How were the studies combined?
A pooled estimate of the rate of PCO for each timeframe was obtained from the observed rates of the eligible studies using the general method of the weighted average. For comparison, the data were pooled using different weights:
the total variance, i.e. weights equal to the inverse of the sum of the study variance and the among-study variance;
no weighting; and
weights equal to the sample size of each study.
In all cases, the 95% confidence intervals (CIs) were calculated using the standard errors derived from the method of moments approach, as proposed by DerSimonian and Laird (see Other Publications of Related Interest no.1).
How were differences between studies investigated?
Chi-squared tests for homogeneity of proportions were performed using the method outlined by Cochran (see Other Publications of Related Interest no.2) for each timeframe.
The authors performed a separate analysis of the four studies that directly compared IOLs with a convex posterior surface. A pooled estimate of the overall relative risk of PCO for the convex posterior versus plano posterior surface IOLs was obtained using the Mantel-Haenszel approach. The Breslow-Day test for homogeneity (see Other Publications of Related Interest no.3) was performed on these four studies.
Using both simple unweighted linear regression models and models weighted by the inverse total variance, the authors examined the relationship of time since surgery and PCO incidence. In these models, the following were tested:
differences in PCO incidence among the three time frames (1,3, and 5 years);
a linear trend in incidence; and
differences in PCO incidence associated with the actual follow-up times reported in the studies.
The authors also investigated whether the actual follow-up time of the studies was associated with the PCO rate within their 1- and 3-year timeframes. In further linear regression models for the 1-year time period, the authors investigated whether other factors could account for a portion of the variability in the reported rates of PCO. The factors investigated included: the type of surgery (phacoemulsification versus standard extracapsular); the type of IOL material (silicone acrylate versus polymethylmethacrylate); the inclusion of a visual acuity threshold for performance of YAG capsulotomy (yes versus no); and the percentage of patients lost to follow-up.