Novel Quantitative Contrast Sensitivity Function Enhances the Prediction of Treatment Outcome and Recurrence in Amblyopia
Jing Liu,Chencui Huang,Susan A Cotter,Lily Y L Chan,Yizhou Yu,Yu Jia,Qingqing Ye,Lei Feng,Ying Yao,Rengang Jiang,Chutong Xiao,Zixuan Xu,Yijing Zhuang,Yunsi He,Yusong Zhou,Xiaolan Chen,Junpeng Yuan,Yun Wen,Wentong Yu,Yangfei Pang,Zhong-Lin Lu,Benjamin Thompson,Jinrong Li
DOI: https://doi.org/10.1167/iovs.65.5.31
2024-05-01
Abstract:Purpose: Although effective amblyopia treatments are available, treatment outcome is unpredictable, and the condition recurs in up to 25% of the patients. We aimed to evaluate whether a large-scale quantitative contrast sensitivity function (CSF) data source, coupled with machine learning (ML) algorithms, can predict amblyopia treatment response and recurrence in individuals. Methods: Visual function measures from traditional chart vision acuity (VA) and novel CSF assessments were used as the main predictive variables in the models. Information from 58 potential predictors was extracted to predict treatment response and recurrence. Six ML methods were applied to construct models. The SHapley Additive exPlanations was used to explain the predictions. Results: A total of 2559 consecutive records of 643 patients with amblyopia were eligible for modeling. Combining variables from VA and CSF assessments gave the highest accuracy for treatment response prediction, with the area under the receiver operating characteristic curve (AUC) of 0.863 and 0.815 for outcome predictions after 3 and 6 months, respectively. Variables from the VA assessment alone predicted the treatment response, with AUC values of 0.723 and 0.675 after 3 and 6 months, respectively. Variables from the CSF assessment gave rise to an AUC of 0.909 for recurrence prediction compared to 0.539 for VA assessment alone, and adding VA variables did not improve predictive performance. The interocular differences in CSF features are significant contributors to recurrence risk. Conclusions: Our models showed CSF data could enhance treatment response prediction and accurately predict amblyopia recurrence, which has the potential to guide amblyopia management by enabling patient-tailored decision making.