Abstract
Purpose :
Two methods are commonly used to evaluate efficacy of myopia control interventions – 1) using continuous data as they were collected and determining whether the treatment effect size meets a pre-specified target, eg, 0.75 D, and 2) dichotomizing the data by applying an arbitrary cut-off for analysis of odds of progression exceeding an amount, eg, 0.50 D. We consider the implications for myopia control efficacy trials and the efficiency of the approaches in detecting clinically important changes in myopia progression.
Methods :
We simulated randomized controlled clinical trials by repeatedly sampling a fixed number of subjects from test and control populations with normally distributed outcomes (changes in refraction and axial length) whose means differed by a fixed effect. Typical progression and Standard Deviation (SD) over time for the control population were estimated from a literature review covering a range of subject ages and races. The empirical power of the clinical trial for detecting a statistically significant difference was computed for a range of effect sizes, progression cut-off values, and sample sizes. Continuous outcomes were assessed by independent sample T-tests and dichotomized outcomes by Chi-square test and Odds Ratio (OR) 95% confidence interval. Significance was determined based on p <0.05 for testing or by an interval failing to cover 1.
Results :
Figure 1 presents an example of the output, plotting empirical power for detecting an efficacy over a range of effect sizes (0.25 to 0.75 D) and progression cut-off values (-0.50 to -1.00 D) for sample sizes ranging between 50 and 200 per group after 36 months of treatment. Predictably, T-test consistently outperformed the Chi-square test and OR for any given sample size to detect any efficacy. Dichotomization requires sample size inflation to achieve comparable power to the continuous approach.
Conclusions :
Arbitrarily dichotomizing continuous outcome measures for analysis of odds without regard to treatment effect size does not sufficiently inform about clinically relevant efficacy, limits generalizability, and is less efficient in design of a myopia control clinical trial.
This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.