* "over" functionality for ipdmetan
* In separate program on advice of Patrick Royston
* created by David Fisher, February 2013
* Release notes for versions prior to v4.0: see end of file
* version 4.0 David Fisher 25nov2020
// no changes to code, other than adding check that -metan- v4.0+ is installed; upversioned to match metan/ipdmetan
// some changes to text in the help file
* version 4.01 David Fisher 12feb2021
// no changes to code; upversioned to match with -ipdmetan-
* version 4.02 David Fisher 20apr2021
// no changes to code; upversioned to match with -ipdmetan-
*! version 4.03 David Fisher 12oct2022
// Added "clear" (and undocumented "clearstack") options
// Added prefix() option so that saved variables are `prefix'_ES etc.
// Fixed bug which made "No. pts" column appear in forestplot with r-class commands even if not explicitly specified (displaying all missing values)
// (note however that a column of missing values DOES still appear on-screen, as a prompt to the user.)
program define ipdover, rclass
version 11
local version : di "version " string(_caller()) ":"
* NOTE: mata requires v9.x
* factor variable syntax requires 11.0
* but neither is vital to ipdover
// Check that -metan- v4.0+ is installed
cap metan
if "`r(metan_version)'"=="" {
nois disp as err "This version of {bf:ipdmetan} requires version 4.00 or higher of {bf:metan}"
exit 499
}
else {
local current_version = 4.05
if `r(metan_version)' < `current_version' {
nois disp as err "{bf:metan} version " as res `r(metan_version)' as err " may not the most recent version available"
nois disp as err "Please check, and consider updating {bf:metan}"
}
}
// ipdmetan has two possible syntaxes:
// "generic" effect measure / Syntax 1 ==> ipdmetan [exp_list] .... : [command] [if] [in] ...
// (calculations based on an estimation model fitted within each study)
// "specific" effect measure / Syntax 2 ==> ipdmetan varlist [if] [in] ... **no colon**
// (raw event counts or means (SDs) within each study using some variation on -collapse-)
// We can let ipdmetan.ado sort out the details; we just want to parse certain options first
// (i.e. those that ipdover.ado will need to use after taking back control from ipdmetan.ado)
// 30th Jan 2018
// But we do need to parse on colon just in case
cap _on_colon_parse `0'
if !_rc {
local 0 `"`s(before)'"'
local after `" :`s(after)'"'
}
syntax [anything(everything)], OVER(string) [STUDY(string) BY(string) PREfix(name local) * ]
if `"`study'"' != `""' {
disp as err `"cannot specify {bf:study()} with {bf:ipdover}; please use {bf:over()} or the {bf:ipdmetan} command"'
exit 198
}
if `"`by'"' != `""' {
disp as err `"cannot specify {bf:by()} with {bf:ipdover}; please use {bf:over()} or the {bf:ipdmetan} command"'
exit 198
}
local options2 `"`options'"'
local 0 `over'
syntax varlist [, Missing]
local over1 `varlist'
local overlen : word count `over1'
local missing1 `missing'
* Parse "over" options and map to study() and by() as appropriate
* (N.B. "over" as a named option is NOT passed to ipdmetan!)
local 0 `", `options2'"'
syntax [, OVER(string) * ]
local options2 `"`options'"'
local 0 `over'
syntax [varlist(default=none)] [, Missing]
local over2 `varlist'
local missing2 `missing'
local 0 `", `options2'"'
syntax [, OVER(string) * ]
if `"`over'"'!=`""' {
di as error `"may not specify more than two {bf:over()} options"'
exit 198
}
local nv2 : word count `over2' // no. of vars in `over2'
if `nv2' > 1 {
disp as err `"cannot specify multiple vars to second {bf:over()} option"'
exit 198
}
local study `over1'
if `"`missing1'"'!=`""' local study `"`over1', `missing1'"'
local by `over2'
if `"`missing2'"'!=`""' local by `"`over2', `missing2'"'
** Now run ipdmetan
// with "ipdover" option; this is a marker that data should not be pooled (i.e. is not a MA)
// ipdmetan uses -preserve- before modifying data
// declare up to 6 tempvars to be used if "`cmdstruc'"=="specific"
// ...and up to 4 more if logrank (note that we don't know if either of these is true yet!!)
tempvar tv1 tv2 tv3 tv4 tv5 tv6 tv7 tv8 tv9 tv10
local outvlist `tv1' `tv2' `tv3' `tv4' `tv5' `tv6'
local lrvlist `tv7' `tv8' `tv9' `tv10'
tempfile ipdfile labfile
cap nois ipdmetan `anything', study(`study') by(`by') prefix(`prefix') ///
ipdover(ipdfile(`ipdfile') labfile(`labfile') outvlist(`outvlist') lrvlist(`lrvlist')) `options' `after' // 30th Jan 2018
if _rc {
if `"`err'"'==`""' {
if _rc==1 nois disp as err `"User break in {bf:ipdmetan}"'
nois disp as err `"Error in {bf:ipdmetan}"'
}
exit _rc
}
// re-load dataset created within ipdmetan
preserve
qui use `ipdfile', clear
// 22nd May 2018: need to collect `counts' here; modified by ipdmetan
// collect "universal" returned statistics
local cmdstruc `r(cmdstruc)'
local effect = cond(`"`r(effect)'"'!="", `"`r(effect)'"', "Effect")
local eform `r(eform)'
local citype `r(citype)'
local lcols `"`r(lcols)'"'
local rcols `"`r(rcols)'"'
local wt `r(wt)' // returned separately from ipdmetan.ado rather than sending straight to forestplot
local nonbeta `r(nonbeta)' // Added Mar 2022
local opts_fplot `"`r(opts_fplot)'"'
return local citype `citype'
return scalar n = r(n)
local totnpts = r(n)
// collect returned statistics specific to particular `cmdstruc'
if "`cmdstruc'"=="generic" {
local estexp `r(estexp)'
local invlist `prefix'_ES `prefix'_seES
return local estexp `"`r(estexp)'"'
return local command `"`r(finalcmd)'"'
return local cmdname `"`r(cmdname)'"'
}
else {
local lrvlist `r(lrvlist)'
local invlist `r(invlist)'
local summstat `r(summstat)'
local log `r(log)'
local usummstat = upper("`summstat'")
return local measure `"`log'`usummstat'"'
}
// Collect variable labels
forvalues h=1/`overlen' {
local varlab`h' `"`r(varlab`h')'"'
}
// Now parse options (either originally specified to -ipdover-, or returned by -ipdmetan-
local 0 `", `r(options)'"'
syntax [, noOVerall noSUbgroup SUMMARYONLY noTABle KEEPAll KEEPOrder ///
EFFIcacy DF(varname numeric) LEVEL(passthru) SAVING(string) noGRaph CLEAR CLEARSTACK * ]
local options_ipdm `"`macval(options)'"'
// July 2021:
// Note: -clearstack- is undocumented; it is shorthand for "clear" + "saving(, stacklabel)", but without the actual saving
if `"`clearstack'"'!=`""' local clear clear
// vaccine efficacy: OR and RR only
if `"`efficacy'"'!=`""' {
cap assert inlist("`summstat'", "or", "rr")
if _rc {
nois disp as err "Vaccine efficacy statistics only possible with odds ratios and risk ratios"
exit _rc
}
}
// use locals to avoid referring to `prefix' going forward
// and thereby simplify code [July 2022]
local outvlist `prefix'_ES `prefix'_seES `prefix'_LCI `prefix'_UCI `prefix'_NN // permanent vars, not tempvars
tokenize `outvlist'
args _ES _seES _LCI _UCI _NN
local _USE `prefix'_USE
// markers of existence of _BY and _OVER
// N.B. only _LEVEL is guaranteed to exist.
// _OVER will only exist if `overlen'>1 (i.e. if there is a need to distinguish)
local _BY
cap confirm var `prefix'_BY
if !_rc local _BY `prefix'_BY
local _OVER
cap confirm var `prefix'_OVER
if !_rc local _OVER `prefix'_OVER
local _LEVEL `prefix'_LEVEL
local _LABELS `prefix'_LABELS
cap confirm numeric var `prefix'_WT
if !_rc local _WT `prefix'_WT
** If raw data, more processing is required to obtain _ES, _seES, _LCI and _UCI
// (processes otherwise done by admetan.ado)
if "`cmdstruc'"=="specific" {
// Amended 22nd May 2018
// cc-checking code (and anything else here) moved to GenEffectVars
// (N.B. contents of `counts' option should have been modified by ipdmetan:
// `group1' etc. removed and made into a "string asis" option, albeit possibly empty)
// Parse options for "specific" effect measure syntax (Syntax 2), required later on
local 0 `", `options_ipdm'"'
syntax [, CC(passthru) noCC2 COUNTS(string asis) LOGRank OEV * ]
local options_ipdm `"`macval(options)'"'
cap nois GenEffectVars `_USE' `invlist', outvlist(`outvlist') ///
summstat(`summstat') `logrank' `cc' `cc2' `level' `options_ipdm'
if _rc {
if `"`err'"'==`""' {
if _rc==1 nois disp as err `"User break in {bf:ipdover.GenEffectVars}"'
else nois disp as err `"Error in {bf:ipdover.GenEffectVars}"'
}
exit _rc
}
// We now have _ES and _seES defined throughout.
// Identify excluded studies
qui replace `_USE'=2 if `_USE'==1 & missing(`_ES', `_seES')
} // end if "`cmdstruc'"=="specific"
// Jan 2018
// Now there should not be any options left un-parsed
// But to get syntax to give usual error messages, there needs to be at least one option specified!
// ==> use cmdstruc(), since we know that is always defined
local 0 `", cmdstruc(`cmdstruc') `options_ipdm'"'
syntax , CMDSTRUC(string)
// Create confidence limit variables if necessary
cap confirm numeric variable `_LCI'
if _rc==7 {
disp as err `"variable {bf:`prefix'_LCI} exists and is string"'
exit _rc
}
else if _rc qui gen double `_LCI' = .
cap confirm numeric variable `_UCI'
if _rc==7 {
disp as err `"variable {bf:`prefix'_UCI} exists and is string"'
exit _rc
}
else if _rc qui gen double `_UCI' = .
// Generate confidence limit values if necessary
// v2.1: ifstmt completely recoded to avoid need for "assert" within GenConfInts
// (N.B. only one of _LCI, _UCI needs to be missing for GenConfInts to be run)
cap nois GenConfInts `invlist' if inlist(`_USE', 1, 3, 5) ///
& !missing(`_ES', `_seES') & missing(`_LCI', `_UCI'), ///
citype(`citype') df(`df') `level' outvlist(`outvlist')
if _rc {
if `"`err'"'==`""' {
if _rc==1 nois disp as err `"User break in {bf:ipdover.GenConfInts}"'
else nois disp as err `"Error in {bf:ipdover.GenConfInts}"'
}
exit _rc
}
local level = r(level)
// MOVED FROM WITHIN PREVIOUS IFSTMT 27th June 2017
// Next, identify excluded studies, and remove them if appropriate, and check that at least one valid estimate remains
// (N.B. otherwise, identify them by "_USE==2")
if `"`keeporder'"'!=`""' local keepall keepall // `keeporder' implies `keepall'
if `"`keepall'"'==`""' {
qui drop if `_USE'==2
summ `_ES', meanonly
if !`r(N)' error 2000
}
// otherwise, maintain original order if requested
if `"`keeporder'"'!=`""' {
tempvar tempuse
qui gen byte `tempuse' = `_USE'
qui replace `tempuse' = 1 if `_USE'==2 // keep "insufficient data" studies in original study order (default is to move to end)
}
else local tempuse `_USE'
** Finish off: return stats & matrices; print to screen; saving/forestplot
// need to sort before forming matrix
// missing values in _BY may cause problems, so need to be careful!
summ `_USE', meanonly
if r(max)==5 {
tempvar use5
qui gen byte `use5' = (`_USE'==5) // marker of _USE==5 to sort on *before* _BY (to get around the issue of missing _BY values)
}
local notuse5 = cond("`use5'"=="", "", `"*(!`use5')"')
// return matrix of coefficients
tempname coeffs
sort `use5' `_BY' `_OVER' `tempuse' `_LEVEL'
mkmat `_OVER' `_BY' `_LEVEL' `_ES' `_seES' `_WT' `_NN' if inlist(`_USE', 1, 2), matrix(`coeffs')
return matrix coeffs = `coeffs'
********************************************
* Print summary info and results to screen *
********************************************
* Print number of studies/patients to screen
// (NB nos. actually analysed as opposed to the number supplied in original data)
disp _n _c
if !missing(`totnpts') local dispnpts = string(`totnpts')
else {
local dispnpts "Unknown"
if "`overall'"!="" local textf " (overall estimation not run)"
}
local dispnpts = cond(missing(`totnpts'), "Unknown", string(`totnpts'))
disp as text "Participants included: " as res "`dispnpts'" as text "`textf'"
* Full descriptions of `summstat', `method' and `re_model' options, for printing to screen
// Build up description of effect estimate type
if "`cmdstruc'"=="generic" {
if `"`exp_list'"'!=`""' local disptxt "Trial subgroup analysis of user-specified effect estimate"
else local disptxt "Trial subgroup analysis of main (treatment) effect estimate"
di _n as text "`disptxt'" as res " `estexp'"
}
else if `"`summstat'"'!=`""' {
local logtext = cond(`"`log'"'!=`""', `"`log' "', `""') // add a space if `log'
if "`summstat'"=="rr" local efftext "`logtext'Risk Ratios"
else if "`summstat'"=="irr" local efftext "`logtext'Incidence Rate Ratios"
else if "`summstat'"=="rrr" local efftext "`logtext'Relative Risk Ratios"
else if "`summstat'"=="or" local efftext "`logtext'Odds Ratios"
else if "`summstat'"=="rd" local efftext " Risk Differences"
else if "`summstat'"=="hr" local efftext "`logtext'Hazard Ratios"
else if "`summstat'"=="shr" local efftext "`logtext'Sub-hazard Ratios"
else if "`summstat'"=="tr" local efftext "`logtext'Time Ratios"
else if "`summstat'"=="wmd" local efftext " Weighted Mean Differences"
else if "`summstat'"=="smd" {
local efftext " Standardised Mean Differences"
if "`method'"=="cohen" local efftextf `" as text " by the method of " as res "Cohen""'
else if "`method'"=="glass" local efftextf `" as text " by the method of " as res "Glass""'
else if "`method'"=="hedges" local efftextf `" as text " by the method of " as res "Hedges""'
}
// Study-level effect derivation method
if "`logrank'"!="" local efftext "Peto (logrank) `efftext'"
else if "`method'"=="peto" local efftext "Peto `efftext'"
di _n as text "Trial subgroup analysis of" as res " `efftext'" `efftextf'
}
** Table of results
if `"`table'"'==`""' {
// find maximum length of labels in LHS column
tempvar vlablen
qui gen long `vlablen' = length(`_LABELS')
if `"`_BY'"'!=`""' {
tempvar bylabels
cap decode `_BY' if inlist(`_USE', 1, 2), gen(`bylabels') // if value label
if _rc qui gen `bylabels' = string(`_BY') if inlist(`_USE', 1, 2) // if no value label
qui replace `vlablen' = max(`vlablen', length(`bylabels'))
drop `bylabels'
}
summ `vlablen', meanonly
local lablen=r(max)
drop `vlablen'
forvalues h=1/`overlen' {
local varlabopt `"`varlabopt' varlab`h'(`"`varlab`h''"')"'
local len = length(`"`varlab`h''"')
if `len'>`lablen' local lablen=`len'
}
local stitle = cond(`overlen'>1, "Subgroup", `"`varlab1'"')
if `"`_BY'"'!=`""' {
local byvarlab : variable label `_BY'
local byvarlab = cond(`"`byvarlab'"'!=`""', `"`byvarlab'"', `"`over2'"')
local stitle `"`byvarlab' and `stitle'"'
}
}
cap nois DrawTableIPD, overlen(`overlen') lablen(`lablen') stitle(`stitle') etitle(`effect') prefix(`prefix') ///
`eform' `varlabopt' `table' `overall' `subgroup'
if _rc {
if `"`err'"'==`""' {
if _rc==1 nois disp as err `"User break in {bf:ipdover.DrawTableIPD}"'
else nois disp as err `"Error in {bf:ipdover.DrawTableIPD}"'
}
exit _rc
}
******************************************
* Prepare dataset for graphics or saving *
******************************************
if `"`saving'"'!=`""' | `"`clear'"'!=`""' | `"`graph'"'==`""' {
quietly {
if `"`saving'"'!=`""' {
// Parse `saving' option first
// use modified version of _prefix_saving.ado to handle `stacklabel' option
my_prefix_savingIPD `saving'
local saving `"`s(filename)'"'
local 0 `", `s(options)'"'
syntax [, STACKlabel * ]
local saveopts `"`options'"'
}
if `"`clearstack'"'!=`""' local stacklabel stacklabel // July 2021; see explanation above
// summaryonly (added Sep 2017 for v2.1)
if `"`summaryonly'"'!=`""' qui drop if inlist(`_USE', 1, 2)
// variable name (titles) for "_LABELS" and "_NN"
local labtitle = cond(`"`summaryonly'"'!=`""' & `"`_BY'"'!=`""', `"`byvarlab'"', `"`stitle'"')
if `"`stacklabel'"'==`""' label variable `_LABELS' `"`labtitle'"'
else label variable `_LABELS' // no title if `stacklabel'
if `"`: variable label `_NN''"'==`""' label variable `_NN' "No. pts"
tempvar strlen
gen `strlen' = length(string(`_NN'))
summ `strlen', meanonly
local fmtlen = max(`r(max)', 3) // min of 3, otherwise title ("No. pts") won't fit
format `_NN' %`fmtlen'.0f // right-justified; fixed format (for integers)
drop `strlen'
** Counts and OE/V
if `"`counts'"'!=`""' {
// Titles
local title1 = cond(`"`group2'"'!=`""', `"`group2'"', `"Treatment"')
local title0 = cond(`"`group1'"'!=`""', `"`group1'"', `"Control"')
tokenize `invlist'
local params : word count `invlist'
// Binary data & logrank HR
if inlist(`params', 2, 4) {
if `params'==4 {
args e1 f1 e0 f0
tempvar n1 n0
qui gen long `n1' = `e1' + `f1'
qui gen long `n0' = `e0' + `f0'
}
else {
tokenize `lrvlist'
args n1 n0 e1 e0
}
qui gen str `prefix'_counts1 = string(`e1') + "/" + string(`n1') if inlist(`_USE', 1, 2, 3, 5)
qui gen str `prefix'_counts0 = string(`e0') + "/" + string(`n0') if inlist(`_USE', 1, 2, 3, 5)
label variable `prefix'_counts1 `"`title1' n/N"'
label variable `prefix'_counts0 `"`title0' n/N"'
local countsvl `prefix'_counts1 `prefix'_counts0
}
// N mean SD for continuous data
// counts = "N, mean (SD) in research arm; N, mean (SD) events/total in control arm"
else {
args n1 mean1 sd1 n0 mean0 sd0
qui gen long `prefix'_counts1 = `n1' if inlist(`_USE', 1, 2, 3, 5)
qui gen str `prefix'_counts1msd = string(`mean1', "%7.2f") + " (" + string(`sd1', "%7.2f") + ")" if inlist(`_USE', 1, 2, 3, 5)
label variable `prefix'_counts1 "N"
label variable `prefix'_counts1msd `"`title1' Mean (SD)"'
qui gen long `prefix'_counts0 = `n0' if inlist(`_USE', 1, 2, 3, 5)
qui gen str `prefix'_counts0msd = string(`mean0', "%7.2f") + " (" + string(`sd0', "%7.2f") + ")" if inlist(`_USE', 1, 2, 3, 5)
label variable `prefix'_counts0 "N"
label variable `prefix'_counts0msd `"`title0' Mean (SD)"'
local countsvl `prefix'_counts1 `prefix'_counts1msd `prefix'_counts0 `prefix'_counts0msd
// Find max number of digits in `_counts1', `_counts0'
summ `prefix'_counts1, meanonly
if r(N) {
local fmtlen = floor(log10(`r(max)'))
format `prefix'_counts1 %`fmtlen'.0f
}
summ `prefix'_counts0, meanonly
if r(N) {
local fmtlen = floor(log10(`r(max)'))
format `prefix'_counts0 %`fmtlen'.0f
}
}
compress `countsvl'
local wt nowt // turn off display of weights (i.e npts) if counts
} // end if `"`counts'"'!=`""'
if `"`oev'"'!=`""' {
if "`logrank'"=="" {
disp as err `"Note: {bf:oev} is not applicable without log-rank data and will be ignored"'
local oev
}
else {
tokenize `invlist'
args _OE _V
label variable `_OE' `"O-E(o)"'
label variable `_V' `"V(o)"'
format `_OE' %6.2f
format `_V' %6.2f
if `"`saving'"'!=`""' | `"`clear'"'!=`""' {
qui rename `_OE' `prefix'_OE
qui rename `_V' `prefix'_V
local _OE `prefix'_OE
local _V `prefix'_V
}
}
} // end if "`oev'"!=""
// Oct 2018:
// Now temporarily multiply _USE by 10
// to enable intermediate numberings for sorting the extra rows
qui replace `_USE' = `_USE' * 10
** Insert extra rows for headings, labels, spacings etc.
// Note: in the following routines, "half" values of _USE are used temporarily to get correct order
// and are then replaced with whole numbers at the end
if trim(`"`_BY'`_OVER'"') != `""' {
tempvar expand
* Subgroup headings (_USE==0) and spacings (_USE==4) for "over" (i.e. `overlen'>1)
if `"`summaryonly'"'==`""' {
bysort `_BY' `_OVER' : gen byte `expand' = 1 + 2*(_n==1)`notuse5'
expand `expand'
replace `expand' = !(`expand' - 1) // `expand' is now 0 if expanded and 1 otherwise
sort `_BY' `_OVER' `expand' `_USE' `_LEVEL'
by `_BY' `_OVER' : replace `_USE' = 0 if !`expand' & _n==2 // row for headings
by `_BY' `_OVER' : replace `_USE' = 45 if !`expand' & _n==3 // row for blank line
if `"`_OVER'"'!=`""' {
drop if `_USE'==0 & missing(`_OVER') // ...but not needed for missing _over
}
drop `expand'
// Extra subgroup headings if both "by" *and* "over"
if `"`_BY'"'!=`""' & `"`_OVER'"'!=`""' {
bysort `_BY' : gen byte `expand' = 1 + 3*(_n==1)`notuse5'
expand `expand'
replace `expand' = !(`expand' - 1) // `expand' is now 0 if expanded and 1 otherwise
sort `_BY' `expand' `_USE' `_LEVEL'
by `_BY' : replace `_USE' = -10 if !`expand' & _n==2 // row for "by" label (title)
by `_BY' : replace `_USE' = -5 if !`expand' & _n==3 // row for blank line below title
by `_BY' : replace `_USE' = 41 if !`expand' & _n==4 // row for blank line to separate "by" groups
drop if `_USE'==41 & missing(`_OVER') // ...but not needed for missing _OVER
replace `_OVER'=. if `_USE'==41
drop `expand'
}
}
else {
summ `_BY', meanonly
bysort `_BY' `_OVER' : gen byte `expand' = 1 + (`_BY'==`r(max)')*(_n==_N)`notuse5'
expand `expand'
replace `expand' = !(`expand' - 1) // `expand' is now 0 if expanded and 1 otherwise
sort `_BY' `_OVER' `expand' `_USE' `_LEVEL'
by `_BY' `_OVER' : replace `_USE' = 45 if !`expand' & _n==2 // row for blank line
drop `expand'
}
}
* Blank out effect sizes etc. in `expand'-ed rows
foreach x of varlist `_LABELS' `_ES' `_seES' `_LCI' `_UCI' `_WT' `_NN' `lcols' `rcols' `countsvl' `_OE' `_V' {
cap confirm numeric var `x'
if !_rc replace `x' = . if !inlist(`_USE', 10, 20, 30, 50)
else replace `x' = "" if !inlist(`_USE', 10, 20, 30, 50)
}
replace `_LEVEL' = . if !inlist(`_USE', 10, 20)
** Now insert label info into new rows
// over() labels
if "`_OVER'"!="" {
forvalues h=1/`overlen' {
replace `_LABELS' = `"`varlab`h''"' if `_USE'==0 & `_OVER'==`h'
label define `_OVER' `h' `"`varlab`h''"', add
}
label values `_OVER' `_OVER'
}
// extra row to contain what would otherwise be the leftmost column heading if `stacklabel' specified
// (i.e. so that heading can be used for forestplot stacking)
else if `"`stacklabel'"' != `""' {
local newN = _N + 1
set obs `newN'
replace `_USE' = -10 in `newN'
if "`use5'"=="" {
tempvar use5 // we need `use5' here, regardless of whether it's needed elsewhere
gen byte `use5' = 0
}
replace `use5' = -1 in `newN'
replace `_LABELS' = `"`labtitle'"' in `newN'
}
// "overall" labels
if `"`overall'"'==`""' {
replace `_LABELS' = "Overall" if `_USE'==50
}
// subgroup ("by") headings & labels
if `"`_BY'"'!=`""' {
qui levelsof `_BY' if `_USE'!=50, missing local(bylist) // _USE!=5 since that will always be missing
local bylab : value label `_BY'
foreach byi of local bylist {
// headings
local bylabi : label `bylab' `byi'
if `"`summaryonly'"'==`""' {
if `"`_OVER'"'!=`""' replace `_LABELS' = "`bylabi'" if `_USE' == -10 & `_BY'==`byi'
else replace `_LABELS' = "`bylabi'" if `_USE' == 0 & `_BY'==`byi'
}
// labels
if `"`subgroup'"'==`""' {
local sglabel = cond(`"`summaryonly'"'!=`""', `"`bylabi'"', `"Subgroup"')
replace `_LABELS' = "`sglabel'" if `_USE'==30 & `_BY'==`byi'
}
}
}
** Sort, and tidy up
if `"`keeporder'"'!=`""' {
qui replace `tempuse' = `_USE'
qui replace `tempuse' = 10 if `_USE'==20 // keep "insufficient data" studies in original study order (default is to move to end)
}
sort `use5' `_BY' `_OVER' `tempuse' `_LEVEL'
cap drop `use5'
if `"`keeporder'"'!=`""' qui drop `tempuse'
replace `_USE' = 0 if `_USE' == -10
replace `_USE' = 60 if `_USE' == 41
replace `_USE' = 30 if `_USE' == 35
replace `_USE' = 50 if `_USE' == 55
replace `_USE' = 40 if inlist(`_USE', -5, 25, 45, 55)
replace `_USE' = `_USE' / 10
// Insert vaccine efficacy
if `"`efficacy'"'!=`""' {
qui gen `prefix'_VE = string(100*(1 - exp(`_ES')), "%4.0f") + " (" ///
+ string(100*(1 - exp(`_LCI')), "%4.0f") + ", " ///
+ string(100*(1 - exp(`_UCI')), "%4.0f") + ")" if inlist(`_USE', 1, 3, 5)
label variable `prefix'_VE "Vaccine efficacy (%)"
qui gen `strlen' = length(`prefix'_VE)
summ `strlen', meanonly
format %`r(max)'s `prefix'_VE
qui compress `prefix'_VE
drop `strlen'
local _VE `prefix'_VE
}
// having added "overall", het. info etc., re-format _LABELS using study names only
gen `strlen' = length(`_LABELS')
if `"`summaryonly'"'==`""' {
summ `strlen' if inlist(`_USE', 1, 2), meanonly
}
else summ `strlen', meanonly // alternative if no study estimates (`summaryonly' should be the only reason for this)
format `_LABELS' %-`r(max)'s // left-justified; length equal to longest study name
drop `strlen'
// Oct 2018: Create _EFFECT (was `estText') here rather than in -forestplot-
// so that e.g. user can insert p-values into results set
if `"`saving'"'!=`""' | `"`clear'"'!=`""' {
// need to peek into `opts_fplot' to extract `dp'
local 0 `", `opts_fplot'"'
syntax [, DP(integer 2) * ]
if `"`eform'"'!=`""' local xexp exp
summ `_UCI', meanonly
local fmtx = max(1, ceil(log10(abs(`xexp'(r(max)))))) + 1 + `dp'
local _EFFECT `prefix'_EFFECT
gen str `_EFFECT' = string(`xexp'(`_ES'), `"%`fmtx'.`dp'f"') if !missing(`_ES')
replace `_EFFECT' = `_EFFECT' + " " if !missing(`_EFFECT')
replace `_EFFECT' = `_EFFECT' + "(" + string(`xexp'(`_LCI'), `"%`fmtx'.`dp'f"') + ", " + string(`xexp'(`_UCI'), `"%`fmtx'.`dp'f"') + ")"
replace `_EFFECT' = `""' if !inlist(`_USE', 1, 3, 5)
replace `_EFFECT' = "(Insufficient data)" if `_USE' == 2
local f: format `_EFFECT'
tokenize `"`f'"', parse("%s")
confirm number `2'
format `_EFFECT' %-`2's // left-justify
label variable `_EFFECT' `"`effect' (`level'% CI)"'
}
} // end quietly
** Save _dta characteristic containing all the options passed to -forestplot-
// so that they may be called automatically using "forestplot, useopts"
// (N.B. _USE, _LABELS and _NN should always exist)
// Modified Mar 2022
if `"`nonbeta'"'==`""' local wgtvar `_NN'
if `"`_WT'"'!=`""' local wgtvar `_WT' // user-defined weights, if supplied
if `"`wgtvar'"'!=`""' { // -ifstmt- is in case of `nonbeta' (see -ipdmetan- )
local wgtopt wgt(`wgtvar')
if `"`wgtvar'"'==`"`_NN'"' local _NNopt `_NN' // include _NN in lcols if wgtvar, but *don't* include user-spec _WT in rcols
}
// end of modified section
local useopts `"use(`_USE') labels(`_LABELS') `wgtopt' nowt `eform' effect(`effect') `keepall' `opts_fplot'"'
if `"`_BY'"'!=`""' local useopts `"`macval(useopts)' by(`_BY')"'
if trim(`"`lcols' `_NNopt' `countsvl' `_OE' `_V'"')!=`""' {
local useopts `"`macval(useopts)' lcols(`lcols' `_NNopt' `countsvl' `_OE' `_V')"'
}
if trim(`"`_VE' `_WT' `rcols'"')!=`""' {
local useopts `"`macval(useopts)' rcols(`_VE' `_WT' `rcols')"'
}
if `"`_WT'"'!=`""' local fpnote `"NOTE: Weights are user-defined"'
else if `"`_NNopt'"'!=`""' local fpnote `"NOTE: Weighting is by sample size"'
if `"`fpnote'"'!=`""' {
local useopts = trim(itrim(`"`macval(useopts)' note(`fpnote')"'))
}
// Store data characteristics
// June 2016: in future, could maybe add other warnings here, e.g. continuity correction??
// NOTE: Only relevant if `saving' (but setup anyway; no harm done)
char define _dta[FPUseOpts] `"`useopts'"'
char define _dta[FPUseVarlist] `_ES' `_LCI' `_UCI'
** Pass to forestplot
if `"`graph'"'==`""' {
tempvar touse
qui gen byte `touse' = 1
// Modified Jan 30th 2018; `wgtopt' is defined earlier
// N.B. if either `counts' or user-defined weights or no _NN, "nowt" is enforced.
// N.B.2. wgt() specifies the weighting variable; nowt suppresses its *display*
cap nois forestplot `_ES' `_LCI' `_UCI', `useopts'
if _rc {
if `"`err'"'==`""' {
if _rc==1 nois disp as err `"User break in {bf:forestplot}"'
else nois disp as err `"Error in {bf:forestplot}"'
}
exit _rc
}
return add
}
** Finally, save dataset
if `"`saving'"'!=`""' | `"`clear'"'!=`""' {
keep `_USE' `_BY' `_OVER' `_LEVEL' `_LABELS' `_ES' `_seES' `_LCI' `_UCI' `_WT' `_NN' `_EFFECT' `lcols' `rcols' `countsvl' `_OE' `_V' `_VE'
order `_USE' `_BY' `_OVER' `_LEVEL' `_LABELS' `_ES' `_seES' `_LCI' `_UCI' `_WT' `_NN' `_EFFECT' `lcols' `rcols' `countsvl' `_OE' `_V' `_VE'
// Oct 2018: Variable labels
if inlist("`summstat'", "or", "rr", "hr") {
label variable `_ES' "Effect size (interval scale)"
}
else label variable `_ES' "Effect size"
label variable `_seES' "Standard error of effect size"
label variable `_LCI' "`level'% lower confidence limit"
label variable `_UCI' "`level'% upper confidence limit"
char define `_LCI'[Level] `level'
char define `_UCI'[Level] `level'
label variable `_NN' "No. pts"
label data `"Results set created by ipdover"'
qui compress
if `"`saving'"'!=`""' {
qui save `"`saving'"', `saveopts'
}
if `"`clear'"'!=`""' {
restore, not
}
}
} // end if `"`saving'"'!=`""' | `"`graph'"'==`""'
end
********************************************
* Program to generate effect size variables
// Based on ProcessInputVarlist and ProcessPoolingVarlist in admetan.ado
// but this version is only for use with raw data in ipdover.ado.
program define GenEffectVars, rclass
syntax varlist(min=3 max=7 default=none), OUTVLIST(namelist min=5 max=8) ///
[SUMMSTAT(string) noINTEGER CC(string) CORnfield LOGRank LEVEL(real 95) ZTOL(real 1e-6)]
// unpack varlists
tokenize `outvlist'
args _ES _seES _LCI _UCI _NN
gettoken _USE invlist : varlist
tokenize `invlist'
local params : word count `invlist'
// generate effect size vars
// Note [Oct 2015]: gen as tempvars for now (to accommodate inverse-variance)
// but will be renamed to permanent variables later if appropriate
// 2 or 3 vars inverse-variance, or logrank HR
// (N.B. summstat may be missing for the former only)
if `params' <= 3 {
foreach opt in cc cornfield {
cap assert `"``opt''"' == `""'
if _rc {
nois disp as err `"Note: Option {bf:`opt'} is not appropriate without 2x2 count data and will be ignored"'
local switchoff `"`switchoff' `opt'"'
}
}
if "`logrank'"!="" {
args oe va
qui replace `_USE' = 2 if `_USE'==1 & sqrt(`va') < `ztol' // insufficient data (`_USE'==2)
qui gen double `_ES' = `oe'/`va' if inlist(`_USE', 1, 3, 5) // logHR
qui gen double `_seES' = 1/sqrt(`va') if inlist(`_USE', 1, 3, 5) // selogHR
}
// Identify studies with insufficient data (`_USE'==2)
else if "`3'"=="" { // input is ES + SE
args _ES _seES
qui replace `_USE' = 2 if `_USE'==1 & missing(`_ES', `_seES')
qui replace `_USE' = 2 if `_USE'==1 & 1/`_seES' < `ztol'
}
else { // input is ES + CI
args _ES _LCI _UCI
qui replace `_USE' = 2 if `_USE'==1 & missing(`_LCI', `_UCI')
qui replace `_USE' = 2 if `_USE'==1 & float(`_LCI')==float(`_UCI')
cap assert `_UCI'>=`_ES' & `_ES'>=`_LCI' if `_USE'==1
if _rc {
nois disp as err "Effect size and/or confidence interval limits invalid;"
nois disp as err `"order should be {it:effect_size} {it:lower_limit} {it:upper_limit}"'
exit _rc
}
// Need to generate _seES
qui gen double `_seES' = (`_LCI' - `_UCI') / (2*invnormal(.5 + `level'/200)) if inlist(`_USE', 1, 3, 5)
qui replace `_USE' = 2 if `_USE'==1 & float(`_seES')==0
}
qui count if inlist(`_USE', 1, 3, 5) & !missing(`_ES', `_seES')
if !r(N) exit 2000
}
// setup for Peto OR method
else if "`method'"=="peto" {
assert `params' == 4
tempvar oe
local a `e1'
qui gen double `oe' = `a' - `ea' // N.B. `ea' was created earlier (as was `va')
qui gen double `_ES' = `oe'/`va' if inlist(`_USE', 1, 3, 5) // logOR or logHR
qui gen double `_seES' = 1/sqrt(`va') if inlist(`_USE', 1, 3, 5) // selogOR or selogHR
if `"`cc'"'!=`""' {
nois disp as err "Note: continuity correction is incompatible with Peto method and will be ignored"
local switchoff `switchoff' cc
}
}
// Binary outcome (OR, RR, RD)
else if `params' == 4 {
assert inlist("`summstat'", "or", "rr", "irr", "rrr", "rd")
args e1 f1 e0 f0 // events & non-events in trt; events & non-events in control (a.k.a. a b c d)
if "`integer'"=="" {
cap {
assert int(`e1')==`e1'
assert int(`f1')==`f1'
assert int(`e0')==`e0'
assert int(`f0')==`f0'
}
if _rc {
di as err "Non integer cell counts found"
exit _rc
}
}
cap assert `e1'>=0 & `f1'>=0 & `e0'>=0 & `f0'>=0
if _rc {
di as err "Non-positive cell counts found"
exit _rc
}
// Find studies with insufficient data (`_USE'==2)
qui replace `_USE' = 2 if `_USE'==1 & missing(`e1', `f1', `e0', `f0')
if "`summstat'"=="or" qui replace `_USE' = 2 if `_USE'==1 & (`e1' + `e0')*(`f1' + `f0')==0
if inlist("`summstat'", "rr", "irr", "rrr") | "`method'"=="peto" {
qui replace `_USE' = 2 if `_USE'==1 & ((`e1'==0 & `e0'==0 ) | (`f1'==0 & `f0'==0))
}
qui replace `_USE' = 2 if `_USE'==1 & (`e1' + `f1')*(`e0' + `f0')==0 // applies to all cases
qui count if inlist(`_USE', 1, 3, 5)
if !r(N) exit 2000
if "`cornfield'"!="" {
if !inlist("`summstat'", "or", "") {
nois disp as err "Note: {bf:cornfield} is only compatible with odds ratios; option will be ignored"'
local switchoff `"`switchoff' cornfield"'
}
else if "`summstat'"=="" {
nois disp as err `"Note: Cornfield-type confidence intervals specified; odds ratios assumed"'
local summstat "or"
}
}
if inlist("`method'", "cohen", "glass", "hedges", "nostandard") {
nois disp as err `"Specified method {bf:`method'} is incompatible with the data"'
exit 184
}
if inlist("`summstat'", "hr", "shr", "tr") {
nois disp as err "Time-to-event outcome types are incompatible with count data"
exit 184
}
if inlist("`summstat'", "wmd", "smd") {
nois disp as err "Continuous outcome types are incompatible with count data"
exit 184
}
/*
if "`summstat'"=="" {
local summstat rr
local effect `"Risk Ratio"'
}
*/
assert "`summstat'"!="" // temp error trap 23rd March 2017
local method = cond("`method'"=="", "mh", "`method'") // default pooling method is Mantel-Haenszel
// 27th March 2017
// tokenize `binvlist'
// args r1 r0
tempvar r1 r0
local type = cond("`integer'"=="", "long", "double")
qui gen `type' `r1' = `e1' + `f1' // total in trt arm
qui gen `type' `r0' = `e0' + `f0' // total in control arm
// qui gen `type' `_NN' = `r1' + `r0' // overall total
// zero-cell adjustments: amended May 2018
local ccval = cond(`"`cc2'"'!=`""', 0, 0.5) // default
if trim(`"`cc'`cc2'"') != `""' {
if `"`cc'"'!=`""' {
local 0 `"`cc'"'
syntax [anything(id="value supplied to {bf:cc()}")] [, OPPosite EMPirical]
if `"`anything'"'!=`""' {
confirm number `anything'
local ccval = `anything'
}
if `"`cc2'"'!=`""' & `ccval' != 0 {
disp as err `"Cannot specify both {bf:cc()} and {bf:nocc}; please choose one or the other"'
exit 198
}
cap assert `ccval'>=0 & `ccval'<1
if _rc {
nois disp as err "Invalid continuity correction: must be in range [0,1)"
exit _rc
}
}
}
local cc = `ccval'
tempvar zeros
qui gen byte `zeros' = `e1'*`f1'*`e0'*`f0'==0
summ `zeros' if `_USE'==1, meanonly
local nz = r(N)
if `nz' & `cc' {
// Sweeting's "opposite treatment arm" correction
if `"`opposite'"'!=`""' {
tempvar cc1 cc0
qui gen `cc1' = 2*`cc'*`r1'/(`r1' + `r0')
qui gen `cc0' = 2*`cc'*`r0'/(`r1' + `r0')
}
// Empirical correction: not valid with ipdover
else if `"`empirical'"'!=`""' {
nois disp as err `"Empirical continuity correction not valid with {bf:ipdover}"'
exit 198
}
else {
local cc1 = `cc'
local cc0 = `cc'
}
tempvar e1_cont f1_cont e0_cont f0_cont t_cont
qui gen double `e1_cont' = cond(`zeros', `e1' + `cc1', `e1')
qui gen double `f1_cont' = cond(`zeros', `f1' + `cc1', `f1')
qui gen double `e0_cont' = cond(`zeros', `e0' + `cc0', `e0')
qui gen double `f0_cont' = cond(`zeros', `f0' + `cc0', `f0')
tempvar r1_cont r0_cont t_cont
qui gen double `r1_cont' = `e1_cont' + `f1_cont'
qui gen double `r0_cont' = `e0_cont' + `f0_cont'
qui gen double `t_cont' = `r1_cont' + `r0_cont'
if `"`opposite'"' != `""' {
drop `cc1' `cc0' // tidy up
}
}
else {
local e1_cont `e1'
local f1_cont `f1'
local e0_cont `e0'
local f0_cont `f0'
local r1_cont `r1'
local r0_cont `r0'
local t_cont `_NN'
}
// now branch by outcome measure
if "`summstat'" == "or" {
if "`method'"=="peto" {
local a `e1'
tempvar c1 c0 ae oe va
qui gen `type' `c1' = `a' + `c' // total events
qui gen `type' `c0' = `b' + `d' // total non-events
qui gen double `ea' = (`r1'*`c1')/ `_NN' // expected events in trt arm
qui gen double `va' = `r1'*`r0'*`c1'*`c0'/( `_NN'*`_NN'*(`_NN' - 1))
qui gen double `oe' = `a' - `ea'
qui gen double `_ES' = `oe'/`va' if inlist(`_USE', 1, 3, 5) // logOR or logHR
qui gen double `_seES' = 1/sqrt(`va') if inlist(`_USE', 1, 3, 5) // selogOR or selogHR
}
else {
// calculate individual ORs and variances using cc-adjusted counts
// (on the linear scale, i.e. logOR)
qui gen double `_ES' = ln(`e1_cont'*`f0_cont') - ln(`f1_cont'*`e0_cont') if inlist(`_USE', 1, 3, 5)
qui gen double `_seES' = sqrt(1/`e1_cont' + 1/`f1_cont' + 1/`e0_cont' + 1/`f0_cont') if inlist(`_USE', 1, 3, 5)
}
} /* end OR */
// setup for RR
else if "`summstat'" == "rr" {
tempvar r s v
qui gen double `r' = `e1_cont'*`r0_cont' / `t_cont'
qui gen double `s' = `e0_cont'*`r1_cont' / `t_cont'
qui gen double `v' = 1/`e1_cont' + 1/`e0_cont' - 1/`r1_cont' - 1/`r0_cont'
qui gen double `_ES' = ln(`r'/`s') if inlist(`_USE', 1, 3, 5) // logRR
qui gen double `_seES' = sqrt(`v') if inlist(`_USE', 1, 3, 5) // selogRR
}
// setup for RD
else if "`summstat'" == "rd" {
tempvar v
qui gen double `v' = `e1_cont'*`f1_cont'/(`r1_cont'^3) + `e0_cont'*`f0_cont'/(`r0_cont'^3)
qui gen double `_ES' = `e1'/`r1' - `e0'/`r0' if inlist(`_USE', 1, 3, 5)
qui gen double `_seES' = sqrt(`v') if inlist(`_USE', 1, 3, 5)
}
} /* end if `params' == 4 */
// N mean SD for continuous data
else {
assert `params' == 6
assert inlist("`summstat'", "wmd", "smd")
args n1 mean1 sd1 n0 mean0 sd0
// input is form N mean SD for continuous data
if "`integer'"=="" {
cap assert int(`n1')==`n1' & int(`n0')==`n0'
if _rc {
nois disp as err "Non integer sample sizes found"
exit _rc
}
}
cap assert `n1'>0 & `n0'>0
if _rc {
nois disp as err "Non positive sample sizes found"
exit _rc
}
foreach opt in cc cornfield {
cap assert `"``opt''"' == `""'
if _rc {
nois disp as err `"Option {bf:`opt'} is not appropriate without 2x2 count data and will be ignored"'
local switchoff `"`switchoff' `opt'"'
}
}
// Find studies with insufficient data (`_USE'==2)
qui replace `_USE' = 2 if `_USE'==1 & missing(`n1', `mean1', `sd1', `n0', `mean0', `sd0')
qui replace `_USE' = 2 if `_USE'==1 & `n1' < 2 | `n0' < 2
qui replace `_USE' = 2 if `_USE'==1 & `sd1'<=0 | `sd0'<=0
qui count if inlist(`_USE', 1, 3, 5)
if !r(N) exit 2000
if "`method'"=="nostandard" & "`summstat'"=="smd" {
nois disp as err `"Cannot specify both SMD and the {bf:nostandard} option"'
exit 184
}
if inlist("`method'", "cohen", "glass", "hedges") & "`summstat'"=="wmd" {
nois disp as err `"Cannot specify both WMD and the {bf:`mdmethod'} option"'
exit 184
}
if inlist("`method'", "mh", "peto") | "`logrank'"!="" {
nois disp as err `"Specified method {bf:`method'} is incompatible with the data"'
exit 184
}
cap assert inlist("`summstat'", "", "wmd", "smd")
if _rc {
nois disp as err "Invalid specifications for combining trials"
exit 184
}
/*
if "`summstat'"=="" {
if "`method'"=="nostandard" { // "nostandard" is a synonym for "wmd"
local summstat "wmd"
local effect `"WMD"'
}
else {
local summstat "smd" // default is standardized mean differences...
local effect `"SMD"'
}
}
*/
assert "`summstat'"!="" // temp error trap 23rd March 2017
local method = cond(inlist("`method'", "", "iv"), "cohen", "`method'") // ...by the method of Cohen
// qui gen long `_NN' = `n1' + `n0' if inlist(`_USE', 1, 2, 3, 5)
if "`summstat'" == "wmd" {
qui gen double `_ES' = `mean1' - `mean0' if inlist(`_USE', 1, 3, 5)
qui gen double `_seES' = sqrt((`sd1'^2)/`n1' + (`sd0'^2)/`n0') if inlist(`_USE', 1, 3, 5)
}
else { // summstat = SMD
tempvar s
qui gen double `s' = sqrt( ((`n1'-1)*(`sd1'^2) + (`n0'-1)*(`sd0'^2) )/( `_NN' - 2) )
if "`mdmethod'" == "cohen" {
qui gen double `_ES' = (`mean1' - `mean0')/`s' if inlist(`_USE', 1, 3, 5)
qui gen double `_seES' = sqrt((`_NN' /(`n1'*`n0')) + (`_ES'*`_ES'/ (2*(`_NN' - 2)) )) if inlist(`_USE', 1, 3, 5)
}
else if "`mdmethod'" == "glass" {
qui gen double `_ES' = (`mean1' - `mean0')/`sd0' if inlist(`_USE', 1, 3, 5)
qui gen double `_seES' = sqrt(( `_NN' /(`n1'*`n0')) + (`_ES'*`_ES'/ (2*(`n0' - 1)) )) if inlist(`_USE', 1, 3, 5)
}
else if "`mdmethod'" == "hedges" {
qui gen double `_ES' = ((`mean1' - `mean0')*(1 - 3/(4*`_NN' - 9))/`s' if inlist(`_USE', 1, 3, 5)
qui gen double `_seES' = sqrt(( `_NN' /(`n1'*`n0')) + (`_ES'*`_ES'/ (2*(`_NN' - 3.94)) )) if inlist(`_USE', 1, 3, 5)
}
drop `s'
}
} /* end else (i.e. if `params' == 6) */
end
*********************************************************************
* Program to generate study-level confidence intervals
// identical subroutine also used in admetan.ado
program define GenConfInts, rclass
syntax varlist(numeric min=2 max=6 default=none) [if] [in], CItype(string) ///
OUTVLIST(varlist numeric) [ DF(varname numeric) LEVEL(real 95) ]
marksample touse, novarlist
// if no data to process, exit without error
return scalar level = `level'
qui count if `touse'
if !r(N) exit
// unpack varlists
tokenize `outvlist'
args _ES _seES _LCI _UCI _WT _NN
local params : word count `varlist' // `varlist' == `invlist'
// Confidence limits need calculating if:
// - not supplied by user (i.e. `params'!=3); or
// - desired coverage is not 95%
if `params'==3 & `level'==95 exit
* Calculate confidence limits for original study estimates using specified `citype'
// (unless limits supplied by user)
if "`citype'"=="normal" { // normal distribution - default
tempname critval
scalar `critval' = invnormal(.5 + `level'/200)
qui replace `_LCI' = `_ES' - `critval'*`_seES' if `touse'
qui replace `_UCI' = `_ES' + `critval'*`_seES' if `touse'
}
else if inlist("`citype'", "t", "logit") { // t or logit distribution
cap confirm numeric variable `df'
if !_rc {
summ `df' if `touse', meanonly // use supplied df if available
cap assert r(max) < .
if _rc {
nois disp as err `"Degrees-of-freedom variable {bf:`df'} contains missing values;"'
nois disp as err `" cannot use {bf:`citype'}-based confidence intervals for study estimates"'
exit 198
}
}
else {
cap confirm numeric variable `_NN'
if !_rc {
summ `_NN' if `touse', meanonly // otherwise try using npts
cap assert r(max) < .
if _rc {
nois disp as err `"Participant numbers not available for all studies;"'
nois disp as err `" cannot use {bf:`citype'}-based confidence intervals for study estimates"'
exit 198
}
tempvar df
qui gen `: type `_NN'' `df' = `_NN' - 2 // use npts-2 as df for t distribution of df not explicitly given
local disperr `"nois disp as err `"Note: Degrees of freedom for {bf:`citype'}-based confidence intervals not supplied; using {it:n-2} as default"'"'
// delay error message until after checking _ES is between 0 and 1 for logit
}
else {
nois disp as err `"Neither degrees-of-freedom nor participant numbers available;"'
nois disp as err `" cannot use {bf:`citype'}-based confidence intervals for study estimates"'
exit 198
}
}
tempvar critval
qui gen double `critval' = invttail(`df', .5 - `level'/200)
if "`citype'"=="t" {
qui replace `_LCI' = `_ES' - `critval'*`_seES' if `touse'
qui replace `_UCI' = `_ES' + `critval'*`_seES' if `touse'
}
else { // logit, proportions only (for formula, see Stata manual for -proportion-)
summ `_ES' if `touse', meanonly
if r(min)<0 | r(max)>1 {
nois disp as err "{bf:citype(logit)} may only be used with proportions"
exit 198
}
qui replace `_LCI' = invlogit(logit(`_ES') - `critval'*`_seES'/(`_ES'*(1 - `_ES'))) if `touse'
qui replace `_UCI' = invlogit(logit(`_ES') + `critval'*`_seES'/(`_ES'*(1 - `_ES'))) if `touse'
}
}
else if inlist("`citype'", "cornfield", "exact", "woolf") { // options to pass to -cci-; summstat==OR only
tokenize `varlist'
args a b c d // events & non-events in trt; events & non-events in control (c.f. -metan- help file)
// sort appropriately, then find observation number of first relevant obs
tempvar obs
qui bysort `touse' : gen long `obs' = _n if `touse' // N.B. MetaAnalysisLoop uses -sortpreserve-
sort `obs' // so this sorting should not affect the original data
summ `obs' if `touse', meanonly
forvalues j = 1/`r(max)' {
qui cci `=`a'[`j']' `=`b'[`j']' `=`c'[`j']' `=`d'[`j']', `citype' level(`level')
qui replace `_LCI' = log(`r(lb_or)') in `j'
qui replace `_UCI' = log(`r(ub_or)') in `j'
}
}
// Now display delayed error message if appropriate
`disperr'
end
*********************************************************************
* Routine to draw output table (ipdover.ado version)
* Could be done using "tabdisp", but doing it myself means it can be tailored to the situation
* therefore looks better (I hope!)
// DF Aug 2016:
// N.B. Code taken directly from latest version of DrawTableAD (as used in admetan.ado), but then modified for use with -ipdover-
// (e.g. "tests and heterogeneity" subroutine is removed, and other small additions/removals)
program define DrawTableIPD
syntax, OVERLEN(integer) ///
[PREfix(name local) LABLEN(integer 0) STITLE(string asis) ETITLE(string asis) ///
EFORM noTABle noOVerall noSUbgroup * ]
tempvar obs
qui gen long `obs'=_n
sort `obs'
// EXTRA LINES FOR USE WITH IPDOVER
cap confirm var `prefix'_BY
if !_rc local _BY `prefix'_BY
cap confirm var `prefix'_NN
if !_rc local _NN `prefix'_NN
cap confirm var `prefix'_OVER
if !_rc local _OVER `prefix'_OVER
foreach x in _USE _LABELS _ES _LCI _UCI {
local `x' `prefix'`x'
}
local swidth = 25 // define `swidth' in case noTAB
if `"`table'"'==`""' {
// EXTRA LINES FOR USE WITH IPDOVER
if `overlen'>1 { // if "over", parse "variable label" options
forvalues h=1/`overlen' {
local 0 `", `options'"'
syntax, VARLAB`h'(string) *
}
}
* Find maximum length of study title and effect title
* Allow them to spread over several lines, but only up to a maximum number of chars
* If a single line must be more than 32 chars, truncate and stop
local uselen = 25 // default (minimum); max is 32
if `lablen'>21 local uselen=min(`lablen', 32)
cap nois SpreadTitle `"`stitle'"', target(`uselen') maxwidth(32) // study (+ subgroup) title
if _rc {
if _rc==1 nois disp as err `"User break in {bf:ipdover.SpreadTitle}"'
nois disp as err `"Error in {bf:ipdover.SpreadTitle}"'
c_local err "noerr" // tell ipdover not to also report an "error in {bf:ipdover.DrawTableIPD}"
exit _rc
}
local swidth = max(`uselen', `r(maxwidth)')
local slines = r(nlines)
forvalues i=1/`slines' {
local stitle`i' `"`r(title`i')'"'
}
cap nois SpreadTitle `"`etitle'"', target(10) maxwidth(15) // effect title (i.e. "Odds ratio" etc.)
if _rc {
if _rc==1 nois disp as err `"User break in {bf:ipdover.SpreadTitle}"'
nois disp as err `"Error in {bf:ipdover.SpreadTitle}"'
c_local err "noerr" // tell ipdover not to also report an "error in {bf:ipdover.DrawTableIPD}"
exit _rc
}
local ewidth = max(10, `r(maxwidth)')
local elines = r(nlines)
local diff = `elines' - `slines'
if `diff'<=0 {
forvalues i=1/`slines' {
local etitle`i' `"`r(title`=`i'+`diff'')'"' // if stitle uses more (or equal) lines ==> line up etitle with stitle
}
}
else {
forvalues i=`elines'(-1)1 { // run backwards, otherwise macros are deleted by the time they're needed
local etitle`i' `"`r(title`i')'"'
local stitle`i' = cond(`i'>=`diff', `"`stitle`=`i'-`diff'''"', `""') // if etitle uses more lines ==> line up stitle with etitle
}
}
* Now display the title lines, starting with the "extra" lines and ending with the row including CI & weight
di as text _n `"{hline `swidth'}{c TT}{hline `=`ewidth'+35'}"'
local nl = max(`elines', `slines')
if `nl' > 1 {
forvalues i=1/`=`nl'-1' {
di as text `"`stitle`i''{col `=`swidth'+1'}{c |} "' %~`ewidth's `"`etitle`i''"'
}
}
di as text `"`stitle`nl''{col `=`swidth'+1'}{c |} "' %~10s `"`etitle`nl''"' `"{col `=`swidth'+`ewidth'+4'}[`c(level)'% Conf. Interval]{col `=`swidth'+`ewidth'+27'}No. pts"'
*** Loop over studies, and subgroups if appropriate
if `"`_BY'"'!=`""' {
qui levelsof `_BY' if `_USE'!=5, missing local(bylist) // _USE!=5 since that will always be missing!
local bylab : value label `_BY'
}
local nby = max(1, `: word count `bylist'')
tempvar touse touse2
qui gen byte `touse' = 1
tempname _ES_ _LCI_ _UCI_
local xexp = cond("`eform'"!=`""', `"exp"', `""')
forvalues i=1/`nby' { // this will be 1/1 if no subgroups
di as text `"{hline `swidth'}{c +}{hline `=`ewidth'+35'}"'
if `"`_BY'"'!=`""' {
local byi : word `i' of `bylist'
qui replace `touse' = (`_BY'==`byi')
summ `_ES' if `touse' & `_USE'==1, meanonly
if !r(N) local nodata `"{col `=`swidth'+4'} (No subgroup data)"'
else local K`i' = r(N)
if `"`bylab'"'!=`""' {
local bylabi : label `bylab' `byi'
}
else local bylabi `"`byi'"'
di as text substr(`"`bylabi'"', 1, `swidth'-1) + `"{col `=`swidth'+1'}{c |}`nodata'"'
local nodata // clear macro
}
// EXTRA LINES FOR USE WITH IPDOVER
forvalues h=1/`overlen' {
gen byte `touse2' = `touse'
if `"`_OVER'"'!=`""' {
qui replace `touse2' = `touse' * (`_OVER'==`h')
}
summ `obs' if `touse2' & inlist(`_USE', 1, 2), meanonly
// EXTRA LINES FOR USE WITH IPDOVER
if `overlen'>1 {
di as text `"{col `=`swidth'+1'}{c |}"'
if !r(N) local nodata `"{col `=`swidth'+4'} (Insufficient data)"'
di as text substr(`"`varlab`h''"', 1, `=`swidth'-1') `"{col `=`swidth'+1'}{c |}`nodata'"'
local nodata // clear macro
}
if r(N) {
forvalues k = `r(min)' / `r(max)' {
local _labels_ = `_LABELS'[`k']
summ `obs' if `touse2' & `_USE'==1 & _n==`k', meanonly // 30th March 2017: can this be improved/made more efficient?
if !r(N) {
di as text substr(`"`_labels_'"', 1, 32) `"{col `=`swidth'+1'}{c |}{col `=`swidth'+4'} (Insufficient data)"'
}
else {
scalar `_ES_' = `_ES'[`k']
scalar `_LCI_' = `_LCI'[`k']
scalar `_UCI_' = `_UCI'[`k']
local _labels_ = `_LABELS'[`k']
di as text substr(`"`_labels_'"', 1, 32) `"{col `=`swidth'+1'}{c |}{col `=`swidth'+`ewidth'-6'}"' ///
as res %7.3f `xexp'(`_ES_') `"{col `=`swidth'+`ewidth'+5'}"' ///
as res %7.3f `xexp'(`_LCI_') `"{col `=`swidth'+`ewidth'+15'}"' ///
as res %7.3f `xexp'(`_UCI_') `"{col `=`swidth'+`ewidth'+26'}"' %7.0f `_NN'[`k']
}
}
}
drop `touse2'
} // end forvalues j=1/`overlen'
* Subgroup effects
if `"`_BY'"'!=`""' & `"`subgroup'"'==`""' {
di as text `"{col `=`swidth'+1'}{c |}"'
local byi: word `i' of `bylist'
summ `obs' if `touse' & `_USE'==3, meanonly
if !r(N) {
di as text `"Effect in subset{col `=`swidth'+1'}{c |}{col `=`swidth'+4'} (Insufficient data)"'
}
else {
scalar `_ES_' = `_ES'[`r(min)']
scalar `_LCI_' = `_LCI'[`r(min)']
scalar `_UCI_' = `_UCI'[`r(min)']
di as text `"Effect in subset{col `=`swidth'+1'}{c |}{col `=`swidth'+`ewidth'-6'}"' ///
as res %7.3f `xexp'(`_ES_') `"{col `=`swidth'+`ewidth'+5'}"' ///
as res %7.3f `xexp'(`_LCI_') `"{col `=`swidth'+`ewidth'+15'}"' ///
as res %7.3f `xexp'(`_UCI_') `"{col `=`swidth'+`ewidth'+26'}"' %7.0f `_NN'[`r(min)']
}
}
} // end forvalues i=1/`nby'
drop `touse' // tidy up
*** Overall effect
if `"`overall'"'==`""' {
di as text `"{hline `swidth'}{c +}{hline `=`ewidth'+35'}"'
summ `obs' if `_USE'==5, meanonly
if !r(N) {
di as text `"Overall effect{col `=`swidth'+1'}{c |}{col `=`swidth'+4'} (Insufficient data)"'
}
else {
scalar `_ES_' = `_ES'[`r(min)']
scalar `_LCI_' = `_LCI'[`r(min)']
scalar `_UCI_' = `_UCI'[`r(min)']
di as text %-20s `"Overall effect{col `=`swidth'+1'}{c |}{col `=`swidth'+`ewidth'-6'}"' ///
as res %7.3f `xexp'(`_ES_') `"{col `=`swidth'+`ewidth'+5'}"' ///
as res %7.3f `xexp'(`_LCI_') `"{col `=`swidth'+`ewidth'+15'}"' ///
as res %7.3f `xexp'(`_UCI_') `"{col `=`swidth'+`ewidth'+26'}"' %7.0f `_NN'[`r(min)']
}
}
di as text `"{hline `swidth'}{c BT}{hline `=`ewidth'+35'}"'
} // end if `"`table'"'==`""'
end
* Subroutine to "spread" titles out over multiple lines if appropriate
// Updated July 2014
// August 2016: identical program now used here, in forestplot.ado, and in admetan.ado
// May 2017: updated to accept substrings delineated by quotes (c.f. multi-line axis titles)
// August 2017: updated for better handling of maxlines()
// March 2018: updated to receive text in quotes, hence both avoiding parsing problems with commas, and maintaining spacing
// May 2018 and Nov 2018: updated truncation procedure
// subroutine of DrawTableIPD
program define SpreadTitle, rclass
syntax [anything(name=title id="title string")] [, TArget(integer 0) MAXWidth(integer 0) MAXLines(integer 0) noTRUNCate noUPDATE ]
* Target = aim for this width, but allow expansion if alternative is wrapping "too early" (i.e before line is adequately filled)
// (may be replaced by `titlelen'/`maxlines' if `maxlines' and `notruncate' are also specified)
* Maxwidth = absolute maximum width ... but will be increased if a "long" string is encountered before the last line
* Maxlines = maximum no. lines (default 3)
* noTruncate = don't truncate final line if "too long" (even if greater than `maxwidth')
* noUpdate = don't update `target' if `maxwidth' is increased (see above)
tokenize `title'
if `"`1'"'==`""' {
return scalar nlines = 0
return scalar maxwidth = 0
exit
}
if `maxwidth' & !`maxlines' {
cap assert `maxwidth'>=`target'
if _rc {
nois disp as err `"{bf:maxwidth()} must be greater than or equal to {bf:target()}"'
exit 198
}
}
** Find length of title string, or maximum length of multi-line title string
// First run: strip off outer quotes if necessary, but watch out for initial/final spaces!
gettoken tok : title, qed(qed)
cap assert `"`tok'"'==`"`1'"'
if _rc {
gettoken tok rest : tok, qed(qed)
assert `"`tok'"'==`"`1'"'
local title1 title1 // specifies that title is not multi-line
}
local currentlen = length(`"`1'"')
local titlelen = length(`"`1'"')
// Subsequent runs: successive calls to -gettoken-, monitoring quotes with the qed() option
macro shift
while `"`1'"'!=`""' {
local oldqed = `qed'
gettoken tok rest : rest, qed(qed)
assert `"`tok'"'==`"`1'"'
if !`oldqed' & !`qed' local currentlen = `currentlen' + 1 + length(`"`1'"')
else {
local titlelen = max(`titlelen', `currentlen')
local currentlen = length(`"`1'"')
}
macro shift
}
local titlelen = max(`titlelen', `currentlen')
// Save user-specified parameter values separately
local target_orig = `target'
local maxwidth_orig = `maxwidth'
local maxlines_orig = `maxlines'
// Now finalise `target' and calculate `spread'
local maxlines = cond(`maxlines_orig', `maxlines_orig', 3) // use a default value for `maxlines' of 3 in these calculations
local target = cond(`target_orig', `target_orig', ///
cond(`maxwidth_orig', min(`maxwidth_orig', `titlelen'/`maxlines'), `titlelen'/`maxlines'))
local spread = min(int(`titlelen'/`target') + 1, `maxlines')
local crit = cond(`maxwidth_orig', min(`maxwidth_orig', `titlelen'/`spread'), `titlelen'/`spread')
** If substrings are present, delineated by quotes, treat this as a line-break
// Hence, need to first process each substring separately and obtain parameters,
// then select the most appropriate overall parameters given the user-specified options,
// and finally create the final line-by-line output strings.
tokenize `title'
local line = 1
local title`line' : copy local 1 // i.e. `"`title`line''"'==`"`1'"'
local newwidth = length(`"`title`line''"')
// if first "word" is by itself longer than `maxwidth' ...
if `maxwidth' & !(`maxlines' & (`line'==`maxlines')) {
// ... reset parameters and start over
while length(`"`1'"') > `maxwidth' {
local maxwidth = length(`"`1'"')
local target = cond(`target_orig', cond(`"`update'"'!=`""', `target_orig', `target_orig' + `maxwidth' - `maxwidth_orig'), ///
cond(`maxwidth', min(`maxwidth', `titlelen'/`maxlines'), `titlelen'/`maxlines'))
local spread = min(int(`titlelen'/`target') + 1, `maxlines')
local crit = cond(`maxwidth', min(`maxwidth', `titlelen'/`spread'), `titlelen'/`spread')
}
}
macro shift
local next : copy local 1 // i.e. `"`next'"'==`"`1'"' (was `"`2'"' before macro shift!)
while `"`1'"' != `""' {
// local check = `"`title`line''"' + `" "' + `"`next'"' // (potential) next iteration of `title`line''
local check `"`title`line'' `next'"' // (amended Apr 2018 due to local x = "" issue with version <13)
if length(`"`check'"') > `crit' { // if longer than ideal...
// ...and further from target than before, or greater than maxwidth
if abs(length(`"`check'"') - `crit') > abs(length(`"`title`line''"') - `crit') ///
| (`maxwidth' & (length(`"`check'"') > `maxwidth')) {
if `maxlines' & (`line'==`maxlines') { // if reached max no. of lines
local title`line' : copy local check // - use next iteration anyway (to be truncated)
macro shift
local next : copy local 1
local newwidth = max(`newwidth', length(`"`title`line''"')) // update `newwidth'
continue, break
}
else { // otherwise:
local ++line // - new line
// if first "word" of new line (i.e. `next') is by itself longer than `maxwidth' ...
if `maxwidth' & (length(`"`next'"') > `maxwidth') {
// ... if we're on the last line or last token, continue as normal ...
if !((`maxlines' & (`line'==`maxlines')) | `"`2'"'==`""') {
// ... but otherwise, reset parameters and start over
local maxwidth = length(`"`next'"')
local target = cond(`target_orig', cond(`"`update'"'!=`""', `target_orig', `target_orig' + `maxwidth' - `maxwidth_orig'), ///
cond(`maxwidth', min(`maxwidth', `titlelen'/`maxlines'), `titlelen'/`maxlines'))
local spread = min(int(`titlelen'/`target') + 1, `maxlines')
local crit = cond(`maxwidth', min(`maxwidth', `titlelen'/`spread'), `titlelen'/`spread')
// restart loop
tokenize `title'
local tok = 1
local line = 1
local title`line' : copy local 1 // i.e. `"`title`line''"'==`"`1'"'
local newwidth = length(`"`title`line''"')
macro shift
local next : copy local 1 // i.e. `"`next'"'==`"`1'"' (was `"`2'"' before macro shift!)
continue
}
}
local title`line' : copy local next // - begin new line with next word
}
}
else local title`line' : copy local check // else use next iteration
}
else local title`line' : copy local check // else use next iteration
macro shift
local next : copy local 1
local newwidth = max(`newwidth', length(`"`title`line''"')) // update `newwidth'
} // (N.B. won't be done if reached max no. of lines, as loop broken)
* Return strings
forvalues i=1/`line' {
// truncate if appropriate (last line only)
if `i'==`line' & "`truncate'"=="" & `maxwidth' {
local title`i' = substr(`"`title`i''"', 1, `maxwidth')
}
return local title`i' `"`title`i''"'
}
* Return values
return scalar nlines = `line'
return scalar maxwidth = min(`newwidth', `maxwidth')
return scalar target = `target'
end
* Modified version of _prefix_saving.ado
// [IPD version] modified so as to include `stacklabel' option
// April 2018, for ipdover v2.2
program define my_prefix_savingIPD, sclass
cap nois syntax anything(id="file name" name=fname) [, REPLACE * ]
if !_rc {
if "`replace'" == "" {
local ss : subinstr local fname ".dta" ""
confirm new file `"`ss'.dta"'
}
}
else {
di as err "invalid saving() option"
exit _rc
}
sreturn clear
sreturn local filename `"`fname'"'
sreturn local options `"`replace' `options'"'
end
**************************************************************
* Release notes prior to v4.00
* originally created by David Fisher, February 2013
* version 1.0 David Fisher 31jan2014
* version 1.01 David Fisher 11aug2016
* August 2016: began adapting code for Syntax 2 of -ipdmetan- (see help file)
* March 2017
// Needs further thought re behaviour of keepall, "study, m" and "by, m"
// when some combinations of study and by do not exist in the data.
// Should they be displayed nevertheless, with the message "(No subgroup data)" or similar?
// (note this is an ipdover issue rather than an ipdmetan/admetan issue)
// (this is for the next version of the package)
* version 2.0 David Fisher 11may2017
* Major update of all parts of the ipdmetan package, including ipdover.ado
* version 2.1 David Fisher 14sep2017
// various bug fixes
// improvements to behaviour if some estimates are missing
* version 3.0 David Fisher 08nov2018
// fixed bug preventing options to [command]
// upversioned to v3.0 to match with ipdmetan and admetan
// added `useopts' functionality
* version 3.1 David Fisher 03dec2018
// fixed bug in RD option
* version 3.2 David Fisher 28jan2019
// no changes to code; upversioned to match admetan/ipdmetan
// but a minor change to text in the help file