Meta-analysis Options

 

Menu location: Analysis_Meta-Analysis_Options.

 

You can set some common calculation and charting options via these functions:-

 

Calculation_Options…

'Try exact calculation': This dictates whether or not conditional maximum likelihood calculations are attempted – the computations can take a long time with large tables that contain a mixture of large and small numbers. In most cases the calculation completes within a few seconds.

 

Continuity correction for sparse tables: This determines the number that is added to each count in a table that contains zero in any cell where a zero would prevent the calculation of the main effect or its variance - for example the odds ratio of a fourfold table (a, b, c, d) is the ratio of cross-products, (a*d)/(b*c), therefore a zero b or c would make it impossible to calculate the odds ratio. The traditional work-around has been to add 0.5 to each count in the table – the so-called 'Haldane correction'. Agresti (1996) recommends adding smaller constants - you may enter your choice. Very small constants bring the Mantel-Haenszel estimate of the pooled odds ratio very close to the conditional maximum likelihood estimate. The Peto method performs well with sparse datasets where the event rate is low, provided there is not a large imbalance between the sizes of the treatment vs. control groups (Bradburn et al., 2007). Alternative empirical continutity corrections (Sweeting et al., 2004) have been introduced to improve calculations where there is a substantial difference between the sizes of the groups being compared. The so called 'treatment arm (sums to 1)' continuity correction performs well with the Mantel-Haenszel pooled estimate. If R is the ratio of control to treatment group sizes then the treatment arm continuity correction for the treatment group is 1/(R+1), and for the control group it is R/(R+1): what is added to each count of a group is in proportion to the size of the group, and the two corrections sum to 1. In the meta-analysis of incidence rates the number is added to the events of both groups of a stratum that has no events in one of its groups; 0.5 is added if the treatment arm correction is selected.

 

'Delay continuity correction': This option works with the fixed effects (Mantel-Haenszel) calculations of the odds ratio, relative risk and risk difference meta-analyses, where it delays the application of any continuity correction that is necessary until the random effects stage: the pooled estimate, its confidence interval and the fixed effects weights are calculated from the counts as they are. The counts as they are must be able to give a pooled estimate: for the odds ratio there must be a study in which neither a nor d is zero, and for the relative risk there must be an event in each group among the studies; otherwise the continuity correction is applied at the fixed effects stage also.

 

Plot_Options…

These options determine the type of bias detection plot used. The vertical axis options are: standard error; precision; 1/sample size; sample size; 1/log(sample size); log(sample size); 1/Mantel-Haenszel weight. If you are considering an incidence study then you should make your selection by reading these options as 'person-time' instead of 'sample size'.

 

The 'include CI if relevant' check box determines whether or not confidence interval lines are drawn on the bias detection plot – these form an inverted triangle on a traditional 'funnel plot'.

 

See bias detection for more information about these plots.