sdm-help-list >

**Exploratory Meta-Regressions with multiple hypotheses and variables**Showing 1-4 of 4 posts

Oct 13, 2022 04:10 PM | Jordan Dejoie

Exploratory Meta-Regressions with multiple hypotheses and variables

Hello All,

We are currently conducting a meta-analysis comparing whole brain activation in cocaine users vs heroin users during cue reactivity tasks. Our initial results from a linear model suggested activation to drug cues was greater in heroin users relative to cocaine users.

We would like to conduct an additional exploratory analysis looking at the role that length of drug use has on these effects. To answer such question we ran a linear model with drug (a binary variable in our table) set to 1 and Length of Use (a continuous variable) set to 0. The analysis did run and we did receive results however, we want to confirm that the way we conducted the analysis is in fact addressing the above question. Our interpretation is that this analysis would be essentially showing whether an effect of drug group (cocaine, heroin) holds when length of use is included as a covariate in the model.

Next, we ran a subsequent linear model with Length of Use as 1 and drug as 0 to examine whether there was a main effect of length of use in certain regions, while accounting for substance type in the model

Lastly, we ran a linear model with an interaction variable (drug * Length of Use), but not including the other variables as 0's in the model, and also received results. The output (i.e., .html tables) from this analysis was organized into two sections (i.e., blobs w SDM z > 3.1, and blobs w SDM z < -3.1); we are unclear how to interpret this output and whether we conducted this analysis correctly to probe an interaction between substance type and mean length of use.

We appreciate any insight anyone has into constructing models that would address the above questions, whether it is possible to run these types of analyses using this software, and any information on how to interpret results of the analyses we ran.

Sincerely,

Jordan

We are currently conducting a meta-analysis comparing whole brain activation in cocaine users vs heroin users during cue reactivity tasks. Our initial results from a linear model suggested activation to drug cues was greater in heroin users relative to cocaine users.

We would like to conduct an additional exploratory analysis looking at the role that length of drug use has on these effects. To answer such question we ran a linear model with drug (a binary variable in our table) set to 1 and Length of Use (a continuous variable) set to 0. The analysis did run and we did receive results however, we want to confirm that the way we conducted the analysis is in fact addressing the above question. Our interpretation is that this analysis would be essentially showing whether an effect of drug group (cocaine, heroin) holds when length of use is included as a covariate in the model.

Next, we ran a subsequent linear model with Length of Use as 1 and drug as 0 to examine whether there was a main effect of length of use in certain regions, while accounting for substance type in the model

Lastly, we ran a linear model with an interaction variable (drug * Length of Use), but not including the other variables as 0's in the model, and also received results. The output (i.e., .html tables) from this analysis was organized into two sections (i.e., blobs w SDM z > 3.1, and blobs w SDM z < -3.1); we are unclear how to interpret this output and whether we conducted this analysis correctly to probe an interaction between substance type and mean length of use.

We appreciate any insight anyone has into constructing models that would address the above questions, whether it is possible to run these types of analyses using this software, and any information on how to interpret results of the analyses we ran.

Sincerely,

Jordan

Oct 19, 2022 04:10 PM | Jordan Dejoie

RE: Exploratory Meta-Regressions with multiple hypotheses and variables

Hi Everyone,

Following up to see if anyone has any helpful suggestions.

Thank you!

Following up to see if anyone has any helpful suggestions.

Thank you!

Nov 1, 2022 01:11 PM | Lydia Fortea -

*Instituto de Investigaciones Biomédicas August Pi i Sunyer (IDIBAPS)*RE: Exploratory Meta-Regressions with multiple hypotheses and variables

Dear Jordan,

I have some questions about your statements:

Firstly, the mean analysis of your meta-analysis is the differences in brain activation between cocaine vs heroin?? or the activation in cocaine and heroine separayely.

1 and 2. The variable drug and length of use what does exactly define? and why did you put the length of use to 0 if that is what you wanted to analyze? I'm not sure if I'm understanding the model you did.

The interacction variable I guess you did it by multiplying the two variables no? The peaks with z>3.1 is the clusters that showed an associated higher activation with your variable, and the ones of z<-3.1, the cluster that showed an associated lower activations with your variable.

Kind regards,

Lydia

I have some questions about your statements:

Firstly, the mean analysis of your meta-analysis is the differences in brain activation between cocaine vs heroin?? or the activation in cocaine and heroine separayely.

1 and 2. The variable drug and length of use what does exactly define? and why did you put the length of use to 0 if that is what you wanted to analyze? I'm not sure if I'm understanding the model you did.

The interacction variable I guess you did it by multiplying the two variables no? The peaks with z>3.1 is the clusters that showed an associated higher activation with your variable, and the ones of z<-3.1, the cluster that showed an associated lower activations with your variable.

Kind regards,

Lydia

Nov 15, 2022 07:11 PM | Jordan Dejoie

RE: Exploratory Meta-Regressions with multiple hypotheses and variables

Lydia,

Thank you for your response.

1. The procedure we followed was to first conduct a mean analysis of each of the subgroups, Cocaine and Heroin. We then ran a linear model with only Drug Type (0 = cocaine, & 1 = heroin) in the model to investigate differences between the two types of studies we included. As an exploratory analysis, we wanted to further investigate the interactive role that average length of drug use might play with respect to neural engagement during cue reactivity as a function of substance type (cocaine, heroin). We calculated the interaction term by multiplying Drug Type & Length of Use (in months) and included this interaction term, along with the main effect regressors of Drug Type and Length of Use in a second linear model.

2. To elaborate, Drug defines whether an included study had cocaine or heroin users as participants and length of use is a continuous variable with the average length of use for the participants in the corresponding study.

3. The interaction variable is described above. We ran one model with length of use set to zero as a baseline, as we removed a few studies from the initial sample that did not have a length of use metric.

If I am understanding you correctly, when running a model with the interaction term as described above, any clusters surviving correction at z > +3.1 would indicate activation that scales with positive values of the interaction term; (i.e., length of use with heroin); any clusters surviving correction at z < -3.1 would indicate activation that scales with zero values of the interaction term (i.e.., length of use with cocaine).

Thank you,

Jordan Dejoie

Thank you for your response.

1. The procedure we followed was to first conduct a mean analysis of each of the subgroups, Cocaine and Heroin. We then ran a linear model with only Drug Type (0 = cocaine, & 1 = heroin) in the model to investigate differences between the two types of studies we included. As an exploratory analysis, we wanted to further investigate the interactive role that average length of drug use might play with respect to neural engagement during cue reactivity as a function of substance type (cocaine, heroin). We calculated the interaction term by multiplying Drug Type & Length of Use (in months) and included this interaction term, along with the main effect regressors of Drug Type and Length of Use in a second linear model.

2. To elaborate, Drug defines whether an included study had cocaine or heroin users as participants and length of use is a continuous variable with the average length of use for the participants in the corresponding study.

3. The interaction variable is described above. We ran one model with length of use set to zero as a baseline, as we removed a few studies from the initial sample that did not have a length of use metric.

If I am understanding you correctly, when running a model with the interaction term as described above, any clusters surviving correction at z > +3.1 would indicate activation that scales with positive values of the interaction term; (i.e., length of use with heroin); any clusters surviving correction at z < -3.1 would indicate activation that scales with zero values of the interaction term (i.e.., length of use with cocaine).

Thank you,

Jordan Dejoie