This longitudinal study enrolled subjects who underwent lens surgery from June 2017 to February 2024 at the Eye and ENT Hospital of Fudan University, Shanghai, China. The Guidelines for Transparent Reporting of a Multivariable Model for Individual Prognosis or Diagnosis (the TRIPOD statement) were followed in this study. The study was approved by the Human Research Ethics Committee of the Eye and ENT Hospital of Fudan University (ChiCTR2000039132). The study adhered to the tenets of the Declaration of Helsinki. All participants provided signed informed consent.

A total of 677 patients (889 eyes) with moderate EL at the Eye and ENT Hospital of Fudan University were prospectively investigated and screened for suitability for this study. Patients missing important examination details, or with microspherophakia, keratoconus, previous history of ocular surgery, uveitis, corneal disease, glaucoma, retinal detachment, or use of contact lenses in the 2 weeks prior to the examinations were excluded from this study. The flow chart of the study is shown in Figure 1. In amblyopia groups, only patients with ametropic amblyopia rather than other types of amblyopia were included. Patients with lens edge uncovered 25 to 50% of the dilated pupil were diagnosed with moderate EL (8).

The family and medical histories were recorded for all subjects before the examinations. All enrolled subjects underwent routine preoperative examinations and comprehensive ocular examinations. The preoperative best-corrected visual acuity (BCVA) was assessed by the same skilled optometrist. Ocular parameters including AL, corneal keratometry [flattest K (K1), steepest K (K2), mean K (Km)], horizontal corneal diameter (WTW), and anterior chamber depth (ACD) were collected by partial coherence interferometry (the IOLMaster 500, Carl Zeiss Meditec, Jena, Germany, was used from June 2017 to March 2018; the IOLMaster 700, Carl Zeiss Meditec, Jena, Germany, was used from March 2018 to February 2024). All measurements were made with a pupil scan of 6 mm in diameter. Subjects were examined by experienced ophthalmologists who were trained in the use of all devices. All values in individual subjects were taken as the means of five repeated measurements for each eye recorded by the equipment. Additionally, because AL and WTW were progressive and age-dependent, to rule out the confounding factors of age, the z-AL and z-WTW were calculated with the formula Z-score = (measured score − normative score)/normative standard deviation (SD), compared with a reference standard (17).

Genetic screening and echocardiography were performed for the patients with CEL, according to our previous study (18). Based on the echocardiography results, cardiac phenotype was categorized into normal (no cardiac manifestations), regurgitation (conditions such as mitral regurgitation or aortic regurgitation), and organic (heart with any kind of organic lesions including aortic root dilation, mitral valve elongation, mitral valve prolapse).

All surgeries were operated by one surgeon (Dr. Y.-X. Jiang). The phacoemulsification procedure, modified capsular tension ring (MCTR) implantation or capsular tension ring and capsular hook (CTR-CH) implantation were conducted as previously described (19). In cases where a peripheral extension or tearing of the capsulorhexis rim occurred, a single-piece foldable IOL was sutured to the sclera using double-strand 9–0 polypropylene (MANI Inc., Tochigi, Utsunomiya, Japan) after anterior lensectomy combined with capsulotomy (23G; Alcon Laboratories Inc., Fort Worth, TX, United States) through a limbal approach (20).

Postoperatively, Levofloxacin Eye Drops (Santen Pharmaceuticals, Inc., Osaka, Japan) and Pred Forte Eye Drops (Allergan Pharmaceuticals, Inc., Dublin, Ireland) were applied three times daily for 1 month. Additionally, 0.1% pranoprofen (Sumika Finechem, Osaka, Japan) was applied three times daily for 1 month, followed by weekly tapering.

The best-corrected visual acuity (BCVA) was assessed by the skilled optometrist at every follow-up visit. Amblyopia was diagnosed according to the most recent edition of the Amblyopia Preferred Practice Pattern (21). After surgery, patients, especially those aged < 7 years old, were recommended to receive treatment for amblyopia. Follow-up visits were scheduled at intervals of 1 month, 6 months, 12 months, and then every 6 months after the surgery. Analysis was conducted separately for the right and left eyes.

The Kolmogorov–Smirnov test was used to confirm the normal distribution of the variables. Normally distributed data were presented as mean ± standard deviation (SD), while skewed data were expressed as median (interquartile range, IQR). Categorical variables were expressed as number and proportion as appropriate. The χ² test and Wilcoxon rank-sum test (Mann–Whitney U test) were used to compare data between the amblyopia and no-amblyopia groups in the training and validation cohorts. Univariable logistic regression analysis was used to describe the relationship between each individual predictor variable and the prediction of amblyopia. The predictor variables showing significant associations with standard outcomes (p < 0.05) were collected for further multivariate logistic regression analysis. Multivariate logistic regression (forward stepwise selection and exclusion criteria of type I error = 0.1 based on likelihood ratio tests) was then performed to build the risk prediction model. All predictor variables were described as odds ratios (ORs) with 95% confidence intervals (CI) and p values were calculated.

Calibration and discrimination were measured to assess the validity of the prediction model. To assess the calibration, we used Hosmer–Lemeshow χ² statistics and calibration curve. To assess the discrimination, the ROC analysis was performed to calculate the area under the ROC curve (AUC) in evaluating the performance of the model. We also examined the net benefit using decision curve analysis (DCA) with regard to clinical usefulness. The training cohort was used to build the clinical ocular prediction model. Then, we verified our model in the validation cohort. p values of less than 0.05 for two-tailed tests were considered statistically significant. GraphPad Prism 9.5 (GraphPad Software, California, United States) and SPSS software version 26.0 (IBM Corp, Armonk, NY) were used for all statistical analyses. R software was used to generate the nomogram, ROC analysis, calibration plots, and DCA curves.

A total of 521 patients (705 eyes) were identified as CEL and randomly divided into a training cohort and a validation cohort at a ratio of 7:3. Table 1 summarizes the baseline characteristics of the participants, including their demographic, ocular, cardiovascular, and genetic data. The training samples were obtained from 148 patients (203 eyes) with ametropic amblyopia and 195 patients (293 eyes) without amblyopia after lens surgery. The validation cohort consisted of 67 patients (76 eyes) with ametropic amblyopia and 111 patients (133 eyes) without amblyopia after lens surgery. The training and validation cohorts were comparable in terms of baseline characteristics (p > 0.05).

As shown in Figure 2, with a follow-up maximum of 3.5 years, the prevalence of amblyopia at various follow-up intervals post-surgery was investigated. Among the pediatric cohort (Figure 2A), there was a noticeable decline in the proportion of the amblyopia population. At 1 month after surgery, the prevalence of amblyopia was 41.13%, which decreased to 16% at 3.5 years after surgery. At 6 months, 1 year, 1.5 years, 2 years, 2.5 years, and 3 years, the prevalence was 33.09, 30.86, 27.67, 20.38, 19.51, and 18.95%, respectively. As for the adults’ cohort, no consistent pattern was identified (Figure 2B).

As shown in Table 1, comparative analysis suggested that there were significant differences in age (p = 0.029), z-AL (p = 0.008), BCVA (p < 0.001), and cardiac phenotype (p = 0.003) between the amblyopia and non-amblyopia groups. Upon thorough evaluation, a prediction model for postoperative amblyopia was constructed using preoperative biometrics. As shown in Table 2, BCVA and cardiac phenotype were selected as the indexes in the prediction model by multivariate logistic regression. The OR of BCVA was 0.015 (95% CI, 0.002–0.131). Compared with the normal phenotype, the OR of the regurgitation phenotype was 0.06 (95% CI, 0.01–0.345), and the OR of the organic phenotype was 0.235 (95% CI, 0.072–0.764).

A nomogram was established based on the variables screened to predict the postoperative amblyopia of CEL patients. Figure 3 shows an example of using the nomogram to predict amblyopia in a given patient. For a patient with a normal cardiac phenotype and BCVA of 0.3, the patient’s total score is −1.29, and the probability of amblyopia is 31.1%. Most patients in the present study had total risk points ranging from-5 to 2. If a patient’s total score is 0, the probability of postoperative amblyopia is more than 60%; therefore, more attention should be paid to its subsequent development and follow-up.

The calibration plots of the model for the training and validation cohorts were well-calibrated (Figures 4A,B). The Hosmer-Lemeshow statistical test of the model in the training cohort supported the goodness-of-fit of the model (χ² = 9.359; p = 0.405). The AUC of the model was 0.846 (95% CI, 0.770–0.922), which was better than each of the two predictors alone, including preoperative BCVA (AUC = 0.731; 95% CI, 0.692–0.786) and cardiac phenotype (AUC = 0.660; 95% CI, 0.565–0.756; Figure 4C). These results indicate favorable discrimination by the prediction model. For the validation cohort, the AUC of the new model was 0.840 (95% CI, 0.735–0.946; Figure 4D). The Hosmer-Lemeshow analysis of the new model in the validation cohort also supported the improved goodness-of-fit of the model (χ² = 7.052; p = 0.632). The DCA curves used to evaluate the clinical benefit of the nomogram are shown in Figures 4E,F, indicating a good clinical utility of the model.