Title
extfunnel -- Graphical augmentations to the funnel plot to illustrate potential impact of a new study on a meta-analysis
Syntax
extfunnel varname1 varname2 [if] [in] [, options]
options Description ------------------------------------------------------------------------- fixedi specifies a fixed effect model using the inverse variance method (default) randomi specifies a random effect model using the method of DerSimonian & Laird cpoints(integer) the number of points to evaluate the contours; a greater value produces smoother contour plots null(real) value of the null hypothesis for effect estimate; default is 0 isquared(numlist max=5) I-squared contours tausquared(numlist max=5) tau-squared contours measure(string) defines the target of inference which can be one of lci/uci/ciwidth loe(numlist max=2) defines the limits of clinical equivalence loeline displays the limits of clinical equivalence newstudycontrol(integer) defines the number of patients in the control arm of a new trial newstudytreatment(integer) defines the number of patients in the treatment arm of a new trial or specifies that odds ratios should be used. Valid only when newstudycontrol and newstudytreatment are specified rr specifies that risk ratios should be used. Valid only when newstudycontrol and newstudytreatment are specified xrange(numlist) defines the range of effect estimates yrange(numlist) defines the range of standard errors sumd display summary diamond sumdposition(real) the vertical coordinate where the summary diamond is placed prediction display prediction interval nonullline suppress display of line of no effect nopooledline suppress display of pooled estimate line noshading suppress display of shaded regions noscatter suppress display of scatter of original study effects nometan suppress display of original meta-analysis results using metan label(string) pass label option to metan eform exponentiates the x-axis labels, valid only when input variables are log-transformed, e.g. log odds ratios scheme(grayscale) colour scheme is grayscale (default) scheme(color) colour scheme is color addplot(string) additional twoway plot level(real) statistical significance level; default is 5 twoway_options pass options to the twoway plot -------------------------------------------------------------------------
Description
extfunnel creates graphical overlay augmentations to the funnel plot. The purpose of these overlays is to display the potential impact a new study would have on an existing meta-analysis, providing an indication of the robustness of the meta-analysis to the addition of new evidence. Thus, these extend the use of the funnel plot beyond assessments of publication biases. Two main graphical displays can be produced; i) statistical significance contours, which define regions of the funnel plot in which a new study would have to be located in order to change the statistical significance of the meta-analysis; and ii) heterogeneity contours, which show how a new study would affect the extent of heterogeneity in a given meta-analysis. extfunnel can also illustrate the impact a new study has on lower and upper confidence interval values, the confidence interval width of the pooled meta-analytic result, and overlays for the impact of a future study on, user defined, limits of clinical equivalence. Inverse variance weighted methods are implemented using both explicit formulae for contour lines, and a simulation approach optimised in Mata.
varname1 is assumed to contain normally distributed effect estimates and varname2 is assumed to contain standard errors of varname1.
Options
fixedi specifies a fixed effect model using the inverse variance method. This is the default.
randomi specifies a random effect model using the method of DerSimonian & Laird, with the estimate of heterogeneity being taken from the inverse-variance fixed-effect model.
cpoints(integer) specifies the number of points to evaluate either the shaded statisticial significance contours and/or the heterogeneity contours. The default number for a fixed and random effect meta-analyses are 3500 and 100, respectively. When a random meta-analysis is invoked, the maximum number of contourpoints is 500. A larger number of cpoints results in a smoother graph, but takes longer to compute (see Remarks for more details).
null(real) is the value of the null hypothesis for effect estimate; default is 0. When measure is specified this is the value which lci/uci/ciwidth is compared to. If lci/uci is specified the value of null is compared to lower/upper confidence interval value of the updated meta-analyses, and colour coded depending on whether the updated estimate is less than or more than the null. If ciwidth is specified then the width confidence interval of the updated meta-analyses are compared to the value define by null.
isquared(numlist) Values that define the I-squared contours. Must be a numlist of maximum length 5 and should have elements in the range 0-100.
tausquared(numlist) Values that define the Tau-squared contours. Must be a vector of maximum length 4 and should have elements in the range 0-Inf.
measure(string) defines the target of inference which can be one of lci/uci/ciwidth.
loe(numlist) defines the limits of clinical equivalence. The default legend assumes a beneficial and detrimental effect in specific directions. The legend can be re-labelled using legend(order(1 "text1" 2 "text2"...)). For further details see Sutton et al. (2007).
loeline displays the limits of clinical equivalence.
newstudycontrol(int) defines the number of patients in the control arm of a new trial. newstudytreatment and newstudycontrol defined together produce a statistical significance contour graph, whereby each possible permutation of results is calculated and analysed within the appropriate meta-analysis model. Odds ratios and risk ratios are supported.
newstudytreatment(int) defines the number of patients in the treatment arm of a new trial.
or specifies that log odds ratios should be used, valid only when newstudytreatment and newstudycontrol are specified. This is the default. Alternatively, rr can be specified for risk ratios.
rr specifies that log risk ratios should be used, valid only when newstudytreatment and newstudycontrol are specified.
xrange(numlist) defines the range of effect estimates.
yrange(numlist) defines the range of standard errors.
sumd display the summary diamond.
sumdposition(real) defines the vertical coordinate where the summary diamond is placed.
prediction displays the prediction interval.
nonullline suppresses the display of the vertical line of no effect.
nopooledline suppresses the display of the vertical line at the pooled effect estimate.
noshading suppresses the display of shaded regions.
noscatter suppresses the display of the scatter of original study effects.
nometan suppress display of original meta-analysis results using metan.
label([namevar=namevar], [yearvar=yearvar]) labels the data by its name, year or both. This is a metan option. Either or both option/s may be left blank. For the table display the overall length of the label is restricted to 20 characters.
eform exponentiates the x-axis labels (valid only when the input variables are log transformed, e.g. log odds ratios or log risk ratios).
scheme(string) specifies the color scheme of the graph. Default is grayscale. Can also specify color which can be useful to distinguish areas when loe is specified.
addplot(string) allows additional twoway plots to be overlayed on the extfunnel plot.
level(real) defines the statistical significance level. Default is 95.
twoway_options see [G] twoway_options.
Remarks
extfunnel uses a simulation based process, optimised in Mata, to produce all statistical significance contour graphs, except when fixedi is used. This process can be slightly computationally intensive when cpoints is large (>100). We recommend that initial plots are built with the default cp(100), with final plots being produced using cp(500). When newstudycontrol and newstudytreatment are used, extfunnel can be very computationally intensive as this form of display has not been optimised in Mata.
Return code 900 or macros overflow can occur when cp is large, this can be remedied by reducing cp slightly.
Examples
Fixed effect funnel plot with statistical significance contours. . extfunnel logOR selogOR
Random effect funnel plot with statistical significance contours, with summary diamond and a prediction interval displayed. . extfunnel logOR selogOR, randomi sumd predict cp(500)
Fixed effect funnel plot with statistical significance contours and I-squared contours. . extfunnel logOR selogOR, isquared(25 35 50)
Fixed effect funnel plot with statistical significance contours and tau-squared contours. . extfunnel logOR selogOR, tausquared(0.5 1 1.5)
Random effect funnel plot with user defined limits of clinical equivalence. . extfunnel ES seES, randomi loe(-0.25 0.25) nometan sumd nopooledline loeline cp(400)
Saved results
extfunnel saves the following in r():
Matrices: r(ESmat) Effect estimate values used to create the plot r(seESmat) Standard error values used to create the plot r(status) Status code for each combination of effect and standard error.
r(status) coding: For the stat. sig. plot; coded 0/1/2 i.e. non-sig/stat. sig greater than null/stat. sig. less than null. Coded 1 to 8 for limits of clinical equivalence, corresponding to default legend.
Note: Matrices are not returned under a fixedi statistical significance contour plot.
Authors
Michael J. Crowther, University of Leicester, United Kingdom. michael.crowther@le.ac.uk.
Dean Langan, Clinical Trials Research Unit (CTRU), University of Leeds, United Kingdom. d.p.langan@leeds.ac.uk.
Alex J. Sutton, University of Leicester, United Kingdom. ajs22@le.ac.uk. Thanks to Rob Herbert and Manuela Ferreira for invaluable suggestions for improvement and testing of the command. Please report any errors you may find.
References
Crowther, M. J., Langan, D. and Sutton, A. J. Graphical augmentations to the funnel plot assess the impact of a new study on an existing meta-analysis: The extfunnel command. The Stata Journal 2012; (In Press).
Langan, D., Higgins, J. P. T., Gregory, W. and Sutton, A. J. Graphical augmentations to the funnel plot assess the impact of additional evidence on a meta-analysis. J. Clin. Epidemiology 2012;65(5):511-9.
Palmer, T. M., J. L. Peters, A. J. Sutton, and S. G. Moreno. Contour enhanced funnel plots for meta-analysis. Stata Journal 2008; 8: 242-254.
Peters, J. L., A. J. Sutton, D. R. Jones, K. R. Abrams, and L. Rushton. Contour-enhanced meta-analysis funnel plots help distinguish publication bias from other causes of asymmetry. Journal of Clinical Epidemiology 2008; 61: 991-996.
Sutton AJ, Cooper NJ, Jones DR, Lambert PC, Thompson JR, Abrams KR. Evidence-based sample size calculations based upon meta-analysis. Statistics in Medicine 2007; 26:2479-2500.
Sutton AJ, Donegan S, Takwoingi Y, Garner P, Gamble C, Donald A. An encouraging assessment of methods to inform priorities for updating systematic reviews. J. Clin. Epidemiol. 2009; 62: 241–251.
Sterne, J. A. C., and R. M. Harbord. Funnel plots in meta-analysis. Stata Journal 2004; 4: 127-141.
Also see
Online: metan (if installed)