{"id":1330,"date":"2022-10-31T16:35:02","date_gmt":"2022-10-31T20:35:02","guid":{"rendered":"https:\/\/blog.uvm.edu\/tbplante\/?p=1330"},"modified":"2023-02-14T13:28:18","modified_gmt":"2023-02-14T18:28:18","slug":"making-subgroup-analysis-figure-in-stata","status":"publish","type":"post","link":"https:\/\/blog.uvm.edu\/tbplante\/2022\/10\/31\/making-subgroup-analysis-figure-in-stata\/","title":{"rendered":"Making a subgroup analysis figure in Stata"},"content":{"rendered":"\n

I was the analyst for the myPACE trial (published here in JAMA Cardiology<\/a>), and needed to put together a subgroup analysis figure. I didn’t find any helpful stock code, so I wrote my own. The code uses the Frames feature that was introduced in Stata 16. It will (a) make a new frame with the required variables but no data, (b) generate a new dichotomous variable from continuous variables, (c) generate labels, (d) grab the Ns, point estimates, and 95% CI from a logistic regression, (e) grab the P-value for interaction for the primary exposure dependent variable*, (f) write the point estimate\/95% CI\/P-value for interaction to the new frame, then (g) switch to the new frame and make this figure. This script uses local macros so needs to be run all at once in a do file, not line by line. <\/strong><\/em><\/p>\n\n\n\n

This uses a stock Stata dataset called “catheter”. The outcome of interest\/dependent variable is “infect”, the primary exposure\/independent variable of interest is “time”, and the subgroups are age, sex, and patient number. This uses logistic regression, but you can easily swap this model out for any other model. <\/p>\n\n\n\n

*You can get more complex code to format the P-values here<\/a>.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Here’s the code!<\/p>\n\n\n\n

frame reset \/\/ drop all existing frames and data\nwebuse catheter, clear \/\/ load analytical dataset\nversion 16 \/\/ need Stata version 16 or newer\n*\n* Make an empty frame with the variables we'll add later row by row\n* The variables \"rowname\" and \"pvalue\" will be strings so when\n* you add to these variables with the --frame post-- command,\n* you need to use quotes. \nframe create subgroup str30 rowname n beta low95 high95 str30 pval\n*\n*** Age\n* Need to generate a dichotomous age variable \n* You don't need to do this if the variable is already dichotomous, \n* ordinal, or nominal\ngenerate agesplit = (age>=50) if !missing(age) \/\/ below 50 is 0, 50 and above is 1\n* \n* Now generate a label for the overall grouping\nlocal label_group \"{bf:Age}\" \n* \n* Group 0, below age 50\n* Generate label for this subgroup\nlocal label_subgroup_0 \"Under 50y\" \n* Now run the model for this subgroup\nlogistic infect time if agesplit==0 \n* Now save the N, beta, and 95% CI as local macros.\n* There's lots you can save after a regression, type --return list--, \n* --ereturn list--, and --matrix list r(table)-- to see what's there\nlocal n_0 = e(N) \nlocal beta_0 = r(table)[1,1] \nlocal low95_0 = r(table)[5,1]\nlocal high95_0 = r(table)[6,1]\n* print above local macros to prove you collected them correctly: \ndi \"For `label_subgroup_0', n=\" `n_0' \", beta (95% CI)=\" %4.2f `beta_0' \" (\" %4.2f `low95_0' \" to \" %4.2f `high95_0' \")\" \n*\n* Group 1, at least 50 years old\nlocal label_subgroup_1 \"At least 50y\" \nlogistic infect time if agesplit==1 \nlocal n_1 = e(N)\nlocal beta_1 = r(table)[1,1] \nlocal low95_1 = r(table)[5,1]\nlocal high95_1 = r(table)[6,1]\ndi \"For `label_group', subgroup `label_subgroup_1', n=\" `n_1' \", beta (95% CI)=\" %4.2f `beta_1' \" (\" %4.2f `low95_1' \" to \" %4.2f `high95_1' \")\" \n* \n* Now run the model with an interaction term between the primary exposure\n* and the subgroup. \nlogistic infect c.time##i.agesplit\n* Grab the p-value as a local macro saved as pval\nlocal pval = r(table)[4,5] \n* Print that local macro to see that you've grabbed it correctly\ndi \"P-int = \" %4.3f `pval'\n* Format that P-value and save that as a new local macro called pvalue\nlocal pvalue \"P=`: display %3.2f `pval''\"\ndi \"`pvalue'\"\n* \n* Now write your local macros to the the \"subgroup\" frame\n* Each \"frame post\" command will add 1 additional row to the frame.\n* We will graph these line by line.\n* First line with overall group name a P-value for interaction:\nframe post subgroup (\"`label_group'\") (.)  (.) (.) (.) (\"`pvalue'\") \n* Now each subgroup by itself:\nframe post subgroup (\"`label_subgroup_0'\") (`n_0')  (`beta_0') (`low95_0') (`high95_0') (\"\") \nframe post subgroup (\"`label_subgroup_1'\") (`n_1')  (`beta_1') (`low95_1') (`high95_1') (\"\") \n* Optional blank line:\nframe post subgroup (\"\") (.) (.) (.) (.) (\"\") \n*\n*** Female sex\n* This is already dichotomous, so don't need to create a new variable\n* like we did for age.\nlocal label_group \"{bf:Sex}\"  \n*group 0, males\nlocal label_subgroup_0 \"Males\" \nlogistic infect time if female==0 \nlocal n_0 = e(N) \nlocal beta_0 = r(table)[1,1] \nlocal low95_0 = r(table)[5,1]\nlocal high95_0 = r(table)[6,1]\n*group 1, females\nlocal label_subgroup_1 \"Females\" \nlogistic infect time if female==1 \nlocal n_1 = e(N) \/\/ N\nlocal beta_1 = r(table)[1,1] \nlocal low95_1 = r(table)[5,1]\nlocal high95_1 = r(table)[6,1]\n*interaction P-value\nlogistic infect c.time##i.female\nlocal pval = r(table)[4,5]\nlocal pvalue \"P=`: display %3.2f `pval''\"\n*write to subgroup frame\nframe post subgroup (\"`label_group'\") (.)  (.) (.) (.) (\"`pvalue'\") \nframe post subgroup (\"`label_subgroup_0'\") (`n_0')  (`beta_0') (`low95_0') (`high95_0') (\"\") \nframe post subgroup (\"`label_subgroup_1'\") (`n_1')  (`beta_1') (`low95_1') (`high95_1') (\"\") \nframe post subgroup (\"\") (.) (.) (.) (.) (\"\") \n*\n*** patient\n* need to generate a patient dichotomous variable\ngenerate patientsplit = (patient>=20) if !missing(patient) \/\/ below 20 is 0, 20 and above is 1\nlocal label_group \"{bf:Patient}\" \n*group 0, below 20\nlocal label_subgroup_0 \"Under 20th patient\" \nlogistic infect time if patientsplit==0 \nlocal n_0 = e(N) \nlocal beta_0 = r(table)[1,1] \nlocal low95_0 = r(table)[5,1]\nlocal high95_0 = r(table)[6,1]\n*group 1, 20 and above\nlocal label_subgroup_1 \"At least the 20th patient\" \nlogistic infect time if patientsplit==1 \nlocal n_1 = e(N) \nlocal beta_1 = r(table)[1,1] \nlocal low95_1 = r(table)[5,1]\nlocal high95_1 = r(table)[6,1]\n*interaction P-value\nlogistic infect c.time##i.agesplit\nlocal pval = r(table)[4,5] \nlocal pvalue \"P=`: display %3.2f `pval''\"\n*write to subgroup frame\nframe post subgroup (\"`label_group'\") (.)  (.) (.) (.) (\"`pvalue'\") \nframe post subgroup (\"`label_subgroup_0'\") (`n_0')  (`beta_0') (`low95_0') (`high95_0') (\"\") \nframe post subgroup (\"`label_subgroup_1'\") (`n_1')  (`beta_1') (`low95_1') (`high95_1') (\"\") \nframe post subgroup (\"\") (.) (.) (.) (.) (\"\") \n*\n*** Now make the figure. You'll have to modify this so the number of rows \n*   in your subgroup frame matches the labels and whatnot\nset scheme s1mono \/\/ I like this scheme\n* Change frame to the subgroup frame\ncwf subgroup\n* Generate a row number by the current order of the data in this frame\ngen row=_n\n* Here's the code to make the figure\ntwoway \/\/\/\n(scatter row beta, msymbol(d) mcolor(black) msize(medium)) \/\/\/\n(rcap low95 high95 row, horizontal lcolor(black) lwidth(medlarge)) \/\/\/\n, \/\/\/\nlegend(off) \/\/\/\nxline(1, lcolor(red) lpattern(dash) lwidth(medium)) \/\/\/\ntitle(\"Title\") \/\/\/\nyti(\"Y Title\") \/\/\/\nxti(\"X Title\") \/\/\/\nyscale(reverse) \/\/\/\nyla( \/\/\/\n1 \"`=rowname[1]'\" \/\/\/\n2 \"`=rowname[2]', n=`=n[2]'\" \/\/\/\n3 \"`=rowname[3]', n=`=n[3]'\" \/\/\/\n4 \" \" \/\/\/ blank since it's a blank row\n5 \"`=rowname[5]'\" \/\/\/\n6 \"`=rowname[6]', n=`=n[6]'\" \/\/\/\n7 \"`=rowname[7]', n=`=n[7]'\" \/\/\/\n8 \" \" \/\/\/ blank since it's a blank row\n9 \"`=rowname[9]'\" \/\/\/\n10 \"`=rowname[10]', n=`=n[10]'\" \/\/\/\n11 \"`=rowname[11]', n=`=n[11]'\" \/\/\/\n12 \" \" \/\/\/ blank since it's a blank row\n, angle(0) labsize(small) noticks) \/\/\/\nxla(0.8(.2)2.2) \/\/\/\ntext(1 1.1 \"`=pval[1]'\", placement(e) size(small)) \/\/\/ these are the p-value labels\ntext(5 1.1 \"`=pval[5]'\", placement(e) size(small)) \/\/\/\ntext(9 1.1 \"`=pval[9]'\", placement(e) size(small)) \n*\n* Now export your figure as a PNG file\ngraph export \"myfigure.png\", replace width(1000)<\/code><\/pre>\n","protected":false},"excerpt":{"rendered":"
I was the analyst for the myPACE trial (published here in JAMA Cardiology), and needed to put together a subgroup analysis figure. I didn’t find any helpful stock code, so I wrote my own. The code uses the Frames feature that was introduced in Stata 16. It will (a) make a new frame with the … Continue reading Making a subgroup analysis figure in Stata<\/span><\/a><\/p>\n","protected":false},"author":4473,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[477491],"tags":[502556,703436,703437],"class_list":["post-1330","post","type-post","status-publish","format-standard","hentry","category-stata-code","tag-stata","tag-subgroup-analysis","tag-subgroup-figure"],"_links":{"self":[{"href":"https:\/\/blog.uvm.edu\/tbplante\/wp-json\/wp\/v2\/posts\/1330","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/blog.uvm.edu\/tbplante\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blog.uvm.edu\/tbplante\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blog.uvm.edu\/tbplante\/wp-json\/wp\/v2\/users\/4473"}],"replies":[{"embeddable":true,"href":"https:\/\/blog.uvm.edu\/tbplante\/wp-json\/wp\/v2\/comments?post=1330"}],"version-history":[{"count":15,"href":"https:\/\/blog.uvm.edu\/tbplante\/wp-json\/wp\/v2\/posts\/1330\/revisions"}],"predecessor-version":[{"id":1379,"href":"https:\/\/blog.uvm.edu\/tbplante\/wp-json\/wp\/v2\/posts\/1330\/revisions\/1379"}],"wp:attachment":[{"href":"https:\/\/blog.uvm.edu\/tbplante\/wp-json\/wp\/v2\/media?parent=1330"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blog.uvm.edu\/tbplante\/wp-json\/wp\/v2\/categories?post=1330"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blog.uvm.edu\/tbplante\/wp-json\/wp\/v2\/tags?post=1330"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}