Continuos NRI and Brier Score

[Federica Guidetti]

Aim of my analysis is to compare the predictive value in term of outcome (cardiovascular death) of a model containing 6, 2 levels, categorical variables (hypotension, EF, NTproBNP, NYHAclass, HF hospitalization, diuretic dode) with all possible 5-variable models that are created by taking one variable out of the 6-variable model at a time. To perform it I would use the continuos NRI and the Brier Score. Can someone help me with the STATA code?

[Rich Goldstein]

there is an official command for the brier score; see

h brier

there is no official command for the NRI but there are at least two user-written commands; use -search- to find and download -nri- and/or -incrisk-

[Federica Guidetti]

dear Rich Thank you for your prompt answer.
I already downloaded "net from http://www.ucr.uu.se/en/images/stories/downloads" however, using this command, I was able to get only the categorical NRI
As you suggested, I was trying to download
.net from http://research.fhcrc.org/content/da...r/files/stata/ However, It was unsuccessful, since I got this message net from https://research.fhcrc.org/content/d...ter/files/stat > a/ https://research.fhcrc.org/content/d...r/files/stata/ either 1) is not a valid URL, or 2) could not be contacted, or 3) is not a Stata download site (has no stata.toc file). current site is still http://www.stata-journal.com/software/sj21-3/ r(679);

[Federica Guidetti]

Dear Rich I also used the Brier score command.

1) I did the LOGISTIC REGRESSION for primary outcome of the 6 VARIABLES MODELS (severeef, severenyha, ntprobnp4000, hfhosp6, hypotension, furequidose80). They are all 2 levels, categorical variables

logit primaryoutcome i.severeef i.severenyha i.ntprobnp4000 i.hfhosp6 i.hypotension i.furequidose80, or cformat(%3.2f)

2) generate the PREDICTED PROBABILITY for the 6 variables model

quietly logistic primaryoutcome i.severeef i.severenyha i.ntprobnp4000 i.hfhosp6 i.hypotension i.furequidose80
predict p1 if e(sample),p

3) Brier Score

brier primaryoutcome p1, group(10)

I found that group(10) is the standard value. In accordance with what needs to be adapted?
attached the Brier score result

[Rich Goldstein]

neither #3 nor #4 are very clear to me; however, for -nri-, try http://personalpages.manchester.ac.uk/staff/mark.lunt and click on "idi"

if there is a question in #4, I don't see it or understand - the parts of the output are described in the manual entry if that is the issue; if not, please clarify

forgot to add that Pepe's lab appears to have moved from Stata to R so -incrisk- might not be available any longer

[Federica Guidetti]

Dear Rich
#3: what is group refering to? (I cannot understand it reding help brier)

#4 I was just showing you the command that I used and the output that I got to have a feedback (I guess that is correct, but I was not sure)

#5 http://personalpages.manchester.ac.uk/staff/mark.lunt and click on "idi" it works: thank you

#6 I was trying to perform the continuous NRI for the outcome primaryoutcome comparing a 6 model variables (severeef, severenyha, severentprobnp, hfhosp, severediuretic, hypotension) with a 5 model variables (severenyha, severentprobnp, hfhosp, severediuretic, severediuretic, hypotension)

nri help suggest:

nri outcomevar [new_marker varlist] [if] [in] [, ignoreold at(numlist) pold(varname) pnew(varname) ]

I was trying to adapt the command to my variable:

nri primaryoutcome[severeef, severenyha, severentprobnp, hfhosp, severediuretic, hypotension], [severenyha, severentprobnp, hfhosp, severediuretic, severediuretic, hypotension]

but I got this error :
weights not allowed
r(101);

The same when I tryed this syntax:

nri primaryoutcome[severeef, severenyha, severentprobnp, hfhosp, severediuretic, hypotension] [, ignoreold severeef]

Do you have some suggestions? I cannot understand what I did wrong.