ABSTRACT
Abstract A challenging problem in meta-analysis is combining studies in which similar medical outcomes are captured in some studies as continuous variables and in others as binary variables. A common approach is to dichotomize the continuous responses and proceed as in the simpler binary case. This approach is practical, but it has limitations. One is that there may be arbitrariness in the choice of the cutoff point. Another is that there is a loss of information. In this paper, we propose a strategy that overcomes both of these difficulties. It is based on assuming that the binary responses are the result of dichotomizing some underlying unobserved continuous variable. Bayesian reconstruction of the unobserved continuous variable preserves the full information from the studies reporting continuous variables and does not require the choice of arbitrary cutoff points.
