Home › Research › End-of-life care
AI on Alzheimer Disease: Large-Scale Reproduction Shows the Field’s Hypotheses Cannot Be Distinguished on Common Data
Ewing MA.. Research Square, 2026-08-27
Topic: Brain imaging, connectivity and assessment
After more than three decades of intensive research, Alzheimer disease care remains largely palliative, and the field has drawn sustained criticism, both for that reason and for pursuing some directions of inquiry over others. Adjudicating such a criticisms fairly requires what the field has never assembled: a comparison of its competing hypotheses on a common dataset, a common sample and a matched protocol. Yet the enormous size and annual volume of the field’s corpus make manual reproduction and comparison humanly impossible. To address this challenge, we introduce AI-mediated Hypothesis Aggregation (AHA), a supervised instrument that reconstructs a field’s competing hypotheses from its literature and re-instantiates their published claims on a shared substrate, pairing a hosted language-model extraction layer with 6 deterministic AI/ML models (5 base classes and 1 ensemble) for all quantitative work. From 95,756 Alzheimer publications (2005–2025) AHA recovered 6 hypothesis camps, identified ADNI as the broadest mature substrate carrying all six, and re-estimated 10,884 papers’ claims on ADNI (≈1.23 million model fits). AHA reproduced the field’s reported performance: its best multivariate panel reached AUC 0.844, within 0.016 of the reported multivariate median of 0.860. Yet on common ground the camps did not differ in predictive capability, resolving onto a single brain-integrity axis. This reframes the criticism. Publication and citation bias does appear present on the common sample under the naive assumption that the dataset carries adequate informative signal: amyloid attracts 32×the publications and 48× the citations of the least-studied camp. However, once conditioning is applied to remove ambiguous signal and noise from the dataset, we find that no camp’s hypothesis retains adjudicating signal, and the common data cannot support any of the camps nor rank them. Judged by that criterion, the research and citation gaps may be rationalizable on factors outside pure scientific comparison. AHA thus neither vindicates nor condemns the research behaviour of the Alzheimer field: rather, it demonstrates that apparent contention may be forced agreement amid common-data limitations.
Key findings
- Yet on common ground the camps did not differ in predictive capability, resolving onto a single brain-integrity axis.
- Publication and citation bias does appear present on the common sample under the naive assumption that the dataset carries adequate informative signal: amyloid attracts 32×the publications and 48× the citations of the least-studied camp.
- However, once conditioning is applied to remove ambiguous signal and noise from the dataset, we find that no camp’s hypothesis retains adjudicating signal, and the common data cannot support any of the camps nor rank them.
- Judged by that criterion, the research and citation gaps may be rationalizable on factors outside pure scientific comparison.
- AHA thus neither vindicates nor condemns the research behaviour of the Alzheimer field: rather, it demonstrates that apparent contention may be forced agreement amid common-data limitations.
Study details
- Effect sizeAUC 0.844
- Evidence strengthLow
How this study scored 16/100
Preprint — not yet peer-reviewed. It scored 26 on the same rubric, but non-peer-reviewed sources are capped at 16, so it shows as 16. Scored zero for study design, sample size, studied in people and randomised / blinded because the record does not state it. We have no data on times cited, citations per year, impact for its field and year, cited by studies here, still being cited, senior author's track record, institution's track record and multi-centre collaboration, so those were left out: it earned 16 of the 62 points we could judge it on, which is 26 out of 100.
| Component | Points | Why |
|---|---|---|
| Journal tier | 2 / 10 | Tier 4 — not a peer-reviewed journal (Research Square) |
| Journal's citation record | 0 / 4 | Not a journal article, so the journal's record does not count |
| Study design | 0 / 12 | No study design recorded — scores zero, not guessed (not stated in the record — scored 0) |
| Strength of evidence | 1 / 5 | Rated “low” in our summary |
| Peer-review validation | 4 / 10 | Preprint — not yet peer-reviewed. Hosted on a preprint server. |
| How recent | 5 / 5 | 2026 — within 1 year(s) of the newest study here |
| Sample size | 0 / 4 | Participant count not stated — scores zero (not stated in the record — scored 0) |
| Effect reported | 3 / 3 | Effect reported with a statistic (interval, p-value, ratio or similar) |
| Studied in people | 0 / 3 | Not stated whether this was studied in people — scores zero (not stated in the record — scored 0) |
| Randomised / blinded | 0 / 2 | No randomisation or blinding stated — scores zero (not stated in the record — scored 0) |
| Times cited | — / 7 | No citation count (not found in OpenAlex) — left out of this score |
| Citations per year | — / 5 | No citation count or no year — left out of this score |
| Impact for its field and year | — / 6 | No field-weighted citation impact (OpenAlex has none yet) — left out of this score |
| Cited by studies here | — / 6 | Not yet checked against this library's citation graph — left out of this score |
| Still being cited | — / 4 | No yearly citation counts — left out of this score |
| Senior author's track record | — / 4 | Senior author's record not available — left out of this score |
| Institution's track record | — / 3 | No institution listed in OpenAlex — left out of this score |
| Multi-centre collaboration | — / 3 | Institutions not listed — left out of this score |
| Indexed in PubMed | 1 / 2 | Not in PubMed, but has a DOI |
| Free to read | 0 / 2 | Behind a paywall — scores zero |
| Points | 16 / 62 | Scored on 12 of 20 parts. We have no data for the rest, so they are left out: 16 of the 62 points we could judge it on is 26 out of 100. |
| Total | 26 / 100 | Capped to 16 because this is not a peer-reviewed article. |
What each of these 20 components measures, and what this score is not →
Read the original
Free copy unknown: no free-copy check has an answer for this record yet. That does not mean it is paywalled.
Cite this study APA, Vancouver, BibTeX · download .bib or .ris
Built from the record alzhub holds for this paper, not looked up. Volume, issue and page numbers are not held in the record. Check the citation against the original before you use it.
APA 7th
Ewing, M. A. (2026). AI on Alzheimer Disease: Large-Scale Reproduction Shows the Field’s Hypotheses Cannot Be Distinguished on Common Data. Research Square. https://doi.org/10.21203/rs.3.rs-10282922/v2
Vancouver
Ewing MA. AI on Alzheimer Disease: Large-Scale Reproduction Shows the Field’s Hypotheses Cannot Be Distinguished on Common Data. Research Square. 2026. doi:10.21203/rs.3.rs-10282922/v2.
BibTeX
@article{ewing2026-e30eb38b,
author = {Ewing, M. A.},
title = {{AI on Alzheimer Disease: Large-Scale Reproduction Shows the Field’s Hypotheses Cannot Be Distinguished on Common Data}},
journal = {Research Square},
year = {2026},
doi = {10.21203/rs.3.rs-10282922/v2},
url = {https://doi.org/10.21203/rs.3.rs-10282922/v2}
}
Related studies in End-of-life care
Also tagged “convolutional”
205 studies in the library carry the topic tag “convolutional”; these are this study's neighbours in a fixed, alphabetical-by-hash tour of them, not the closest in content. All 205 tagged studies.