gms | German Medical Science

Background: In a systematic review on the effect of a metric exposure on a binary outcome, one may find very different summaries and effect measures that seem hard to meta-analyze. Odds ratios may be given as per unit of exposure, per standard deviation of exposure, per logarithmic unit or the standard deviation of that, or per one of two or more classes. And those classes may be delimited by quantiles with the extremes mentioned or not. Here the focus is on how data were pre-processed for mixed logistic regression, the meta-analysis, when event counts were available for quantile classes.

Methods: First, the parameters of the assumed lognormal distribution of the metric exposure were estimated from whatever quantiles there were available, using formulae like those of Wan et al. [1]. From these, extremes could be estimated as (1/(n+1))-quantiles. Such class limits are needed for both, the semi-parametric trapezoidal rule, and the parametric estimation of class means, to determine values representing the classes delimited by quantiles. They were used to explain the numbers of events reported for these classes in a logistic regression. Random study effects were considered for both, a transformation of incidence, and for logarithm of odds ratio. Considering both requires rich data that may not be available for all research questions. Some judgment is needed then, which of the random effects should vary more given the specified selection of participants. As precision of regression coefficients depends not only on sample size, but also on the regressor range, the results were plotted less like a forest plot and more suggestive of a regression. More than one symbol per study, with size indicating case number, were overlaid with the individual and the summary regression lines, with line width proportional to case numbers.

Application: Methods are illustrated by the meta-analysis following a systematic review conducted by Sina Röhrs. It studies the association between major cardiac adverse events and blood levels of trimethylamine-N- oxide in patients with diabetes.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.