Simple robust procedures for combining risk differences in sets of 2 × 2 tables

Abstract

Meta-analyses often use a random-effects model to incorporate unexplained heterogeneity of study results. Trimmed versions of meta-analytic estimators for the risk difference, adapted from procedures designed for a random-effects analysis, can resist the impact of a few anomalous studies. A simulation study compared untrimmed and trimmed versions of four meta-analytic procedures that give weighted averages of risk differences. An adaptation of Winsorized estimates of components of variance gains some resistance to anomalous studies when estimating variability. The simulations found that a modified version of the DerSimonian-Laird estimator is attractive when risk differences reveal the added variability described by a random-effects model, and that a 20 per cent trimmed, weighted version of this procedure offers resistance against the impact of highly anomalous results. Among four trimmed procedures considered, the trimmed version of the modified DerSimonian-Laird estimator offers the best performance over a wide range of simulation designs and sample sizes. None of the methods, whether trimmed or untrimmed, is uniformly preferable. A published meta-analysis of a vaccination against TB provides data that serve to illustrate differences among the eight procedures.

Eating disorders in females with type 1 diabetes: an update of a meta‐analysis

“…This leads to energy purging by glucosuria. Only a few publications with sufficient detailed (Emerson et al, 1996) information have been located (Affenito et al, 1997;Bryden et al, 1999;Cantwell & Steel, 1996;Fairburn et al, 1991;Jones et al, 2000;Rodin et al, 1991;Takii et al, 1999). Data and findings are shown in Table 7 using OR and in Table 8 using RD as a measure of effect.…”

“…The overall robust finding is a five times increase in OR for retinopathy in eating disordered versus non-eating disordered IDDM females. If the risk difference-RD (Cytel Software Corporation, 2000)-is used as the parameter (see Table 6) the overall finding is a mean RD of 0.33, no matter whether a non-weighted or an n-weighted mean is used (Emerson et al, 1996). This translates into an average 33 per cent (95%CI: 25 to 42%) increase in retinopathy in eating disordered IDDM females when compared with similar groups of noneating disordered IDDM females.…”

“…There seems to be a strong relationship between eating disorder and intentional underdosing of insulin in females with type 1 diabetes. Overall common OR is 12 and the n-weighted mean RD is 0.40 (the n-weighted mean is used as the studies differ significantly in size) (Emerson et al, 1996). Forty per cent more eating disordered IDDM females (95%CI: 29 to 50%) than non-eating disordered IDDM females misuse insulin to achieve weight loss.…”

“…Corporation, 2000) incorporates the principles outlined in Santner and Snell (1980) and Berger and Boos (1994), and an idea in Berger and Boos (1994) is extended from the level of hypothesis testing to confidence intervals. The combination of findings using RD as a measure of effect follow Emerson, Hoaglin, and Mosteller (1996).…”

Synthesizing medical evidence: systematic reviews and metaanalyses

“…1,283,308,334,369 Any of the three can form the basis for a metaanalysis. [113][114][115]122 The odds ratio has attractive mathematical properties, but the relative risk is more easily understood by clinicians; the two measures are similar for "rare events" (event rate Յ 10%), but when event rates are high the odds ratio can be substantially larger than the relative risk, and may give an exaggerated impression of the treatment effect. 308 The risk difference between groups is also easily understood by clinicians.…”