Statistics from Altmetric.com
I would like to congratulate Abugreen et al1 for publishing an interesting article on the relation between choroidal neovascularisation (CNV) in one eye and age related maculopathy (ARM) in the opposite eye. However, the required data to support some of the authors’ conclusions were not given in the paper.
The authors stated in the abstract that “the area occupied by the CNV in the first eye also influenced severity of ARM changes in the fellow eye.” In the results section it was stated that “age, sex, cardiovascular disorder, and smoking status were not significant predictors for ARM severity in this model.” In the statistical analysis section of Methods, as well as in the footnotes to Table 5, it was mentioned that the “CNV subtype” was the dependent variable and not the “ARM severity in the fellow eye.” Therefore, the model was attempting to predict the “type of CNV in the same eye,” and not the “ARM severity in the fellow eye” using the independent variables age, sex, cardiovascular disorder, etc. In addition, there appear to be typographical errors in the p values in Table 5; 0.57 should read 0.057, 0.19 should read 0.019.
Also in Table 5, the odds ratio of stage 3 ARM (soft indistinct drusen or reticular drusen with pigmentary irregularities) being 9.48 times more than no ARM predicting an occult CNV over a classic CNV was also misleading. This is because the individual coefficients of the independent variables reflect the contribution of these factors (including area of the CNV lesion) to the variance of the dependent variable (the CNV subtype)! In order to examine the effect of CNV subtype in one eye on the severity of ARM in the other eye, perhaps the authors should collapse the five possible values for the severity of ARM into two and designate this the dependent variable in the logistic regression model. Since the logistic regression model predicts the log odds that an observation will have an indicator equal to 1, to facilitate interpretation of odd ratios, it is crucial to specify which of the response condition (“occult CNV” or “classic CNV”) is designated as 1 (the counterpart being 0). Alternatively, they could perform “multiple linear regression” and designate “severity of ARM” as the dependent variable with five possible values. In either of these two alternatives, area of lesion (in the eye with CNV), age, etc could be additional independent variables.
If the authors were to compare only “severity of ARM” with “CNV subtype” (four possible values), Kendall’s rank correlation is also a reasonable approach.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.