Different keratoconus definitions can lead to substantial prevalence disparities in population-based studies

Abstract

This report explores the prevalence of keratoconus in a population-based cohort of adults aged 40 or older according to ten different definitions. All Rotterdam Study participants with reliable Pentacam scans and no prior corneal refractive surgery were cross-sectionally analysed (n = 2660). First, we applied a novel evidence-based definition. Suspected keratoconus was defined as having at least one eye with a final D-index (BAD-D) ≥ 2.6. Manifest keratoconus was defined as having at least one eye with: (1) BAD-D ≥ 2.6; and (2) a score of at least 4/10 on the novel Rotterdam Keratoconus Scale (RKS); and (3) a confirming assessment of the relevant Pentacam maps; and (4) meeting Holladay's criteria in case of recent contact lens usage. Using this proposed definition, 72 participants (2.71%, 95%CI: 2.16-3.40%) had suspected keratoconus, while 10 participants (0.38%, 95%CI: 0.20-0.69%) had manifest keratoconus. To assess reproducibility, two specialists independently applied the proposed definition, with a substantial inter-observer agreement (Kappa = 0.74). Interestingly, 6(60%) patients were unaware of having keratoconus. Applying nine alternative definitions from similar screening studies produced prevalence estimates ranging from 0.19 to 9.29% in the same cohort. Moreover, counting solely on a BAD-D cutoff of 2.6 to define keratoconus was unreliable, with a low positive predictive value of 14%. These findings explain partially the large heterogeneity in the reported keratoconus prevalences, underscoring the need for a standardized definition.

Fig. 1

Filtering process according to the proposed keratoconus definition.

Fig. 2

Keratoconus prevalence in The Rotterdam Study according to other published definitions (2a), and their respective agreement levels (2b). ^a The study lists two keratoconus definitions. However, Only the definition based on the Topographic Keratoconus index (TKC) was readily applicable to our dataset. The heatmap was generated with the “pheatmap” package in R. All the reported kappa coefficients were statistically significant (P < 0.001). NZ, New Zealand.

Fig. 3

Unawareness rates reported in different screening studies.