Abstract
Purpose :
Contrast sensitivity (CS) is important for patient safety, as well as distance perception and driving (Ginsburg 2003). Presbyopia-correcting intraocular lenses (PCIOLs) increase spectacle independence, at the expense of small losses in CS. It is important to show that CS differences of PCIOLS compared to monofocals are small, traditionally taken to be 0.30 log units or less. Standard deviation of CS measurements is traditionally expected to be 0.40. In this study, we look at variability in CS measurements, different criteria to determine safety limits, and the impact in terms of number of subjects needed.
Methods :
Population variability of a commercially available CS measurement system (M&S) was estimated from retrospective analysis of 80 conditions with a total of 5840 CS values from different studies. Monte Carlo simulations were used to estimate the percentage of times a study would be expected to result in a hypothetical PCIOL exceeding the criteria. Two different criteria were compared: first when a statistically significant difference of at least 0.30 on at least two out of four spatial frequencies were found (2SF), and second when a statistically significant difference of at least 0.30 was found on any of the four spatial frequencies (1SF).
Results :
The retrospective analysis had 90% of the conditions normally distributed, with standard deviations ranging between 0.21 and 0.48. PCIOLs did not have higher standard deviations than monofocal IOLs. If data from different studies with the same IOL were pooled, normality only held for 56% of the conditions, indicating the preferability of within-studies rather than between-studies comparisons.
The Monte Carlo simulations revealed several insights. For example, running a study with 80 subjects would be sufficiently powered to accurately determine a true difference of 0.25 to be safe with a power of 0.80 in the 2SF criterion. With the 1SF criterion, the power would drop to 0.37, and a maximum allowed difference of 0.20 is needed to have a power of 0.80. To avoid a type 2 error with a true difference of 0.35 in 95% of the cases, 60 subjects sufficed for all criteria.
Conclusions :
Variability of CS estimates and which criteria are used have a profound impact on the number of subjects needed and which true differences to aim for when studying PCIOLs with the goal of showing CS differences lower than 0.30.
This is a 2020 ARVO Annual Meeting abstract.