How we score a study
Every study in our library carries a number out of 100. This page is the whole rubric behind that number — what the 20 parts measure, why some entries are capped, and what the score deliberately is not.
- 27,780studies scored
- 46.0median score
- 9–93full range
- 26,746peer-reviewed articles
- 1,034not peer-reviewed, so capped at 16
- 15,506scored on fewer than 20 parts, because we have no data for some
The 20 things we score
The 100 points are split 20 ways, in 4 groups. The split is a judgement, and here it is in full.
- How the study was done and where it was published: 58 points, 10 components. How it was built and which journal published it carry the most, because those two tell you most about whether a finding will hold up. The journal counts twice over: its tier, and how much its papers are cited. How recent the study is, how big it was, whether it reports the size of its effect, whether it was done in people and whether it was randomised and blinded carry less, because size and recency do not make up for a weak design.
- How other researchers have used it: 28 points, 5 components. Times cited, citations a year, citations against its field, citations from studies in this library, and citations in the last three years.
- Who did it: 10 points, 3 components. The senior author's and the institution's track record, and how many institutions took part.
- Whether a reader can find and read it: 4 points, 2 components. Indexed in PubMed, and free to read.
When we have no data. If a study's record does not say something, such as how many people took part, that part scores zero, but if we simply have no data for a part, such as how often the study has been cited, we leave that part out and score the study on the parts we could check. 15,506 of the 27,780 studies are scored on fewer than all 20 parts, and each of those study cards says how many.
Versions. Version 1.0 had 6 components: journal, design, strength of evidence, peer review, recency, and a combined size-and-precision score. Version 2.0 (8 October 2026) had 9: it replaced size-and-precision with sample size, and added effect reported, studied in people, and randomised or blinded. Version 3.0 (10 October 2026) had 19: it added how often other researchers cite the study, who did it, and whether a reader can find and read it, and gave the first nine smaller ceilings. Version 4.0 (29 October 2026), the current one, has 20: it added the journal's own citation record, and stopped counting a gap in our data as a zero. Under version 2.0, 5,682 of 15,290 records (37%) had the same score on every component, because most records leave design, size and effect blank, which is why version 3.0 added parts that read data 94–100% of records had.
Where the points come from
The library is strongest on peer-review validation (97.7% of the points on offer) and weakest on randomised / blinded (3.8%).
This chart is wider than your screen — drag it sideways, or open the table underneath, which has every number in it.
Every number in this chart, as a table
| Component | Average earned | Maximum | Share of the maximum | Studies scored on it | Studies with no data (left out) |
|---|---|---|---|---|---|
| Journal tier | 7.2 | 10 | 72.5% | 27,780 | 0 |
| Journal's citation record | 2.2 | 4 | 53.9% | 15,451 | 12,329 |
| Study design | 2.2 | 12 | 18.4% | 27,780 | 0 |
| Strength of evidence | 1.8 | 5 | 36.3% | 27,523 | 257 |
| Peer-review validation | 9.8 | 10 | 97.7% | 27,780 | 0 |
| How recent | 2.3 | 5 | 46.3% | 27,776 | 4 |
| Sample size | 0.7 | 4 | 17.7% | 27,780 | 0 |
| Effect reported | 0.4 | 3 | 13.7% | 27,780 | 0 |
| Studied in people | 1.1 | 3 | 36.8% | 27,780 | 0 |
| Randomised / blinded | 0.1 | 2 | 3.8% | 27,780 | 0 |
| Times cited | 3.8 | 7 | 54.4% | 15,000 | 12,780 |
| Citations per year | 2.3 | 5 | 46.6% | 14,996 | 12,784 |
| Impact for its field and year | 3.2 | 6 | 52.8% | 14,744 | 13,036 |
| Cited by studies here | 1.4 | 6 | 23.9% | 15,249 | 12,531 |
| Still being cited | 1.7 | 4 | 41.5% | 15,000 | 12,780 |
| Senior author's track record | 2.6 | 4 | 66.4% | 13,459 | 14,321 |
| Institution's track record | 1.4 | 3 | 47.5% | 14,597 | 13,183 |
| Multi-centre collaboration | 1.4 | 3 | 45% | 15,317 | 12,463 |
| Indexed in PubMed | 1.9 | 2 | 96% | 27,780 | 0 |
| Free to read | 0.8 | 2 | 38.3% | 23,265 | 4,515 |
Read across each row: the pale track is the most that component can be worth, and the filled part is what the average study in our library actually earns. The sentence under each one is the explanation a real record in the library carries, as the score wrote it, with the names of authors and institutions taken out.
-
Journal tier up to 10 points
Which journal published it, placed in ordered tiers — never a numeric impact factor.
Across the 27,780 studies it is scored on, it averages 7.3 of 10 (72.5% of the points on offer). For 3 studies the record does not state it, so it scores zero there — and the card says so.
The scoreable level nearest that average is 7 of 10, which 21,033 of those 27,780 studies sit on. One of them explains itself like this:
Tier 3 — peer-reviewed journal (Journal of Alzheimer's disease : JAD)
. -
Journal's citation record up to 4 points
How much the journal's papers are cited over its whole life (its h-index in OpenAlex). It grades the many journals our tiers lump together. Old, large journals do best, so it is a track record, not a guarantee.
Across the 15,451 studies it is scored on, it averages 2.2 of 4 (53.9% of the points on offer). For 12,329 studies we have no data for it, so it is left out of their score.
The scoreable level nearest that average is 2 of 4, which 4,515 of those 15,451 studies sit on. One of them explains itself like this:
Alzheimer's & Dementia: journal h-index 228 (OpenAlex, 2026-10-29)
. -
Study design up to 12 points
How the study was built: a pooled review of many trials tells you more than one snapshot.
Across the 27,780 studies it is scored on, it averages 2.2 of 12 (18.4% of the points on offer). For 20,234 studies the record does not state it, so it scores zero there — and the card says so.
The scoreable level nearest that average is 3 of 12, which 20 of those 27,780 studies sit on. One of them explains itself like this:
A computer model rather than new measurements
. The level the most studies actually sit on is a different one — 0 of 12, shared by 20,234 of the 27,780:No study design recorded — scores zero, not guessed
. -
Strength of evidence up to 5 points
How firm the finding is in our own summary of it, from preliminary up to high.
Across the 27,523 studies it is scored on, it averages 1.8 of 5 (36.3% of the points on offer). For 257 studies we have no data for it, so it is left out of their score.
The scoreable level nearest that average is 2 of 5, which 64 of those 27,523 studies sit on. One of them explains itself like this:
Rated “early” in our summary
. The level the most studies actually sit on is a different one — 1 of 5, shared by 19,189 of the 27,523:Rated “low” in our summary
. -
Peer-review validation up to 10 points
Whether other scientists checked it before publication.
Across the 27,780 studies it is scored on, it averages 9.8 of 10 (97.7% of the points on offer).
The scoreable level nearest that average is 10 of 10, which 26,746 of those 27,780 studies sit on. One of them explains itself like this:
Peer-reviewed. Published as an article in a named peer-reviewed journal
. -
How recent up to 5 points
How close it is to the newest study we carry.
Across the 27,776 studies it is scored on, it averages 2.3 of 5 (46.3% of the points on offer). For 4 studies we have no data for it, so it is left out of their score.
The scoreable level nearest that average is 2 of 5, which 4,795 of those 27,776 studies sit on. One of them explains itself like this:
2020 — within 10 years of the newest study here
. The level the most studies actually sit on is a different one — 1 of 5, shared by 11,795 of the 27,776:2015 — within 20 years of the newest study here
. -
Sample size up to 4 points
How many people were in it, in bands: a blank count scores nothing.
Across the 27,780 studies it is scored on, it averages 0.7 of 4 (17.7% of the points on offer). For 18,260 studies the record does not state it, so it scores zero there — and the card says so.
The scoreable level nearest that average is 1 of 4, which 3,028 of those 27,780 studies sit on. One of them explains itself like this:
20 participants
. The level the most studies actually sit on is a different one — 0 of 4, shared by 18,260 of the 27,780:Participant count not stated — scores zero
. -
Effect reported up to 3 points
Whether it reports how big the effect was, and with an interval or test rather than a bare number.
Across the 27,780 studies it is scored on, it averages 0.4 of 3 (13.7% of the points on offer). For 22,790 studies the record does not state it, so it scores zero there — and the card says so.
The scoreable level nearest that average is 0 of 3, which 22,792 of those 27,780 studies sit on. One of them explains itself like this:
No effect size reported — scores zero
. -
Studied in people up to 3 points
Whether the record shows it was done in people, not in mice or a dish.
Across the 27,780 studies it is scored on, it averages 1.1 of 3 (36.8% of the points on offer). For 16,118 studies the record does not state it, so it scores zero there — and the card says so.
The scoreable level nearest that average is 0 of 3, which 17,563 of those 27,780 studies sit on. One of them explains itself like this:
Not stated whether this was studied in people — scores zero
. -
Randomised / blinded up to 2 points
Whether the study itself says it was randomised, and whether it was blinded.
Across the 27,780 studies it is scored on, it averages 0.1 of 2 (3.8% of the points on offer). For 24,647 studies the record does not state it, so it scores zero there — and the card says so.
The scoreable level nearest that average is 0 of 2, which 26,177 of those 27,780 studies sit on. One of them explains itself like this:
No randomisation or blinding stated — scores zero
. -
Times cited up to 7 points
How many times other papers have cited it, as counted by OpenAlex, in bands.
Across the 15,000 studies it is scored on, it averages 3.8 of 7 (54.4% of the points on offer). For 12,780 studies we have no data for it, so it is left out of their score.
The scoreable level nearest that average is 4 of 7, which 1,872 of those 15,000 studies sit on. One of them explains itself like this:
Cited 66 times (OpenAlex, 2026-10-10)
. The level the most studies actually sit on is a different one — 5 of 7, shared by 3,939 of the 15,000:Cited 116 times (OpenAlex, 2026-10-10)
. -
Citations per year up to 5 points
Citations per year since it came out, so a new paper is not judged against a thirty-year head start.
Across the 14,996 studies it is scored on, it averages 2.3 of 5 (46.6% of the points on offer). For 12,784 studies we have no data for it, so it is left out of their score.
The scoreable level nearest that average is 2 of 5, which 2,841 of those 14,996 studies sit on. One of them explains itself like this:
3.0 citations a year over 3 years since 2025
. The level the most studies actually sit on is a different one — 3 of 5, shared by 3,888 of the 14,996:8.3 citations a year over 3 years since 2025
. -
Impact for its field and year up to 6 points
OpenAlex's field-weighted citation impact: its citations against the average for papers of the same field and year.
Across the 14,744 studies it is scored on, it averages 3.2 of 6 (52.8% of the points on offer). For 13,036 studies we have no data for it, so it is left out of their score.
The scoreable level nearest that average is 3 of 6, which 4,459 of those 14,744 studies sit on. One of them explains itself like this:
Cited 4.7× the average for papers of its field and year (OpenAlex FWCI)
. -
Cited by studies here up to 6 points
How many other studies in this library cite it, counted from the reference lists of the studies we hold.
Across the 15,249 studies it is scored on, it averages 1.4 of 6 (23.9% of the points on offer). For 12,531 studies we have no data for it, so it is left out of their score.
The scoreable level nearest that average is 1 of 6, which 2,581 of those 15,249 studies sit on. One of them explains itself like this:
Cited by 1 other study in this library
. The level the most studies actually sit on is a different one — 0 of 6, shared by 5,533 of the 15,249:No other study in this library cites it
. -
Still being cited up to 4 points
How often it was cited in the last three calendar years: is the field still using it?
Across the 15,000 studies it is scored on, it averages 1.7 of 4 (41.5% of the points on offer). For 12,780 studies we have no data for it, so it is left out of their score.
The scoreable level nearest that average is 2 of 4, which 4,126 of those 15,000 studies sit on. One of them explains itself like this:
Cited 16 times in 2024–2026 (2026 still in progress; OpenAlex)
. -
Senior author's track record up to 4 points
The last-listed (usually senior) author's career h-index in OpenAlex. A track record, not a guarantee.
Across the 13,459 studies it is scored on, it averages 2.7 of 4 (66.4% of the points on offer). For 14,321 studies we have no data for it, so it is left out of their score.
The scoreable level nearest that average is 3 of 4, which 4,477 of those 13,459 studies sit on. One of them explains itself like this:
The last-listed author has a career h-index of 94 (OpenAlex)
. -
Institution's track record up to 3 points
The h-index of the strongest institution on the paper, from OpenAlex. A track record, not a guarantee.
Across the 14,597 studies it is scored on, it averages 1.4 of 3 (47.5% of the points on offer). For 13,183 studies we have no data for it, so it is left out of their score.
The scoreable level nearest that average is 1 of 3, which 5,227 of those 14,597 studies sit on. One of them explains itself like this:
The strongest institution on the paper has an h-index of 1,477 (OpenAlex)
. -
Multi-centre collaboration up to 3 points
How many institutions took part. Work repeated across several centres is harder to get by chance.
Across the 15,317 studies it is scored on, it averages 1.4 of 3 (45% of the points on offer). For 12,463 studies we have no data for it, so it is left out of their score.
The scoreable level nearest that average is 1 of 3, which 4,897 of those 15,317 studies sit on. One of them explains itself like this:
OpenAlex lists 2 institutions in 1 country
. The level the most studies actually sit on is a different one — 2 of 3, shared by 5,462 of the 15,317:OpenAlex lists 4 institutions in 1 country
. -
Indexed in PubMed up to 2 points
Whether PubMed indexes it, or failing that it has a DOI, so a reader can find and check it.
Across the 27,780 studies it is scored on, it averages 1.9 of 2 (96% of the points on offer). For 96 studies the record does not state it, so it scores zero there — and the card says so.
The scoreable level nearest that average is 2 of 2, which 25,655 of those 27,780 studies sit on. One of them explains itself like this:
Indexed in PubMed (PMID 31413141)
. -
Free to read up to 2 points
Whether anyone can read the full paper for free. This is about access, not quality.
Across the 23,265 studies it is scored on, it averages 0.8 of 2 (38.3% of the points on offer). For 4,515 studies we have no data for it, so it is left out of their score.
The scoreable level nearest that average is 0 of 2, which 14,352 of those 23,265 studies sit on. One of them explains itself like this:
Behind a paywall — scores zero
.
Why some entries stop at 16
A conference abstract, a preprint, a company press release and a regulator's notice are all worth reading. None of them has been through peer review. So the rubric caps every one of them at 16 out of 100 and gives it a warning chip, which means no non-peer-reviewed entry can ever outrank a peer-reviewed paper in our library — not as a convention we try to hold to, but as arithmetic.
How the scores are spread
Every study in the library, in 5-point bins. Median 46, range 9–93.
This chart is wider than your screen — drag it sideways, or open the table underneath, which has every number in it.
Every number in this chart, as a table
| Score range | Studies |
|---|---|
| 0–4 | 0 |
| 5–9 | 1 |
| 10–14 | 18 |
| 15–19 | 1015 |
| 20–24 | 31 |
| 25–29 | 363 |
| 30–34 | 711 |
| 35–39 | 4652 |
| 40–44 | 5768 |
| 45–49 | 4206 |
| 50–54 | 3412 |
| 55–59 | 2709 |
| 60–64 | 1927 |
| 65–69 | 1364 |
| 70–74 | 902 |
| 75–79 | 463 |
| 80–84 | 191 |
| 85–89 | 44 |
| 90–94 | 3 |
| 95–99 | 0 |
The number 16 is not a round guess. The lowest-scoring peer-reviewed article we carry scores 21, so the cap sits below it with room to spare. If a peer-reviewed study ever scored 16 or less, the scoring step would stop and publish no new scores until the cap was fixed. Of the 1034 entries that are not peer-reviewed, 997 scored high enough on the other components to be pulled down by the cap; the rest were already below it. On those 997 study cards we print the raw total as well as the capped one, so you can see exactly what the cap cost.
What the library actually looks like
A score tells you about one study. These two charts tell you about the pile: which subjects we hold strong evidence on, and — the thing a single number hides — what kinds of study that evidence is actually made of.
Which subjects rest on stronger evidence
Each subject's median score, with the middle half of its studies as a bar and its full range as a line.
This chart is wider than your screen — drag it sideways, or open the table underneath, which has every number in it.
Every number in this chart, as a table
| Subject | Studies | Lowest | 25th percentile | Median | 75th percentile | Highest | Peer-reviewed |
|---|---|---|---|---|---|---|---|
| Clinical trials | 162 | 16 | 43 | 61 | 76 | 88 | 145 |
| How common it is | 522 | 16 | 50 | 57 | 63 | 85 | 490 |
| Laboratory science | 46 | 12 | 48.2 | 56.5 | 62.8 | 80 | 43 |
| Prevention and risk | 1004 | 16 | 46 | 55 | 65 | 87 | 979 |
| Getting a diagnosis | 815 | 16 | 46 | 54 | 61 | 79 | 789 |
| Long-term cohort studies | 542 | 16 | 48 | 54 | 60 | 80 | 526 |
| Lewy body dementia | 566 | 16 | 45.2 | 51 | 56 | 71 | 547 |
| Treatments | 1489 | 10 | 40 | 50 | 64 | 90 | 1432 |
| Caregiving | 834 | 16 | 43 | 50 | 60 | 83 | 816 |
| Cognitive tests and assessment | 881 | 10 | 43 | 49 | 57 | 84 | 849 |
| Disparities | 833 | 16 | 42 | 49 | 57 | 79 | 808 |
| Mild cognitive impairment | 701 | 16 | 44 | 49 | 57 | 77 | 685 |
| Biomarkers | 2195 | 11 | 40 | 48 | 58 | 91 | 2102 |
| Prevention trials | 835 | 16 | 40 | 48 | 67 | 93 | 821 |
| Brain tissue and pathology | 612 | 10 | 41 | 48 | 54 | 80 | 587 |
| TREM2 and immune genetics | 701 | 16 | 43 | 47 | 53 | 85 | 684 |
| Genetics | 1616 | 10 | 40 | 46 | 55 | 86 | 1528 |
| Brain scans and imaging | 1459 | 16 | 39 | 46 | 53 | 86 | 1385 |
| Digital health tools | 1094 | 16 | 41 | 46 | 54 | 85 | 1044 |
| Costs and economics | 199 | 16 | 40 | 46 | 57 | 82 | 180 |
| Tau biomarkers | 690 | 16 | 39 | 45 | 52 | 88 | 661 |
| Clinical safety | 630 | 16 | 37 | 45 | 55 | 86 | 616 |
| Inflammation and the immune system | 868 | 16 | 40 | 43 | 49 | 81 | 852 |
| End-of-life care | 855 | 13 | 37 | 43 | 53 | 86 | 839 |
| Risk genes beyond APOE | 847 | 10 | 39 | 43 | 47 | 73 | 824 |
| Hippocampus size on brain scans | 492 | 36 | 36 | 43 | 52 | 84 | 492 |
| Risk factors | 2174 | 11 | 37.2 | 42 | 53 | 89 | 2065 |
| Vascular contributions | 546 | 12 | 38 | 42 | 52 | 84 | 529 |
| Disease mechanisms | 1381 | 12 | 36 | 40 | 43 | 79 | 1334 |
| Drug development | 873 | 15 | 36 | 40 | 43 | 89 | 823 |
| Health systems and care delivery | 654 | 9 | 35 | 40 | 47 | 86 | 631 |
| Down syndrome and Alzheimer's | 664 | 16 | 35 | 39 | 43 | 74 | 640 |
A high median here means the studies we carry on that subject score well. It does not mean the subject is settled, and it does not mean we have all of it. We hold 32 subjects and 46–2195 studies in each.
What kinds of study each subject rests on
Laboratory science is the most observational: 20 of its 46 studies watch what happens rather than test a change.
This chart is wider than your screen — drag it sideways, or open the table underneath, which has every number in it.
Every number in this chart, as a table
| Subject | Studies | Systematic review | Meta-analysis | Randomised controlled trial | Phase 2 or phase 3 trial | Cohort study | Case-control | Cross-sectional | Other or not recorded |
|---|---|---|---|---|---|---|---|---|---|
| Laboratory science | 46 | 0 | 0 | 1 | 1 | 15 | 4 | 1 | 24 |
| Biomarkers | 2195 | 0 | 22 | 0 | 1 | 24 | 0 | 219 | 1929 |
| Costs and economics | 199 | 2 | 2 | 6 | 0 | 10 | 0 | 11 | 168 |
| Disparities | 833 | 0 | 16 | 0 | 0 | 0 | 0 | 79 | 738 |
| Caregiving | 834 | 2 | 43 | 13 | 0 | 2 | 0 | 73 | 701 |
| Long-term cohort studies | 542 | 0 | 13 | 0 | 0 | 0 | 0 | 46 | 483 |
| How common it is | 522 | 0 | 21 | 0 | 0 | 12 | 0 | 31 | 458 |
| End-of-life care | 855 | 0 | 7 | 0 | 0 | 0 | 0 | 66 | 782 |
| Cognitive tests and assessment | 881 | 0 | 14 | 0 | 0 | 0 | 0 | 68 | 799 |
| Hippocampus size on brain scans | 492 | 0 | 19 | 0 | 0 | 0 | 0 | 37 | 436 |
| Getting a diagnosis | 815 | 4 | 18 | 2 | 0 | 5 | 0 | 52 | 734 |
| Prevention and risk | 1004 | 1 | 79 | 12 | 1 | 45 | 0 | 25 | 841 |
| Vascular contributions | 546 | 0 | 12 | 0 | 0 | 0 | 0 | 37 | 497 |
| Health systems and care delivery | 654 | 0 | 6 | 0 | 0 | 0 | 0 | 44 | 604 |
| Brain scans and imaging | 1459 | 0 | 10 | 0 | 0 | 0 | 0 | 94 | 1355 |
| Tau biomarkers | 690 | 0 | 8 | 0 | 0 | 0 | 0 | 42 | 640 |
| Risk factors | 2174 | 0 | 70 | 0 | 0 | 0 | 0 | 124 | 1980 |
| Mild cognitive impairment | 701 | 0 | 12 | 0 | 0 | 0 | 0 | 37 | 652 |
| Genetics | 1616 | 0 | 74 | 1 | 0 | 14 | 11 | 49 | 1467 |
| Digital health tools | 1094 | 0 | 16 | 0 | 0 | 0 | 0 | 46 | 1032 |
| Prevention trials | 835 | 0 | 42 | 0 | 0 | 0 | 0 | 30 | 763 |
| Down syndrome and Alzheimer's | 664 | 0 | 1 | 0 | 0 | 0 | 0 | 19 | 644 |
| Clinical safety | 630 | 0 | 24 | 0 | 0 | 0 | 0 | 18 | 588 |
| Brain tissue and pathology | 612 | 0 | 8 | 0 | 0 | 0 | 0 | 17 | 587 |
| Lewy body dementia | 566 | 0 | 0 | 0 | 0 | 0 | 0 | 11 | 555 |
| Treatments | 1489 | 2 | 45 | 3 | 4 | 6 | 1 | 15 | 1413 |
| TREM2 and immune genetics | 701 | 0 | 23 | 0 | 0 | 0 | 0 | 8 | 670 |
| Disease mechanisms | 1381 | 0 | 7 | 0 | 0 | 0 | 0 | 14 | 1360 |
| Drug development | 873 | 0 | 17 | 0 | 0 | 0 | 0 | 7 | 849 |
| Clinical trials | 162 | 0 | 1 | 2 | 17 | 0 | 0 | 1 | 141 |
| Risk genes beyond APOE | 847 | 0 | 7 | 0 | 0 | 0 | 0 | 4 | 836 |
| Inflammation and the immune system | 868 | 0 | 12 | 0 | 0 | 0 | 0 | 3 | 853 |
This is the chart that a score cannot show you. Two subjects can have the same median and rest on completely different kinds of evidence: one on trials that changed something and measured what happened, the other on observation, which can tell you two things travel together but never that one caused the other. Where the bar is grey, the record does not state a design, and the rubric scores it zero rather than guessing.
What this score is not
It is not an official measure of quality. It is alzhub's own editorial rubric. No journal, publisher, funder or ranking body has reviewed it or endorsed it.
It is not only a citation metric. 28 of the 100 points come from citations: how often the study has been cited, how often per year, how that compares with papers of the same field and year, how many studies in this library cite it, and whether it is still being cited. Another 4 come from how much the journal's papers are cited. The other 68 points come from how the study was done, where it was published, who did it and whether a reader can find it. It is not an impact factor. Journals are placed in ordered tiers, never given an impact factor — we could not verify current impact factors from a source we control, and an invented one would be worse than none. The journal's citation record is its h-index as OpenAlex publishes it, stored with the date we read it. If you want the citation counts on their own, they are a separate page.
It is not a verdict on whether a study is right. A low score means we could see less of the things this rubric looks for — not that the finding is wrong. An early-phase trial in a small journal can be the study that matters most, and a 90-point paper can still turn out to be mistaken.
A null result is not a failure. We carry 23 studies whose finding is that something does not work, scoring between 16 and 87. They are in the library on purpose. Someone deciding whether to buy a supplement is served by a rigorous trial showing it does nothing, and our rubric ranks a null result above an early positive finding for exactly that reason — the rigour is already counted by the design and journal components, so demoting a null result would punish good science for reporting an inconvenient answer.
It is not a substitute for reading the study. Every card links to the original. The score is there to help you decide what to open, not to tell you what to think about it.
The full tables, so you can check the arithmetic
These are the lookup values the score uses: one table for each of the 20 components. A study's score is the points it earned out of the points it could be judged on, turned into a number out of 100. For a study we have data on for all 20 parts, that is simply the sum. Where a table says “left out”, that part is not counted for that study (see when we have no data). Anything that is not peer-reviewed is then capped at 16. Nothing else feeds the score.
| Tier | What it is | Points |
|---|---|---|
| Tier 1 | top-tier general journal | 10 |
| Tier 2 | leading specialty journal | 9 |
| Tier 3 | peer-reviewed journal | 7 |
| Tier 4 | not a peer-reviewed journal | 2 |
| Design | Points |
|---|---|
| A systematic review of all the studies on the question | 12 |
| A meta-analysis pooling many studies | 12 |
| A randomised controlled trial | 11 |
| A large phase 3 trial | 11 |
| A mid-stage phase 2 trial | 8 |
| A cohort study following people over time | 7 |
| A prospective cohort study, followed forward from enrollment | 7 |
| A retrospective cohort study, reconstructed from past records | 7 |
| A longitudinal study following the same people over time | 7 |
| A case-control study | 6 |
| A snapshot of people at one point in time | 4 |
| A computer model rather than new measurements | 3 |
| No design recorded | 0 |
| Rating in our summary | Points |
|---|---|
| High | 5 |
| Moderate | 3 |
| Null result | 3 |
| Early | 2 |
| Preliminary | 2 |
| Low | 1 |
| No rating recorded | left out |
| Status | What it means | Points |
|---|---|---|
| Peer-reviewed | A research article published in a peer-reviewed journal: other scientists reviewed it before it was published. | 10 |
| Preprint — not yet peer-reviewed | Posted publicly by its authors before any journal review. The findings may change or may never be published. | 4 |
| Report — not peer-reviewed | An analysis published by an organisation under its own name. Useful, but it did not go through journal peer review. | 3 |
| Conference abstract — not peer-reviewed | A short summary presented at a conference. Conference abstracts are screened, not peer-reviewed, and often contain interim numbers that change before any full paper appears. | 3 |
| Press release — not a study | An announcement written by a company about its own results. There is no independent review and no full dataset to check. | 1 |
| Regulatory decision — not a study | A decision or notice from a regulator (FDA, EMA, CMS, SEC). It is an official action, not a study — the evidence behind it lives in the trials it cites. | 1 |
| Retracted — do not rely on it | The journal has withdrawn this paper, or this entry is the retraction notice itself. Retraction means the findings should not be relied on, so it is scored like a non-peer-reviewed source. | 0 |
| The journal's h-index (OpenAlex) | Points |
|---|---|
| 500 or more | 4 |
| 250 or more | 3 |
| 150 or more | 2 |
| 75 or more | 1 |
| Below 75 | 0 |
| Not a journal article | 0 |
| No journal record | left out |
A journal's h-index is the largest number h such that h of its papers have each been cited at least h times.
| Publication year | Points |
|---|---|
| Within 1 year of the newest study here | 5 |
| Within 3 years of the newest study here | 4 |
| Within 6 years of the newest study here | 3 |
| Within 10 years of the newest study here | 2 |
| Within 20 years of the newest study here | 1 |
| Older than 20 years | 1 |
| No year recorded | left out |
| Participants | Points |
|---|---|
| 10,000 or more participants | 4 |
| 1,000 or more participants | 3 |
| 100 or more participants | 2 |
| 1 or more participants | 1 |
| Participant count not stated | 0 |
| What the record states | Points |
|---|---|
| Effect size with a statistic (interval, p-value, ratio or similar) | 3 |
| Effect size as a number only | 2 |
| No effect size, or one given in words only | 0 |
| What the record states | Points |
|---|---|
| A participant count, or a PubMed type used only for studies in people | 3 |
| The title names an animal or lab model, or nothing shows it was done in people | 0 |
| What the record states about the study itself | Points |
|---|---|
| Randomised | 1 |
| Blinded or placebo-controlled | 1 |
| Neither stated, or a review of other studies | 0 |
The two points add together.
| Times cited (OpenAlex) | Points |
|---|---|
| 1,000 or more | 7 |
| 300 or more | 6 |
| 100 or more | 5 |
| 40 or more | 4 |
| 15 or more | 3 |
| 5 or more | 2 |
| 1 or more | 1 |
| Never cited | 0 |
| No OpenAlex count | left out |
| Citations a year since publication | Points |
|---|---|
| 50 or more a year | 5 |
| 20 or more a year | 4 |
| 8 or more a year | 3 |
| 3 or more a year | 2 |
| 1 or more a year | 1 |
| Fewer than 1 a year | 0 |
| No count | left out |
| Citations against the average for its field and year | Points |
|---|---|
| 40× the average or more | 6 |
| 20× the average or more | 5 |
| 8× the average or more | 4 |
| 3× the average or more | 3 |
| 1× the average or more | 2 |
| Above 0 | 1 |
| Zero | 0 |
| No figure from OpenAlex | left out |
| Studies in this library that cite it | Points |
|---|---|
| 200 or more | 6 |
| 100 or more | 5 |
| 30 or more | 4 |
| 10 or more | 3 |
| 3 or more | 2 |
| 1 or more | 1 |
| None | 0 |
| Not yet checked | left out |
| Citations in the last three calendar years | Points |
|---|---|
| 200 or more | 4 |
| 50 or more | 3 |
| 15 or more | 2 |
| 3 or more | 1 |
| Fewer than 3 | 0 |
| No yearly counts | left out |
| Last-listed author's career h-index (OpenAlex) | Points |
|---|---|
| 100 or more | 4 |
| 50 or more | 3 |
| 20 or more | 2 |
| 5 or more | 1 |
| Below 5 | 0 |
| No record | left out |
An OpenAlex profile holding fewer than 10 works is left out, because new papers are often filed under a split-off profile.
| Strongest institution's h-index (OpenAlex) | Points |
|---|---|
| 2,500 or more | 3 |
| 1,500 or more | 2 |
| 700 or more | 1 |
| Below 700 | 0 |
| None listed | left out |
| Institutions on the paper (OpenAlex) | Points |
|---|---|
| 10 or more | 3 |
| 4 or more | 2 |
| 2 or more | 1 |
| One | 0 |
| Not listed | left out |
A study PubMed indexes as multicentre gets at least 2.
| Where a reader can find it | Points |
|---|---|
| Indexed in PubMed | 2 |
| Not in PubMed, but has a DOI | 1 |
| Neither | 0 |
| Access | Points |
|---|---|
| Free to read | 2 |
| Behind a paywall | 0 |
| Not recorded | left out |
Rubric version 4.0. Recency is measured against 2027, the newest year in the library, and not against today's date, so the same data always produces the same scores. Both the scores and the figures on this page are regenerated from the published data, where you can download both.
Back to the research library → Newest studies first What the library's numbers show