The research library over time, by year and category

Studies per year, 1932–2027

N = 27,376 studies with a publication year (4 more have none and are not plotted).

Read this before the charts

The jump at the right-hand end is not a research boom. Our library was assembled recently and leans toward new papers: it holds 2,086 studies from 2025 and 1 from 2027, against roughly 500–580 in each of 2019–2024. That is when we collected, not how much was published. The shaded band on every chart marks 2025 onward. Use the years before it for trends.

Data as of 2026-11-01 (newest study date in the library); library file last written 2026-11-03.

Studies published that year
2025 onward: library-collection spike

Reading: the library holds a few dozen studies a year before 1990, climbs to about 1,368 a year by 2021, then jumps to 3,835 in 2025 and 1 in 2027. The jump is when the library was built, not a research boom.

Show the numbers

Year	Studies
1932	1
1935	1
1958	1
1960	1
1961	1
1963	2
1964	1
1967	1
1968	2
1969	1
1970	1
1972	1
1974	2
1975	3
1976	2
1977	7
1978	4
1979	4
1980	8
1981	5
1982	16
1983	17
1984	15
1985	22
1986	33
1987	41
1988	57
1989	54
1990	84
1991	81
1992	88
1993	107
1994	125
1995	139
1996	184
1997	218
1998	200
1999	232
2000	306
2001	314
2002	346
2003	352
2004	442
2005	452
2006	524
2007	518
2008	558
2009	604
2010	700
2011	632
2012	716
2013	806
2014	781
2015	946
2016	892
2017	1,065
2018	1,133
2019	1,217
2020	1,316
2021	1,368
2022	1,345
2023	1,145
2024	1,214
2025	3,835
2026	2,086
2027	1

Mean study score per year, 1985–2027

N = 27,278 studies, in years with at least 20 studies. 24 earlier years (98 studies) have too few for a stable average and are not drawn.

Average score of that year’s studies
2025 onward: library-collection spike

Reading: the mean score moves between 33.6 and 51.5 points across 1985-2027; it was 44.7 in 1985 and 33.6 in 2026. Years before 1985 have fewer than 20 studies each, so no mean is drawn. The score is our own rubric (how we score), a measure of how much evidence a study gives, not of whether it is right.

Show the numbers

Year	Studies	Mean score
1985	22	44.7
1986	33	45.4
1987	41	45.2
1988	57	44.1
1989	54	41.6
1990	84	44.1
1991	81	44.7
1992	88	46.5
1993	107	44.8
1994	125	45.2
1995	139	44.5
1996	184	45.2
1997	218	45.7
1998	200	46.5
1999	232	45.4
2000	306	47.2
2001	314	46.6
2002	346	46.2
2003	352	48.1
2004	442	47.4
2005	452	47.7
2006	524	47.9
2007	518	48.4
2008	558	48.8
2009	604	49.2
2010	700	48.5
2011	632	49.4
2012	716	48.9
2013	806	49.5
2014	781	49.1
2015	946	49.0
2016	892	49.4
2017	1,065	50.6
2018	1,133	50.6
2019	1,217	50.3
2020	1,316	50.4
2021	1,368	51.5
2022	1,345	51.1
2023	1,145	51.0
2024	1,214	51.3
2025	3,835	45.2
2026	2,086	33.6

Predictor categories, year by year

You asked for the categories that predict Alzheimer’s. Four of the library’s categories are about that: risk factors, prevention, genetics and biomarkers (tagged below), and they are also four of its six largest. For comparison we add the next two largest, treatments and diagnosis. Every panel uses the same vertical scale (0–300), so a taller line really is more studies. Panels start in 1990, because earlier years hold a handful of studies. Narrower subjects such as risk genes beyond APOE and tau biomarkers are separate categories and are not added into these.

Risk factorspredictor

N = 2,142 studies, 1990–2027 (2,174 in all years).

Peak before the collection spike: 165 in 2024; 258 in 2025 and 0 in 2027.

Preventionpredictor

N = 999 studies, 1990–2027 (1,004 in all years).

Peak before the collection spike: 58 in 2018; 208 in 2025 and 0 in 2027.

Geneticspredictor

N = 1,614 studies, 1990–2027 (1,615 in all years).

Peak before the collection spike: 86 in 2024; 170 in 2025 and 0 in 2027.

Biomarkerspredictor

N = 1,780 studies, 1990–2027 (1,798 in all years).

Peak before the collection spike: 182 in 2024; 242 in 2025 and 0 in 2027.

Treatments

N = 1,486 studies, 1990–2027 (1,489 in all years).

Peak before the collection spike: 103 in 2024; 215 in 2025 and 0 in 2027.

Diagnosis

N = 791 studies, 1990–2027 (815 in all years).

Peak before the collection spike: 33 in 2011; 180 in 2025 and 0 in 2027.

Average score by category and period

Study count (N) and mean score. A dash means fewer than 20 studies, too few for a stable average.

Category	2000–2009	2010–2019	2020–2026
Risk factors	252 · 44.5	634 · 46.4	1,173 · 44.2
Prevention	147 · 53.7	386 · 56.8	446 · 53.4
Genetics	251 · 48.6	553 · 51.1	723 · 44.1
Biomarkers	90 · 41.0	435 · 47.5	1,211 · 50.9
Treatments	281 · 55.2	412 · 54.5	745 · 48.3
Diagnosis	190 · 57.5	248 · 57.5	282 · 45.9

Reading: each cell is “studies · mean score”. The 2020–2026 column includes the collection spike.

What the library is made of

N = 27,380 studies in 32 categories, all years.

Predictor category
Other category

Reading: 32 categories. The four predictor categories (dark bars) hold 6,591 of 27,380 studies (24%); a study carries one category, and some are narrower parts of a topic (risk genes beyond APOE, TREM2 and immune genetics, tau biomarkers).

Show the numbers

Category	Studies
Risk factors	2,174
Biomarkers	1,798
Genetics	1,615
Treatments	1,489
Brain scans and imaging	1,459
Disease mechanisms	1,380
Digital health tools	1,094
Prevention	1,004
Cognitive tests and assessment	881
Drug development	873
Inflammation and the immune system	868
End-of-life care	855
Risk genes beyond APOE	847
Prevention trials	835
Caregiving	833
Disparities	833
Diagnosis	815
TREM2 and immune genetics	701
Mild cognitive impairment	701
Tau biomarkers	690
Down syndrome and Alzheimer's	664
Health systems and care delivery	654
Clinical safety	630
Brain tissue and pathology	612
Lewy body dementia	566
Vascular contributions	546
Long-term cohort studies	542
How common it is	522
Hippocampus size on brain scans	492
Costs and economics	199
Clinical trials	162
Laboratory science	46

How these were checked

Before drawing, the page checks its own sums: the per-year counts add to 27,376 (27,380 studies minus 4 with no year), and the category counts add to 27,380. If a sum ever failed, the page would refuse to draw rather than show a wrong chart. Nothing is estimated or filled in. This describes our library, which is not every Alzheimer’s paper ever written.