GAMLj3 Logistic Model: Subgroup-Specific Odds Ratios from an Interaction Model

[hanamura.toru]

Title: GAMLj3 Logistic Model: Subgroup-Specific Odds Ratios from an Interaction Model

Dear GAMLj developers and jamovi community,

I am using the Logistic model under “Generalized Linear Models” in the GAMLj3 module.

I would like to ask whether GAMLj3 can report subgroup-specific odds ratios, confidence intervals, and p-values derived from a single logistic regression model containing an interaction term.

For example, suppose I fit the following model:

Outcome ~ Treatment × Biomarker + covariates

From this single interaction model, it should be possible to estimate the treatment effect within each biomarker subgroup, or the biomarker effect within each treatment group, by appropriately combining the main-effect and interaction coefficients.

These would be conditional subgroup-specific odds ratios derived from the single interaction model, rather than odds ratios obtained by fitting separate logistic regression models within each subgroup.

For comparison, I am also using “Multivariable Survival Analysis” in the jSurvival module. In its Cox proportional hazards model, jSurvival reports a table called “Within-Subgroup Hazard Ratios.” These subgroup-specific hazard ratios, confidence intervals, and p-values are derived from the single Cox interaction model rather than from separate models fitted within each subgroup.

However, in the GAMLj3 Logistic model, I have not found an equivalent table that directly reports subgroup-specific odds ratios, confidence intervals, and p-values from a single logistic interaction model.

Could you please clarify whether:

reporting conditional subgroup-specific odds ratios from a single logistic interaction model is not recommended for statistical reasons, or

this functionality has simply not yet been implemented in GAMLj3?

If this functionality is already available in GAMLj3, I would be grateful if you could tell me which options I should select.

If it has not yet been implemented, would it be reasonable to consider adding a table of subgroup-specific odds ratios, confidence intervals, and p-values derived from the fitted interaction model, similar to the “Within-Subgroup Hazard Ratios” table in jSurvival?

My main interest is predictive biomarker analysis in clinical trials, where the treatment effect within biomarker-defined subgroups—or the biomarker effect within treatment groups—is often clinically important.

Thank you very much for your guidance.

Best regards,

Toru Hanamura

[mcfanda@gmail.com]

hi
If I understood correctly, what you're are looking for are called *simple effects*, and they are available in GAMLj3. In your example outcome~ treatment X biomaker, the simple effect of treatment across biomaker is the effect, odd ratio, etc, of treatment estimated at each level of biomaker.

In pratical terms, you set up your interaction model

Screenshot from 2026-07-17 18-15-14.png (51.8 KiB) Viewed 268 times

Then you go to "simple effects" and declare the simple effects variable (say treatment) and the moderator variable (the one for which level you want the effects of the other variable for).

Screenshot from 2026-07-17 18-20-13.png (11.42 KiB) Viewed 268 times

In the output, you find two tables, one for omnibus test and one with coefficients, odd ratios and test, for the effect of treatment for each biomaker level.

Screenshot from 2026-07-17 18-21-58.png (38.11 KiB) Viewed 268 times

Hope this helps

[hanamura.toru]

Dear Marcello,

Thank you very much for your prompt and helpful reply.

That was exactly what I was looking for. I had not realized that the "Simple Effects" option provided subgroup-specific odds ratios derived from the single interaction model.

Your explanation was very clear, and it will be very useful for my predictive biomarker analyses in clinical trials.

Thank you again for your kind support and for developing such a useful module.

Best regards,

Toru Hanamura