Tracking myopia development through axial length progression: a retrospective longitudinal study

Abstract

Background

Current prediction models for myopia progression remain limited in their ability to provide personalized risk assessments.

Objective

To examine the predictive role of axial length (AL) progression in progressive myopia.

Introduction

This retrospective cohort study analysed longitudinal ocular biometric data collected through repeated measurements in a school-based population.

Subjects

The study population comprised a longitudinal cohort of 1697 Chinese students aged 6–14 years, with ocular biometric data collected between December 2017 and May 2019 (18-month follow-up period).

Method

The dataset was randomly partitioned into development and validation cohorts at a 2:1 ratio, with two-thirds of the data allocated for nomogram model construction and the remaining one-third reserved for validation. The axial length-to-corneal radius (AL/CR) ratio was defined as the AL divided by the mean CR value measured in 90° meridians and 180° meridians: AL/CR = 2AL/(CR_90° + CR_180°). The primary outcome is progressive myopia, defined as an annualized spherical equivalent (SE) progression rate ≥0.75 DS/year over 1.5 years. The key predictor, first-visit AL progression, reflects 6-month axial elongation from baseline.

Results

The baseline subjects (N = 1697) were divided into the training set (N = 1132) and the validation set (N = 565). Multivariate logistic regression analysis indicated AL/CR in baseline (OR = 70.414, 95%CI: 22.795–217.511, p < .001) and first-visit AL progression (OR = 12845.569, 95%CI: 2915.219–56602.490, p < .001) significantly contributed to the risk of progressive myopia. Accordingly, baseline AL/CR and first-visit AL progression were treated as the main factor to build the nomogram model. The model showed good predictive performance (AUC = 0.785 in training set/0.771 in validation set) with well-calibrated slopes (approaching 1) and clinically useful thresholds (0.20–0.80).

Conclusions

This study develops a personalized prediction model for progressive myopia, grounded on factors of the first visit AL progression and baseline AL/CR. The model offers a dynamic and reliable foundation for selecting effective myopia control measures in future stages.

Introduction

The World Health Organization’s global eye health strategy highlights refractive errors, including myopia, as a priority due to their impact on visual health and quality of life, especially among children and adolescents [1]. A report by the Brien Holden Vision Institute projects that, by 2050, approximately 49.8% of the global population is expected to be affected by myopia [2]. Children and adolescents in East Asia face the highest prevalence and severity of myopia globally. The prevalence of myopia in China increased from 35.9% in 1995 to 60.04% in 2019, with rates of 35.6%, 71.1% and 80.5% observed among primary, middle and high school students, respectively [3].

The refractive status of the eye is determined by a combination of anatomical variables, including corneal power, nuclear opalescence and axial length (AL). Among these anatomical variables, AL is the most significant determinant of refractive error [4]. Therefore, ocular biometry, particularly AL measurement, has demonstrated significant value in myopia screening and follow-up efforts worldwide in recent years. The effectiveness of myopia screening depends on the use of simple and practical screening indicators to accurately identify individuals at high risk of myopia. This enables the provision of evidence-based intervention strategies aimed at reducing the incidence of myopia at an early stage. Recent research has increasingly highlighted the predictive value of AL and the axial length-to-corneal radius ratio (AL/CR; CR: corneal radius) in myopia risk assessment [5]. Findings from these studies underscore the critical role of AL in myopia screening. Moreover, when combined with the stable measurement outcomes provided by automated refractometers, this approach not only minimizes the labour and time costs associated with cycloplegic refraction but also substantially improves the accuracy of predicting myopia onset risk.

On the other hand, eye care professionals need to pay greater attention in their daily practice to individuals experiencing myopia progression, particularly those classified as having progressive myopia. According to the Expert Consensus on Myopia Management White Paper (2022), myopia with an annual increase in spherical equivalent (SE) refractive error of ≥0.75 diopters (D) is defined as progressive myopia. For these patients, in addition to routine refractive correction, more proactive and scientifically grounded myopia control interventions are required. Examples of such interventions include orthokeratology lenses, multifocal soft contact lenses, pharmacological treatments such as atropine eye drops, and specially designed defocus spectacle lenses [6]. In China, it is generally recommended to perform refractive examinations and ocular biometric measurements every 6 months for progressive myopia [7]. However, adherence to regular follow-up visits tends to decline over time among myopic patients, leading to potential challenges in effective disease management [8,9]. Effectively utilizing early follow-up ocular biometric data to predict the onset and progression of myopia may positively impact follow-up compliance. Chen et al., based on a large cohort study, identified an annual AL progression threshold of 0.2 mm as a predictive marker for myopia progression, achieving a high predictive accuracy with an area under the curve (AUC) of 0.88 [10]. Nevertheless, current prediction models for myopia progression remain limited in their ability to provide personalized risk assessments, highlighting the need for more individualized and precise predictive frameworks.

Nomogram model is a predictive tool that translates risk factors into quantitative contributions, transforming complex statistical models into intuitive graphical representations. This enables clear visualization of the relative influence of each predictive variable on myopia progression. For example, Guo et al. in a large-sample cohort study conducted in Wenzhou, introduced a nomogram to visually evaluate the risks associated with myopia onset [11]. However, the impact of early AL changes during follow-up on the rate of myopia progression over a defined period has not been systematically investigated, which is critical for implementing timely and precise myopia control strategies in clinical practice. Building on prior research, this 1.5-year retrospective longitudinal study develops a nomogram-based predictive model for progressive myopia risk, incorporating AL changes over six months. The aim is to provide a more evidence-based and rational framework for myopia prevention and control in patients undergoing their initial re-examination.

Method

Research design and participants

This retrospective study analysed data from a longitudinal cohort of 1697 Chinese students aged 6–14 years from three Shenzhen schools between December 2017 and May 2019 (18-month follow-up period) [12]. The study protocol included myopia screening and ocular biometric measurements conducted at 6-month intervals across four visits. This study, conducted in accordance with the principles of the Declaration of Helsinki, utilized data derived from our previous research and was approved by the Institutional Research Ethics Committee of Shenzhen Eye Hospital (Approval No. 2025KYPJ034), with a waiver of informed consent granted for secondary data analysis.

Exclusion criteria

Individuals were excluded according to the following criteria, based on the electronic health records during the follow-up period from December 2017 to May 2019: (1) incomplete data (missing data from one or more of the four follow-up visits); (2) strabismus and nystagmus; (3) colour perception abnormalities; (4) fundus diseases.

Refractive examination

The refractive data of participants was assessed through cycloplegic refraction, which was performed following procedures. A standard logarithmic tumbling E chart (WB-1112E, Wenbang, Shanghai, China) was positioned 5 m away for vision test. Cycloplegia was induced using three drops of tropicamide phenylephrine (Santen, Osaka, Japan) at 10-minute intervals, ensuring adequate mydriasis. At least 10 min after the final drop, cycloplegic refraction was measured using an automated refractometer (KR-8800, Topcon, Tokyo, Japan). The mean of five readings was automatically calculated, and the spherical diopter, cylindrical diopter and axis were electronically recorded.

Ocular biometric parameter measurements

Optical biometry (IOL-Master 700, Carl Zeiss, Jena, Germany) was used to measure the OB of the enrolled students. The subjects were asked to blink before the measurement to ensure that the tear film covered the entire cornea. Five measurements were performed for each subject, and the final value was determined by taking the average of the five readings. Among the recorded data, AL and CR were included in the final analysis.

Definition

Following the International Myopia Institute (IMI) white paper guidelines, myopia was defined as SE ≤ −0.50 diopters (DS) under relaxed accommodation [13]. In accordance with the Expert Consensus on Myopia Management White Paper (2022), progressive myopia is defined as SE progression greater than or equal to 0.75 diopters (D) per year [6]. In this study, progressive myopia was defined as an annualized SE progression rate ≥0.75 DS/year during the entire 1.5-year follow-up period. (Data annualization method: The total SE change (in DS) over the follow-up duration was divided by the actual follow-up time (in years) to calculate the annualized progression rate.) The AL/CR ratio was defined as the AL divided by the mean CR value measured in 90° meridians and 180° meridians: AL/CR = 2AL/(CR_90° + CR_180°). The key predictor, first-visit AL progression, was derived from biometry measurements, representing 6-month axial elongation since baseline.

Data analysis

Baseline characteristics between training and validation cohorts were compared using the Wilcoxon rank sum test for continuous variables, and Fisher’s exact test or Chi-square tests for categorical variables. Due to the non-normal distribution of the data, the association between first-visit AL progression and SE progression was quantified using Spearman’s rank correlation analysis. Univariate and multivariate logistic regression after stepwise selection identified risk factors for progressive myopia. Meanwhile, receiver operating characteristic (ROC) curves, calibration curves and decision curve analysis (DCA) were employed to evaluate the predictive performance and clinical utility of the nomogram across different age groups (6–8 years, 9–11 years and 12–14 years). All analyses were conducted in R (version 4.2.2; R Foundation for Statistical Computing, Vienna, Austria) with statistical significance set at two-tailed p < .05.

Results

This study utilizes a secondary analysis of validated data from our team’s previously published work [12]. The dataset consisted of 1697 Chinese children (3394 eyes). Two researchers independently performed a dual-blind verification process (Figure 1). A third researcher served as an arbiter to resolve any discrepancies. This retrospective investigation aligns conceptually with the prior study in examining myopia progression patterns, while its exclusion criteria represent a subset of the original study’s comprehensive screening protocol. Consequently, no participants were excluded due to incomplete data, strabismus, colour vision deficiency or fundus patho­logies. Since the SE between left and right eyes was highly correlated (Spearman’s rank correlation: baseline, r = 0.82; 0.5 year, r = 0.84; 1 year, r = 0.85; 1.5 years, r = 0.84), the right eye’s SE was used for analysis.

Figure 1.

Flowchart of cohort derivation and dataset partitioning.

Characteristics of the participants

To enhance the reliability of our model and minimize the risk of data overinterpretation, we randomly divided the 1697 subjects into a training set and an independent validation set at a ratio of 2:1, without replacement. The baseline characteristics of the subjects in the training set (N = 1132) and the validation set (N = 565) are presented in Table 1. There were no significant differences between the two groups in terms of age, sex, school, grade, SE or AL at baseline (all p > .05).

Table 1.

Baseline characteristics of the children in the training and validation cohorts.

Variables	Total (N = 1697)	Training cohort (n = 1132)	Validation cohort (n = 565)	p Value
Continuous variables
Age, years	10.00 ± 2.22	10.04 ± 2.21	9.94 ± 2.24	.433
SE (baseline), D	−1.25 ± 1.54	−1.29 ± 1.57	−1.17 ± 1.48	.441
AL (baseline), mm	24.03 ± 1.10	24.06 ± 1.11	23.97 ± 1.08	.112
Refraction state				.078
Non-myopic (SE > −0.5 D)	747 (44.02%)	481 (42.49%)	266 (47.08%)
Myopic (SE ≤ −0.5 D)	950 (55.98%)	651 (57.51%)	299 (52.92%)
Discrete variables
Gender, n (%)				.354
Male	892 (52.56%)	586 (51.77%)	306 (54.16%)
Female	805 (47.44%)	546 (48.23%)	259 (45.84%)
School, n (%)				.483
Nanshan Experimental School	342 (20.15%)	219 (19.35%)	123 (21.77%)
Nanshan Second Foreign Language School	624 (36.77%)	418 (36.93%)	206 (36.46%)
Songping School	731 (43.08%)	495 (43.73%)	236 (41.77%)
Grade				.895
Grade 1	276 (16.26%)	181 (15.99%)	95 (16.81%)
Grade 3	339 (19.98%)	223 (19.70%)	116 (20.53%)
Grade 5	431 (25.40%)	287 (25.35%)	144 (25.49%)
Grade 7	651 (38.36%)	441 (38.96%)	210 (37.17%)

D: diopter; SE: spherical equivalent; AL: axial length.

Correlation between first-visit AL progression and myopia development

The SE progression between the 6- and 12-month follow-up intervals was calculated to quantify myopia progression over a 0.5-year period following the first visit. Similarly, the SE progression from 6 to 18 months was used to represent cumulative progression over a 1-year period after the first visit. Spearman’s rank correlation coefficients were calculated between first-visit AL progression and SE progression over the next 0.5 year and 1 year. First-visit AL progression showed significant correlations with both 6–12-month SE progression (training set: r = −0.235, p < .001; validation set: r = −0.248, p < .001) and 6–18-month SE progression (training set: r = −0.368, p < .001; validation set: r = −0.314, p < .001). Figure 2 shows the correlation between the two factors in all the subjects; first-visit AL progression showed significant correlations with both 6–12-month SE progression (Figure 2(A), r = −0.240, p < .001), 6–18-month SE progression (Figure 2(B), r = −0.350, p < .001) and annualized SE progression (Figure 2(C), r = −0.495, p < .001).

Figure 2.

Correlation analysis of first-visit AL progression and SE progression.

The risk factors of progressive myopia

As shown in Table 2, logistic regression analysis was used for evaluation of the risk factors of progressive myopia. Univariate analysis indicated that AL/CR, SE at baseline, and first-visit AL progression are significant risk factors for myopia, while age, sex and school were not considered significant risk factors for progressive myopia in training set. Besides, grade 3 and grade 5 were identified as significant factors influencing whether participants would be classified as being at the stage of progressive myopia. Specifically, SE at baseline (OR = 0.752, p < .01), AL/CR at baseline for 1 unit (OR = 63.521, p < .01) and first-visit AL progression for 1 mm (OR = 13831.851, p < .01) all contributed to a higher risk of progressive myopia in the following 1 year. Multivariate logistic regression analysis after stepwise indicated that AL/CR, and first-visit AL progression are significant risk factors for progressive myopia, with AL/CR at baseline (OR = 70.414, 95%CI: 22.795–217.511, p < .001) and first-visit AL progression (OR = 12845.569, 95%CI: 2915.219–56602.490), p < .001).

Table 2.

Logistic regression analysis of risk factors for progressive myopia in school-age children (n = 1132).

Variables	Univariate		Multivariate
Independent	p Value	OR (95%CI)	p Value	OR (95%CI)
Age	.760	0.992 (0.939–1.047)
Sex
Male		1 (Reference)
Female	.888	0.983 (0.769–1.256)
School
Nanshan Experimental School		1 (Reference)
Nanshan Second Foreign Language School	.141	0.776 (0.553–1.088)
Songping School	.346	0.855 (0.618–1.184)
Grade
Grade 1		1 (Reference)
Grade 3*	.010*	1.717 (1.135–2.597)
Grade 5*	.002*	1.848 (1.246–2.741)
Grade 7	.587	1.111 (0.761–1.622)
Ocular biometry
SE at baseline*	<.001*	0.752 (0.692–0.816)
AL/CR at baseline*	<.001*	63.521 (22.627–178.320)	<.001*	70.414 (22.795–217.511)
First-visit AL progression*	<.001*	13831.851 (3181.121–60142.346)	<.001*	12845.569 (2915.219–56602.490)

OR: odds rate; SE: spherical equivalent; AL: axial length; CR: corneal radius.

Progressive myopia was defined as an annualized spherical equivalent (SE) progression rate ≥0.75 DS/year during the entire 1.5-year follow-up period.

Constructing nomogram model for predicting progressive myopia

According to the result in the logistic regression analysis, AL/CR at baseline and first-visit AL progression were treated as the main factors to build the nomogram model. Each 0.1 unit increase at baseline AL/CR was associated with a nearly five-point increase in the score, while each 0.1 mm increase in first-visit AL progression corresponded to a nearly 10-point increase, as total score is 130 points (Figure 3). The predictive performance of the model was evaluated using ROC analysis. In the training set, the AUC was 0.785 for the entire cohort, while age-specific analyses yielded AUCs of 0.811 (6–8 years), 0.803 (9–11 years) and 0.739 (12–14 years). Consistent results were found in the validation set, with corresponding AUCs of 0.771, 0.836, 0.764 and 0.735, respectively (Figure 4(A,B)). The calibration curve demonstrated good concordance between the actual and predicted probabilities in both the training and validation sets (Figure 4(C,D)). The DCA showed that the model in both training set and validation set demonstrated favourable performance within a wide threshold range (0.20–0.80) (Figure 4(E,F)).

Figure 3.

Nomogram for predicting the risk of progressive myopia over 1.5-year follow-up. AL/CR: axial length-to-corneal radius ratio; first-visit AL progression, 6-month axial elongation since baseline.

Figure 4.

Evaluation of the nomogram for predicting the risk of progressive myopia over 1.5-year follow-up. (A) ROC curve of training set; (B) ROC curve of validation set; (C) calibration curve of training set; (D) calibration curve of validation set; (E) DCA curve of training set; (F) DCA curve of validation set. ROC: receiver operator characteristic; DCA: decision curve analysis

Discussion

Myopia has evolved into a global public health crisis, with a worldwide prevalence of 22.9% in 2020 and over 52.7% among Chinese children/adolescents, projected to affect half the global population by 2050 [2,14]. Progressive myopia requires timely intervention to prevent transition to high myopia, as pathologic high myopia increases the risks of blinding complications including glaucoma, cataract, myopic macular degeneration and retinal detachment [15,16].

According to a Chinese epidemiological study, myopia prevalence surges from 38.16% at ages 6–12 (primary school) to 77.52% and peaks at 84.00% by ages 12–15 (secondary school) [17]. Thus, ages 6–14 represent a critical period for investigating biometric characteristics of myopia progression [18]. To overcome the challenge of poor follow-up adherence, this study developed a clinically actionable nomogram model based on a cohort of 1697 children by innovatively integrating cycloplegic refraction and short-term (6-month) axial elongation parameters for risk stratification. This innovation provides a quantified, real-time prediction of myopia progression risk (>0.75 DS per year), significantly streamlining clinical decision-making. Furthermore, by expressing risk as an intuitive percentage, the model bridges the communication gap between clinicians and parents, increasing acceptance of and compliance with personalized treatment plans.

The progression of myopia in adolescents is predominantly characterized by continuous axial elongation. Existing studies have confirmed a significant positive correlation between AL and SE, indicating synchronous changes between these parameters [19,20]. This study further investigates their temporal relationship through longitudinal analysis. Given the 1.5-year follow-up duration in this study, annualizing SE progression data minimized the potential impact of selection bias on the reliability of the results.

Results demonstrate that the first-visit AL progression significantly correlates with subsequent SE changes at both 6-month and 12-month intervals, suggesting AL’s longitudinal predictive value for refractive progression. Notably, AL progression shows stronger correlation with 6–18-month SE progression than with 6–12-month outcomes, potentially attributable to measurement sensitivity differences: AL is measured with 0.01 mm precision via biometric devices, whereas SE is typically recorded in 0.25 D increments. Consequently, AL changes manifest earlier in monitoring, while accurate detection of SE progression trends requires longer observation periods.

Logistic regression identified baseline AL/CR and SE as independent predictors of rapid myopia progression, highlighting the critical role of baseline ocular biometrics in myopia development. This finding aligns with prior studies demonstrating the high discriminative power of AL/CR for myopia diagnosis [12,21]. In contrast, demographic variables (age, gender) and environmental factors (school) showed no significant association with progression rates, diverging from Mu et al.’s findings on gender’s role in myopia incidence [22]. This suggests that gender may influence myopia onset but not its progression rate.

Age is an unavoidable confounding factor in childhood myopia progression. The results demonstrated robust predictive performance in children aged 6–8, 9–11 and 12–14 years, while a slight decline of AUC was observed in the 12–14 age subgroup. The calibration and DCA results are also similar. As shown in Figure 5, Spearman’s correlation analysis revealed a significant negative correlation between first-visit AL progression and age in the overall sample (r = −0.259, p < .01), training cohort (r = −0.275, p < .01) and validation cohort (r = −0.217, p < .01), indicating that AL progression decreases with increasing age. Due to the significant correlation between first-visit AL progression and age, age was not included as a separate variable in the final nomogram model with minimum Akaike information criterion (AIC). Given the ongoing need for myopia control among Chinese children and adolescents aged 12–14, this age subgroup was retained in the model which remains clinically valuable. Besides, univariate analysis revealed that grade 3 and grade 5 were significantly associated with progressive myopia status, suggesting that changes in academic pressure may influence myopia progression. However, since not all grades were included in this study, grade level was not incorporated into the final predictive model. Based on these findings, the final predictive model incorporated baseline AL/CR and first-visit AL progression as key predictors.

Figure 5.

Correlation between first-visit AL progression and age in different cohorts. (A) Total sample. (B) Training set. (C) Validation set.

The incorporation of longitudinal ocular biometric parameters enhances the predictive dimensionality of current myopia prediction models. Each 0.1 increase in baseline AL/CR and 0.1 mm increase in first-visit AL progression corresponded to 5-point and 10-point rises in risk score, respectively, demonstrating that first-visit AL progression carried the highest predictive weight for identifying ‘progressive myopia’ in recent 1.5 years. The model exhibited robust stability, with ROC analysis revealing comparable AUC values between training and validation cohorts (0.785 vs. 0.771, respectively), further corroborated by DCA and calibration curve. Compared with Guo et al.’s myopia onset prediction model (1 year, AUC = 0.8570 in training set vs. AUC = 0.8257 in validation set), the current model’s performance suggests opportunities for refinement [11]. This underscores the necessity to integrate genetic determinants (e.g. parental myopia history) and behavioural metrics (e.g. near-work duration, anticipated academic workload) in future studies to establish a comprehensive myopia progression prediction framework.

Although this study developed a personalized prediction tool for dynamic myopia progression, several limitations should be noted. First, the exclusive focus on urban children in Shenzhen limits geographical and ethnic diversity. Given that significant differences in AL progression exist across regions, ethnicities and age groups among paediatric populations, this geographical homogeneity may restrict the generalizability of our findings [23]. Second, the 1.5-year follow-up period insufficiently captures longitudinal impacts of axial elongation. Finally, the nomogram incorporated only cycloplegic refraction and 6-month AL parameters; future studies should integrate genetic markers and environmental exposure metrics to systematically elucidate myopia progression mechanisms and refine intervention timing.

Conclusions

This study develops a personalized prediction model for progressive myopia that dynamically integrates 6-month AL progression and baseline AL/CR. The proposed dynamic probability visualization framework translates complex biometric trajectories into clinically interpretable risk profiles, providing ophthalmologists with an updatable decision-support tool. Compared to static risk categorization systems, this approach allows for precision adjustment of control strategies according to individual progression patterns observed in longitudinal monitoring.

Funding Statement

This study was supported by Sanming Project of Medicine in Shenzhen (No. SZZYSM202411007).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are available from the corresponding author, Hongwei Deng, upon reasonable request.