All ML models outperformed the MMSE alone. 5 key findings →
Design: longitudinal study Participants: 1,331 Evidence: moderate
Full summary, score breakdown and related studies → Original source Free full text
Research
27,380 Alzheimer's and dementia studies in plain English, each scored out of 100.
In total, 41 participants consented (n=10, 24% participants with Alzheimer disease; n=11, 27% participants with Lewy body disease; and n=20, 49% controls). 5 key findings →
Design: longitudinal study Participants: 41 Evidence: moderate
Full summary, score breakdown and related studies → Original source Free full text
Recent evidence suggests that Web-based interventions have the potential to support caregivers by decreasing caregiver stress and burden. 3 key findings →
Design: randomized controlled trial Evidence: high
Full summary, score breakdown and related studies → Original source Free full text
Care networks ranged from 1 to 8 members. 5 key findings →
Design: randomized controlled trial Evidence: high
Full summary, score breakdown and related studies → Original source Free full text
This study included 33,355 patients in the training, 14,294 in the internal validation, and 6711 in the external validation sets. 5 key findings →
Design: cohort study Participants: 33,355 Evidence: moderate
Full summary, score breakdown and related studies → Original source Free full text
BrainCheck Memory assessments were significantly correlated to the SLUMS exam, with coefficients ranging from .5 to .7. 5 key findings →
Participants: 398 Evidence: moderate
Full summary, score breakdown and related studies → Original source Free full text
A thematic analysis revealed factors that motivate household members to use these technologies. 3 key findings →
Evidence: low
Full summary, score breakdown and related studies → Original source Free full text
The findings of this study suggest that predictive modeling could inform patients and clinicians about ASCVD at risk for dementia. 3 key findings →
Participants: 29,561 Evidence: low
Full summary, score breakdown and related studies → Original source Free full text
We identified several key themes: surrogates more often chose comfort care for care recipients with dementia, particularly at advanced stages. 5 key findings →
Participants: 3,931 Evidence: low
Full summary, score breakdown and related studies → Original source Free full text
The Extra Trees model achieved up to 99% accuracy for personalized wristband data, while the multilayer perceptron model performed best in general models with 98% accuracy. 5 key findings →
Evidence: low
Full summary, score breakdown and related studies → Original source Free full text
The presented results suggest the suitability of ecological handwriting analysis for the all-around monitoring of MCI, from early screening to disease progression evaluation. 4 key findings →
Participants: 57 Evidence: low
Full summary, score breakdown and related studies → Original source Free full text
The Neural Network classifier achieved an accuracy of 91% with a 95% CI ranging from 0.79-0.97 and an AUC of 94%, on the test set after identifying the most significant features for AD prediction. 5 key findings →
Participants: 174 Evidence: low
Full summary, score breakdown and related studies → Original source Free full text
After screening 3670 articles for eligibility, the review includes 12 studies. 5 key findings →
Evidence: low
Full summary, score breakdown and related studies → Original source Free full text
A total of 63 studies were included in this study. 5 key findings →
Participants: 63 Evidence: low
Full summary, score breakdown and related studies → Original source Free full text
38/100
In phase 1, the mean system usability scores increased from 65.4 (SD 11.8) in round 1 to 73.8 (SD 15.9) in round 3, exceeding the benchmark of 68. 5 key findings →
Design: clinical trial Evidence: low
Full summary, score breakdown and related studies → Original source Free full text
Our analysis revealed a diverse range of topics reflecting community members' varying needs and concerns of individuals affected by AD/ADRD. 5 key findings →
Evidence: low
Full summary, score breakdown and related studies → Original source Free full text
The calibration models for photopic lux and correlated color temperature demonstrated strong accuracy, with an adjusted R² of 0.858 and 0.982, respectively, ensuring reliable sensor measurements. 5 key findings →
Evidence: low
Full summary, score breakdown and related studies → Original source Free full text
The TLC study used an appropriate design and data collection procedures, along with acceptable recruitment capability. 5 key findings →
Participants: 15 Evidence: low
Full summary, score breakdown and related studies → Original source Free full text
=3.40; P=.04). Greater cognitive improvements were associated with younger age, higher functional status, and lower baseline cognitive function. 3 key findings →
Evidence: low
Full summary, score breakdown and related studies → Original source Free full text
Fifty-four participants (~8% response rate) from the MindCrowd web-based cohort (mean age of 62.39 years; 39 females; and 23 APOE ε4 carriers) completed 75 trials of Super G. Results show that Super G performance was significantly associated with each of the targeted risk factors. 5 key findings →
Evidence: moderate
Full summary, score breakdown and related studies → Original source Free full text
This ultra-brief questionnaire enables rapid screening for cognitive impairment (in <1 min) by family members or as part of community geriatric assessments. 2 key findings →
Evidence: low
Full summary, score breakdown and related studies → Original source Free full text
Where “Free full text” comes from
Free full text links and the Full text may need a subscription label come from Unpaywall, Europe PMC and OpenAlex records of free copies, checked 10 Oct 2026. A study with neither is unknown, not paywalled.
How the 0–100 score works — and why some entries are capped
This is alzhub's own score. It is not an official metric, not an impact factor, and no journal, publisher or ranking body endorses it. We built it so that a reader who is not a scientist can tell a landmark trial from a conference abstract without having to already know the difference. Every point comes from one of 20 components, all of them printed on the study itself, so you can disagree with our arithmetic rather than having to trust the total.
- Journal tier up to 10 pts
- Journal's citation record up to 4 pts
- Study design up to 12 pts
- Strength of evidence up to 5 pts
- Peer-review validation up to 10 pts
- How recent up to 5 pts
- Sample size up to 4 pts
- Effect reported up to 3 pts
- Studied in people up to 3 pts
- Randomised / blinded up to 2 pts
- Times cited up to 7 pts
- Citations per year up to 5 pts
- Impact for its field and year up to 6 pts
- Cited by studies here up to 6 pts
- Still being cited up to 4 pts
- Senior author's track record up to 4 pts
- Institution's track record up to 3 pts
- Multi-centre collaboration up to 3 pts
- Indexed in PubMed up to 2 pts
- Free to read up to 2 pts
Journals are placed in ordered tiers, never given a numeric impact factor — we could not verify current impact factors from a source we control, and an invented one would be worse than none. Tier 1 is the top-tier general journals (NEJM, The Lancet, JAMA, Nature, Science, Nature Medicine); tier 2 is the leading specialty journals; tier 3 is every other peer-reviewed journal. A separate part scores how much the journal's own papers are cited, which tells the tier-3 journals apart.
Two rules we don't bend. First, we never fill in a missing value: if a record does not state a sample size or a design, that component scores zero, and if we simply have no data for a part, such as a citation count, that part is left out and the study is scored on the parts we could check. The study card says which. Second, anything that is not a peer-reviewed article is capped at 16/100 and carries a warning chip — so a conference abstract or a company press release can never outrank a peer-reviewed paper, no matter how good it looks.
We rank a rigorous null result — a good trial showing that something does not work — above an early positive finding. Someone deciding whether to buy a supplement is served by that, and the rigour is already counted by the design and journal components.
The score is computed from the published data alone and is reproducible: the same file always yields the same numbers.
See the whole rubric drawn out → — four charts covering all 27,380 studies: how the scores are spread and where the cap bites, which of the 20 components the library is strong and weak on, which subjects rest on stronger evidence, and what kinds of study each subject is actually built from.
What this library is — and isn't
This is a large sample of the Alzheimer's literature, not an exhaustive index of it. It holds 27,380 records, built mostly from automated pulls of Europe PMC (60.6% of the library) and OpenAlex (18.6%), plus smaller pulls from PubMed, Crossref and medRxiv. If a study, a drug, or a finding isn't here, that is not evidence it doesn't exist in the literature — it may simply not have been pulled in yet.
1,032 records (3.8% of the library) are not peer-reviewed — 962 of those are preprints posted before any journal review, and the rest are conference abstracts, regulatory notices and reports. Every study card on this page states its status plainly, and the score above is capped for anything that isn't a peer-reviewed article.
See the full, machine-readable breakdown of this library's coverage →