Baricitinib for COVID-19, named by a knowledge-graph search and confirmed in randomised trials
A search of BenevolentAI's machine-learning-built biomedical knowledge graph named the rheumatoid arthritis drug baricitinib as a candidate COVID-19 treatment within weeks of the outbreak; randomised trials run by groups unconnected to BenevolentAI later found it lowered 28-day mortality in hospitalised patients, from 14% to 12% among 8,156 patients in RECOVERY.
- Lab
- BenevolentAI
- Model
- BenevolentAI biomedical knowledge graph with machine-learning literature extraction
- Field
- Medicine
- Date
- 2020-02-04
- Human collaborators
- Peter Richardson, Justin Stebbing
Sources
Original work
What was found
The Lancet letter describes the knowledge graph as a repository of structured medical information including connections extracted from the literature by machine learning, queried by algorithms with customisations for the new virus. The team looked for approved drugs that might block viral entry through AAK1, a regulator of the endocytosis the virus was expected to use. Of 378 AAK1 inhibitors in the graph, 47 were approved drugs and six bound AAK1 with high affinity; the oncology drugs among them needed doses the authors judged unsafe, while baricitinib reaches AAK1-inhibiting plasma concentrations at its normal 2 mg or 4 mg dose and also inhibits JAK1/2, so the letter proposed it to reduce both viral entry and inflammation. The company's later account calls the method human-guided iterative queries of the graph. Three randomised trials followed: NIAID's ACTT-2 (1,033 patients) found baricitinib plus remdesivir shortened median recovery from 8 to 7 days; Eli Lilly's COV-BARRIER (1,525 patients) missed its primary endpoint but found 28-day mortality of 8% against 13% on placebo; and the UK RECOVERY platform trial (8,156 patients) found 12% against 14%, an age-adjusted rate ratio of 0.87.
Novelty check
Europe PMC, baricitinib AND (coronavirus OR 2019-nCoV OR COVID*), first published up to 10 February 2020: the Lancet letter of 4 February is the earliest record naming baricitinib for the disease; the other early hits are general repurposing reviews. JAK inhibitors as a class were soon tested for COVID-19, and RECOVERY's meta-analysis of nine trials pools baricitinib with other JAK inhibitors (a 20% mortality reduction), so the benefit belongs to the class as much as to this pick; the specific early proposal of baricitinib was this group's.
Caveats and known objections
The AI's part was retrieval and ranking in a graph that humans queried iteratively; the mechanistic argument and the decision to publish were human, which is why autonomy is ai-assisted. The letter's case rested on blocking viral entry through AAK1, but the clinical benefit is usually attributed to baricitinib's anti-inflammatory JAK inhibition, and no trial isolated an antiviral effect. COV-BARRIER was sponsored by the drug's maker and missed its primary endpoint, with mortality a secondary outcome. RECOVERY's 13% proportional reduction was smaller than earlier, smaller trials had suggested.
Independent checks
Kalil et al., ACTT-2 Study Group, NIAID (New England Journal of Medicine, 2021): Double-blind trial in 1,033 hospitalised adults: baricitinib plus remdesivir shortened median time to recovery to 7 days from 8 with remdesivir alone (rate ratio 1.16), with fewer serious adverse events. · link ↗
RECOVERY Collaborative Group, University of Oxford (The Lancet, 2022): Open-label platform trial in 8,156 hospitalised patients, 95% already on corticosteroids: 28-day mortality 12% with baricitinib against 14% with usual care, age-adjusted rate ratio 0.87 (95% CI 0.77 to 0.99). · link ↗
Disagree with these grades?
Bring a citation: a grade moves on evidence, not on argument.
Or on GitHub: submit a check challenge the grade send a correction or send a pull request
Flag this for triage
Signals order the review queue and nothing else. They are never published, and they never move a grade: that takes a citation.
Entry history (1 event)
- AddedEntered the registry graded Independently checked and AI-assisted.
Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.
Graded independently checked for verification and ai-assisted for autonomy. What these mean.
Cite this entry
whataifound.org. (2020). Baricitinib for COVID-19, named by a knowledge-graph search and confirmed in randomised trials. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2020-02-04-baricitinib-covid19BibTeX
@misc{whataifound-benevolentai-2020-covid19,
title = {Baricitinib for COVID-19, named by a knowledge-graph search and confirmed in randomised trials},
author = {{whataifound.org}},
year = {2020},
howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
note = {Result by BenevolentAI. Verification: Independently checked. Autonomy: AI-assisted.},
url = {https://whataifound.org/finding/2020-02-04-baricitinib-covid19}
}