net reclassification index (NRI) and integrated discrimination index (IDI)*computing confidence intervals

[tejas singh]

Hi All,

I am having difficulties obtaining 95% CI's from the NRI and IDI command in stata.
I have pasted below some sample code and output. The output reports only the p-value.
Does anyone have any ideas on how I can obtain 95% CI's from this. Thanks alot, Tejas.


. idi MACE height $xlist

Integrated Discrimination Improvement for height = 0.0005
Standard Error = 0.0005
z = 0.8680
P-value (one-sided) = 0.1927
P-value (two-sided) = 0.3854


*Note: $xlist simply refers to a custom varlist i have created for purposes of IDI; MACE, major adverse cardiovascular events.

[Maarten Buis]

idi is a community contributed package. It is good that you tell us that and where you got it from. Sometimes there are various versions floating around in cyberspace. You can imagine the confusion that occurs when we talk about different versions without realizing it.

A quick glance at the literature suggests (but correct me if I am wrong) that the test is based on a standard normal distribution. So the 95% confidence interval is:

// open example data
sysuse nlsw88, clear

// prepare the data

gen byte black = race == 2 if !missing(race)
label variable black "respondent's race"
label define black 0 "not black" ///
                   1 "black"
label value black black

idi union black grade south collgrad
di "lb = " r(idi) - invnormal(.975)*r(se)
di "ub = " r(idi) + invnormal(.975)*r(se)

[tejas singh]

Hi Maarten, Thanks for your rapid help. Much appreciated.

[Gunnar Austad]

Hello,

I use the NRI package from Mark Lunt (net from http://personalpages.manchester.ac.uk/staff/mark.lunt).

I am calculating the categorical NRI for the new risk factor "plaqueburden". I set the risk tresholds at 5%, 10% and 15%.

I use the following syntax:
nri priorCVD plaqueburden sex age smk lpa diabetes CVDinfamily, at(0.05 0.10 0.15)

Results are as follows:
In subjects with priorCVD == 1, 18.64% had increased probabilities
In subjects with priorCVD == 1, 15.25% had decreased probabilities

In subjects with priorCVD == 0, 6.07% had increased probabilities
In subjects with priorCVD == 0, 10.86% had decreased probabilities

Net Reclassification Improvement for plaqueburden = 0.0818
Standard Error = 0.0542
z = 1.5109
P-value (one-sided) = 0.0654
P-value (two-sided) = 0.1308

However, I have trouble obtaining the 95% CI. Maybe I should manage this by the information given in the thread above, but I am not. Would very much appreciate if anyone could provide the correct syntax.