This study presents a machine learning model that predicts cycloplegic refraction in adults with myopia using standard non-cycloplegic eye measurements, aiming to reduce the need for cycloplegic drops while still identifying patients who require them.
Age 18 to 60 years, of either sex;
Spherical equivalent between -0.50 diopters and -10.00 diopters, with myopia in one or both eyes, and with cylinder of 4.00 diopters or less;
Best-corrected visual acuity of 20/25 or better in each eye;
Clear cornea, no keratoconus, corneal scarring, or other pathologies; clear lens;
Incomplete clinical data to support the diagnosis;
Ocular conditions such as subclinical keratoconus, keratoconus, or moderate-to-severe corneal haze or leukoma;
Allergy or contraindication to cycloplegic agents;
Refusal to participate in the study.
Each subject underwent a comprehensive preoperative examination, including cycloplegic and non-cycloplegic refractions, Pentacam, etc.
Other
Cross-sectional
Machine learning model for predicting cycloplegic refraction
Diagnostic testThe machine learning model was applied to each participant's non-cycloplegic parameters to predict cycloplegic spherical equivalent.
Accuracy of predicted cycloplegic spherical equivalent
Accuracy of the machine learning model in predicting cycloplegic spherical equivalent in the validation dataset, evaluated by mean absolute error, root mean square error, and coefficient of determination, expressed for spherical equivalent in diopters.
Diagnostic performance for identifying patients requiring cycloplegic refraction
Area under the receiver operating characteristic curve, sensitivity, and specificity of the model for classifying patients with an absolute difference of 0.50 diopters or more between non-cycloplegic and cycloplegic spherical equivalent in the validation dataset.
Agreement between predicted and measured cycloplegic refraction
Agreement between predicted and measured cycloplegic spherical equivalent assessed by Bland-Altman analysis with mean bias and 95% limits of agreement, and by the intraclass correlation coefficient in the validation dataset.
Click on the lead sponsor to view all of their trials.
