Comparative effectiveness of bevacizumab and ranibizumab in the comparison of age-related macular degeneration treatments trials.

Comparative Effectiveness of Bevacizumab andRanibizumab in the Comparison of Age-Related Macular Degeneration Treatments Trials To theEditor In2012,Martinetal1publishedtheresultsat2years of theComparisonofAge-RelatedMacularDegenerationTreatments Trials and found a nonsignificant difference between bevacizumabandranibizumabaccordingtothecontinuousend point of the change in letters of visual acuity (mean difference forbevacizumabvs ranibizumab,−1.4 letters; 95%CI,−3.7 to +0.8 letters;P = .21).More recently, Ying et al2 carried out a multivariate analysis of the same study data in which the bevacizumab groupwas shown to have higher odds of losing 15 or more letters than the ranibizumab group (odds ratio, 1.83; 95% CI, 1.07 to 3.14; P = .03). If the crude rates of failures reported by Ying and colleagues are examined by univariate analysis (37 of 501 patients for bevacizumab vs 24 of 528 patients for ranibizumab), the odds ratio is 1.67 (95% CI, 0.99 to 2.84; P = .05), which is very close in terms of magnitude and statistical significance to that obtained by multivariate statistics. Hence, while these univariate results and the results from themultivariate analysis are similar, they suggest thepossibilityof adifferent conclusion from the primary results of the Comparison of Age-RelatedMacular Degeneration Treatments Trials.1 In interpreting their results, Ying and colleagues postulated that, as comparedwith theprevious resultsbyMartinand colleagues, themore pronounceddifference betweenbevacizumab and ranibizumab was related to the multivariate designof their analysis and to the consequent adjustment for covariates. However, the observation that univariate and multivariate results were similar suggests another explanation: the inferiority of bevacizumab vs ranibizumab could be a spurious result related to the adoptionof adichotomousend point in substitution for the continuous one. In general, when the same data set is analyzed by using a continuous or a dichotomous variable, the most information is contained in the analysis of the continuous variable. Expressing the results according to thedichotomousvariable (defined according to a cut point) candetermine amore effective communication of the results, but some information is unavoidably lost. This loss of information can have unpredictable effects, particularly if there is a considerable amount of data near the cut point. For this reason, the inferiority of bevacizumabvs ranibizumab foundbyYingandcolleaguescould be an artifact related to converting the continuous end point into a dichotomous one. To exclude this artifact, a reanalysis could be made by the authors in which the end point is kept as a continuous variable. If this reanalysis confirms the inferiority of bevacizumab, the clinical relevance of this important finding would be strongly enhanced.