricities (1, 3, 6, 10, and 14 mm) in four quadrants (temporal, superior, nasal, and inferior). A generalized linear mixed model evaluated the effects of risk group (low risk vs. high risk, based on baseline hyperopic reserve and parental myopia), eccentricity, quadrant, visit, axial length (AXL), and age. A separate linear mixed model was used to assess the preonset ChT in children who developed myopia.

Results

After adjustment for age and AXL, ChT decreased with eccentricity and varied by quadrant (P < 0.001). ChT increased across visits (P < 0.001), whereas greater AXL was independently associated with a thinner ChT (P < 0.001). No main effect of risk group was observed (P > 0.20); however, a significant Risk group × Eccentricity × Quadrant interaction (P < 0.001) indicated distinct spatial profiles with low-risk children, who exhibited thicker choroids, whereas children at high risk showed steeper thinning. In the preonset analysis, the mean ChT did not differ between future myopes and persistent nonmyopes (P = 0.16), but a significant Myopia status × Eccentricity interaction (P < 0.01) indicated thinner choroids before onset.

Conclusions

Choroidal remodeling in early childhood is spatially patterned. Although the global ChT did not differ by risk group, altered spatial profiles were evident in high-risk children and future myopes before clinical onset. These results support a potential role for wide-field spatial metrics in the identification of early structural susceptibility to myopia.

Myopia prevalence is increasing worldwide, with projections indicating that by 2050 more than one-half of the global population will be myopic and nearly 1 billion people will have high myopia.¹ This surge increases the burden of sight-threatening complications such as glaucoma, cataract, retinal detachment, and myopic maculopathy,² significantly impacting socioeconomic and personal well-being. Although several interventions can slow myopia progression in children, none prevent its onset or fully halt eye growth.^3–5 A major barrier to improved treatments is the limited understanding of the underlying mechanisms of eye growth regulation and myopia development. A detailed characterization of these early structural changes, before myopia develops, is critical for clarifying the pathways of myopia susceptibility and preventing this disease.

The choroid plays a central role in emmetropization and eye-growth regulation, acting as a highly vascular and biomechanically active layer capable of rapid thickness changes in response to optical defocus,⁶ which help to fine-tune the retinal position, guide scleral remodeling, and relay retina-derived growth signals to the sclera, reinforcing its key role in coordinating emmetropization.⁷

Cross-sectional studies consistently show that the central (subfoveal) choroid is thinner in adults with myopia and thicker in hyperopic eyes, and choroidal thickness (ChT) correlates with axial length (AXL).^8–10 However, these cross-sectional data do not capture how ChT evolves during the critical period when myopia first develops and do not allow drawing causative conclusions. Longitudinal studies in adults indicate that ChT changes with age, but the direction and magnitude of these changes vary across study cohorts. For example, Lee et al.⁹ reported a gradual increase in ChT (∼1.3 µm/year) between ages 20 and 28 years, and found that a thicker baseline ChT predicted a slower myopic refractive progression, suggesting that age-related ChT change reflects more than simply differences in refractive error. Other adult longitudinal studies, however, have documented decreases in ChT over time, more prominently in the subfoveal and parafoveal areas.¹⁰ Duan et al.¹⁰ also observed a similar subfoveal and parafoveal choroidal thinning over 2 years in young adults, but they found that emmetropic eyes exhibited greater thinning than myopic eyes, and that the AXL did not significantly predict ChT change. Together, these studies indicate that (1) ChT may demonstrate longitudinal changes in adults, (2) the pattern of thinning or thickening is not uniform across retinal eccentricities, and (3) the relationship between axial elongation and choroidal thinning, which is well-established cross-sectionally, is not consistently supported longitudinally. These findings in adults suggest ChT dynamics reflect a complex interplay of aging, baseline choroidal structure, refractive state, and ocular growth patterns, as illustrated by conflicting results across adult cohorts.^9,10

Similar cross-sectional patterns have been reported in children.^11–14 Jin et al.¹³ found that the central ChT varied significantly with refractive status in children, with myopic eyes exhibiting thinner choroids and hyperopic eyes exhibiting thicker choroids than emmetropic eyes. A negative association between myopia and ChT has also been reported, as the subfoveal and parafoveal (1–2 mm) ChT was significantly greater in emmetropic compared with moderate myopic eyes.^11,12 Longitudinal studies in children more consistently demonstrate that eyes developing or progressing into myopia exhibit faster or more pronounced choroidal thinning than emmetropic eyes.^14–16 In contrast, Read et al.¹⁵ followed 101 children aged 10 to 15 years over 18 months and found a significant overall increase in the subfoveal ChT, with myopic children having thinner choroids than nonmyopic children at baseline. Although the change in ChT over time did not differ significantly between the myopic and nonmyopic children, faster axial eye growth was associated with less choroidal thickening, and some children with faster axial growth showed choroidal thinning.¹⁵ In a 2-year longitudinal study of school-aged children with myopia, Xu et al.¹⁷ reported significant choroidal thinning at the fovea as well as across parafoveal areas extending up to 1.5 mm from the fovea. Shen et al.¹⁸ followed children aged 3 to 14 years and reported that the subfoveal ChT increased in nonmyopic children with stable refraction but decreased in children showing refractive progression, further supporting an association between choroidal thinning and myopia development. Jin et al.¹⁶ followed 118 children aged 7 to 12 years at baseline for 1 year and measured ChT in the subfoveal, parafoveal, and perifoveal areas (1, 3, and 6 mm). They found that children who developed myopia showed significant thinning in all eccentricities tested, with the greatest thinning in the perifoveal (more peripheral) area, whereas children without a myopic shift showed no significant change. Consistent with these longitudinal studies, Fontaine et al.¹⁹ showed that myopic eyes exhibit progressive subfoveal choroidal thinning over time, whereas nonmyopic eyes demonstrate choroidal thickening, with ChT closely related to the AXL. These findings align with reports that faster choroidal thinning is associated with greater axial elongation and myopia progression. However, not all longitudinal studies in children agree; for example, Xu et al.¹⁷ reported variability in ChT changes with eccentricity, with some premyopic children showing less consistent thinning and weaker associations with axial elongation, highlighting inconsistencies in ChT–eye growth relationship.

These studies indicate that the choroidal thinning associated with myopia is not spatially uniform. Some studies report that subfoveal and parafoveal areas are the most affected, while others find larger changes in more peripheral (perifoveal) areas.^10,16 Yet, most pediatric longitudinal studies have examined only the subfoveal and parafoveal and perifoveal areas, typically within 6 mm of the fovea, leaving longitudinal ChT trends at wider eccentricities poorly understood in children.^16,17 Whether early myopia-related ChT changes originate centrally, peripherally, or as global shifts in the choroidal spatial profile remains unknown.

Most existing research has focused on the subfoveal or parafoveal areas, with very limited data describing how the choroid varies across the wider retina, and they do not report data before myopia onset. Understanding structural changes before myopia develops is essential for identifying early biomarkers of susceptibility. PICNIC (Preventing Myopia: Investigating Contributing Factors to Nearsightedness in Children) addressed this critical gap by following young emmetropic children longitudinally before, during, and after the onset of myopia, using a wide-field optical coherence tomography (OCT) protocol that captures structural information across the posterior pole. This design allows the evaluation of whether spatial thickness profiles, not just central subfoveal thickness, differentiate children at high risk (HR) from those at low-risk (LR) from myopia, and whether early deviations in choroidal or retinal morphology precede refractive changes. By characterizing the wide-field structural development over time, this work aimed to identify the early structural indicators of susceptibility that may inform future myopia prediction, prevention, and intervention strategies.

Methods

Participants and Study Design

Young healthy children aged 6 to less than 9 years at baseline recruited for PICNIC (n = 102) participated in these experiments.^20–25 Study visits were scheduled every 6 months (±2 weeks) over a 3-year follow-up period, resulting in a maximum of seven visits per participant. For the present longitudinal analyses, we only used those participants who had data available for a minimum of three visits (n = 92); therefore, 10 participants were not included. Informed consent/assent for study participation was obtained in writing from parents and children. The study followed the tenets of the Declaration of Helsinki and was approved by the Institutional Review Board at the New England College of Optometry.

A comprehensive ophthalmic examination was performed to confirm eligibility for study participation. The inclusion criteria at the baseline visit were (1) ages between 6 and less than 9 years, (2) uncorrected distance visual acuity of +0.10 logMAR or better in each eye, and (3) refractive error (spherical equivalent [SE], determined by cycloplegic open-field autorefraction) of functional emmetropia, that is, no less than 0.00 diopter (D) and no more than +2.25 D in the right eye, with less than or equal to 0.75 D of anisometropia or 0.75 D of astigmatism in either eye (determined by cycloplegic open-field autorefraction).²⁰ General exclusion criteria were (1) a history of ocular surgery or disease that may have resulted in visual consequences, (2) using ocular or systemic drugs that may affect their vision, (3) binocular or accommodative dysfunction, (4) a history of seizures or epilepsy, (5) the need for optical correction, and (6) being unable to provide verbal or written informed assent.²⁰

The flow of the examination for PICNIC followed the same order for all participants: ocular history, undilated refraction and entrance testing, undilated anterior ocular health examination, IOP, accommodation experimental tests,^23,24 peripheral aberrations measurements,²¹ psychophysical blur perception,²² cyclopentolate drop instillation, OCT, break/snack, ocular biometry, dilated refraction, and dilated ocular health evaluation. For feasibility reasons, OCT was conducted immediately after the instillation of the second drop of cyclopentolate (within 2 minutes). Cycloplegia was induced with two drops of 1% cyclopentolate, administered 5 minutes apart in both eyes. Cycloplegic autorefraction was measured using an open-field Grand-Seiko autorefractor (WAM-5500, Shigiya Machinery Works Ltd., Hiroshima, Japan) 30 minutes after the second drop instillation. Five measurements were taken for each eye and averaged to determine the SE, calculated as the sphere power plus one-half of the cylindrical power. Ocular biometric parameters, including the AXL, were measured using an optical biometer (Lenstar, LS 900, Hagg-Streit, Germany) after cycloplegic autorefraction. Five measurements were averaged for each eye. Ocular health was evaluated with a slit lamp and binocular indirect ophthalmoscopy and the IOP with an iCare tonometer (iCare, Raleigh, NC, USA). Testing times were not standardized across visits within participants. However, most participants were tested during the early afternoon hours (approximately 3:00–4:00 pm), shortly after school.

At the baseline visit, children were classified into HR or LR groups for future myopia development based on a combination of parental myopia and age-specific cycloplegic refractive error criteria, as described previously.^20–22 Briefly, children with lower hyperopic reserve for their age were classified as HR, whereas children with more hyperopic reserve were classified as LR. Classification thresholds also varied according to the presence or absence of parental myopia. Parental myopia was obtained from the parents’ eyeglasses prescription and/or undilated open-field autorefractor measurements at 6 m.

Myopia onset during the study was defined as a cycloplegic SE that was more negative than −0.25 D at any follow-up visit, accompanied by a corresponding reduction in distance visual acuity. This specific threshold was selected to optimize sensitivity to the earliest detectable refractive and structural shifts toward myopization, consistent with the diagnostic criteria established by major longitudinal pediatric eye growth studies.^20,26 Participants who met these criteria during the study period were classified as future myopes, while those who maintained an SE of −0.25 D or greater across all seven visits were classified as persistent nonmyopes.

Procedures: Wide-Field OCT Imaging

Wide-field spectral domain OCT (Spectralis, Heidelberg Engineering, Heidelberg, Germany) scans of the posterior pole were obtained within 2 minutes of applying the second eyedrop of cyclopentolate. The imaging protocol included a 12-line, 55° radial (RAD) scan (using an attached wide-field lens), acquired using the high-resolution setting (1536 A-scans/B-scan), with 16 frames averaged per B-scan (Fig. 1). To improve visualization of the choroid–sclera interface, images were acquired in enhanced depth imaging mode. Images were captured once; however, scans were repeated when image quality was insufficient for reliable visualization, including cases of low contrast, motion artefacts, or incomplete visualization of the choroid–sclera interface. The transverse/lateral scaling for each eye was determined using individualized ocular biometry, calculated using a three-surface schematic eye; the digital axial resolution of the device is was 3.5 µm/pixel.^27,28

Figure 1.

Mapping and segmentation of the total retinal thickness and ChT. (Left) Wide-field fundus map overlaid with an ETDRS-based grid at 1-, 3-, 6-, 10-, and 14-mm diameters. The topographic thickness (µm) was generated via scattered interpolation following manual correction of automated segmentations. (Right) Representative spectral domain OCT B-scan illustrating the three primary segmentation boundaries: internal limiting membrane (ILM, green), Bruch's membrane (BM, yellow), and the choroid–sclera interface (CSI, red). The topographic analysis yielded 20 spatial locations across four quadrants for longitudinal thickness extraction. Scale bar, 200 µm

Wide-Field OCT Image Analysis

Image analyses were performed using custom programs written in MATLAB (The MathWorks Inc, Natick, MA, USA).^29–34 In brief, we used a previously described neural network trained on a DeepLab v3+ network with a ResNet-50 backbone to identify the internal limiting membrane, Bruch's membrane, and the choroid–sclera junction.^30,31,33,34 ChT measurements were calculated on an A-scan–by–A-scan basis as the axial distance between Bruch's membrane and the choroid–sclera junction. As such, the measurements represent the axial thickness rather than the curvature-corrected distances normal to the retinal contour. Curvature and local tilt in widefield OCT images were not corrected for and may therefore influence the peripheral thickness estimates, particularly at higher eccentricities. Although efforts were made to obtain well-centered scans with minimal tilt, perfect alignment is sometimes difficult to obtain in pediatric widefield imaging.

Because the neural network was trained on 15° to 30° scans and not retrained on the 55° widefield scans, its application in this study assumes that the underlying A-scan–level reflectivity patterns used for layer boundary detection remained consistent across the scan geometries. Although differences in the field of view may introduce a domain shift, segmentation outputs were evaluated through agreement between the RAD and raster scan protocols, and all segmentations were manually corrected as necessary. The results should be interpreted as robust to moderate differences in scan width, while acknowledging that performance in extreme peripheral areas may be influenced by scan geometry and signal quality. Segmentation errors were manually corrected by a single trained masked grader, and total retinal and choroid thickness maps were generated for the scanned images using scattered interpolation.

ChT was extracted as the mean thickness averaged across all the sampled locations within each area. This mapping scheme was defined by five annular eccentricities (with outer borders at 1 mm, 3 mm, 6 mm, 10 mm, and 14 mm) across four quadrants (temporal, superior, inferior, and nasal), yielding 20 distinct spatial areas averages per visit (Fig. 1). The subfoveal area corresponds with the center of the fovea, parafoveal areas to 1 to 3 mm from the center of the fovea, and perifoveal areas to 3 to 6 mm, with the wider eccentricities representing the more peripheral retinal locations. This nomenclature reflects functional OCT-based spatial sampling. The peripheral 10 mm and 14 mm eccentricities were also included to evaluate ChT beyond the macula and into the near periphery. So, eccentricities between 6 and 10 mm encompass the perifoveal-to-near peripheral retina, including the arcuate vascular areas, while measurements beyond 10 mm sample the more peripheral areas. Although the imaging extending further to the equator would be of interest, the images beyond 14 mm were not consistently visible in all OCT scans. The optic nerve head (ONH) area was identified and excluded from ChT averaging by assigning not a number values to those locations. Because the RAD scan protocol sparsely sampled the posterior pole, the ONH was not consistently captured in all scans and was therefore not manually delineated on the corresponding image. Excluding the ONH values minimized potential confounding from the absence of choroidal tissue within the ONH.

Cross-Protocol Validation and Segmentation Agreement

This validation analysis was important to verify its performance of the neural network on the wider field structural maps used in this study (55° vs. 30°) before evaluating the longitudinal trends. Due to the lack of identical repeat RAD scans across all sessions, cross-protocol agreement was evaluated by comparing widefield RAD scans against high-density horizontal raster (VOL; 45° x 55°, 17 B-scans) scans acquired during the same visit. Statistical agreement between these modalities was quantified at the overlapping eccentricities (1 mm, 3 mm, and 6 mm) using single- and average-measure intraclass correlation coefficients (ICCs) alongside Bland–Altman analysis to evaluate systematic bias, limits of agreement, and potential proportional bias, as described elsewhere in this article.

Statistical Analysis

To evaluate the longitudinal changes in ChT across eccentricities, quadrant, and risk group, a generalized linear mixed model with a Gamma distribution and log link was implemented. Fixed effects included risk group (LR, HR), eccentricity (1, 3, 6, 10, or 14 mm), quadrant (temporal, superior, inferior, or nasal), and visit (1–7). Covariates included the AXL (mm, time varying) and age (years, time varying). Visit number was used as the primary temporal variable to model within-subject longitudinal change because all participants were examined at standardized 6-month intervals, making visit a direct measure of within-subject longitudinal change. Age was included as a time-varying covariate to account for developmental effects independent of the measurement sequence. This approach avoids conflating within-subject temporal change with between-subject age differences that can occur when age is used as the sole time metric in cohorts with variable baseline ages and narrow age ranges. The model included the full four-way interaction (Risk group × Eccentricity × Quadrant × Visit) was specified. Notably, age did not contribute significantly to model fit once the AXL was included, supporting the use of visit as the principal longitudinal index.

Participant ID was entered as a random intercept to account for repeated observations within individuals. Because each participant contributed repeated measurements across multiple eccentricities (EccIndex), quadrants, and visits, a composite repeated-measures index (RepIndex) was constructed to uniquely identify each observation within participant: RepIndex = (Visit − 1) × 20 + (EccIndex − 1) × 4 + Quadrant, where the EccIndex ranged from 1 to 5 for eccentricities ordered by increasing distance (mm). The RepIndex was modeled as a repeated factor within participant using a scaled-identity covariance structure, which assumes independence across spatial locations and visits while allowing shared variance at the participant level. Pairwise contrasts were Bonferroni adjusted for multiple comparisons. Model diagnostics confirmed an appropriate distributional fit, link function behavior and dispersion based on deviance, Pearson χ² test, and examination of the deviance residuals. Because the measurements were repeated across visits for each participant, eccentricities and quadrants were nested within each participant–eye; the models accounted for within-subject correlation using a scaled-identity covariance structure. This approach was selected because the repeated observations represent a crossed spatial–temporal design rather than a single ordered longitudinal series. The scaled-identity structure provides a parsimonious adjustment for within-subject dependence without imposing assumptions regarding the ordering or decay of correlations among retinal locations.

To assess whether the spatial ChT profiles differed before myopia onset, a separate linear mixed model was fitted, including myopia status (future myope vs. nonmyope), eccentricity (1, 3, 6, 10, and 14 mm), and their interaction as fixed effects, with the AXL included as a time-varying covariate. Participant ID was included as a random intercept to account for repeated observations within individuals across visits and eccentricities. Unlike the primary longitudinal model, visit was not included as an explicit fixed effect because the preonset analysis was designed to evaluate the overall spatial ChT profile differences between the future myopes and the persistent nonmyopes across the premyopic period, rather than to model longitudinal changes over time. In addition, visits occurred at standardized 6-month intervals for all participants, making visit and time effectively equivalent measures in this cohort. Repeated observations within participants were therefore accommodated through the mixed-effects structure, which accounts for within-subject correlation across visits and spatial locations. Residual diagnostics confirmed approximate normality, supporting the use of a Gaussian identity link. For consistency with the prior pediatric myopia literature and to reduce collinearity during the preonset window, the AXL was retained in the model in place of age.

To evaluate the performance of the neural network on the wider field 55° images, agreement between ChT measurements obtained from wide-field RAD scans and horizontal raster scans (VOL) was assessed in the three overlapping retinal eccentricities (1, 3, and 6 mm). Note that repeated RAD scans were not consistently available across visits; therefore, comparison with raster scans acquired during the same session was used as a pragmatic assessment of measurement agreement. This analysis was important to test interprotocol agreement—whether the A-scan spacing was comparable enough to obtain adequate segmentation performance in the widefield images. Three analyses were performed: (1) ICCs using a two-way mixed-effects absolute-agreement model, (2) a Bland–Altman analysis to estimate the mean bias and 95% limits of agreement, and (3) an assessment of proportional bias using linear regression of the RAD–VOL difference against the mean of both measurements. All analyses were performed in SPSS (v29, IBM, Armonk, NY, USA), with supplemental plotting conducted in R (v4.3; The R Foundation for Statistical Computing, Vienna, Austra). The ICC values were interpreted as follows: less than 0.50 indicated poor agreement, 0.50 to 0.75 indicated moderate agreement, 0.75 to 0.90 indicated good agreement, and greater than 0.90 indicated excellent agreement. Effect estimates from the mixed models were interpreted based on both statistical significance and magnitude of change across visits, eccentricities, and risk groups.

Results

The baseline demographic and clinical metrics for the analyzed cohort (N = 92) are summarized in Table. The cohort consisted of 44 children classified as HR and 48 children classified as LR for myopia. At baseline, LR children were significantly older than HR children (mean age: HR, 7.04 ± 1.60 years vs. LR, 8.13 ± 1.35 years; P = 0.002). Consequently, the baseline cycloplegic SE was significantly lower (less hyperopia) in the HR group (+0.88 ± 0.50 D) compared with the LR group (+1.25 ± 0.63 D; P < 0.001). However, no statistically significant baseline differences were observed between the risk groups for AXL (mean: HR, 22.78 ± 0.63 mm; LR, 22.69 ± 0.68 mm; P = 0.465) or distance visual acuity (median: HR, +0.02 D [interquartile range, −0.04 to +0.08 D] vs. LR, +0.02 D [interquartile range, −0.02 to +0.06 D]; P = 0.588). Sex distribution, ethnic background, and baseline astigmatism were closely balanced between groups (P > 0.050 for all) (Table). Over the 3-year follow-up period, a total of 14 children developed incident myopia.

Table.

Baseline Demographic and Clinical Metrics for the Studied Overall Cohort as Well as the Subgroups of Children at HR and LR for Myopia

	Overall Cohort (N = 92)	HR ( n = 44)	LR (n = 48)	P Value
Refraction (SE) (D, mean ± SD)	+1.00 ± 0.63	+0.88 ± 0.50	+1.25 ± 0.63	<0.001
Astigmatism (cylinder) (D, (mean ± SD)	−0.25 ± 0.50	−0.25 ± 0.50	−0.25 ± 0.50	0.715
AXL (mm, mean ± SD)	22.73 ± 0.65	22.78 ± 0.63	22.69 ± 0.68	0.465
Distance visual acuity (logMAR, median [IQR])	+0.02 [−0.04 to +0.08]	+0.02 [−0.02 to +0.06]	+0.02 [−0.02 to +0.02]	0.588
Age years, (mean ± SD)	7.42 ± 1.75	7.04 ± 1.60	8.13 ± 1.35	0.002
Sex (n, %)
Female	51	24	27	N/A
Male	41	20	21	N/A
Ethnicity (n, %)
Hispanic/Latino	22	13	9	N/A
Not Hispanic or Latino	70	31	39	N/A
Race (n, %)
White	49	17	32	N/A
Asian	9	9	0	N/A
Black/African American	12	6	6	N/A
Other/mixed	22	12	10	N/A

N/A, not available.

Data provided include cycloplegic SE in diopters (D), astigmatism (cylinder power, D), AXL in millimeters and distance (6 m) logMAR visual acuity for the right eye; also, age in years, percentage of males and females (at birth) and racial identity.

Cross-Protocol Agreement and Automated Segmentation Validation

The initial validation analysis conducted to establish the reliability of the automated deep-learning network on wide-field scans showed a strong correlation between the RAD and VOL scans, yielding ICCs values of 0.914 at 1 mm, 0.938 at 3 mm, and 0.971 at 6 mm (all P < 0.001), and average ICCs exceeding 0.95 across all overlapping areas. Bland–Altman analyses also revealed minimal systematic mean biases, with a central bias of +6.64 µm at 1 mm, which corresponds with an axial variance of less than two pixels and demonstrates highly acceptable performance. However, the 95% limits of agreement were relatively wide in the center (−45.75 to +59.03 µm at 1 mm). This greater variability in the central areas is consistent with differences in spatial sampling between the RAD and VOL scan protocols. Within the central 1-mm region, the six RAD scans converge near the fovea and therefore provide relatively dense sampling, whereas the lower-density raster protocol may contribute only a single B-scan through this small central region. Thus, although the two protocols yield similar mean ChTs, the markedly different number and spatial distribution of samples contributing to the central estimate can increase between-protocol variability. At larger eccentricities, the raster scan contributes progressively more B-scans to each annular region, reducing this sampling imbalance, even though RAD scan sampling becomes relatively less dense with increasing eccentricity.³⁵ Accordingly, between-protocol variability was substantially lower at the 3- and 6-mm eccentricities (95% limits of agreement, −24.90 to +28.04 µm at both locations) than at the central 1-mm eccentricity. This pattern is therefore consistent with the expected effects of scan geometry and sampling density and supports the use of RAD scans for circular annular analyses when the raster scan density is limited. Overall, the validation results demonstrate sufficient agreement between the two acquisition protocols to support the application of the automated segmentation approach to the wide-field RAD scans.

Spatial Profiles and Longitudinal Changes in ChT

Across the cohort, ChT exhibited a strong and systematic spatial variation. The thickness decreased significantly from the central retina toward the periphery (main effect of eccentricity, F = 1,829.20, P < 0.001), confirming robust central-to-peripheral thinning (Fig. 2). The large eccentricity effect (partial η² ≈ 0.58) reflects the well-established anatomical gradient of the decreasing ChT from the fovea toward the periphery. At the central 1-mm eccentricity, the estimated marginal mean ChT across all visits was 343.5 µm (95% confidence interval [CI], 270.3–436.5 µm) among LR participants and 322.9 µm (95% CI, 245.1–410.3 µm) among HR participants, whereas at the outermost 14–mm ring, thickness values declined significantly (LR, 258.2 µm [95% CI, 203.2–328.1 µm]; HR, 252.5 µm [95% CI,198.7–320.9 µm]). The intermediate eccentricities followed a uniform spatial track of this gradient (3 mm: LR, 336.5 µm [95% CI, 264.8–427.6 µm]; HR, 316.8 µm [95% CI, 249.3–402.5 µm]; 6 mm: LR, 317.1 µm [95% CI, 249.5–402.9 µm]; HR, 300.9 µm [95% CI, 236.8–382.3 µm]; and 10 mm: LR, 276.1 µm [95% CI, 217.2–350.8 µm]; HR, µm 267.4 [95% CI, 210.4–339.8 µm]), confirming a gradual and highly systematic spatial thinning across the posterior pole.

Figure 2.

The mean ChT as a function of retinal eccentricity (in mm) in children at LR or HR for myopia. The solid lines denote the estimated mean ChT, and the shaded areas represent the 95% CIs.

A significant main effect of quadrant (temporal, superior, inferior, or nasal) on ChT was observed (F = 5,984.16, P < 0.001; partial η² ≈ 0.64), indicating variation in thickness across the quadrants (Fig. 3). The temporal and superior quadrants showed higher ChT values, while the nasal and inferior quadrants showed lower ChT values across all eccentricities. These quadrant differences were larger in the center (1–3 mm), with mean differences of approximately 40 to 60 µm, and smaller at peripheral eccentricities (10–14 mm), with differences generally of less than 25 µm. The quadrant effects remained significant after adjustment for age and AXL.

Figure 3.

ChT for each of the five eccentricities (1, 3, 6, 10, and 14 mm) at each quadrant (superior, nasal, inferior, temporal) for children at LR (green) and HR (orange) for myopia. Data points represent the group means, and the straight lines connect the adjacent quadrants to illustrate the topographical variation without implying interpolation. Lower ChTs at the 10- to 14-mm eccentricities reflect proximity to the ONH.

A significant longitudinal effect was observed after adjustment for age and AXL (main effect of visit, P < 0.001). The model-adjusted estimated marginal means increased across follow-up. In the ring-averaged model allowing visit trajectories to differ by risk group, the adjusted ChT increased from 285.6 µm at visit 1 to 319.6 µm at visit 7 in the LR group (change = +34.0 µm) and from 269.9 µm to 306.7 µm in the HR group (change = +36.9 µm). The estimated marginal means were evaluated at common values of age and AXL; therefore, these adjusted changes differed from the unadjusted longitudinal profiles shown descriptively in Figure 4.

Figure 4.

Observed (unadjusted) longitudinal changes in the mean ChT by myopia risk group. The mean ChT profiles across the retinal eccentricities are shown for children at LR of myopia (green, left) and HR of myopia (orange, right) across study visits. The lines represent the mean ChT at each eccentricity, and the line types indicate the visit. All measurements were corrected for lateral magnification and aligned using the optic nerve–fovea angle to ensure anatomical correspondence. The error bars represent ±1 standard deviation of the mean. The values shown are descriptive raw means and therefore differ from the model-adjusted estimated marginal means reported in the Results.

Despite this longitudinal increase in ChT, the AXL remained a strong independent predictor of ChT (F = 305.09; P < 0.001), such that longer eyes exhibited thinner choroids across all eccentricities and quadrants (Fig. 5). This finding is of interest given the narrow range of the AXL found in these young children with functional emmetropia. Over the 3 years of the study, the mean AXL increased from 22.73 ± 0.65 mm at visit 1 to 23.24 ± 0.64 mm at visit 7, indicating ongoing ocular growth. In contrast, age was not a significant predictor once the AXL was included in the models (F = 0.001; P = 0.980), indicating that structural scaling related to eye size rather than age accounted for most of the developmental variance.

Figure 5.

Correlations of ChT with (A) the AXL and (B) age. (A) A significant inverse correlation between the AXL and ChT is noted across all eccentricities (1, 3, 6, 10, and 14 mm), with the strongest associations (R² = 0.10–0.14) occurring in the central areas. (B) ChT remains relatively stable across the 6- to 12-year age range (R² = ≈ 0), indicating that ocular elongation was more associated with choroidal thinning than age. Both the LR (green) and HR (orange) groups exhibit similar distribution patterns, suggesting that the structural relationship between the AXL and ChT is consistent regardless of the baseline myopia risk.

Collectively, these findings demonstrate that ChT varies along eccentricity, quadrant, AXL, and visit, with the effect sizes largest for eccentricity and quadrant, intermediate for AXL, and smaller for longitudinal change.

Risk Group Differences in Spatial Choroidal Profiles

There was no main effect of risk group on the mean ChT averaged across visits, eccentricities, or quadrants (F = 1.14; P = 0.280). In the ring-averaged analysis, LR children exhibited thicker choroids than HR children across most eccentricities, with the largest absolute differences observed centrally at the 1-mm eccentricity ring (ranging from +16.01 µm to +23.35 µm across the quadrants) compared with the 14-mm periphery (ranging from −5.51 µm to +8.07 µm). However, none of these individual post hoc pairwise contrasts attained statistical significance after adjusting for multiple comparisons (P ≥ 0.054). Conversely, a highly significant Risk group × Eccentricity × Quadrant interaction was observed (F = 338.59; P < 0.001), indicating distinct, spatially dependent differences in the topographical ChT profiles between the two risk groups (Fig. 3). Inspection of the estimated marginal means indeed revealed a consistent topographical pattern in which children at LR exhibited thicker choroids than children at HR across most eccentricities and quadrants combinations, with the greatest absolute variations centrally. At the 1-mm eccentricity, the mean adjusted differences (LR minus HR) ranged from +16.01 to +23.35 µm (∼6%–10% relative difference) across the quadrants, although individual post hoc pairwise contrasts did not attain statistical significance after strict multiple-comparison adjustments (P ≥ 0.142). At the 3-mm and 6-mm eccentricities, differences ranged from +10 to +20 µm (P ≥ 0.054), with the strongest localized trend observed in the 3-mm nasal quadrant. Beyond 10 mm, group differences compressed to less than 10 µm (P ≥ 0.410). Topographically, the temporal and superior quadrants had the largest absolute differences, while the nasal quadrant showed the smallest. These point-by-point comparisons indicate that while the three-way interaction confirms significant divergence in the overall spatial profiles (P < 0.001), individual eccentricity/quadrant values lack the localized statistical power to clear conservative post hoc adjustments.

Premyopia Differences in Future Myopes

In the preonset of myopia analyses, children who later developed myopia (N = 14) showed thinner choroids across all eccentricities before myopia development compared with those who remained nonmyopic (n = 78). Although there was no significant difference in the overall average ChT, there was a significant interaction between group and eccentricity (P < 0.01). This finding indicates that the spatial distribution of ChT differed between groups, with the profiles of future myopes and persistent nonmyopes gradually deviating as the distance from the fovea increased (Fig. 6).

Figure 6.

Spatial ChT profiles across eccentricities in children before myopia onset. Future myopes (red) demonstrated thinner mean ChT values across eccentricities compared with children who remained nonmyopic (black). Both groups exhibited progressive thinning toward the peripheral retina, as expected. A significant Myopia × Eccentricity interaction indicates different spatial patterns of ChT change between the groups. Solid lines represent regression fits, shaded areas indicate 95% CIs, and data points represent the group mean ChT values at each eccentricity.

The model-adjusted estimated marginal mean ChT was lower in future myopes than in persistent nonmyopes across eccentricities, with between-group estimated marginal means differences of 28.9 µm at 1 mm, 26.0 µm at 3 mm, and 23.0 µm at 6 mm. The corresponding estimated marginal means were 317.0 µm (95% CI, 287.7–346.3 µm) versus 345.9 µm (95% CI, 333.1–358.8 µm) at 1 mm, 312.9 µm (95% CI, 283.6–342.1 µm) versus 338.9 µm (95% CI, 326.0–351.8 µm) at 3 mm, and 296.9 µm (95% CI, 267.6–326.2 µm) versus 319.9 µm (307.0–332.8 µm) at 6 mm for future myopes and persistent nonmyopes, respectively.

Based on the observed variances, sensitivity analyses indicated a greater than 90% power to detect localized ChT differences of approximately 20 µm or greater at the central 1-mm eccentricity under the multiplicity-corrected α. The sensitivity decreased for smaller effects and toward the peripheral retina, with less than 50% power to detect differences of 10 µm or less in the far periphery. Thus, the study had a high sensitivity for central differences, on the order of 20 µm or greater, whereas the sensitivity was lower at the peripheral eccentricities where effects of 10 µm or less could not be reliably excluded.

Discussion

In this longitudinal cohort of young children, ChT demonstrated pronounced spatial variation, decreasing systematically from the central retina toward the periphery and varying significantly across quadrants. In addition, ChT was strongly and negatively associated with the AXL, even within this cohort of children with functional emmetropia at baseline and a relatively narrow range of ocular dimensions. These findings are consistent with previous studies in both adults and children reporting thinner choroids with increasing retinal eccentricity and a longer AXL.^11,13,19,36 The present study extends these observations by demonstrating that these spatial patterns are maintained over multiple years of follow-up and can be characterized across a substantially wider retinal area than has been evaluated in previous pediatric studies.

A notable finding was the significant influence of retinal location on ChT. Beyond the expected center-to-periphery thinning, substantial differences were observed among the quadrants, with the temporal and superior regions generally exhibiting a greater thickness the than nasal and inferior regions. These circumferential differences were most apparent at the central eccentricities and became less pronounced toward the periphery. Together, these findings indicate that ChT is characterized by a complex spatial organization that cannot be fully described by subfoveal or ring-averaged measurements alone. The magnitude of the eccentricity and quadrant effects observed in this study is consistent with the known anatomical topography of the choroid and highlights the value of wide-field imaging approaches for characterizing regional variation. Previous studies have reported similar nonuniform spatial distributions of ChT, although most have been limited to the macular region or to substantially narrower fields of view.^{12,13,16,36,37} By extending the measurements to a 14-mm eccentricity, the present study demonstrates that these spatial gradients remain evident well beyond the parafoveal and perifoveal regions typically evaluated in pediatric myopia studies.

The model-adjusted longitudinal increase in ChT across visits provides additional context for understanding choroidal development during childhood. Previous studies have reported choroidal thickening during normal childhood growth, whereas children who develop myopia or experience faster myopic progression tend to exhibit a reduced thickening or progressive thinning over time.^{15,16,18,19,38} In the present cohort, ChT showed a significant positive association with visit number after adjustment for age and AXL. The physiological mechanisms underlying this longitudinal change cannot be determined from the present analysis. Although developmental and refractive processes may contribute to changes in ChT during childhood, their relative contributions remain uncertain and require further investigation. Experimental and theoretical models proposed by Nickla and Wallman⁶ and Summers⁷ suggest that the choroid may participate in ocular growth regulation through biomechanical and biochemical signaling pathways linking the retina and sclera. Although the present data cannot establish the mechanisms underlying this longitudinal effect, they support the concept that the choroid remains a dynamic tissue throughout childhood.^6,15,18,38

When children were categorized according to myopia risk status at baseline (LR or HR), no significant differences were observed in the overall mean ChT. However, significant differences emerged in the spatial distribution of ChT across eccentricities and quadrants. Children at LR for myopia generally exhibited thicker choroids than children at HR, particularly at central and near-peripheral eccentricities and most notably within the temporal and superior quadrants. These differences diminished toward the far periphery and were less evident nasally. Although none of the individual location-specific comparisons remained significant after correction for multiple comparisons, the significant Risk group × Eccentricity × Quadrant interaction indicates that the overall spatial configuration of ChT differed between groups. Importantly, the absence of significant post hoc contrasts limits the ability to determine which specific retinal locations contributed most strongly to this interaction, particularly given the multiplicity correction and relatively modest localized differences. Larger studies will therefore be needed to provide greater statistical power for location-specific comparisons and to characterize more precisely the spatial pattern of ChT differences associated with myopia risk. These findings suggest that differences associated with myopia risk may be more apparent in the spatial distribution of ChT across the posterior pole than in global thickness measures alone. Similar observations have been reported in previous pediatric studies, where regional choroidal differences associated with refractive status or myopia progression varied by retinal location rather than being fully captured by average or subfoveal thickness alone.^11,16,17

A similar pattern was observed in the premyopia analyses. Children who later developed myopia exhibited thinner choroids across all eccentricities before refractive error onset, with the greatest differences observed centrally and progressively smaller differences toward the retinal periphery. The mean ChT averaged across the eccentricities did not differ significantly between future myopes and nonmyopes; however, the significant Myopia status × Eccentricity interaction indicates that the spatial profile of ChT differed before clinical onset. The magnitude of these differences was modest, and individual eccentricity-specific comparisons were not statistically significant after adjustment for multiple testing. Nevertheless, the consistency of the pattern across retinal locations suggests that spatial variation in ChT may accompany the ocular changes that occur before the development of measurable myopia. Previous longitudinal studies have similarly reported thinner choroids and greater rates of choroidal thinning in children with myopia or who exhibit faster progression.^16–19 The present findings extend these observations to children with premyopia and suggest that such differences may not be uniform across the posterior pole, but may instead be expressed as alterations in the choroidal topography.

The mechanisms underlying these spatial differences remain incompletely understood. Experimental studies have demonstrated that the choroid responds dynamically to optical defocus, accommodation, and changes in the visual environment, while clinical studies have consistently reported associations among ChT, refractive error, and axial elongation.^{6,11,36,39,40} The present findings are broadly consistent with this literature and further demonstrate that these relationships extend beyond the central macula into more peripheral retinal locations. However, because the present study is observational, the findings do not establish whether spatial differences in ChT contribute directly to subsequent ocular growth or instead accompany other structural changes associated with myopia development. Additional longitudinal and experimental studies will be required to clarify the biological significance of these spatial patterns and their relationship to ocular growth mechanisms.

The AXL was associated with ChT throughout the study, with longer eyes exhibiting thinner choroids across all retinal eccentricities and quadrants. This finding is consistent with previous reports linking a longer AXL to a reduced ChT in both children and adults.^11,13,14,36 Notably, this relationship was observed in a cohort of children with functional emmetropia at baseline and no significant differences in the AXL between the LR and HR groups. Visit remained positively associated with ChT after adjustment for concurrent AXL and age, indicating that the longitudinal visit effect was not fully accounted for by differences in ocular size. Conversely, the strong inverse AXL–ChT association remained after adjustment for visit and age, indicating that ocular size was an important independent determinant of ChT. Together, these findings suggest that the longitudinal variation in ChT and axial ocular growth are related but statistically separable components of structural development. Although these findings do not establish a causal role for ChT in ocular growth, they support previous observations that the choroidal structure remains closely linked to axial elongation throughout childhood.

Several limitations should also be considered. First, although the overall cohort was relatively large for a longitudinal imaging study, only 14 children developed myopia during the follow-up, limiting the statistical power for preonset analyses. Second, while the wide-field RAD scan protocol provided extensive spatial coverage, measurements were obtained at predefined scan orientations and therefore do not represent continuous volumetric sampling of the entire posterior pole. Third, the observed differences between groups were modest relative to interindividual variability, which likely contributed to the absence of significant pairwise comparisons at specific retinal locations. Several physiological factors have been reported to influence short-term ChT measurements, including diurnal rhythms, accommodation, light exposure, and caffeine intake.^41–45 Although caffeine intake was not permitted on testing days and OCT imaging was performed at a consistent stage of the study protocol, testing times were not strictly standardized and these physiological influences may have contributed additional variability to ChT measurements. Finally, agreement analyses demonstrated excellent correspondence between the RAD and raster scan protocols, although the limits of agreement were wider centrally than peripherally. This observation should be considered when interpreting the longitudinal ChT changes, particularly within the central retina, where physiological and measurement variability may be greater and where the observed changes may differ from the model-adjusted changes after accounting for ocular growth.

Conclusions

ChT in young children exhibited substantial spatial variation across both eccentricity and quadrant and was associated with the AXL. Although the overall ChT did not differ significantly between the risk groups or between future myopes and nonmyopes, both analyses revealed differences in the spatial distribution of ChT across the posterior pole. These findings indicate that wide-field assessment of choroidal topography may provide additional information regarding ocular structure during childhood beyond central thickness measurements alone. Further longitudinal studies with larger numbers of incident myopes will be important to determine the reproducibility, biological significance, and potential clinical relevance of these spatial patterns.