Discussion: View Thread

I am familiar with Forest Plots for showing consistency of between-treatment differences across studies (or centers) in meta-analyses. However, this term is used for similar plots for comparison of between-treatment differences across various levels of demographic or baseline characteristics in clinical literature, especially for display of hazard ratios in oncology. However, I have seen no references in the statistical literature. Is anyone familiar with references in the statistical literature? 
For the meta-analyses, there is one line for each study. For the latter display, there will be multiple lines for each variable (such as two lines for Gender: one displaying the between-treatment difference for Females and one for Males; there may be several lines such as results for each level of ECOG PS for oncology studies). Should such plots just be used to examine if the treatment effect is observed to be the same at each level of the variable? Am I making too much of this (especially concern of "interaction" when the treatment effects differ)?
Thanks,
David

------------------------------
David Bristol
Statistical Consulting Services, Inc.
------------------------------

I've wondered about the history of the use of "forest plots" for displaying subgroup effects so I hope someone can give some history.

Note that forest plots for subgroups do not represent good statistical practice because

1. Researchers frequently create improper subgroups by binning continuous baseline characteristics such as age
2. The estimates are not covariate adjusted
3. The estimates are incredibly easy to misinterpret, as opposed to fully covariate-adjusted interaction tests

For more, see Sections 10.10.6 and 13.6.1 in  http://hbiostat.org/doc/bbr.pdf

Hbiostat		remove preview
View this on Hbiostat >

http://hbiostat.org/doc/bbr.pdf


------------------------------
Frank Harrell
Vanderbilt University School of Medicine
------------------------------

Original Message:
Sent: 06-22-2019 02:35
From: David Bristol
Subject: Forest Plots

I am familiar with Forest Plots for showing consistency of between-treatment differences across studies (or centers) in meta-analyses. However, this term is used for similar plots for comparison of between-treatment differences across various levels of demographic or baseline characteristics in clinical literature, especially for display of hazard ratios in oncology. However, I have seen no references in the statistical literature. Is anyone familiar with references in the statistical literature?
For the meta-analyses, there is one line for each study. For the latter display, there will be multiple lines for each variable (such as two lines for Gender: one displaying the between-treatment difference for Females and one for Males; there may be several lines such as results for each level of ECOG PS for oncology studies). Should such plots just be used to examine if the treatment effect is observed to be the same at each level of the variable? Am I making too much of this (especially concern of "interaction" when the treatment effects differ)?
Thanks,
David

------------------------------
David Bristol
Statistical Consulting Services, Inc.
------------------------------

This note is off-topic, but the textbook online looks excellent (Biostatistics for Biomedical Research). I enjoyed browsing it for a few minutes. I downloaded to browse more later. If it is free, that is a very generous gesture.

Thanks,

Nayak

------------------------------
Nayak Polissar
Principal Statistician
The Mountain-Whisper-Light Statistics
------------------------------

Original Message:
Sent: 06-22-2019 07:23
From: Frank Harrell
Subject: Forest Plots

I've wondered about the history of the use of "forest plots" for displaying subgroup effects so I hope someone can give some history.

Note that forest plots for subgroups do not represent good statistical practice because

1. Researchers frequently create improper subgroups by binning continuous baseline characteristics such as age
2. The estimates are not covariate adjusted
3. The estimates are incredibly easy to misinterpret, as opposed to fully covariate-adjusted interaction tests

For more, see Sections 10.10.6 and 13.6.1 in  http://hbiostat.org/doc/bbr.pdf

Hbiostat		remove preview
View this on Hbiostat >

http://hbiostat.org/doc/bbr.pdf


------------------------------
Frank Harrell
Vanderbilt University School of Medicine

Original Message:
Sent: 06-22-2019 02:35
From: David Bristol
Subject: Forest Plots

I am familiar with Forest Plots for showing consistency of between-treatment differences across studies (or centers) in meta-analyses. However, this term is used for similar plots for comparison of between-treatment differences across various levels of demographic or baseline characteristics in clinical literature, especially for display of hazard ratios in oncology. However, I have seen no references in the statistical literature. Is anyone familiar with references in the statistical literature?
For the meta-analyses, there is one line for each study. For the latter display, there will be multiple lines for each variable (such as two lines for Gender: one displaying the between-treatment difference for Females and one for Males; there may be several lines such as results for each level of ECOG PS for oncology studies). Should such plots just be used to examine if the treatment effect is observed to be the same at each level of the variable? Am I making too much of this (especially concern of "interaction" when the treatment effects differ)?
Thanks,
David

------------------------------
David Bristol
Statistical Consulting Services, Inc.
------------------------------

Frank.
Thank you for this document. I found the suggested sections very informative and helpful for subgroup analyses, interactions, etc,
Now, if we could discover more information and history of any statistical development of forest plots, or maybe they are just so inappropriate that there is no such development.
Thanks,
David

------------------------------
David Bristol
Statistical Consulting Services, Inc.
------------------------------

Original Message:
Sent: 06-22-2019 07:23
From: Frank Harrell
Subject: Forest Plots

I've wondered about the history of the use of "forest plots" for displaying subgroup effects so I hope someone can give some history.

Note that forest plots for subgroups do not represent good statistical practice because

1. Researchers frequently create improper subgroups by binning continuous baseline characteristics such as age
2. The estimates are not covariate adjusted
3. The estimates are incredibly easy to misinterpret, as opposed to fully covariate-adjusted interaction tests

For more, see Sections 10.10.6 and 13.6.1 in  http://hbiostat.org/doc/bbr.pdf

Hbiostat		remove preview
View this on Hbiostat >

http://hbiostat.org/doc/bbr.pdf


------------------------------
Frank Harrell
Vanderbilt University School of Medicine

Original Message:
Sent: 06-22-2019 02:35
From: David Bristol
Subject: Forest Plots

I am familiar with Forest Plots for showing consistency of between-treatment differences across studies (or centers) in meta-analyses. However, this term is used for similar plots for comparison of between-treatment differences across various levels of demographic or baseline characteristics in clinical literature, especially for display of hazard ratios in oncology. However, I have seen no references in the statistical literature. Is anyone familiar with references in the statistical literature?
For the meta-analyses, there is one line for each study. For the latter display, there will be multiple lines for each variable (such as two lines for Gender: one displaying the between-treatment difference for Females and one for Males; there may be several lines such as results for each level of ECOG PS for oncology studies). Should such plots just be used to examine if the treatment effect is observed to be the same at each level of the variable? Am I making too much of this (especially concern of "interaction" when the treatment effects differ)?
Thanks,
David

------------------------------
David Bristol
Statistical Consulting Services, Inc.
------------------------------