Paired observations of refractive error after cataract surgery

To quantify the design effect if pairing between eyes is ignored when reporting refractive outcome following cataract surgery.

A retrospective review was conducted of biometry prediction error in 612 patients who had bilateral phacoemulsification with intraocular lens (IOL) implantation. The levels of uncertainty (confidence interval and variance) of the estimates for the biometry prediction error were compared when intra-cluster correlation between ocular measurements of paired eyes within each patient was either taken into account or ignored. The design effect (the ratio of the two variances) was calculated.

The correlation between the two eyes was weak, as indicated by the intra-cluster correlation coefficient of 0.22 (confidence limits 0.15-0.30). The mean biometry prediction error was 0.73 dioptres (D) and the 95% confidence limits for the mean in the two situations were similar at 0.69-0.76D and 0.69-0.77D, respectively. The variance of the biometry prediction error if pairing of eyes was ignored (0.00033D) was slightly less than if pairing was taken into account (0.00041D). This yielded a design effect of 1.2.

A high correlation between data collected from pairs of eyes is not inevitable. If data on both eyes are used to estimate biometry prediction error, the usual statistical procedures will underestimate the standard errors and the variance, but the effect will be trivial and the confidence intervals only negligibly smaller. The maximum variance is expected to be 1.2 times higher than that given by simple statistical methods that ignore the pairing of eyes.