Limitations & competing views
A research project is only as trustworthy as its willingness to argue against itself. This page is the deliberate counter-weight to the rest of the site: what our framing gets wrong, what serious scientists would argue instead, and what the numbers can and cannot claim.
What this map does and does not claim
- The convergence funnel describes a subtype, not all depression. Elevated inflammation is present in roughly 25–40% of patients. The neuroinflammation + HPA + oxidative-stress model should be read as a subtype account, not a universal one.
- "Depression" itself is a contested construct. DSM criteria allow 227+ symptom combinations; major depression is probably not one illness with one mechanism. Our single causal map inherits that limitation.
- Biology can launder politics. Poverty and adversity are real causes of the causes; routing them straight into cortisol and cytokines risks relocating a structural problem into the individual. Structural and political-economic drivers deserve their own place.
- Prevalence is modelled, not measured. The world map uses estimates built largely on Western-validated instruments; cross-country comparisons reflect detection and reporting as much as true burden.
- Knowing a cause is not knowing a cure. Decades of biomarker psychiatry have barely shifted outcomes (first-line remission ~30%; ~67% cumulative after four steps). No biomarker yet guides routine treatment selection.
Competing paradigms we hold alongside ours
Social / political-economy
Depression as a reasonable response to bad material conditions; prevalence tracks inequality. Many "causes" are downstream of political choices.
Network / symptom view
There may be no latent "depression" — symptoms (sleep, rumination, anhedonia) cause one another. Treatment then targets high-centrality symptoms, not a presumed substrate.
Stop averaging (biotypes / RDoC)
Averaging across patients hides real biological subtypes. Study dimensions and subgroups, not the DSM category as if it were a natural kind.
Computational psychiatry
Depression as pathological precision-weighting of negative prior beliefs — a distinct, testable mechanism with different treatment implications.
Evolutionary view
Low mood as a partly adaptive response to unsolvable problems — a caution against treating all depressive affect as pathology.
The translation critique
The strongest systemic critique: biological richness has not yet produced biomarker-guided care. Epistemic humility is warranted.
On the numbers: association, causation & population impact
Three rules keep the cause map honest:
- Population attributable fractions (PAFs) do not sum. Eliminating every modifiable cause might prevent on the order of 40–60% of cases — not 100% — because causes overlap.
- Population impact ≠ individual risk. A rare high-risk exposure (childhood sexual abuse, OR≈2.7) can contribute a smaller population share than a common low-risk one (poor sleep; air pollution).
- Don't double-count mediators. Smoking → inflammation and ultra-processed food → obesity → inflammation share a pathway; their effects can't simply be added.
Where we can, claims should carry a robustness indicator (an E-value: the confounding strength needed to explain an association away) and a causal tier (genetics/MR + trial > MR or natural experiment > observational).