Machine learning‐based nomogram to predict poor response to overnight orthokeratology in Chinese myopic children: A multicentre, retrospective study

Abstract Purpose To develop and validate an effective nomogram for predicting poor response to orthokeratology. Methods Myopic children (aged 8–15 years) treated with orthokeratology between February 2018 and January 2022 were screened in four hospitals of different tiers (i.e. municipal and provincial) in China. Potential predictors included 32 baseline clinical variables. Nomogram for the outcome (1‐year axial elongation ≥0.20 mm: poor response; <0.20 mm: good response) was computed from a logistic regression model with the least absolute shrinkage and selection operator. The data from the First Affiliated Hospital of Chengdu Medical College were randomly assigned (7:3) to the training and validation cohorts. An external cohort from three independent multicentre was used for the model test. Model performance was assessed by discrimination (the area under curve, AUC), calibration (calibration plots) and utility (decision curve analysis). Results Between January 2022 and March 2023, 1183 eligible subjects were screened from the First Affiliated Hospital of Chengdu Medical College, then randomly divided into training ( n = 831) and validation ( n = 352) cohorts. A total of 405 eligible subjects were screened in the external cohort. Predictors included in the nomogram were baseline age, spherical equivalent, axial length, pupil diameter, surface asymmetry index and parental myopia ( p < 0.05). This nomogram demonstrated excellent calibration, clinical net benefit and discrimination, with the AUC of 0.871 (95% CI 0.847–0.894), 0.863 (0.826–0.901) and 0.817 (0.777–0.857) in the training, validation and external cohorts, respectively. An online calculator was generated for free access ( http://39.96.75.172:8182/#/nomogram ). Conclusion The nomogram provides accurate individual prediction of poor response to overnight orthokeratology in Chinese myopic children.