Schizophrenia's Evidence Network Has Two Disconnected Research Islands

The M_MED series: mapping epistemic structure in pharmaceutical evidence

This is the fourth experiment in IRDME's pharmaceutical evidence topology series. Each experiment treats a medical evidence chain as a multilayer graph: nodes are theoretical claims, empirical findings, and regulatory outcomes; edges encode three epistemic relations.
• justifies -- claim A provides the theoretical or mechanistic basis for claim B
• selects_endpoints -- A determines or constrains what outcome measures are used when evaluating B
• cites_as_support -- A is cited as empirical evidence in support of B

Previous experiments found:
• M_MED2 (vaccines, healthy control): r = +0.75, p = 0.014. Different hubs dominate the justification layer vs the citation layer. Structurally healthy differentiation.
• M_MED1 (antidepressants): r = +0.41, not significant. Monoamine hypothesis rank #1 in all three layers. Mild circularity.
• M_MED3 (opioids): r = +0.14, not significant. pain_undertreated_claim rank #1 in both justification and citation. Strong circularity. Porter and Jick citation anomaly detected by topology alone without reading any content.

M_MED4 tests the schizophrenia evidence chain. Two frameworks coexist in this domain.

The dopamine hypothesis (D2 receptor hyperactivity causes positive symptoms) has been dominant since the 1960s. It is supported by PET imaging evidence showing elevated striatal dopamine in schizophrenia patients, and by the clinical fact that D2-blocking drugs reliably reduce positive symptoms. The empirical grounding is real.

The NMDA receptor hypofunction hypothesis (Olney, Coyle) explains what the dopamine hypothesis cannot: negative symptoms, cognitive deficits, and the full schizophrenia syndrome produced by ketamine and PCP in healthy subjects. It is mechanistically upstream: NMDA hypofunction on GABAergic interneurons causes disinhibition of dopamine neurons, producing the dopamine excess detected by PET. The NMDA hypothesis explains the dopamine finding rather than competing with it. Despite this, no NMDA-targeted drug has achieved regulatory approval for schizophrenia.

Pre-registration

All four hypotheses were committed to a public repository with a SHA-256 hash before any analysis was run. Hash: 6631aad9. Timestamp: 2026-06-03T15:55:18Z. Records: github.com/vladi160/preregistrations, experiment M_MED4_v1.

Results

h2 CONFIRMED -- dopamine_hyperactivity_hypothesis is rank #1 in the justification layer (degree = 4). It directly justifies the D2 blocking mechanism, which justifies the endpoint selection, which justifies the clinical trials, which justify regulatory approval. Treatment guidelines loop back to justify it further. It is the structural root of the evidence chain.

h3 CONFIRMED -- dopamine_hyperactivity_hypothesis is rank #1 in the citation layer (degree = 4). The RCTs cite it, the PET imaging cites it, the treatment guidelines cite it, and the hypothesis cites the guidelines back in return. The founding hypothesis dominates both the theoretical and citation structure. Same structural circularity found in antidepressants (M_MED1) and opioids (M_MED3).

h1 DENIED -- r(justifies, selects_endpoints) = -0.128. This is new. The endpoint selection layer (PANSS - the scale used to measure treatment outcomes in schizophrenia trials) has a negative correlation with the theoretical justification layer. PANSS is rank #1 in the endpoint layer but rank #5 in justification. The theoretical debate between dopamine and NMDA is structurally invisible at the measurement layer. PANSS is used regardless of which hypothesis is correct. This is a new structural class: measurement-layer dominance. The endpoint has become institutionally autonomous from the theoretical structure that originally motivated it.

The M_MED structural spectrum

Four experiments, four structural regimes:
• M_MED2 vaccines: r = +0.75 -- healthy differentiation
• M_MED1 antidepressants: r = +0.41 -- mild circularity
• M_MED3 opioids: r = +0.14 -- strong circularity
• M_MED4 schizophrenia: r = -0.13 -- measurement-layer dominance

The gradient from healthy to distorted is now a measurable spectrum.

The most important finding: paradigm islands

When the analysis completed, the IRDME engine flagged: graph_connected = False.

This was not a data error. It was the key structural finding.

The seven nodes in the dopamine branch (dopamine_hyperactivity_hypothesis, d2_receptor_blocking_mechanism, panss_bprs_endpoint, chlorpromazine_haloperidol_rct, dopamine_pet_imaging, fda_antipsychotic_approval, treatment_guidelines_apa) and the three nodes in the NMDA branch (nmda_hypofunction_hypothesis, ketamine_pcp_schizophrenia_model, negative_symptom_gap) are completely disconnected across all three layers.

Zero cross-paradigm edges in the justification layer. Zero in the endpoint layer. Zero in the citation layer.

The two research communities have never formally engaged each other in the structural evidence network. Neither paradigm cites the other, challenges the other's evidence, or shares an endpoint. They are not competing within the same evidence space -- they are parallel research islands.

This is not suppression. In suppression, the NMDA hypothesis would be present in the citation network but ranked low. Here it is topologically absent from the dopamine citation space entirely, and the dopamine paradigm is absent from the NMDA citation space. They coexist without contact.

The NMDA hypothesis does drop from rank #2 in justifications to rank #4 in citations. The nodes between it and dopamine in the citation layer are treatment_guidelines_apa and chlorpromazine_haloperidol_rct -- both are products of the dopamine paradigm, not independent evidence. The dominant paradigm insulates itself through its own institutional outputs.

But the deeper finding is structural isolation, not just rank suppression.

What IRDME cannot say

This analysis reads graph structure, not scientific content. It cannot:
• Determine which hypothesis is scientifically correct
• Evaluate the clinical evidence for dopamine or NMDA hypotheses
• Say whether current antipsychotic treatments are appropriate
• Predict which theory will be validated by future neuroscience

What it can say: the published evidence network for schizophrenia has the structural topology of two isolated research islands, with an entrenched measurement endpoint that is no longer coupled to the theoretical competition between them.