95% Confidence Interval (LR+ / LR-)

[Carolina Hincapie]

Dear Stata community,


I am currently working on Stata 18 version, and I can't find the way to calculate the 95% conf. interval of the LR+ and LR -

diagtest Test Outcome
(8,050 observations deleted)

           |        Outcome
  Test     |         0          1 |     Total
-----------+----------------------+----------
         0 |       385         42 |       427
         1 |        86         35 |       121
-----------+----------------------+----------
     Total |       471         77 |       548

True D defined as mort ~= 0                           [95% Conf. Inter.]
-------------------------------------------------------------------------
Sensitivity                     Pr( +| D)  45.45%      41.29%   49.62%
Specificity                     Pr( -|~D)  81.74%      78.51%   84.98%
Positive predictive value       Pr( D| +)  28.93%      25.13%   32.72%
Negative predictive value       Pr(~D| -)  90.16%      87.67%   92.66%
-------------------------------------------------------------------------
Prevalence                      Pr(D)      14.05%      11.14%   16.96%
-------------------------------------------------------------------------

. roctab  Outcome Test  

Detailed report of sensitivity and specificity
------------------------------------------------------------------------------
                                           Correctly
Cutpoint      Sensitivity   Specificity   classified          LR+          LR-
------------------------------------------------------------------------------
( >= 0 )          100.00%         0.00%       14.05%       1.0000    
( >= 1 )           45.45%        81.74%       76.64%       2.4894       0.6673
( >  1 )            0.00%       100.00%       85.95%                    1.0000
------------------------------------------------------------------------------


                      ROC                     Asymptotic normal  
           Obs       area     Std. err.      [95% conf. interval]
     ------------------------------------------------------------
           548     0.6360       0.0299        0.57734     0.69461

Does someone knows how to perform the 95% conf. interval of the LR+ and LR - on Stata?



Thanks in advance

[Bruce Weaver]

Hello Carolina Hincapie. Like risk ratios, your positive and negative likelihood ratios are ratios of two independent proportions. Therefore, I would use the MOVER-R method Robert Newcombe describes for risk ratios in this article. I don't know if anyone has written Stata code for this method. Perhaps someone else knows.

[Felix Bittmann]

You could also use bootstrapping to generate CIs, if this makes sense from a statistical point of view.

[Carolina Hincapie]

Thanks for the answers.


Some one recommended this webs sites for the 95% CI:

MedCalc's Diagnostic test evaluation calculator

Epidat 3.1 (English) - Consellería de Sanidade - Servizo Galego de Saúde (sergas.es)

In both they give me the same 95% CI between them, but different than the ones given by Stata (Post #1):

Not sure if I should concern about this.

[Vivek Gupta]

You can try the user contributed command diagt from net sj 4-4 sbe36_2
http://fmwww.bc.edu/RePEc/bocode/d/diagt.hlp

 diagt truth test 

Truth.   |
           |      Test
           |      Pos.       Neg. |     Total
-----------+----------------------+----------
  Abnormal |        17          6 |        23 
    Normal |         4         54 |        58 
-----------+----------------------+----------
     Total |        21         60 |        81 
True abnormal diagnosis defined as truth = 1 


                                                  [95% Confidence Interval]
---------------------------------------------------------------------------
Prevalence                         Pr(A)       28%       19%      39.5%
---------------------------------------------------------------------------
Sensitivity                      Pr(+|A)     73.9%     51.6%     89.8%
Specificity                      Pr(-|N)     93.1%     83.3%     98.1%
ROC area               (Sens. + Spec.)/2      .835      .738      .933 
---------------------------------------------------------------------------
Likelihood ratio (+)     Pr(+|A)/Pr(+|N)      10.7      4.04      28.4 
Likelihood ratio (-)     Pr(-|A)/Pr(-|N)       .28       .14      .559 
Odds ratio                   LR(+)/LR(-)      38.3      9.94       147 
Positive predictive value        Pr(A|+)       81%     58.1%     94.6% 
Negative predictive value        Pr(N|-)       90%     79.5%     96.2% 
---------------------------------------------------------------------------