Summary Evidence map Program selection Companion vs laboratory Methods Limitations Verify this work

Systematic review

Brief survey of animal-to-human concordance in the biomedical literature

We reviewed studies directly comparing a finding in live animals against the same finding in humans, then collected the evidence.

Summary

Each row is one finding: a sentence in which a paper compared an animal with a human. A: treatment outcomes, B: toxicity and safety, C: disease biology.

Findings by evidence level
Evidence level	Studies	Findings	Concordant	Discordant	Mixed
A — Intervention outcomes	162	366	233	103	30
B — Toxicity and safety	20	80	48	18	14
C — Disease biology	259	747	516	158	69

Caveats

Dominated by evidence level C — how similar the biology looks, not what happened when a disease was treated.

Quantities reported across multiple studies are not pooled, in general.

Publications generally biased toward positive findings.

Evidence map

Studies per disease area and species; the superscript is the highest level in that cell. Companion and laboratory dogs and cats are counted separately. The grid holds the 992 results that name a species. The other 201 are not on it: they state a finding about animal models as a class rather than about a named species, so there is no species to place.

Disease area	mouse	rat	guinea pig	rabbit	pig	sheep/goat	non-human primate	laboratory dog	companion dog	companion cat	horse	zebrafish	All
oncology	46^A	2^C			1^C		1^B	3^A	35^A	1^C		4^A	88
neurology	43^A	8^A	1^C		3^C	1^C	7^A		4^A			1^C	60
immunology/inflammation	38^A	3^A	1^C			1^C	3^A	2^C	3^C			1^C	48
cross-cutting toxicology	14^A	5^A		2^A	3^C	1^A	7^A	3^A				5^A	25
cardiovascular	12^A	3^A	2^A		6^A	1^C	4^A	8^A				4^A	36
liver/GI	21^A	5^A	1^C		1^C		1^B	1^A	1^C				27
infectious disease	12^A	1^C		1^C	2^A		4^A						19
pain/musculoskeletal	10^A	8^A				1^C		1^C	2^A		2^C		19
psychiatry/addiction	11^A	2^A					4^A						16
metabolic/endocrine	10^A	2^A			3^C	1^C	1^C					1^C	15
ophthalmology	2^C						1^C	4^A	1^C			1^C	6
respiratory	6^A	1^C											7
reproductive/developmental	1^C	2^C											3
hematology	1^C						1^C						2
renal	3^C			1^B									4
dermatology	2^C												2
other	10^C	1^A					1^C	1^C				2^C	14
all	231	41	5	4	17	6	33	23	46	1	2	17	419

Fewer studiesMore (46 at most)· Superscript = highest evidence level in the cell (click a cell to see those findings)

Browse all 1,193 results →

Choosing Nori's first program

How can this collection of concordance evidence help us select our first candidate? We collected pairs of drugs with similar mechanisms targeted toward both human and companion animals, then asked whether the clinical outcomes pointed in the same direction.

Concordance counts only pairs that could be called either way. Mixed and indeterminate pairs are shown but excluded from the percentage.

Primary (drugs, biologics, vaccines, cell/gene therapies)

82% (76%–87%)
of 185 that could be called

586
pairs

Primary: disease-modifying use

83% (75%–89%)
of 116 that could be called

407
pairs

Primary: symptomatic or preventive use

81% (70%–90%)
of 69 that could be called

179
pairs

Primary: human approval before the veterinary evidence

83% (76%–89%)
of 149 that could be called

440
pairs

Primary: human approval after the veterinary evidence

100% (59%–100%)
of 7 that could be called

19
pairs

Primary: no US human approval

72% (53%–87%)
of 29 that could be called

127
pairs

Primary, disease-modifying: dog

82% (72%–90%)
of 79 that could be called

316
pairs

Primary, disease-modifying: cat

79% (58%–93%)
of 24 that could be called

63
pairs

Excluded from primary: supplements, minerals, devices, regimens, non-treatments

73% (59%–84%)
of 52 that could be called

201
pairs

Class analogues (separate)

75% (19%–99%)
of 4 that could be called

15
pairs

ConcordantDiscordantMixedIndeterminate

A second model re-judged 40 random primary pairs blind: 88% agreement, Cohen's kappa 0.764.

How to read this

Most are human medicines adopted later in veterinary practice, so agreement records adoption, not prediction.

"Discordant" means all evidence points the opposite way.

Browse all pairs →

We found that 477 of 799 drug pairs were approved for humans first, 22 the reverse, and a median gap 30 years between human drug approval and vet drug approval.

Filtering 1,949 human programs on where dog biology corresponds leaves 54 candidates.

Program selection →

Companion animals vs laboratory models

Where companion animals and laboratory models agreed with the human outcome, across the 108 pairs with a clear answer on all three sides

	Laboratory matched	Laboratory did not
Companion matched	86	4
Companion did not	14	4

83%

companion animals matched humans

93%

laboratory models matched humans

pairs where only one agreed

Of those, companion animals were right in 4 (22%, 95% CI 6%–48%).

where the two disagreed, laboratory models matched the human outcome more often than companion animals.

Where the comparison is informative

Subset	Pairs	Companion matched	Laboratory matched
human result positive	104	89/104 (86%)	100/104 (96%)
human result negative tests prediction	4	1/4 (25%)	0/4 (0%)
approved in humans before the veterinary evidence	90	78/90 (87%)	84/90 (93%)
approved in humans after the veterinary evidence	2	2/2 (100%)	2/2 (100%)
no US human approval	16	10/16 (62%)	14/16 (88%)

Counting a laboratory side only when every study agrees: 65 pairs, companion 85%, laboratory 94%. concordance with human outcomes did not differ detectably between companion animals with naturally occurring disease and laboratory models of the same drug and condition.

Laboratory side read from abstracts by gemini-2.5-flash, audited 79% correct (42 of 53 reads).

Extreme class imbalance

104 of 108 pairs have a positive human result and the laboratory literature is almost uniformly positive, so agreement is near-automatic.

On the 4 negative-human pairs, laboratory models matched 0 and companion animals 1.

Methods

Eligible: a finding in live animals reported alongside the human one. Two independent retrieval mechanisms, two screening stages, extraction from open-access full text with sentence and page, verification by a second model, then adjudication of every result — including those verification rejected. Counts are computed from the records, not summarized by a model.

eligible studies with a final result: 363 (seeds excluded: 37; not from expanded retrieval: 0)
found by citation chasing: 167; by queries: 191; by both: 32
estimated eligible studies reachable (Chapman): 976 (95% CI 708–1244)
estimated recall of the union: 33% (95% CI 26%–46%)

Assumes the two mechanisms find eligible studies independently; positive dependence makes this an upper bound on recall.

Of results verification accepted, 34% were later dropped or corrected; of those it rejected, 10% were reinstated. 602 of 1,193 had a single unblinded adjudicator, oversampled by the spot-check.

Limitations

~33% coverage (CI 26–46%), itself an upper bound.
Open-access full text only; veterinary side largely from abstracts.
Findings within a paper are not independent of one another.
Quantities grouped only where metric and unit match.
Selective publishing inflates apparent agreement.
Agreement is not prediction.
Adjudication partly single-reviewer and unblinded.
201 results are kept off the evidence map: nearly all carry a grouped label ("animals", "rodents") rather than a species. A minority record the human side of the comparison in the animal field. Results naming several species now count toward each.

All 99 limitations, the protocol, data and code: repository.

Verify this work

Forty seeded results, each with its source, page and sentence. Open the spot-check →