This hospital-based study was carried out at the Zhongshan Ophthalmic Center (ZOC). The protocol was reviewed and approved by the ZOC’s Institutional Ethics Committee (2012KYNL002), with all procedures performed in compliance with the principles of the Declaration of Helsinki. All participants and their parents or guardians signed documented informed consent forms.

High myopia included in this study was considered as spherical power ≤ −6.0 diopters (D). Of the 426 participants with high myopia initially enrolled between 2011 and 2012. Follow-up visits were conducted every 2 years thereafter, for a total of 8 years. Participants with secondary myopia—resulting from systemic or ocular conditions such as diabetes-related hyperglycemia—were excluded. Individuals who had worn defocus or orthokeratology lenses, undergone corneal refractive or intraocular surgery, or received atropine treatment or repeated low-level red-light therapy were also excluded. Participants with ocular diseases other than MM or with severe systemic disorders were not included.

At each visit, participants underwent comprehensive ocular assessments using standardized protocols. AL was measured prior to cycloplegia using a Lenstar LS900 (Haag-Streit AG). For subjects with AL exceeding 32 mm—which exceeds the measurement range of the Lenstar LS900—switch to the IOLMaster (Carl Zeiss Meditec) for measurement. The two devices have been shown to yield highly consistent measurements. ^, The Goldmann applanation tonometer was employed for the measurement of intraocular pressure (IOP). After confirming complete mydriasis (pupil diameter ≥ 6 mm and absence of light reflex), refraction was determined by KR-8800 autorefractor (Topcon Corporation). ETDRS LogMAR E chart (Precision Vision) was applied to assess Best-corrected visual acuity (BCVA). Fundus photographs were taken of every fully dilated eye by CX-1 camera (Canon Inc.), centered on macula and optic disc, respectively.

MM was graded into five categories through the classification system developed by the Meta-analysis for Pathologic Myopia Study Group (Meta-PM): C0, no myopic retinal changes; C1, tessellated fundus; C2, diffuse chorioretinal atrophy; C3, patchy chorioretinal atrophy; and C4, macular atrophy. Additionally, three “plus” lesions—Fuchs spot, choroidal neovascularization (CNV), and lacquer cracks (LCs)—were evaluated. Pathologic myopia (PM) was defined as the MM classification more severe than C1 or the existence of “plus” disease.

MM progression was considered as the onset of a new MM category, the appearance of a new “plus” lesion, existing atrophic lesions significantly enlarged, or the increased number of existing “plus” lesions.

Baseline fundus photographs were graded initially, followed by the grading of photographs obtained after an 8-year follow-up. All fundus photographs were graded by senior ophthalmologists with specialized training in MM classification, following a standardized Meta-PM grading manual. Inter-grader reliability was assessed using Cohen’s kappa coefficient, with median unweighted κ values of 0.84 (range, 0.71-0.85) for C0/C1 lesion differentiation, 0.80 (range, 0.79-0.84) for C2 or more severe MM lesions, and 0.73 (range, 0.72-0.79) for LCs.

Progression was assessed by side-by-side comparison, with graders aware of the baseline grading. Discrepancies among graders were resolved through committee discussion to assign a final grading for each eye. For details on the grading of fundus photographs, please refer to the previous articles. ^,

Logistic regression models incorporating generalized estimating equations (GEE) were employed to evaluate and screen potential predictive factors. Univariate regression analyses were first performed, and factors with a P -value <.10 were subsequently selected for the multivariate model. Baseline characteristics were assessed using multivariable models: Model 1 was designed with baseline age, PM, and AL, while Model 3 consisted of baseline age, PM, and SE. Two-year change rate variables were then added to create Model 2 (baseline age, AL, PM, and 2-year AL change rate) and Model 4 (baseline age, SE, PM, and 2-year SE change rate) to cf predictive performance. To quantify the discriminative performance of each prediction model, the area under the receiver operating characteristic curve (AUC) was employed as the key evaluation metric. The DeLong test was used to cf the AUCs of correlated ROC curves. Considering predictive value, only PM, AL, and the 2-year AL change rate were ultimately selected as predictors. The nomogram was finally developed based on these predictive factors, showing acceptable performance. Calibration of the model was examined through calibration curves and supplemented with the Hosmer-Lemeshow test. A significant outcome signals that the prediction model’s calibration is imperfect. Decision curve analysis was utilized for assessing the prediction model’s net clinical benefit across different threshold probabilities. Internal validation of the model was conducted using bootstrap resampling. To enhance the clinical operability of the predictive model, we determined the optimal cut-off value for the 2-year AL change rate using Youden’s Index derived from the ROC curve.

Data from both eyes of participants were included. Spherical power plus half the cylindrical power equals SE. The 2-year change rate was calculated using measurements obtained at baseline and the first follow-up visit, which was conducted after a 2-year interval.

The annualized change rate was calculated as: annualized change rate= (AL at 2-year visit − AL at baseline)/ Time interval between the two visits.

For example, if the baseline AL was 26.80 mm, the AL at the 2-year visit was 27.40 mm, and the follow-up interval was 2.0 years, the AL annualized change rate would be: (27.40-26.80)/2.0 = 0.30 mm/year.

To improve the clinical interpretability and avoid inflated effect estimates, the 2-year AL change rate was rescaled from per 1 mm/year to per 0.1 mm/year in Model 2, a unit that is clinically meaningful and consistent with typical ocular growth patterns in children and adolescents with high myopia.

Continuous data were described as mean and SD (SD). Categorical data were described using frequencies and corresponding percentages. GEE with exchangeable correlation matrices was employed to account for interocular correlation within participants. Predictive modeling and logistic regression methods have been described in the preceding sections. Statistical analyses were performed using R software (version 4.5.1). Statistical significance was set at P (two-tailed) <.05.