AI Case Study
Sun Yat-sen University researchers predict high myopia for young adults at clinically acceptable levels 8 years in advance
Researchers from Zhongshan Ophthalmic Centre at Sun Yat-sen University use a random forest algorithm to predict young adults at risk of high myopia as far out as 8 years at clinically acceptable levels. This allows targeted therapy at those particularly at risk for high myopia in the hopes that early intervention can help.
Industry
Public And Social Sector
Education And Academia
Project Overview
The researchers conduct the first "utilisation of large-scale electronic medical record data to generate a random forest algorithm for predicting disease prognosis, which, in our analysis, was the risk of developing high myopia in adulthood. Furthermore, this algorithm exhibited high accuracy in a predicting future trait, i.e., the dioptre value at 18 years of age. Our data suggest that this prediction can be performed as early as 8 years prior to an individual turning 18 years old.
Identifying “severe myopia” in younger children is of major clinical importance but poses a significant challenge. The severity of myopia is often estimated as the degree of SE, with an SE of −6.00 dioptres chosen as the cutoff to define high myopia. High myopia carries a much greater risk of developing other ocular complications, including retinal detachment, glaucoma, and pathological myopia."
Reported Results
"For the comparative analysis, the random forest algorithm outperformed the generalised estimating equation and the mixed-effects model in the detection of high myopia." The method "achieved stable performance for high myopia detection... With respect to predicting the presence of high myopia, our algorithm provided clinically acceptable prediction over 3 years (the AUC ranged from 0.940 to 0.985), 5 years (the AUC ranged from 0.856 to 0.901), and even 8 years (the AUC ranged from 0.801 to 0.837)."
Technology
"Predictors included age at examination, spherical equivalent (SE), and annual progression rate. Cycloplegic refraction was performed according to a standard protocol in each centre. The right eye was arbitrarily chosen to represent a specific individual. Using these predictors, we aimed to develop an algorithm to predict SE and presence of high myopia in the subsequent 10 years (with each year as a predictive time point). Random forest is an ensemble learning method that operates by constructing a multitude of decision trees at training time and outputting the class that is the mode of the classes (classification) or mean prediction (regression) of the individual trees. Here, we employed the random forest algorithm for the development of the prediction algorithm... Each decision tree in the random forest was built using a bootstrap sample with replacement from the original data. This bootstrap aggregation and random feature selection helped reduce the variance of the algorithm and avoided over-fitting."
Function
R And D
Core Research And Development
Background
"Myopia has reached epidemic levels among young adults in East and Southeast Asia, affecting an estimated 80%–90% of high school graduates, with approximately 20% of them having high myopia. Various interventions, including atropine eyedrops and orthokeratology, have been proposed to control myopia progression; however, these approaches confer significant side effects. Identifying those at greatest risk who should undergo targeted therapy is the most important clinical challenge faced by ophthalmologists and optometrists."
Benefits
Data
"This study analysed 687,063 longitudinal electronic medical records from the largest ophthalmic centres in China. Eight ophthalmic centres were included in the study, including Zhongshan Ophthalmic Centre (ZOC), the Haizhu Optometry Department (HZD), the Huangpu Optometry Department (HPD), the Panyu Optometry Department (PYD), the Dongguan Guangming Ophthalmic Hospital (DGC), the Optometry Centre in Huizhou City (HZC), the Haikou Longhua Optometry Department (LHD), and the Xiuying Optometry Department in Haikou City (HKC). This study also included 2 datasets collected from population-based cohort studies: the Guangzhou Outdoor Activity Longitudinal Trial (GOAL) [5] and the Refractive Error Longitudinal Study (RELS). These 8 ophthalmic centres and 2 cohorts from South China collectively composed a representative medical big data sample for children of Chinese ethnicity. This sample could be generalisable to Chinese children living in Hong Kong, Taiwan, and Singapore, where myopia is similarly a common public health problem in children."