[R] error in lrm( )

[David Winsemius]

On Dec 9, 2010, at 8:06 AM, 笑啸 wrote:

> Dear Sir or Madam：
>
>
>
> I am a doctor of urology,and I am engaged in developing a nomogram  
> of bladder cancer. May I ask for your help on below issue?
>
> I set up a dataset which include 317 cases. I got the Binary  
> Logistic Regression model by SPSS.And then I try to reconstruct the  
> model
>
> （lrm(RECU~Complication+T.Num+T.Grade+Year+TS)） by R-Project,and try  
> to internal validate the model through using the function  
> “validate( )”,and get the ROC through the function  
> “plot.roc( )”.The outcomes like this: At last I want to get the  
> Logistic model ,and get the prediction accuracy .Now the “Area under  
> the curve”(0.6931) is not too bad,

As a doctor with some experience using logistic regression and looking  
at the output of ROC analyses, I would ask why you think an AUC is  
"not too bad"? Just flipping a fair coin would give you an expected  
AUC of 0.5.

> but the “Dxy”(I think it as the prediction accuracy probability) is  
> too low.

If I'm not mistaken there is a 1-1 relationship between AUC and Dxy,  
so I suspect you have an incorrect scale for judging the merits of one  
or the other of those figure of merit numbers. (You may want to  
purchase Harrell's text "Regression Modeling Strategies" in which I  
just checked my memory on this point and found that I was correct.)

-- 
David.

> And I don’t know which reason lead to the outcomes.Maybe I have a  
> mistake understanding on the function “lrm( )”,and apply it wrong.
>
> Could you please give me some idea on how to resulve this problem?  
> Thanks in advance for your kind support.

>
> warmly regards,
>  Ding
> --------------------------------------- 
> outcomes 
> ----------------------------------------------------------------------------
> Logistic Regression Model
> lrm(formula = RECU ~ Complications + T.Num + T.Grade + Year + TS, x  
> = TRUE, y = TRUE)
>
>                Model Likelihood                     
> Discrimination                     Rank Discrim.
>                  Ratio Test                                     
> Indexes                                    Indexes
> Obs    317   LR chi2     37.78                 R2    
> 0.154                         C      0.693
>  0     201   d.f.         5                               g     
> 0.876                           Dxy    0.386
>  1     116   Pr(> chi2)   <0.0001              gr     
> 2.400                           gamma  0.408
> max |deriv| 2e-09                                   gp    
> 0.183                           tau-a  0.180
>                                                                       Brier 
>  0.207
>
>                                           Coef               
> S.E.             Wald Z            Pr(>|Z|)
> Intercept                           -2.3566         0.3819        
> -6.17                <0.0001
> Complications                 1.6807         0.6005         
> 2.80                0.0051
> T.Num                                0.6481         0.2503          
> 2.59               0.0096
> T.Grade                            0.4276           0.1820         
> 2.35                 0.0188
> Year                                   0.5759             
> 0.2849       2.02               0.0432
> TS                                        0.6313           
> 0.2750      2.30                    0.0217
>
>> validate(f,B=200)
>
>         index.orig   training   test     optimism index.corrected   n
>
> Dxy       0.3861     0.4081    0.3699  0.0382     0.3479        200
> R2        0.1537     0.1716    0.1378  0.0339     0.1198        200
> Intercept  0.0000      0.0000    -0.0585 0.0585     -0.0585        200
> Slope     1.0000      1.0000    0.8835  0.1165     0.8835        200
> Emax     0.0000      0.0000    0.0375  0.0375     0.0375        200
> D        0.1160      0.1315    0.1030  0.0285     0.0875        200
> U        -0.0063     -0.0063    0.0021  -0.0084    0.0021        200
> Q        0.1223      0.1378    0.1010  0.0369     0.0855        200
> B        0.2073      0.2035    0.2114  -0.0079     0.2153        200
> g        0.8755      0.9415    0.8170   0.1244     0.7511        200
> gp       0.1833      0.1920    0.1728   0.0192     0.1641        200
>
>> plot.roc(RECU,l)
>
> Call:
>
> plot.roc.default(x = RECU, predictor = l)
>
> Data: l in 201 controls (response 0) < 116 cases (response 1).
>
> Area under the curve:  
> 0.6931______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.

David Winsemius, MD
West Hartford, CT