------------------------------------------------------------------------------- help for meta_lr ------------------------------------------------------------------------------- Graph of positive and negative likelihood ratios in diagnostic test

meta_lr {theta1} {se_theta1} {theta2} {se_theta2} {stratavariables} [if > exp] [inrange] [, stratify combine weighting id(strvar) fix eform ylab graphoption]

Description

meta_lr does mainly two things: it can do stratified meta-analysis of i > ndividual estimates theta1 and theta2 by stratavariables, save the stratified pooled estima > tes and their confidence intervals in new data set and draw them in one graph; > if stratified variables are not given, meta_lr draw theta1 and theta2 and their confidence interval by > individual study in one graph, with or without weighting. Combined meta-analysis results > are shown in star.

This command is mainly used to draw graph in diagnostic test.The user p > rovides the effect estimate as thetas (e.g., log positive likelihood ratio and log negative likelihood ratio) > . Likewise, the user provides the standard error of thetas. Commands meta and metareg must be installed before using this command.

options for stratified analysis

stratify must be given when do the stratified analysis. Meanwhile, at l > east one stratavariable must be given.

fix indicates statified meta-analysis using fixed model. Default is ran > dom effect model.

eform is the same option as in meta. It requests that the output of str > atified meta-analysis is exponentiated. When this option is chosen, option xscale(log) is sug > gested to use. ylab gives labels on y axis. These labels are the strata variables. graphoption the following graph options of STATA 8 can be used in this > command. symbol() msize() mcolor() xscale() xline() xlabel() blpattern() blcolor() ytitle() xtitl > e() scheme()

options for not stratified analysis

Option stratify must not be given.

id(strvar) is a character variable which is used to label the studies. > A string variable must be given. combine provide the combined estimates of theta1 and theta2 with their > confidence intervals. In graph, they are shown in star with two vertical lines to show the combined point es > timates. combine can be chosen together with fix which means that fixed meta-analysis model is used, random model is > the default one. weighting gives individual study weights. If fix is chosen, weights are > reverse of variance of theta1 and theta2. If fix is not chosen, weights are 1/(tau^2 + variance). fix define that combined estimates are obtained from fixed meta-analysi > s and weighting is reverse of variance. Default is that combined estimates are from random effect model and weights are > 1/(tau^2 + variance). graphoption the same as for stratified analysis. Required input variables

theta1 the first effect estimate (for example log negative likelihood > ratio) se_theta1 the corresponding standard error

theta2 the second effect estimate (for example log positive likelihoo > d ratio) se_theta2 the corresponding standard error theta1 and theta2 can be the same effect estimates. This means that str > atified meta-analysis will be done only on one effect estimate and the graph only shows for one es > timate. Hence, this command also can be used in meta-analysis of randomised controlled trails which will show t > he similar graph as meta. However meta_lr provide more options, with or without combine, weighting and mo > re graph options. The disadvantage is that the user needs to type the same theta and its corresponding standa > rd error twice. stratavariable are the stratified variables. They must be discrete or b > inary variables. These variables are re-grouped before drawing the graph. For example, if the user has a variable calle > d test which can be 0, 0.5 or 1. This variable is re-grouped to 1, 2 and 3. This change is made in order to label y ax > is. But the user's data set will not be changed at all. Example

. meta_lr llrpos llrposse llrneg llrnegse test funding, stratify fix ef > orm xscale(log) xline(1) . meta_lr llrpos llrposse llrneg llrnegse, id(author) combine weighting . meta_lr logitsens se_sens logitsens se_sens, id(author) combine Author

Aijing Shang Institute of Social and Preventive Medicine, University Berne, Switzerl > and Email: shang@ispm.unibe.ch Also see

On-line: help for meta (if installed), metan (if installed), metareg (i > f installed)