Discussion: View Thread

This message has been cross posted to the following eGroups: Statistical Consulting Section and Statistics in Epidemiology Section .
-------------------------------------------
My apologises for cross posting.

Dear All,

I wonder what is the interpretation of confidence intervals for sensitivity and specificity measures. are they just measure of precision or meaning sth else?
Most of the papers report these two measures but almost never talk about the interpretation of their confidence interval.
Also, if we can calculate the confidence interval, we can do a test of hypothesis and report a p-value?!! If so, what will be the interpretation of p-value? what will be tested?
More and more discussion concerning the issue will be appreciated.
Many thanks for involving.

Rgds,
Amir





-------------------------------------------
Amir Kasaeian
PhD Student in Biostatistics
Tehran University of Medical Sciences (TUMS)
amir_kasaeian@yahoo.com
akasaeian@razi.tums.ac.ir
-------------------------------------------

Confidence intervals for proportions such as sensitivity and specificity have the same meaning as confidence intervals have for any other parameter. A 95 percent confidence interval for specificity for example means that if the experiment were repeated many times and 95% confidence intervals are computed each time, they have the property that approximately 95% of those intervals will contain the "true" value for specificity and 5% will not.

Whenever you are able to construct confidence intervals like this there is the corresponding hypotehsis test with the null hypothesis that the specificity is some value p0 versus the alternative that it is different from p0.  The hypothesis can be tested by generating a 95% confidence interval for specificity and rejecting the null hypothesis if and only if p0 does not fall inside the confidence interval.
-------------------------------------------
Michael Chernick
Director of Biostatistical Services
Lankenau Institute for Medical Research
-------------------------------------------






Original Message:
Sent: 04-19-2012 13:38
From: Amir Kasaeian
Subject: confidence interval for sensitivity and specificity

This message has been cross posted to the following eGroups: Statistical Consulting Section and Statistics in Epidemiology Section .
-------------------------------------------
My apologises for cross posting.

Dear All,

I wonder what is the interpretation of confidence intervals for sensitivity and specificity measures. are they just measure of precision or meaning sth else?
Most of the papers report these two measures but almost never talk about the interpretation of their confidence interval.
Also, if we can calculate the confidence interval, we can do a test of hypothesis and report a p-value?!! If so, what will be the interpretation of p-value? what will be tested?
More and more discussion concerning the issue will be appreciated.
Many thanks for involving.

Rgds,
Amir





-------------------------------------------
Amir Kasaeian
PhD Student in Biostatistics
Tehran University of Medical Sciences (TUMS)
amir_kasaeian@yahoo.com
akasaeian@razi.tums.ac.ir
-------------------------------------------

Dear Amir, You may need to provide additional detail regarding your goals in order to obtain more precise feedback. In general, depending on the area of application, there may be expectations regarding a minimum sensitivity or specificity necessary to, say, commercialize a new medical test. So, the confidence limits could support decisions based on the observed data. On a related note, you may want to review the literature covering ROC curves as these are a helpful visualization of the relationship between sensitivity and specificity which by their nature work in opposition. The "optimal" relationship can be assessed with an ROC curve (or curves). Best regards, David ------------------------------------------- David Reasner Albemarle Scientific Consulting -------------------------------------------
Original Message: Sent: 04-19-2012 14:15 From: Michael Chernick Subject: confidence interval for sensitivity and specificity Confidence intervals for proportions such as sensitivity and specificity have the same meaning as confidence intervals have for any other parameter. A 95 percent confidence interval for specificity for example means that if the experiment were repeated many times and 95% confidence intervals are computed each time, they have the property that approximately 95% of those intervals will contain the "true" value for specificity and 5% will not. Whenever you are able to construct confidence intervals like this there is the corresponding hypotehsis test with the null hypothesis that the specificity is some value p0 versus the alternative that it is different from p0. The hypothesis can be tested by generating a 95% confidence interval for specificity and rejecting the null hypothesis if and only if p0 does not fall inside the confidence interval. ------------------------------------------- Michael Chernick Director of Biostatistical Services Lankenau Institute for Medical Research -------------------------------------------
Original Message: Sent: 04-19-2012 13:38 From: Amir Kasaeian Subject: confidence interval for sensitivity and specificity This message has been cross posted to the following eGroups: Statistical Consulting Section and Statistics in Epidemiology Section . ------------------------------------------- My apologises for cross posting. Dear All, I wonder what is the interpretation of confidence intervals for sensitivity and specificity measures. are they just measure of precision or meaning sth else? Most of the papers report these two measures but almost never talk about the interpretation of their confidence interval. Also, if we can calculate the confidence interval, we can do a test of hypothesis and report a p-value?!! If so, what will be the interpretation of p-value? what will be tested? More and more discussion concerning the issue will be appreciated. Many thanks for involving. Rgds, Amir ------------------------------------------- Amir Kasaeian PhD Student in Biostatistics Tehran University of Medical Sciences (TUMS) amir_kasaeian@yahoo.com akasaeian@razi.tums.ac.ir -------------------------------------------

Yes, on the ROC curves but pay heed to my comment that Colin Begg has shown that if you construct an ROC curve based on a fitted model the AUC is not a good statistic to use to compare two classification methods via formal hypothesis testing.


-------------------------------------------
Michael Chernick
Director of Biostatistical Services
Lankenau Institute for Medical Research
-------------------------------------------






Original Message:
Sent: 04-20-2012 08:36
From: David Reasner
Subject: confidence interval for sensitivity and specificity

Dear Amir,

You may need to provide additional detail regarding your goals in order to obtain more precise feedback. In general, depending on the area of application, there may be expectations regarding a minimum sensitivity or specificity necessary to, say, commercialize a new medical test. So, the confidence limits could support decisions based on the observed data. On a related note, you may want to review the literature covering ROC curves as these are a helpful visualization of the relationship between sensitivity and specificity which by their nature work in opposition. The "optimal" relationship can be assessed with an ROC curve (or curves).

Best regards,
David


-------------------------------------------
David Reasner
Albemarle Scientific Consulting
-------------------------------------------

Dear Michael and David,

Thanks for your informative comments. They were helpful and and exclude doubt for me!
I would be also very heppy to know more about waht Colin Begg has shown. Could you please let me know more deatils?

Cheers,
Amir 



-------------------------------------------
Amir Kasaeian
PhD Student in Biostatistics
Tehran University of Medical Sciences (TUMS)
amir_kasaeian@yahoo.com
akasaeian@razi.tums.ac.ir
-------------------------------------------






Original Message:
Sent: 04-20-2012 11:21
From: Michael Chernick
Subject: confidence interval for sensitivity and specificity


Yes, on the ROC curves but pay heed to my comment that Colin Begg has shown that if you construct an ROC curve based on a fitted model the AUC is not a good statistic to use to compare two classification methods via formal hypothesis testing.


-------------------------------------------
Michael Chernick
Director of Biostatistical Services
Lankenau Institute for Medical Research
-------------------------------------------






Original Message:
Sent: 04-20-2012 08:36
From: David Reasner
Subject: confidence interval for sensitivity and specificity

Dear Amir,

You may need to provide additional detail regarding your goals in order to obtain more precise feedback. In general, depending on the area of application, there may be expectations regarding a minimum sensitivity or specificity necessary to, say, commercialize a new medical test. So, the confidence limits could support decisions based on the observed data. On a related note, you may want to review the literature covering ROC curves as these are a helpful visualization of the relationship between sensitivity and specificity which by their nature work in opposition. The "optimal" relationship can be assessed with an ROC curve (or curves).

Best regards,
David


-------------------------------------------
David Reasner
Albemarle Scientific Consulting
-------------------------------------------