Prediction of cycloplegic refraction for noninvasive screening of children for refractive error

Machine-learning random forest algorithms predict post-cycloplegic myopic corrections from noncycloplegic clinical data

SIGNIFICANCE

Machine learning random forest algorithms were used to predict objective refractive outcomes after cycloplegic refraction using noncycloplegic clinical data. A classification model predicted post-cycloplegic myopia and could be useful in screening, and a second regression model predicted post-cycloplegic refractive and could provide a useful objective starting point in noncycloplegic subjective refractions.

PURPOSE

A classification model sought to predict post-cycloplegic myopia using noncycloplegic clinical data to enhance myopia screening accuracy, whereas the regression model looked to predict objective refraction outcomes after cycloplegia for use as a starting point for noncycloplegic subjective refraction.

METHODS

A cross-sectional study included data from 2483 eyes. Pre-refraction measurements, such as uncorrected visual acuity, axial length, and corneal curvature radius, were recorded. After cycloplegia, the spherical equivalent was measured. Random forest-based classification and regression models were established with input variables including age, gender, axial length, corneal curvature radius, axial length-to-corneal curvature radius ratio, spherical equivalent, and uncorrected visual acuity. Model performance was assessed using various metrics.

RESULTS

The random forest classification model achieved high out-of-bag validation accuracy (92%), cross-validation accuracy (93%), external validation accuracy (94%), and precision (95%). The external validation sensitivity was 93%, and specificity was 95%. The regression model internal validation showed an out-of-bag validation R2 of 0.86, root mean square error (RMSE) of 0.66, and mean absolute error of 0.49. The 10-fold cross-validation R2 was 0.87, the RMSE was 0.64, and the mean absolute error was 0.48. In the external validation, R2 was 0.88, the RMSE was 0.63, and the mean absolute error was 0.48.

CONCLUSIONS

By analyzing noncycloplegic clinical data, the classification model enables earlier detection of myopia, supporting timely intervention and management. The regression model aims to accurately predict post-cycloplegia myopic corrections, providing reliable initial data for subjective refraction. This could help optometrists perform noncycloplegic subjective refraction more efficiently and is particularly relevant in China, where retinoscopy is not yet fully popularized and many school students decline cycloplegic refraction due to academic pressures and limited free time, primarily because it requires a follow-up the next day.

Machine Learning Models for Predicting Cycloplegic Refractive Error and Myopia Status Based on Non-Cycloplegic Data in Chinese Students

Purpose

To develop and validate machine learning (ML) models for predicting cycloplegic refractive error and myopia status using noncycloplegic refractive error and biometric data.

Methods

Cross-sectional study of children aged five to 18 years who underwent biometry and autorefraction before and after cycloplegia. Myopia was defined as cycloplegic spherical equivalent refraction (SER) ≤-0.5 Diopter (D). Models were evaluated for predicting SER using R2 and mean absolute error (MAE) and myopia status using area under the receiver operating characteristic (ROC) curve (AUC). Best-performing models were further evaluated using sensitivity/specificity and comparison of observed versus predicted myopia prevalence rate overall and in each age group. Independent data sets were used for training (n = 1938) and validation (n = 1476).

Results

In the validation dataset, ML models predicted cycloplegic SER with high R2 (0.913-0.935) and low MAE (0.393-0.480 D). The AUC for predicting myopia was high (0.984-0.987). The best-performing model for SER (XGBoost) had high sensitivity and specificity (91.1% and 97.2%). Random forest (RF), the best-performing model for myopia, had high sensitivity and specificity (92.2% and 96.9%). Within each age group, difference between predicted and actual myopia prevalence was within 4%.

Conclusions

Using noncycloplegic refractive error and ocular biometric data, ML models performed well for predicting cycloplegic SER and myopia status. When measuring cycloplegic SER is not feasible, ML may provide a useful tool for estimating cycloplegic SER and myopia prevalence rate in epidemiological studies.

Translational Relevance

Using ML to predict cycloplegic refraction based on noncycloplegic data is a powerful tool for large, population-based studies of refractive error.

Prediction of refractive error and its progression: a machine learning-based algorithm

OBJECTIVE

Myopia is the refractive error that shows the highest prevalence for younger ages in Southeast Asia and its projection over the next decades indicates that this situation will worsen. Nowadays, several management solutions are being applied to help fight its onset and development, nonetheless, the applications of these techniques depend on a clear and reliable assessment of risk to develop myopia.

METHODS AND ANALYSIS

In this study, population-based data of Chinese children were used to develop a machine learning-based algorithm that enables the risk assessment of myopia's onset and development. Cross-sectional data of 12 780 kids together with longitudinal data of 226 kids containing age, gender, biometry and refractive parameters were used for the development of the models.

RESULTS

A combination of support vector regression and Gaussian process regression resulted in the best performing algorithm. The Pearson correlation coefficient between prediction and measured data was 0.77, whereas the bias was -0.05 D and the limits of agreement was 0.85 D (95% CI: -0.91 to 0.80D).

DISCUSSION

The developed algorithm uses accessible inputs to provide an estimate of refractive development and may serve as guide for the eye care professional to help determine the individual best strategy for management of myopia.

Refractive Error and Axial Length and Their Related Factors in 8-Year-Old Japanese Children: The Yamanashi Adjunct Study of the Japan Environment and Children's Study (JECS)

PURPOSE

To investigate the distribution of visual acuity, refractive error, and axial length in 8-year-old children who participated in an additional survey in Yamanashi Prefecture of the Japan Environmental Children's Study (hereafter referred to as JECS-Y) conducted from 2019 to 2021.

PARTICIPANTS AND METHODS

Eight-year-old children who participated in the JECS-Y study were subjected to noncycloplegic measurements of refractive error and axial length. If the uncorrected visual acuity was less than 20/20, the best corrected visual acuity was evaluated in accordance with the autorefraction data. A questionnaire was administered regarding the parent's history of eyeglass wear or contact lens use.

RESULTS

Among the 400 participating children, the rate of uncorrected visual acuity of 20/20 or better in both eyes was 70.4%. The mean equivalent spherical equivalent error for both eyes was -0.366 ± 1.016 D. The mean axial length was 23.08 ± 0.225 mm in all patients. The males showed significantly longer axial length than the females despite no differences in body height. There was a significant correlation between axial length, spherical refractive, and uncorrected visual acuity. The children of parents with a history of wearing eyeglasses or contact lenses showed a significantly more myopic equivalent refractive error than those without a history.

CONCLUSIONS

This study clarified the current state of refractive error in 8-year-old children and the association of inheritance with refractive error. In addition, the axials were significantly longer in male patients.

IMI 2023 Digest

Myopia is a dynamic and rapidly moving field, with ongoing research providing a better understanding of the etiology leading to novel myopia control strategies. In 2019, the International Myopia Institute (IMI) assembled and published a series of white papers across relevant topics and updated the evidence with a digest in 2021. Here, we summarize findings across key topics from the previous 2 years. Studies in animal models have continued to explore how wavelength and intensity of light influence eye growth and have examined new pharmacologic agents and scleral cross-linking as potential strategies for slowing myopia. In children, the term premyopia is gaining interest with increased attention to early implementation of myopia control. Most studies use the IMI definitions of ≤-0.5 diopters (D) for myopia and ≤-6.0 D for high myopia, although categorization and definitions for structural consequences of high myopia remain an issue. Clinical trials have demonstrated that newer spectacle lens designs incorporating multiple segments, lenslets, or diffusion optics exhibit good efficacy. Clinical considerations and factors influencing efficacy for soft multifocal contact lenses and orthokeratology are discussed. Topical atropine remains the only widely accessible pharmacologic treatment. Rebound observed with higher concentration of atropine is not evident with lower concentrations or optical interventions. Overall, myopia control treatments show little adverse effect on visual function and appear generally safe, with longer wear times and combination therapies maximizing outcomes. An emerging category of light-based therapies for children requires comprehensive safety data to enable risk versus benefit analysis. Given the success of myopia control strategies, the ethics of including a control arm in clinical trials is heavily debated. IMI recommendations for clinical trial protocols are discussed.

Prediction of spherical equivalent difference before and after cycloplegia in school-age children with machine learning algorithms

Purpose

To predict the need for cycloplegic assessment, as well as refractive state under cycloplegia, based on non-cycloplegic ocular parameters in school-age children.

Design

Random cluster sampling.

Methods

The cross-sectional study was conducted from December 2018 to January 2019. Random cluster sampling was used to select 2,467 students aged 6-18 years. All participants were from primary school, middle school and high school. Visual acuity, optical biometry, intraocular pressure, accommodation lag, gaze deviation in primary position, non-cycloplegic and cycloplegic autorefraction were conducted. A binary classification model and a three-way classification model were established to predict the necessity of cycloplegia and the refractive status, respectively. A regression model was also developed to predict the refractive error using machine learning algorithms.

Results

The accuracy of the model recognizing requirement of cycloplegia was 68.5-77.0% and the AUC was 0.762-0.833. The model for prediction of SE had performances of R^2 0.889-0.927, MSE 0.250-0.380, MAE 0.372-0.436 and r 0.943-0.963. As the prediction of refractive error status, the accuracy and F1 score was 80.3-81.7% and 0.757-0.775, respectively. There was no statistical difference between the distribution of refractive status predicted by the machine learning models and the one obtained under cycloplegic conditions in school-age students.

Conclusion

Based on big data acquisition and machine learning techniques, the difference before and after cycloplegia can be effectively predicted in school-age children. This study provides a theoretical basis and supporting evidence for the epidemiological study of myopia and the accurate analysis of vision screening data and optometry services.

Ocular biometrics and uncorrected visual acuity for detecting myopia in Chinese school students

The study is to evaluate the performance of ocular biometric measures and uncorrected visual acuity (UCVA) for detecting myopia among Chinese students. Among 5- to 18-year-old Chinese students from two cities of China, trained eye-care professionals performed assessment of ocular biometrics (axial length (AL), corneal curvature radius (CR), anterior chamber depth) under noncycloplegic conditions using NIDEK Optical Biometer AL-Scan, distance visual acuity using retro-illuminated logMAR chart with tumbling-E optotypes, and cycloplegic refractive error using NIDEK autorefractor with administration of 0.5% tropicamide. Spherical equivalent (SER) in diopters (D) was calculated as sphere plus half cylinder, and myopia was defined as SER ≤ - 0.5 D. Performances of ocular biometrics and UCVA (individually and in combination) for detecting myopia were evaluated using sensitivity and specificity, predictive values, and area under ROC curve (AUC) in both development dataset and validation dataset. Among 3436 students (mean age 9.7 years, 51% female), the mean (SD) cycloplegic SER was - 0.20 (2.18) D, and 1269 (36.9%) had myopia. Cycloplegic SER was significantly correlated with AL (Pearson Correlation coefficient r = - 0.82), AL/CR ratio (r = - 0.90), and UCVA (r = 0.79), but was not correlated with CR (r = 0.02, p = 0.15). The AL/CR ratio detected myopia with AUC 0.963 (95% CI 0.957-0.969) and combination with UCVA improved the AUC to 0.976 (95% CI 0.971-0.981). Using age-specific AL/CR cutoff (> 3.00 for age < 10 years, > 3.06 for 10-14 years, > 3.08 for ≥ 15 years) as myopia positive, the sensitivity and specificity were 87.0% (95% CI 84.4-89.6%) and 87.8% (86.0-89.6%), respectively, in the development dataset and 86.4% (95% CI 83.7-89.1%) and 89.4% (95% CI 87.3-91.4%), respectively, in the validation dataset. Combining AL/CR and UCVA (worse than 20/32 for age < 10 years, and 20/25 for ≥ 10 years) provided 91.9% (95% CI 90.4-93.4%) sensitivity and 87.0% (95% CI 85.6-88.4%) specificity, positive value of 80.6% (95% CI 78.5-82.6%) and negative value of 94.8% (95% CI 93.8-95.8%). These results suggest that AL/CR ratio is highly correlated with cycloplegic refractive error and detects myopia with high sensitivity and specificity,  AL/CR ratio alone or in combination with UCVA can be used as a tool for myopia screening or for estimating myopia prevalence in large epidemiological studies with limited resources for cycloplegic refraction.

Prediction for Cycloplegic Refractive Error in Chinese School Students: Model Development and Validation

Purpose

To predict cycloplegic refractive error using measurements obtained under noncycloplegic conditions.

Method

Refractive error was measured in 5- to 18-year-old Chinese students using a NIDEK autorefractor before and after administration of 0.5% tropicamide. Spherical equivalent (SER) in diopters (D) was calculated as sphere plus half cylinder. A multivariable prediction model for cycloplegic SER was developed using data from students in Jinyun (n = 1938) and was validated using data from students in Hangzhou (n = 1498). The performance of the prediction model was evaluated using R², mean difference between predicted and measured cycloplegic SER, and sensitivity and specificity for predicting myopia (cycloplegic SER ≤ −0.5 D).

Results

Among 3436 students (mean age, 9.7 years; 51% female), the mean (SD) noncycloplegic and cycloplegic SER values were −1.12 (1.97) D and −0.20 (2.19) D, respectively. The prediction model that included demographics, noncycloplegic SER, axial length/corneal curvature radius ratio, uncorrected visual acuity (UCVA), and intraocular pressure predicted cycloplegic SER with R² of 0.93 in the development dataset and 0.92 in the validation dataset. The mean (SD) differences between predicted and measured cycloplegic SER were 0.0 (0.55) D in the development dataset and 0.06 (0.64) D in the validation dataset. In both the development and validation datasets, the combination of predicted SER and UCVA yielded high sensitivity (91.4% and 91.9%, respectively) and specificity (95.0% and 90.1%, respectively) for detecting myopia.

Conclusions

Cycloplegic refractive error can be predicted using measurements obtained under noncycloplegic conditions. The prediction model could potentially be used to correct the myopia prevalence in epidemiological studies in which administering cycloplegic agent on all participants is not feasible.

Translational Relevance

The prediction model may provide a tool for correcting the overestimation of myopia from noncycloplegic refractive error in future epidemiological studies in which administering cycloplegic agent on all participants is not feasible.