We read with great interest the article by Raymond et al.
1 on a randomized controlled study comparing refractive outcomes after cataract surgery using applanation ultrasound (US) or partial coherence laser interferometry with the IOLMaster (Carl Zeiss Meditec, Dublin, CA). The purpose of the study was to assess whether these methods of measurement of axial length have a difference in precision of refractive outcomes. There are two aspects of the design of this study that compromise its conclusions.
The authors state that the trial was powered to detect a difference of 0.24 D in mean absolute error (MAE), without explaining the reasons or providing any evidence of why a difference of <0.24 D is not clinically significant. We can only assume that a level of 0.24 D was selected because of evidence supporting that a change of 0.25 D in spherical equivalent has an impact on unaided visual acuity.
2 A level of 0.24 D in MAE can actually have a big impact on refractive outcomes. For example, Olsen
3 discovered a difference at 0.23 D in MAE between applanation US and IOLMaster biometry (0.65 D vs. 0.43 D). This result translated to improved refractive outcomes from 45.5% and 77.3% for applanation US to 62.5% and 92.4% for IOLMaster for deviations of ±0.5 and ±1.0 D from the expected outcome (
P < 0.00001).
3 According to the criteria set for the study by Raymond et al.,
1 this level of improvement in refractive outcomes is not clinically significant. There have been no clinical studies validating a specific level of clinical significance for MAE in the setting of refractive outcomes after cataract surgery.
MAE is a measure of the spread (precision) of a distribution assuming a mean numerical error (MNE) of 0. When the MNE is not 0, the MAE is increased, and it no longer quantifies spread (precision) alone but is also affected by inaccuracy. The authors' decision not to use optimized IOL constants but to use those recommended by the manufacturer (118.9 for IOLMaster and 118.7 for applanation US) could have introduced systematic errors from high MNEs and further compromised the power of the study.4 Nevertheless, the resulting MNEs were small (−0.10 and +0.12 for the IOLMaster and applanation US, respectively). It is not advisable though to omit such an important step when investigating refractive outcomes, especially in the setting of a randomized controlled study.
In conclusion, this study is underpowered to answer the question of whether applanation US biometry offers the same precision in refractive outcomes as the IOLMaster. Optimizing the IOL constant could have compromised the study further.