From Patient Reports to a Pharmacodiagnostic Corpus
The knowledge-production infrastructure this chapter describes — N-of-1 experimentation, registries, online communities — has a specific and underused application. Individual patients run many drug trials. Each trial is a probe of one biological node. If the reports are captured in a comparable form, the pooled set becomes a population-level instrument for ranking which mechanisms are rate-limiting in ME/CFS — the aggregation extension of the single-patient pharmacodiagnostic method in Medication Response Reference: From Drug Response to Mechanism Identification (The Population Pharmacodiagnostic Corpus - Pooling Drug Responses Across Patients).
This is the intersection of two arguments the paper makes separately. The patient-generated-knowledge argument says patients produce epistemically valuable data. The pharmacodiagnostic argument says drug responses are mechanistic constraints. Combined: a pool of patient drug-response reports is a large, naturally occurring constraint-satisfaction problem over the ME/CFS hypothesis space. The corpus can rank candidate pathways and design the confirmatory trial; it cannot confirm mechanism — the same hypothesis-generation/conclusion boundary this chapter draws for community knowledge generally (Community Error Amplification).
The chapter’s N-of-1 and registry infrastructure already captures much of what a pharmacodiagnostic corpus needs, but not in poolable form. The candidate minimum fields are: drug, dose, dose band, duration, response domain, side-effect type, and stop reason — though whether this set is sufficient remains unresolved (The Minimum Data Schema for a Poolable Corpus). The aggregation logic is multi-level Bayesian pooling of single-patient trials with responder-subgroup analysis rather than a single average (Schmid and Yang 2022) (Zucker, Ruthazer, and Schmid 2010) (Hendrickson et al. 2020). The direct precedent is the Eckey 2025 ME/CFS and long-COVID treatment survey (n=3,925, >150 treatments), which ranks treatments by perceived benefit but does not rank mechanisms (Eckey et al. 2025).
(Certainty: 0.35 — a design proposal assembled from established methods; feasibility in ME/CFS is untested. Origin: brainstorm.)
Falsifiable prediction: A corpus built to this schema reproduces at least one known drug-response association and yields a per-drug responder ranking whose agreement across independent cohort halves exceeds a pre-specified threshold (rank correlation ρ > 0.5), and whose between-drug differences track dose and target rather than community discussion volume. Falsified if the ranking is unstable across splits, if no pooled association survives adjustment for indication and treatment popularity, or if the ranking tracks discussion volume rather than pharmacology. On the last condition the proposal is abandoned, not refined (Alternative Explanations That Could Explain the Corpus Entirely).
Consequence: The drug experiences patients already accumulate could become a research asset — but only if the community adopts a shared schema that records dose and duration, not just whether a drug helped.
A pharmacodiagnostic corpus built from patient and clinician reports inherits the epistemic limits this chapter already documents. It is self-selected (The Severity-Epistemic Gradient; see also Selection Bias Samples the Wrong End of the Severity Gradient); it is vulnerable to community error amplification, where a plausible but wrong treatment signal converges through social reinforcement (Community Error Amplification); it is unblinded, so open-label placebo and expectation contaminate reported benefit (Flávio-Reis et al. 2025) (Nikles et al. 2022); and it under-represents severe patients, who are least able to contribute (The Severity-Epistemic Gradient). Two further limits travel with any observational drug corpus: confounding by indication, since patients do not try drugs at random, and spectral degeneracy, if the accessible drug set cannot distinguish the competing hypotheses — the latter being the one limit that adding patients cannot fix. These limits do not make the corpus worthless — they define it as a hypothesis-generation instrument, not a conclusion.
Consequence: A pooled patient corpus is only as trustworthy as its stated limits; presenting a community treatment signal as a confirmed mechanism would repeat the exact epistemic error — treating testimony as proof — that this chapter argues against.
One implication follows specifically from this chapter’s evidence base. The reporting asymmetries here run opposite to intuition: positive reports are inflated by open-label placebo (Flávio-Reis et al. 2025), community enthusiasm (Community Error Amplification), and publication bias, while null reports appear less inflated by expectation and are systematically absent from the published record — though fluctuation and regression to the mean offset this, so whether their net bias is smaller is a question the corpus test estimates, not an assumption. The corpus-specific form of this asymmetry — the proposal that pooled nulls may be the more informative stratum, with its own falsifiable test — is developed at Pooled Nulls May Be More Informative Than Pooled Positives. If it holds, a patient who tried a drug and saw nothing has contributed the most valuable report in the registry, reversing how communities currently weight their data.
The first test of this argument needs no new recruitment. The Eckey 2025 survey is itself a large pool of patient drug-response reports — n=3,925, more than 150 treatments (Eckey et al. 2025). It ranks treatments, not mechanisms, but its treatment-to-target mapping can be projected onto the hypothesis space and the resulting mechanism ranking checked for split-half stability, which is the corpus proposal’s own replication test run on data that already exist (eckey mechanism reanalysis; The Eckey 2025 Benefit Ranking Can Be Projected Onto a Mechanism Ranking That Replicates). What such a re-analysis inherits is stated in ch33: it can only support the mechanism ranking if it survives the severity, popularity, and many-to-many-target alternatives Severity, Popularity, and Many-to-Many Mapping Can Manufacture a Mechanism Ranking. The interpretive stakes cut both ways. If the re-analysis works, patient-reported drug experience — the kind of data this chapter argues is underused — becomes a mechanism-ranking instrument. If it fails, the failure is consistent with three possibilities the test cannot separate on its own: the projection’s assumptions are inadequate, the reports carry no recoverable mechanism signal, or the survey instrument itself is too unstable to measure either (No Reliability Standard Yet Justifies the ρ > 0.5 Replication Threshold). Only the second of these bears on this chapter’s epistemic claim; the third is evidence about the survey, not about patient knowledge. A failed re-analysis is therefore at most weak evidence against the claim, and the chapter’s own null (What If Patient Knowledge Has No Unique Epistemic Content?) remains unresolved by any single re-analysis.
Consequence: The Eckey survey lets the community test the “patient reports as research asset” claim on data it already owns, converting this chapter’s epistemic argument into a falsifiable test — though the test is immediate only to start, not to trust, since the validation design it needs is itself unresolved.