Eyeworld

26 | EYEWORLD | FALL 2026 ASCRS NEWS The sample is also relatively large for a U.S. pediatric CXL series. Eighty pediatric eyes sub- stantially exceed the sample sizes in many prior pediatric reports, and the inclusion of multiple age groups creates an opportunity to explore whether treatment effect varies across the age spectrum. The racial and ethnic diversity of the population similarly creates an opportunity to investigate populations historically under- represented in long-term North American CXL studies. Finally, the authors do not conceal the limitations created by attrition. They explicitly state that later observations represent progres- sively smaller, non-identical subsets of returning eyes and acknowledge that this may generate both reduced power and selection bias. That transparency considerably improves interpreta- tion of the results. What this study convincingly tells us The strongest conclusion is relatively conserva- tive: Among patients with progressive kerato- conus who returned for follow-up after epithe- lium-off CXL, average visual and tomographic outcomes were favorable for several years, including in a substantial pediatric population. Most Kmax flattening appears to occur during the first 2 years, and approximately 10% of eyes demonstrated >1 D worsening at the 12-month timepoint. The data are reassuring regarding pediatric CXL and add useful North American experience to the literature. They are considerably less definitive regarding whether pediatric patients actually respond better than adults, whether treatment remains equally effective through 5 years in the original cohort, or whether out- comes differ by race/ethnicity. The finding that Kmax was no longer significantly different from baseline at year 5 is particularly interesting. It should not be interpreted as evidence that the CXL effect disappears after 4 years—the sample has fallen to only 45 eyes, with consequently reduced sta- tistical power—but neither should the finding simply be dismissed. Longer and more complete follow-up is needed to determine whether this represents statistical noise, attrition, or the be- ginning of late loss of treatment effect. The investigators also performed a baseline- referenced analysis in which each postoperative observation was compared with the baseline measurement from the same eyes. This is an important methodological improvement over simply graphing raw cohort means as though they represented a single continuously observed population. The resulting pattern was similar: early improvement in CDVA, initial corneal thin- ning followed by partial recovery, and maximum Kmax flattening around year 2. Age-stratified analyses suggested favorable outcomes across all age groups, with the pediat- ric and younger adult groups showing apparent- ly larger or more persistent effects. In patients younger than 18, CDVA remained statistically improved through 48 months, while Kmax was significantly flatter at 12 and 24 months. The authors interpret these results as suggesting a particularly durable response in younger patients. Race/ethnicity subgroup analyses similarly demonstrated statistically significant early im- provements in several outcomes among White, Hispanic, and Asian patients, although smaller subgroup sizes limited statistical power in other groups. Statistical strengths A major strength is the authors' recognition of the statistical problems created by incomplete longitudinal data. Rather than simply compar- ing each year's aggregate mean with the original baseline mean, they calculated interval-specific paired changes using only eyes represented at both baseline and that postoperative timepoint. This avoids one particularly misleading form of longitudinal analysis in which changing cohort composition masquerades as within-patient change. The use of a nonparametric paired test is also reasonable given the relatively small sam- ples at later intervals and the likely non-normal distributions of several keratoconus measure- ments. The analysis preserves within-eye pairing rather than treating baseline and postoperative observations as independent samples. Another statistical strength is presenta- tion of 95% confidence intervals in the figures rather than relying exclusively on P values. This provides at least some information regarding precision and makes the increasing uncertainty at later follow-up visually apparent. continued from page 25

• Cover