Meta-analysis of skewed data: combining results reported on log-transformed or raw scales.
Higgins JP, White IR, Anzures-Cabrera J. · 2008
Read Meta-analysis of skewed data: combining results reported on log-transformed or raw scales. by Higgins JP, White IR, Anzures-Cabrera J. (2008) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc2978323.txt.
Introduction
Higgins, White, and Anzures-Cabrera worked on a pooling problem in a 2008 methods study. The task is combining trial results when outcomes are skewed. Some papers report raw values, while others report logarithms. Leaving studies out wastes evidence, so a common scale is needed.
They review and extend ways to convert between the two reporting styles for two-group studies, including pooled ratios of geometric means. Simulations and a worked example favour a method assuming log-normal raw data, which held up across varied true shapes, with fresh formulas for its uncertainty. Assuming equal spread in both groups helps only when that guess is right.
Catalog record
| CREATOR | Higgins JP, White IR, Anzures-Cabrera J. |
|---|---|
| YEAR | 2008 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Meta-analysis of skewed data: Combining results reported on log-transformed or raw scales By Julian P T Higgins, Ian R White, Judith Anzures-Cabrera Published in: Statistics in Medicine DOI: 10.1002/sim.3427 ABSTRACT When literature-based meta-analyses involve outcomes with skewed distributions, the best available data can sometimes be a mixture of results presented on the raw scale and results presented on the logarithmic scale. We review and develop methods for transforming between these results for two-group studies, such as clinical trials and prospective or cross-sectional epidemiological studies. These allow meta-analyses to be conducted using all studies and on a common scale. The methods can also be used to produce a meta-analysis of ratios of geometric means when skewed data are reported on the raw scale for every study. We compare three methods, two of which have alternative standard error formulae, in an application and in a series of simulation studies. We conclude that an approach based on a log-normal assumption for the raw data is reasonably robust to different true distributions; and we provide new standard error approximations for this method. An assumption can be made that the standard deviations in the two groups are equal. This increases precision of the estimates, but if incorrect can lead to very misleading results. Copyright © 2008 John Wiley & Sons, Ltd. INTRODUCTION Meta-analyses of clinical trials, epidemiological studies and other types of study may involve continuous outcome data. Continuous data can be skewed, typical examples being concentrations (e.g. of plasma triglycerides), other ratio or reciprocal measures (e.g. percentage reduction), measures related to resource use (e.g. recovery time) or assessment scales when there is a large proportion of ‘normal’ participants with scores towards one extreme of the scale (e.g. measures of cognition in population-based studies). Standard inferences on the means of skewed data are valid for large sample sizes due to the central limit theorem, which determines that the mean of the outcome measurements is approximately normally distributed with standard deviation given by the standard error (obtained by dividing the standard deviation by the square-root of the sample size). Since standard meta-analytic methods assume normality in the distribution of the means, but not the raw data [1], they are valid when sample sizes within individual studies are sufficient to enable the central limit theorem to hold approximately. Focussing on the raw outcome measurements is problematic when the sample size is small, as the standard deviation and mean are affected by extreme values in one direction. It may also lead to loss of efficiency regardless of the sample size. This is well recognized by authors of primary research studies. A common approach to dealing with skewed outcome data is to take a logarithmic transformation of each observation and to conduct the analysis using log-transformed values. This yields, for example, a mean of the log-concentration levels together with a standard deviation of the log-concentration levels, leading directly to a confidence interval for the mean log-concentration level. The mean of the logs can be readily transformed to a geometric mean along with a confidence interval. A logarithmic transformation can offer further advantages, including a focus of the analysis on clinically more appropriate measures of effect [2]. A practical complication in meta-analyses based on summary data is that some studies may present means and standard deviations on the log scale (or geometric means on the raw scale), while other studies present means and standard deviations on the raw scale. It can be difficult to determine exactly which data have been presented. It may, for example, be unclear whether a mean is an arithmetic mean or a geometric mean. Furthermore, some papers may present inappropriate results, such as the exponential of the standard deviation of log-transformed values. Even when the results have been correctly interpreted; however, there remains the problem of combining results on different scales. Here, we present straightforward and approximate transformations that enable meta-analyses on either the raw or the log-transformed scale, irrespective of how results are presented. We do assume; however, that the nature of all results extracted from papers is known, and we focus on making inferences concerning the comparison of two groups. SOME TYPES OF PRESENTATION OF CONTINUOUS OUTCOME DATA Consider first a single group; say an intervention or a control group from a clinical trial, or a specific exposure group in an observational epidemiological study. Let n be the sample size in this single group. Let and sx represent the arithmetic mean and standard deviation of raw (not log-transformed) measurements. Lower and upper limits of a 95 per cent confidence interval for the mean, are obtained as where t is the 97.5 percentage point of the t-distribution with (n − 1) degrees of freedom. Let and sz represent the arithmetic mean and standard deviation of log-transformed measurements. Lower and upper limits of a 95 per cent confidence interval for are obtained as The geometric mean may be obtained as . A 95 per cent confidence interval for the geometric mean is given by Data available to a meta-analyst might be in one of the following formats, although the list is not exhaustive: Mean and standard deviation of raw measurements ( and sx).Mean and standard error for raw measurements and sx/).Mean and confidence interval for raw measurements (, and ).Mean and standard deviation of log-transformed measurements ( and sz).Mean and standard error for log-transformed measurements ( and sz/).Mean and confidence interval for log-transformed measurements (, and ).Geometric mean and confidence interval (g, gl and gu).Geometric mean and incorrect standard deviation (g and exp(sz)). The formulae above can be used to convert any of (1)–(8) to either the mean () and standard deviation (sx) for raw measurements orthe mean () and standard deviation (sz) for log-transformed measurements. This should be undertaken before applying the transformation methods below. METHODS FOR TRANSFORMING SUMMARY DATA For variable X with a log-normal distribution, such that it is a standard result that the mean and variance of X are given by and We consider three methods for transforming between log-transformed and raw scales, that is, for estimating the mean and variance of X from the sample mean and variance of Z, or vice versa. The first two methods exploit the result above. In Method 1, we transform the mean and standard deviation within each group, and then make the comparison across groups. The standard deviations are thus allowed to differ in the two groups. Method 2 follows the same approach as Method 1, but assumes a common standard deviation underlying both groups. This assumption of common standard deviation could be made on either the raw or the log-transformed scale; we choose the latter as a generally more plausible assumption. Method 3 targets the difference between the groups rather than the group means separately. It does not assume a log-normal distribution for the raw data, and is applicable to other transformations as well as the log transformation. We also derive expressions for the standard errors of the estimators. One possibility for Methods 1 and 2 is to apply standard methods to the converted means and standard deviations for the two groups to obtain a difference in means and its standard error: we call this the ‘ad hoc’ approach. However, estimators based on the