2010-11-10 10:13

Posting: # 6135
Views: 20,546

 Dissolution similarity [Dissolution / BCS / IVIVC]

Dear friends,

I have received a regulatory query to check disso similarity.

Query is:

To do a statistical comparison of the dissolution profiles of the 10mg and 20mg strength at PH 6.8 according to the - "Model Independent Multivariate Confidence Region Procedure”. Calculate confidence interval of Mahalanobis distance (DM) and a respective similarity limit (RD).

(Reference: Dissolution Testing of Immediate Release Solid Oral Dosage Forms, U.S. Department of Health and Human Services, Food and Drug Administration, Center for Drug Evaluation and Research (CDER), August 1997) as described in the revised module 3.2.P.2.2).

My data is:
I have two software: SAS 9.1.3 (but not having SAS/IML) and WinNonlin 5.3.
SAS 9.1.3 also not supporting to Hotelling's T² method.

Is there any other SAS program to calulate above model with SAS/STAT?
Pl. guide me to resolve the query.

Thank You.


Edit: Category changed. [Helmut]

2011-04-21 10:27

@ Nirali
Posting: # 6930
Views: 15,454

 Dissolution similarity

Dear Friends,

I am awaiting for response. Pl. guide me if any idea regarding above mentioned post.

Thanks & Regards,

Nirali Mehta

2011-04-28 13:14

@ Nirali
Posting: # 6966
Views: 15,418

 Dissolution similarity


As this question comes up from time to time, I found the following reference:

I'll write a function in R, but I'm short in time for the moment.
Maybe after the weekend.

Good luck


PS: for those who want to play with the data in R:
m.ref <-
structure(list(V1 = c(42.06, 44.16, 45.63, 48.52, 50.49, 49.77
), V2 = c(59.91, 60.18, 55.77, 60.39, 61.82, 62.73), V3 = c(65.58,
67.17, 65.56, 66.51, 69.06, 69.77), V4 = c(71.81, 70.82, 70.5,
73.06, 72.85, 72.88), V5 = c(77.77, 76.11, 76.92, 78.54, 78.99,
80.18), V6 = c(85.67, 83.27, 83.91, 84.99, 86.86, 84.2), V7 = c(93.14,
88.01, 86.83, 88, 89.7, 88.88), V8 = c(94.23, 89.59, 90.12, 93.43,
90.79, 90.47)), .Names = c("V1", "V2", "V3", "V4", "V5", "V6",
"V7", "V8"), class = "data.frame", row.names = c(NA, -6L))

m.test <-
structure(list(V1 = c(19.99, 22.08, 21.93, 22.44, 25.67, 26.37
), V2 = c(36.7, 39.29, 38.54, 39.46, 42.35, 41.34), V3 = c(47.77,
49.46, 47.76, 49.72, 52.68, 51.01), V4 = c(55.08, 56.79, 55.14,
58.67, 59.71, 57.75), V5 = c(65.69, 67.22, 65.25, 69.21, 71.51,
69.44), V6 = c(81.37, 82.42, 83.49, 84.93, 86.61, 85.9), V7 = c(92.39,
89.93, 90.19, 94.12, 93.8, 94.45), V8 = c(97.1, 95.62, 95.62,
95.51, 96.7, 98.07)), .Names = c("V1", "V2", "V3", "V4", "V5",
"V6", "V7", "V8"), class = "data.frame", row.names = c(NA, -6L

cov(m.ref)  #Variance-covariance matrix of reference
cov(m.test) #Variance-covariance matrix of test

Berlin, Germany,
2011-04-29 13:14

@ Aceto81
Posting: # 6973
Views: 15,692

 Mahalanobis distance - R code

Dear Ace,

there is a function mahalanobis() in the stats package, i.e in the base installation of R.
But unfortunately it gives no confidence interval out of the box.
And this is the tricky part I think and for that I wish you a nice weekend :-D.

Beside the Tsong paper which is cryptic in this respect and doesn't give ready to program advice have a look at this online resource for another method:
B. Reiser
Needs some postscript reader.

Have also a look at the function ginv() of the MASS package, which implements the Moore-Penrose inverse which is necessary in calculating the Mahalanobis distance if the estimated variance-covariance matrix is not invertible.

If you are interested, I have some code ruins.



NanKing, China,
2011-10-06 14:03
(edited by yicaoting on 2011-10-06 14:18)

@ Nirali
Posting: # 7434
Views: 16,381

 DDSolver, Mahalanobis distance and its 90% CI

Dear Nirali,

I am the developer of DDSolver, an Excel add-in for analysis of drug dissolution data.

Have you tried DDSolver? It can be accessed here.

This program was developed during my Ph.D study years. It is able to calculate Ms's distance and it's 90% confidence interval. Hope it can help you.

Although users are forbidden to see the code inside the program, it is not important for pharma scientists.

If you want to use DDSolver to calculate Ma's distance, I can promise that I have compared the results of DDSolver with that of the famous paper which might be practically recommended by FDA. Moreover, I have compared DDSolver's Ma's distance with that of Mathematica. Mathematica is a famous numerical analysis software.
Both comparisons are high satisfactor.

Finally, let me use your data to calc Ma's distance and it's 90%CI. The results are as follows:

Some explanations: the conclusion of Reject is obtained based on Max Difference of 10% in between mean dissolution profiles through all the sampling time points as a similarity limit. I used 10mg-formuation as a Reference formulation.

Alternatively, you can specify a certain Ma's distance value as similarity limit which can be obtained from two bathes of your stardard formulations. To do this, you can use DDSolver's MSD Determination for Fration-Time Data module.

I am very glad if someone can compare DDSolver's results with that of other programs.

2012-07-03 05:40

@ yicaoting
Posting: # 8879
Views: 14,353

 DDSolver, Mahalanobis distance and its 90% CI

Dear All,

Thank you for response. I have installed DDSolver and it is very helpful to me.


Nirali Mehta

Berlin, Germany,
2012-07-06 08:54

@ yicaoting
Posting: # 8900
Views: 14,676

 DDSolver, Mahalanobis distance and its 90% CI

Dear yicaoting,

» If you want to use DDSolver to calculate Ma's distance, I can promise that I have compared the results of DDSolver with that of the famous paper which might be practically recommended by FDA. Moreover, I have compared DDSolver's Ma's distance with that of Mathematica, Mathematica is a famous numerical analysis software.
» Both comparisons are high satisfactor.

if I got it right (I can't see your code) you calculate the 90% CI of the MD simply as:
    MD + sqrt(Fval/K)
with K the scaling factor and Fval the critical F value from the Tsong et. al. paper.

Is this really that simple? How does this fit into the lagrange multiplier method mention in that paper?

How does this compare to the method of calculating CI's of MD elaborated in
Reiser, B. (2001)
"Confidence Intervals for the Mahalanobis Distance"
Communications in Statistics. Simulation and Computation, 30, 37-45
online resource

implemented in SAS (needs module IML) described here.

BTW: The Tsong et al paper can be found in downloadable form via this link.



2019-03-07 18:23

@ yicaoting
Posting: # 20008
Views: 4,459

 DDSolver, Mahalanobis distance and its 90% CI

I am using DD solver, however it is online giving me following statics
p (sampling points)
K (scaling factor)
Hotelling's T^2
Mahalanobis Distance (MSD)

I am not able to see Max_MSD.
can you please guide me how to calculate this from the same data that is shared above.

Also, help me understand why am I not getting Max MSD calculations in the MSD for Fraction-time data
 Admin contact
20,348 posts in 4,275 threads, 1,405 registered users;
online 20 (1 registered, 19 guests [including 9 identified bots]).
Forum time (Europe/Vienna): 09:06 UTC

When you meet the Buddha, you kill him;
when you meet the patriarchs, you kill them.    Wumen Huikai

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz