Influence of a single study in meta-analysis estimation -------------------------------------------------------
^metaninf^ varlist [^if^ exp] [^in^ range] [, ^fixed fixedi random randomi^ ^rr or rd peto cornfield hedges cohen glass nosta^ndard ^label(^label_vars^) il^evel^(^#^) ol^evel^(^#^) notable nograph^ ^sav^e^(^varname [^, replace^]^)^ ]
^metaninf^ investigates the influence of each individual study on the overall meta-analysis summary estimate. The command presents a table and a graph of the results of an influence analysis in which the meta-analysis is reestimated omitting each study in turn.
The table numerically provides the results, with the rows of the table being the meta-analysis of all studies except the "omitted" study named in that row, and the usual, full meta-analysis (omitting none of the studies) being given as the "combined" results at the bottom of the table.
The graph visually provides the same results in a plot, naming the omitted study on the left margin and presenting the resulting "omitted" meta-analytic summary statistics as a horizontal confidence interval. The full, "combined" results are shown as the solid vertical lines.
No formal test of influence is given; rather, the program provides and displays results to which some general guidelines can be applied to assess influence. One such guideline is that an individual study is suspected of excessive influence if the point estimate of its "omitted" analysis lies outside the confidence interval of the "combined" analysis. Another quideline is that a study is excessively influential if its "omitted" meta-analytic estimate differs in significance relative to the "combined" analysis. Neither of these quidelines provides definitive proof that such a study should, or should not, be removed from the analysis; they merely provide a suggestion that some attention be paid to potential reasons for its influence.
^metaninf^ uses program ^metan^ as its meta-analysis engine and its options are a direct subset of ^metan^'s. It is expected that users will merely edit a ^met > an^ command line by adding the "inf" suffix to the program name to run ^metaninf^.
Like ^metan^, ^metaninf^ requires either four or six variables to be declared. When four variables are specified, analysis of binary data is performed. When six variables are specified, the data are assumed continuous.
Binary data is presented as the 4 cell counts (a b c d) of a 2x2 table where a and b are the number of event & no-event subjects in the intervention group and c and d are similar numbers for the control group.
. ^metaninf a b c d^
Continuous data is presented as 6 values: the n, mean and sd for the treatment (or intervention) group followed by similar values for the control group.
. ^metaninf nt mt sdt nc mc sdc^
Scaling and pooling options ---------------------------
Options for binary data
^rr^ pools risk ratios (the default).
^or^ pools odds ratios.
^rd^ pools risk differences.
^fixed^ specifies a fixed effect model using the method of Mantel & Haenszel (the default).
^fixedi^ specifies a fixed effect model using the inverse variance method.
^peto^ specifies that Peto's assumption free method is used to pool odds ratios.
^random^ specifies a random effects model using the method of DerSimonian & Laird, with the estimate of heterogeneity being taken from the Mantel-Haenszel model.
^randomi^ specifies a random effects model using the method of DerSimonian & Laird, with the estimate of heterogeneity being taken from the inverse variance fixed effect model.
^cornfield^ computes confidence intervals for odds ratios by the method of Cornfield, rather than the (default) Woolf method.
Options for continuous data
^cohen^ pools standardized mean differences by the method of Cohen (the default).
^hedges^ pools standardized mean differences by the method of Hedges.
^glass^ pools standardized mean differences by the method of Glass.
^nostandard^ pools unstandardized mean differences.
^fixed^ specifies a fixed effect model using the inverse variance method (the default).
^random^ specifies a random effects model using the DerSimonian & Laird method.
General output options ----------------------
^label(^[^namevar^=namevar] [^, yearvar^=yearvar]^)^ labels the data by its name, year or both. However, neither option is required. For the table display the overall length of the label is restricted to 16 characters.
^ilevel()^ specifies the significance level (eg 90,95,99) for the individual trial confidence intervals.
^olevel()^ specifies the significance level (eg 90,95,99) for the overall (pooled) confidence intervals.
^ilevel^ and ^olevel^ need not be the same, and by default are equal to the significance level specified using the ^set level^ command. ^notable^ prevents display of the table of results.
^nograph^ prevents display of the graph. ^save(^varname [^, replace^]^)^ saves the jackknifed estimates.
To run ^metaninf^, the ^metan^ command [STB-44: sbe24] must be installed.
^metaninf^ is a direct rip-off of ^metainf^ (by Aurelio Tobias; see STB-47: sbe > 26; STB-56: sbe26.1). It blatently uses ^metan^ (by Michael J Bradburn, Jonathan J > Deeks and Douglas G Altman; see STB-44: sbe24) as its meta-analysis calculation engine and ^mhplot^, a slight variation of ^hplot^ (by Nicolas Cox), as its graphic engine. The current author is indebted to the above authors for all their hard work, thus allowing this program to be written with no apparent personal intellectual input. Thanks guys!
. ^metaninf a b c d, fixedi or label(namevar=studyid, yearvar=year)^ . ^metaninf nt mt sdt nc mc sdc, cohen notable^
Thomas J Steichen, firstname.lastname@example.org
Also see --------
On-line: help for @metan@, and if installed: @metainf@, @meta@, @metareg@, @me > tacum@, @metabias@, @metatrim@, @metap@, @galbr@, @funnel@, @labbe@, and @hpl > ot@.