chstock / dtcompair Goto Github PK

Hello Christian,

I have been using DTComPair a lot to analyse real data from two paired studies. It's been extremely helpful!

I have stumbled upon what could be considered a corner-case: if all subjects are diseased (d=1) or non-diseased (d=0), tab.paired stops with an error:

Error in eval(substitute(d), data, parent.frame()) : 
  numeric 'envir' arg not of length one

What do you think about changing the check

if (identical(sort(unique(d)), as.integer(c(0, 1))) == FALSE)

with something like

if (all(unique(d) %in% c(0, 1)) == FALSE)

?

(In a similar fashion, one could consider modifying the checks for y1 and y2).

I am a bit hesitant because this would cause errors downstream (eg, in sesp.rel, but not in tpffpf.rel), and making the necessary changes to all functions would be a lot of work. At the same time, for example in the example below, calculating the rTPF would be ok...

> df <- data.frame(
+   d = rep(1, 10),
+   y1 = rbinom(10, 1, 0.3),
+   y2 = rbinom(10, 1, 0.7)
+ )
> paired.data <- tab.paired2(d, y1, y2, df)
> paired.data
Two binary diagnostic tests (paired design)

Test1: 'y1'
Test2: 'y2'

Diseased:
           Test1 pos. Test1 neg. Total
Test2 pos.          2          6     8
Test2 neg.          0          2     2
Total               2          8    10

Non-diseased:
           Test1 pos. Test1 neg. Total
Test2 pos.          0          0     0
Test2 neg.          0          0     0
Total               0          0     0

> tpffpf.rel(paired.data)
$tpf
       rel.tpf se.log.rel.tpf    lcl.rel.tpf    ucl.rel.tpf   pval.rel.tpf 
    4.00000000     0.61237244     1.20450229    13.28349491     0.02358585 

$fpf
       rel.fpf se.log.rel.fpf    lcl.rel.fpf    ucl.rel.fpf   pval.rel.fpf 
           NaN            NaN            NaN            NaN            NaN 

$alpha
[1] 0.05

(tab.paired2 is a modified version of tab.paired where I made the suggested change above to the if condition).

What do you think?
Thank you for taking the time to consider this.

Andrea

See: Yougui Wu (2023). Simple methods for comparing two predictive values with incomplete data. Journal of Biopharmaceutical Statistics, DOI: 10.1080/10543406.2023.2188925.

Hello Christian,

Thank you for this very useful package!

If you don't mind, I'd have two feature requests for you to consider:

1. Add a function to compare TPF and TNF on a relative scale [1] (and/or FPF, since relative FPF is estimable also from screen-positive paired data, unlike relative TNF)

2. Calculate and return the estimated covariance $\hat\sigma_{PN}$ [2] (pv.rpv() function)

Thank you for considering these requests,
Andrea

[1] Cheng, H. and Macaluso, M., 1997. Comparison of the accuracy of two tests with a confirmatory procedure limited to positive results. Epidemiology, pp.104-106.

[2] Moskowitz, C.S. and Pepe, M.S., 2006. Comparing the predictive values of diagnostic tests: sample size and analysis for paired study designs. Clinical trials, 3(3), pp.272-279.

Hello Christian,

Sorry for bothering you with another issue. I have been using DTComPair for some actual data analysis and I have noticed that:

• sesp.rel and tpffpf.rel both compute Test2/Test1: I coded them so to be consistent with sesp.diff.ci, sesp.mcnemar, and sesp.exactbinom's behaviour (ie, Test2-Test1)
• pv.rpv computes Test1/Test2
• pv.gs and pv.wgs compute abs(Test1-Test2) (as per R code), but the help file says Test2-Test1. Is there a reason for taking the absolute value?

I believe it would be good to harmonise the test used as referent across the package's functions. Let me know what you think.

(PS: one could consider also harmonising what is returned by the package's functions: either always a list of named vectors, like sesp.diff.ci, or a list of lists, like sesp.mcnemar.)

Thank you for considering this!

Implement 2 d.f. chi-square test by Lachenbruch and Lynch (Stat Med, 1998, 17: 2207-17), as a 'generalized' McNemar's test, for a joint comparison of sensitivity and specificity.

See: Yougui Wu (2023). Joint comparison of the predictive values of multiple binary diagnostic tests: an extension of McNemar’s test, Journal of Biopharmaceutical Statistics, 33:1, 31-42. DOI: 10.1080/10543406.2022.2065500.