BMI as a Potential Mediator Rather than Moderator

[davebridges]

Gap 2: BMI at diagnosis is likely a mediator, not a moderator — stratified "interactions" may reflect mediation + collider bias

Labels: reviewer-response analysis causal-inference priority-high

Reviewer summary

The manuscript frames obesity as a moderator of Cushing's effects (stratify by BMI ≥30, interpret stratum differences in Figure 2 / Table 3 as interaction). The reviewer argues this treats obesity as exogenous, but:

• Hypercortisolism is a known causal driver of obesity (11β-HSD1 → visceral adiposity, metabolic syndrome pathways) [1][2].
• The manuscript itself notes that 70–95% of Cushing's patients gain weight and 32–41% become obese.
• Table 4's dose-response by obesity class is consistent with mediation (more cortisol → more obesity → more downstream damage), not pure moderation.

Risks this creates:

1. Collider bias when conditioning on BMI (a post-exposure variable) in stratified or adjusted analyses.
2. Misinterpretation of the headline narrative — "synergistic for liver, additive for glucose, protective for BP" could partly reflect mediation geometry (where on the causal chain BMI sits for each outcome) rather than biological moderation.
3. Undermines the stated goal of separating obesity-dependent from obesity-independent Cushing's effects, because stratification alone cannot do that when the stratifier is on the causal pathway.

What needs to change

A. Manuscript text

1. Methods — add a short "Causal Framework" subsection with a DAG (Cushing's → BMI → outcomes; Cushing's → outcomes; confounders → everything). State explicitly that BMI at diagnosis is plausibly a mediator.

2. Methods — justify the stratification as descriptive. Add:

   "We stratified participants by obesity status to characterize effect heterogeneity; however, because weight gain and obesity are classic manifestations of Cushing's disease, BMI at diagnosis may lie on the causal pathway rather than act purely as an effect modifier."

3. Results / Discussion — reconcile with Table 4's non-linear BMI row. Add:

   "Interpretations that rely on BMI as an effect modifier should be read alongside mediation analyses and the causal assumptions laid out in Methods. Although the primary stratified analysis supports the interpretation that 'blood glucose increases with obesity and Cushing's disease are largely additive,' non-linear BMI modeling (Table 4) identified statistically significant interactions under continuous