We demonstrate that the numerical values are nearly identical for a data example. For comparing more than two treatments, we explain the SUCRA statistic and introduce a quantity, called P-score, that can be considered as a frequentist analogue to SUCRA. We use a simple analytical argument to show that the probability of being best can be misleading if we compare only two treatments. In this article, we intend a critical appraisal of ranking, considering both the Bayesian and the frequentist perspective. WinBUGS code for obtaining rank probabilities is given in the supplementary information of. They also introduced several graphical presentations of ranking, such as rankograms, bar graphs and scatterplots, and a numerical summary of the rank distribution, called the Surface Under the Cumulative RAnking curve (SUCRA) for each treatment. Salanti et al., introducing a rank statistic, extended the consideration to the probabilities that a treatment out of n treatments in a network meta-analysis is the best, the second, the third and so on until the least effective treatment. Within the Bayesian framework, authors have noted that it is not sufficient and can be misleading to solely look at the probability of being best, as it does not take uncertainty into account. By contrast, from a frequentist perspective, treatment effects are thought as fixed parameters and thus, strictly speaking, a concept like ‘the probability that A is better than B’ does not make sense. Thus statements such as ‘treatment A is superior to treatment B with probability 60 %’ or ‘Treatment A ranges under the three best of ten treatments with probability 80 %’ are possible. From a Bayesian perspective, parameters such as those describing the relative effectiveness of two treatments are random variables and as such have a probability distribution. One outstanding feature of the Bayesian approach often noted is that it allows to rank the treatments according to their comparative effectiveness. It has been argued that ‘Bayesian methods have undergone substantially greater development’. Bayesian methods are often preferred in network meta-analysis for their greater flexibility and more natural interpretation. The methodology of network meta-analysis has developed quickly and continues to be refined using both Bayesian and frequentist approaches. However, neither SUCRA nor P-score offer a major advantage compared to looking at credible or confidence intervals.Īn increasing number of systematic reviews use network meta-analysis to compare three or more treatments to each other even if they have never been compared directly in a clinical trial. Like the SUCRA values, P-scores induce a ranking of all treatments that mostly follows that of the point estimates, but takes precision into account. Ranking treatments in frequentist network meta-analysis works without resampling. Using case studies of network meta-analysis in diabetes and depression, we demonstrate that the numerical values of SUCRA and P-Score are nearly identical. They measure the mean extent of certainty that a treatment is better than the competing treatments. P-scores are based solely on the point estimates and standard errors of the frequentist network meta-analysis estimates under normality assumption and can easily be calculated as means of one-sided p-values. For comparing treatments in a network meta-analysis, we propose a frequentist analogue to SUCRA which we call P-score that works without resampling. The treatments can then be ranked by the surface under the cumulative ranking curve (SUCRA). Within a Bayesian framework, for each treatment the probability of being best, or, more general, the probability that it has a certain rank can be derived from the posterior distributions of all treatments. Network meta-analysis is used to compare three or more treatments for the same condition.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |