Clinical predictors of retinal detachment in high myopia

Paper title: Early-warning predictive model of retinal detachment in high myopia: a comprehensive construction and validation study

Authors: Wang B (1), Zhao D (1), Li R (1), Wang Q (1), Jiang F (1), Yang J (1), Shi B (1)

1. Department of Ophthalmology, Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, Hubei Province, 430022, China.

Date: Published online January 16, 2026

Reference: Wang B, Zhao D, Li R, Wang Q, Jiang F, Yang J, Shi B. Early-warning predictive model of retinal detachment in high myopia: a comprehensive construction and validation study. BMC Ophthalmol. 2026 Jan 16;26(1):73.

[Link to open access paper]

Retinal detachment is one of the most serious complications of high myopia, with risk increasing with axial elongation, leading to retinal thinning, traction, and degenerative change. The standard "passive diagnosis and treatment" model means that patients often only present after the onset of floaters, flashes, or visual field loss, meaning treatment outcomes may be limited if detachment has already occurred. While OCT provides detailed views of the macula, it may not fully capture peripheral pathology, and widefield imaging lacks integrated tools for quantifying lesion severity. This study aimed to develop a robust early warning model for myopic retinal detachment (MRD) by combining clinical data, ocular biometrics, and quantitative imaging features to support earlier identification and targeted monitoring.

A retrospective cohort of 352 patients (mean age 42 years) with high myopia (spherical equivalent of –6.00 D or more and axial length ≥ 26.0 mm) was analyzed, with participants split into a training set (n = 246) and a validation set (n = 106). Three prediction models were developed (logistic regression, support vector machine, and random forest) and their performance compared.

Key points were as follows:

• Mean refraction was just over -8.00D and axial length around 27.5mm; around 11% developed MRD within 6 months
• Axial length, level of myopia, central macular thickness, degree of vitreous liquefaction (graded from mild to severe), area of retinal degeneration, and number of retinal breaks (>1) were identified as independent risk factors for MRD within 6 months.
• Clinical signs which were not shown to be risk factors for MRD were retinal NFL and choroidal thickness; degree of vitreous opacity; staphyloma size and pigment epithelial detachment. Age, sex, family history, intraocular pressure, corneal curvature and lens thickness were also not predictive.
• The random forest analysis model was able to accommodate a wider range of input types and capture more complex patterns in the data, making it the most effective approach overall.
• Within this model, area of retinal degeneration was the most important predictor of MRD, followed by axial length ≥28mm, then macular thickness <240µm, then number of retinal breaks (>1), then degree of vitreous liquefaction.
• The model uses multi-dimensional indicators - from biometry to OCT and wide-field imaging to clinical inputs - to provide clinically applicable, risk-based monitoring. Further development and application of this tool into clinical practice workflows could support earlier identification and/or intervention in high-risk cases, particularly in patients with longer axial lengths or extensive retinal degeneration.

This study offers a practical framework for identifying which high myopia patients may be at greater risk of retinal detachment, before symptoms appear. The model brings together clinical measures already available in practice - such as axial length, macular thickness, and vitreous changes - alongside quantitative imaging of central and peripheral retinal degeneration, supporting a more structured, data-informed approach to risk assessment.

Of the indicators identified, area of retinal degeneration and axial length ≥28mm had the greatest influence, supporting more frequent monitoring of patients with extensive peripheral changes or longer eyes. Considering the additional risk factors identified here - central macular thickness <240µm, number of retinal breaks (>1) and degree of vitreous liquefaction - helps to build the overall picture of risk. The model’s strength lies in its ability to combine multiple predictors, rather than relying on a single measure, making it more adaptable to real-world variability between patients.

While this tool isn’t currently available for direct use in clinical practice, its findings can still help guide decisions, for example, when determining review intervals or identifying / referring  patients for further imaging or prophylactic treatment. Patients with large areas of retinal degeneration and/or axial lengths exceeding 28mm indicate those at most benefit from closer surveillance, even in the absence of symptoms. Patients with thinner maculae are also at risk. Although vitreous liquefaction is not routinely graded, (with posterior vitreous detachment being the end-stage), this study suggests that more advanced liquefaction may also indicate elevated risk, particularly when present alongside other structural changes.

While this study presents a structured approach to identifying retinal detachment risk in high myopia, further evidence is needed to support its clinical application. The model was developed from retrospective data at a single centre and its generalisability to broader patient populations and clinical settings remains uncertain. Large, multi-centre prospective studies would help determine whether these risk indicators can be consistently identified and applied across different patient populations and clinical settings.

The study also focused on short-term outcomes, with detachment occurrence tracked over a six-month period. Long-term predictive performance remains unknown, and future work could explore whether the same risk factors hold over extended follow-up. Additionally, while the model demonstrates the potential value of incorporating quantitative imaging features, these tools are not yet standardized in routine care and/or diagnostic instrument software. Further research is needed to determine how best to integrate this type of analysis into clinical workflows and how predictive modelling could become a more established part of risk-based monitoring for high myopia.

Objective: To develop an early warning prediction model integrating clinical indicators, ocular structural parameters, and quantitative fundus imaging features for myopic retinal detachment (MRD) to enhance the accuracy and clinical utility of MRD risk screening, thereby providing a quantitative basis for intervention in high-risk populations.

Methods: A retrospective analysis was performed on 352 patients with high myopic admitted the ophthalmology department between January 2020 and December 2023. The patients were randomly divided into a training set and a validation set at a 7:3 ratio. Twenty indicators were collected, including clinical and demographic data, ocular biometric parameters and fundus imaging and lesion assessment. Univariate analysis was used to screen MRD-related indicators, followed by least absolute shrinkage and selection operator (LASSO) regression for variable compression. Independent risk factors were identified through multivariate logistic regression. Random forest (RF), support vector machine (SVM), and logistic regression (LR) models were constructed using Python 3.9.0 and the sklearn library. Model performance was evaluated based on the receiver operating characteristic (ROC) curve and area under the curve (AUC), sensitivity, and specificity, with the optimal model selected and key predictive indicators analyzed.

Results: There was no statistically significant difference in the baseline data of patients between the training set and the validation set (P > 0.05). Through univariate analysis and logistic multivariate regression analysis, it was found that axial length (AL), central macular thickness, degree of vitreous liquefaction, area of retinal degeneration, and number of retinal breaks were independent risk factors for the occurrence of MRD (P < 0.05). The AUC of the RF model (0.890) was significantly higher than that of the LR model (0.813) and the SVM model (0.872), making it the optimal model.

Conclusion: This study successfully constructed and verified an MRD early warning prediction model based on multi-dimensional indicators. The study identified AL, central macular thickness, degree of vitreous liquefaction, area of retinal degeneration, and number of retinal breaks as independent risk factors for MRD. This model provides clear guidance for clinicians to conduct risk stratification and formulate follow-up strategies, demonstrating high clinical applicability and promotion potential.

[Link to open access paper]