TOPIC
To determine the relative merits of spherical equivalent refraction (SER) vs axial length (AL) as the primary measure for monitoring childhood myopia progression and predicting adult myopia-related pathology.
CLINICAL RELEVANCE
Myopia prevalence is rising worldwide, but most patients only reach low-to-moderate myopia. Many guidelines emphasize AL over SER using fixed thresholds to define “high risk” (eg, ≥26 mm), yet a large proportion of myopia-related pathology occurs in eyes that never reach those lengths.
METHODS
We conducted a systematic review and meta-analysis of population-based observational studies (≥200 eyes; PubMed and Embase searched from 1990 to July 2025) reporting AL and/or SER with demographics and/or retinal, cataract, or primary open-angle glaucoma (POAG) outcomes. Random effects meta-analyses with Hartung-Knapp adjustment were performed when k ≥ 3; other results were synthesized descriptively. Risk of bias was assessed with the Newcastle-Ottawa Scale and certainty of evidence with Grading of Recommendations Assessment, Development and Evaluation.
RESULTS
Seventy studies met inclusion criteria. Normal AL varied strongly with age, sex, height, and ethnicity, whereas SER showed much weaker dependence. A small tail of emmetropes reached AL ≥26 mm. Retinal pathology rose steeply with longer AL (odds ratio [OR] 3.85 per +1 mm), implying risk ratio ≈25 at AL 26 mm relative to AL 23.6 mm (pooled gradient), but this long-eye threshold applied to <10% of eyes. SER data, although less frequent, showed consistent retinal risk increases with OR ≈1.5 to 1.8 per–1 diopter and OR 2 to 12 for high myopia. For cataract prevalence, SER-defined moderate/high myopia yielded ORs of 3.09 for nuclear and 4.58 for posterior subcapsular cataract, whereas AL-defined effects were null or modest. For POAG prevalence, AL per +1 mm showed OR 1.37, whereas SER-defined moderate/high myopia showed OR 2.95. Certainty of evidence was moderate for prevalence analyses and lower for incidence and descriptive blocks.
CONCLUSION
Absolute AL thresholds clearly identify a small tail of very long eyes at high retinal risk, but SER better captures cataract and POAG risk gradients across the myopic range that most patients will reach. SER should be the primary progression and risk metric for monitoring childhood myopia control, with AL used selectively to monitor the minority of children with greater axial elongation.
INTRODUCTION
M yopia prevalence is projected to approach 50% of the global population and high myopia around 10% by 2050. Myopia substantially increases the risk of sight-threatening sequelae, including myopic maculopathy, rhegmatogenous retinal detachment, primary open-angle glaucoma (POAG), and cataract, so “myopia control” is fundamentally about disease prevention, not just refractive correction. , As optical, pharmacologic, and behavioral myopia-control interventions transition into routine pediatric eyecare, ,, there is a pressing need for outcome measures that both quantify slowed progression in childhood and map credibly onto lifetime risk of myopia-related pathology. ,
Axial length (AL) has gained prominence as a structural endpoint. Influential reports and guidelines emphasize AL’s relative robustness to accommodation/diurnal effects and anterior-segment reshaping and increasingly prioritize AL change over spherical equivalent refraction (SER), ,, reinforced by adult cohort data showing that very long AL (eg, ≥26 mm) is associated with sharply increased odds of uncorrectable visual impairment, myopic maculopathy, and other complications. ,,, However, normal AL varies systematically by sex, height, ethnicity, and birth cohort ,,,,,,,,,,,,,,, ; thus, in children, AL percentiles or growth-curve trajectories trending toward a fixed adult cutpoint (eg, 26 mm) can represent very different positions within the “normal” distribution across individuals. ,, In population-based data, AL-risk curves for visual impairment and myopic maculopathy are typically “flat then exponential,” with steep risk escalation largely confined to a sparsely populated high-myopia tail. ,, Most myopic adults never reach these extreme AL values, even though myopia-associated pathology also occurs with low to moderate myopia. ,,,,,,,,,,,,,,, In contrast, across the refractive range most patients will reach, SER shows consistent dose-response risk increases for myopic maculopathy, POAG, and cataract. ,
Conceptually, these measures describe the same biology from different vantage points: pathology reflects posterior elongation relative to anterior optics—a functional mismatch. Absolute AL measures globe length but is masked to corneal and lenticular power and is confounded by normal demographic variation, ,,,,,,,,,,,,,,, whereas SER encodes the net optical mismatch targeted by myopia control and more directly tracks clinically meaningful risk gradients for several outcomes , despite susceptibility to accommodative and anterior-segment influences.
To address these gaps, we registered a systematic review, meta-analysis, and Bayesian network meta-analysis (NMA) in PROSPERO (CRD420251123893), with a statistical analysis plan (SAP) specifying a priori comparisons of SER and absolute AL thresholds for predicting myopia-related pathology. The review and SAP prespecified 3 linked analytical components: (1) describing normal AL variation by sex, height, ethnicity, and related factors; (2) examining how AL and SER relate to each other and to visual function, including the possibility of very long emmetropic eyes; and (3) comparing SER- and AL-based metrics as predictors of retinal, lenticular, and glaucomatous pathology. , Planned methods included random effects meta-analysis, bivariate diagnostic accuracy models, and Bayesian network meta-analysis (Surface Under the Cumulative Ranking [SUCRA] curve), alongside risk-of-bias and Grading of Recommendations Assessment, Development and Evaluation (GRADE) ; diagnostic accuracy and network meta-analyses ultimately proved infeasible owing to limited compatible reporting, as detailed below.
METHODS
PROTOCOL AND REPORTING
This systematic review and meta-analysis followed a PROSPERO registration (CRD420251123893) and prespecified SAP and is reported in accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA). Full search strategies and additional prespecified methods are provided in the Supplement. IRB approval and informed consent were not required because only published, aggregate data were used.
DATA SOURCES AND STUDY SELECTION
PubMed and Embase were searched for English-language, population-based observational studies (cross-sectional or cohort) published January 1, 1990–July 31, 2025. Baseline data from interventional trials were eligible if analyzed observationally. Studies included ≥200 eyes (or participants if one eye/person) and reported (1) AL with demographics; (2) SER with visual acuity; or (3) AL and/or SER with prevalence or incidence (≥1-year follow-up) of myopia-related retinal pathology, cataract, or POAG (Supplementary Table S1). We excluded case reports, clinic-based series, nonhuman studies, and reports without relevant AL/SER and covariate/outcome data. Two reviewers independently screened titles/abstracts and full texts, extracted data, and assessed risk of bias; disagreements were resolved by consensus. Reference lists of included studies were screened. Figure 1 summarizes study selection.
Preferred Reporting Items for Systematic Reviews and Meta-Analyses flowchart of included studies.
DATA ITEMS AND DEFINITIONS
We extracted design, sampling frame, recruitment era, age, sex (defined by biological sex [male/female] as reported by the studies), ethnicity, and urban/rural status; ocular measurements (AL and SER with variance, biometry/refraction methods, cycloplegia, lens status); and nonocular covariates (height, weight, BMI, genetic risk measures where available). We preferentially extracted adjusted regression coefficients (and SEs or CIs) for AL or SER associations with demographic variables and for AL-SER relationships, harmonizing units before synthesis (Supplement). When only means and SDs were available, we approximated whether AL ≥26 mm lay within 2 to 3 SDs of the emmetropic mean.
RISK OF BIAS AND CERTAINTY
Risk of bias for observational studies was assessed using the Newcastle-Ottawa Scale (representativeness, exposure/outcome measurements, including biometry method and cycloplegia status, and adequacy of adjustment for confounders). No study was excluded solely on the basis of risk-of-bias ratings, but these assessments informed sensitivity analyses (eg, optical vs ultrasound biometry; phakic-only vs mixed lens status). Certainty of evidence was assessed qualitatively with GRADE.
ANALYSES
Normative AL/SER variation
In population-based cohorts reporting AL and/or SER with ≥1 demographic covariate, we summarized age and sex distributions and synthesized associations of AL and SER with age, sex, height, and AL-SER coupling. For longitudinal pediatric and adolescent cohorts, we extracted ΔAL and ΔSER and computed β = ΔSER/ΔAL (diopters [D]/mm), defining “weak coupling” as |β| <1.35 D/mm (50% of the ≈2.7 D/mm schematic-eye expectation ). To evaluate whether very long but emmetropic eyes occur, we extracted AL distributions in emmetropic subgroups (SER −0.50 to +0.50 D) and summarized the upper-tail evidence descriptively.
Pathology prediction
Retinal outcomes were grouped as “any myopic retinopathy,” encompassing diffuse or patchy chorioretinal atrophy, lacquer cracks, myopic maculopathy, and macular atrophy, using the authors’ definitions (often using the Meta-Analysis for Pathologic Myopia classification system [META-PM] or similar tool). Cataract outcomes were analyzed by subtype—cortical, nuclear, and posterior subcapsular (PSC)—based on the Lens Opacities Classification System III (LOCS III) or comparable grading, with separate incidence and prevalence analyses. Glaucoma analyses were restricted to POAG as defined by International Society of Geographical and Epidemiological Ophthalmology (ISGEO) or comparable population-based criteria.
SER was considered in 3 ways: (1) continuous per-unit change (usually per −1 D, coded so more negative values indicate greater myopia); (2) thresholds contrasting any myopia with emmetropia or nonmyopia (eg, SER ≤−0.5 vs >−0.5 D or ≤−1.0 D vs >–1.0 D); and (3) severity categories contrasting moderate or high myopia (typically ≤−3 to −6 D) with emmetropia, or with milder myopia in some incidence analyses. AL was analyzed either as a continuous predictor (per +1 mm) or using study-defined long-eye thresholds near 26 mm vs shorter eyes (most cohorts used AL ≥26.0 mm; the Hisayama cohort reported sex-specific ROC-derived cutpoints [25.9 mm in men and 25.3 mm in women ), or, in one cataract study, the longest vs shortest AL quintile. Exposure coding was harmonized so that odds ratios (ORs), risk ratios (RRs), or hazard ratios (HRs) >1 reflect higher risk with longer AL or more myopic SER.
Meta-analysis
For each prespecified “block” (pathology family × prevalence/incidence × exposure metric), we pooled when k ≥3 comparable studies using random effects meta-analysis (inverse-variance weighting; DerSimonian-Laird τ²; Hartung-Knapp CIs), summarizing heterogeneity with I ² and τ². For incidence outcomes, RRs/HRs were analyzed on the log-RR scale. We performed influence and sensitivity analyses. When fewer than 3 studies contributed to a block, we rescaled effect estimates to common units (per −1 D SER or per +1 mm AL) and described ranges and patterns narratively.
Software
All analyses were conducted in R 4.5.1 using prespecified packages (meta, metafor).
Analyses prespecified but not feasible
Diagnostic accuracy analyses (SER/AL thresholds vs visual acuity) and Bayesian network meta-analysis (SUCRA ranking) were prespecified but not performed because too few studies reported compatible visual-acuity data with both SER and AL, and exposure definitions lacked shared comparators; synthesis was confined to pairwise meta-analysis and descriptive comparisons. Small-study effects were not assessed because all syntheses had k <10.
Data availability
Deidentified extraction worksheets and analysis code are available from the corresponding author on reasonable request.
RESULTS
CHARACTERISTICS OF INCLUDED STUDIES
The search identified 2432 records; after deduplication 1747 titles/abstracts were screened, 76 full texts were assessed, and 70 studies were included ( Figure 1 ). Overall risk of bias was low to moderate by Newcastle-Ottawa Scale (Supplementary Table S3). Across all analytic components, the review included multiple population-based cohorts from East Asia, Central and Northern Europe, Central India, Mongolia, the Caribbean, and North America, spanning childhood through older adulthood ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, (Supplementary Table S2). Adult normative distributions yielded a pooled mean AL of ≈23.6 mm and mean SER around −1.7 D. ,,,,,,,,,,,,,,,,,, Pathology analyses incorporated large, well-known cohorts such as Beaver Dam, Blue Mountains, , Barbados, the Visual Impairment Project, Tajimi, Korean National Health and Nutrition Examination Survey (KNHANES), Singapore Malay Eye Study (SiMES), , Singapore Indian Eye Study (SINDI), , and others.
NORMAL VARIATION IN AL AND SER
In population-based data, AL varies predictably with age, sex, and body size, whereas SER is much less sensitive to these nonocular factors ( Table 1 ).
TABLE 1
Normative Axial Length and Spherical Equivalent Refraction Associations.
| Association (unit) | Age Group | Pooled Effect (95% CI) | I (%) | k | Prediction Interval | Clinical Interpretation |
|---|---|---|---|---|---|---|
| AL-Age (mm per year) | Adults | +0.014 (0.006 to 0.021) | 93 | 4 | −0.003 to +0.030 | AL stable in adulthood (≈0.14 mm/decade) |
| AL-Age (mm per year) | Children | +0.12 (0.08 to 0.17) | 97 | 3 | +0.04 to +0.21 | ≈1.2 mm increase in AL over 10-y school-age span |
| AL-Height (mm per 0.10 m) | Adults | +0.11 (0.08 to 0.15) | 85 | 8 | +0.03 to +0.20 | Each 10 cm taller ≈0.11 mm longer AL |
| AL-Sex (female-male, mm) | Children | −0.38 (−0.47 to −0.29) | 98 | 6 | −0.61 to −0.15 | Female eyes are ≈0.3-0.4 mm shorter than male eyes at the same age |
| AL-SER (mm per 1 D) | Adults | −0.26 (−0.34 to −0.17) | 100 | 3 | −0.42 to −0.09 | More myopic SER ≈ longer AL; ≈0.25-0.30 mm per–1 D |
| SER-AL (D per 1 mm) | Children | −0.55 (−1.24 to +0.13) | 100 | 3 | −1.92 to +0.81 | Directionally negative but imprecise |
| SER-Sex (female-male, D) | Children | −0.17 (−0.29 to −0.05) | 82 | 4 | −0.41 to +0.07 | Females are ≈0.2 D more myopic than males |
AL = axial length, D = diopter, I ² = proportion of variability due to heterogeneity, k = number of independent estimates, SER = spherical equivalent refraction.
Age
During childhood, AL lengthens by ≈0.08 to 0.17 mm per year, ,, so a 10-year age difference across school years corresponds to a ≈1.2-mm difference in “normal” AL. Longitudinal pediatric/adolescent cohorts report mean AL growth ≈0.25 to 0.30 mm/y with mean refractive about −0.3 D/y, but with wide variability across cohorts. ,,,,,,, The coupling between elongation and refraction change is often weak; 1-mm AL change corresponds to ≈1 D SER change, and ≈60% of cohorts fall within the prespecified “weak-coupling” band (|β| <1.35 D/mm). Non–East Asian cohorts tend to show slower AL growth but uniformly weak coupling, ,, whereas East Asian cohorts show faster AL growth and more negative β, yet still have weak coupling in about half of entries. ,,,, In contrast, adult eyes change very little (pooled slope ≈ 0.014 mm/y), with prediction intervals that include almost no structural change. ,,,
Sex, height, and ethnicity
In children, girls’ eyes are about 0.3 to 0.4 mm shorter than boys’ at the same age, ,,,,, with similar sex gaps in adults where reported. , Taller adults have longer eyes: ≈0.11 mm AL per +0.10 m height ,,,,,,, ; children show comparable height-AL slopes (≈0.10-0.16 mm per +0.10 m). , By contrast, BMI and body weight show weak or negligible associations with AL (typically ≤0.1 mm per 10 kg/m² or near-zero per 10 kg). ,,, Ethnic differences can match or exceed sex differences; for example, in multiethnic cohorts from Xinjiang, Han children aged 12 to 18 years had axial lengths ≈1.0 mm longer than Kyrgyz peers at similar ages and with similar refractive distributions. SER-height associations in adults are essentially null.
SER vs AL (adults)
When AL and SER are modeled together in large adult cohorts, each extra −1.0 D of myopia is associated with ≈0.25 to 0.30 mm longer AL on average (pooled AL-SER slope −0.26 mm per 1 D; 95% CI −0.34 to −0.17), ,, implying that a 1-mm difference in AL maps to only a few diopters’ difference in SER, and this mapping varies between cohorts.
SER-AL coupling (children)
Girls are ≈0.2 D more myopic than boys despite shorter eyes. ,,, Cross-sectional SER-AL slopes are directionally negative but highly heterogeneous (roughly −0.2 to −1.1 D/mm), with distribution-based fits in one large Russian cohort around −1.3 D/mm. ,, Pooled estimates are imprecise and the prediction interval includes near-zero slopes.
Summary
Absolute AL varies by ≈1 mm or more across normal age, sex, height, and ethnic differences, on the same scale as many “high-risk” millimeter cutpoints, whereas SER is comparatively demography-agnostic and only partly reflects AL elongation.
LONG AXIAL LENGTH IN EMMETROPIC, NONPATHOLOGIC EYES
Across child and adult cohorts, most emmetropic eyes had AL in the low to mid-20s, with a small but consistent tail of very long yet emmetropic eyes (Figure 2). In children, emmetropic means typically lay between ≈22 and 24 mm, increasing with age (eg, ≈22.4 mm in 3-6-year-old Chinese preschoolers in Shenzhen and Chengdu vs ≈23.5 mm in 11-17-year-old German adolescents), and the bulk of the distribution remained well below 26 mm. ,,,,, Adult emmetropic AL distributions were similar. Across diverse populations, including rural Central India, Mongolia, Yunnan, Beijing, Singapore, ,, Hisayama, Tehran, Los Angeles Latino adults, and the Blue Mountains, mean AL clustered near ≈23 mm, with interpopulation differences generally ≤1 mm. Men and taller participants had longer eyes, but the mean AL in even the “longest” cohorts rarely exceeded 24 mm.
In both age groups, AL distributions were approximately Gaussian with a positively skewed long-eye tail. Histograms and ranges from European-ancestry Australian, Northern Irish, Beijing, Singaporean, ,, and other cohorts showed eyes extending into the 26- to 27-mm range and beyond, typically representing the top 1% to 3% of the sample; a minority of these very long eyes remained within the emmetropic band (SER −0.50 to +0.50 D). In such individuals—often taller males—corneal curvature and lens power appear to compensate for axial elongation, maintaining emmetropia despite AL values near or above 26 mm.
SER AND AL AS PREDICTORS OF PATHOLOGY
Retinal pathology
For the prevalence of any myopic retinopathy, AL showed the strongest and most consistent signal ( Table 2 ). Across 4 population-based cohorts, each additional +1 mm of AL was associated with an OR of 3.85 (95% Hartung-Knapp CI 1.70-8.72; I ² ≈ 69%). ,,, Long-eye thresholds near 26 mm produced very large but imprecise effects. ,, Two cohorts used AL ≥26.0 mm, whereas the Hisayama cohort contributed sex-stratified ROC-derived thresholds (25.9 mm men; 25.3 mm women—hence 2 Hisayama entries in Supplementary Figure S2B ). SER-based retinal data were less frequently reported; available cohorts suggested OR ≈1.5 to 1.8 per −1 D, , and high-myopia categories showed ORs ≈2 to 12, , with one outlier estimate ≈54 with extremely wide CIs. Overall, the retinal data support AL, particularly very long AL, as a powerful structural flag for severe myopic retinopathy, whereas SER provides a consistent but less precisely quantified gradient.
TABLE 2
Prevalence of Any Myopic Retinopathy.
| Association (unit) | Effect Measured | Pooled Effect (95% CI) | k | I (%) | Clinical Interpretation |
|---|---|---|---|---|---|
| AL per +1 mm | Odds ratio | 3.85 (1.70-8.72) | 4 | ≈69% | ≈4-fold higher odds for each +1 mm of axial length |
| AL threshold near 26 mm vs shorter (study-defined) | Odds ratio | 121.62 (2.31-6394.29) | 3 | ≈88% | Very long AL conveys very high risk |
| SER per–1 D (descriptive) | Odds ratio | ≈1.5-1.8 | <3 | — | Each–1 D of myopia increases odds by 50%-80% |
| SER high myopia vs emmetropia (descriptive) | Odds ratio | ≈2-12 (1 outlier ≈54) | <3 | — | High SER-defined myopia conveys high risk |
AL = axial length, D = diopter, I ² = proportion of variability due to heterogeneity, k = number of independent estimates, SER = spherical equivalent refraction.
Large estimates reflect sparse outcomes and should be interpreted cautiously. Thresholds were study-defined. Two cohorts used AL ≥26.0 mm; Hisayama (2019) used sex-specific ROC cutpoints (25.9 mm men; 25.3 mm women).
Cataract (by subtype)
Cataract analyses were stratified by subtype—cortical, nuclear, posterior subcapsular (PSC)—and by incidence vs prevalence, with SER providing the main pooled evidence ( Table 3 ). Incidence analyses using SER thresholds did not show clear associations for cortical ,, or nuclear cataract ,,, and only borderline evidence for PSC. ,, Pooled random effects RRs were 1.29 (95% CI 0.80-2.08; k = 3; I ² ≈ 78%) for cortical, 34,38,44 1.29 (0.75-2.21; k = 4; I ² ≈ 88%) for nuclear, ,,, and 1.44 (0.97-2.14; k = 3; I ² ≈ 43%) for PSC, ,, and Hartung-Knapp intervals plus leave-1-out analyses confirmed that none of these incidence blocks provided a definitively positive signal.
TABLE 3
Cataract Incidence and Prevalence by Subtype—Nuclear, Cortical, and Posterior Subcapsular.
| Association (unit) | Cataract Subtype | Effect Measured | Pooled Effect (95% CI) | I (%) | k | Clinical Interpretation |
|---|---|---|---|---|---|---|
| Incidence: high myopia vs no myopia (SER) | Cortical | Relative risk | 1.29 (0.80-2.08) | 78 | 3 | No convincing signal |
| Nuclear | Relative risk | 1.29 (0.75-2.21) | 88 | 4 | No convincing signal | |
| PSC | Relative risk | 1.44 (0.97-2.14) | 43 | 3 | Borderline risk | |
| Prevalence: moderate or high myopia vs no myopia (SER) | Cortical | Odds ratio | 1.07 (0.67-1.71) | 9 | 4 | Reproducible null |
| Nuclear | Odds ratio | 3.09 (1.38-6.91) | 49 | 4 | ≈3-fold higher prevalence | |
| PSC | Odds ratio | 4.58 (1.35-15.55) | 67 | 4 | ≈4-5-fold higher prevalence | |
| Prevalence: any myopia vs no myopia (SER) | Cortical | Odds ratio | 0.87 (0.74-1.03) | 4 | 5 | Reproducible null |
| Nuclear | Odds ratio | 2.08 (1.07-4.02) | 92 | 5 | ≈2-fold higher prevalence | |
| PSC | Odds ratio | 1.62 (1.14-2.30) | 48 | 5 | ≈1.5-fold higher prevalence | |
| Prevalence: AL per +1 mm (descriptive) | Cortical | Odds ratio | <1.0 | — | <3 | Slightly protective |
| Nuclear | Odds ratio | Near-null | — | <3 | No convincing signal | |
| PSC | Odds ratio | ≈1.1-1.3 | — | <3 | Modest, imprecise effect |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
