Niacin moved HDL in stroke patients but not their recovery, and then the funding ran out

What was tried

Henry Ford Health System ran a single-site, randomised, quadruple-masked, placebo-controlled Phase 2 trial of extended-release niacin in people recovering from ischemic stroke. Patients entered 72 hours to 7 days after symptom onset with an NIHSS score of 4 to 21. Three arms received niacin 500 mg daily, 1000 mg daily, or placebo for 24 weeks. The primary outcome was participants with a serious adverse event. Functional recovery on the NIHSS was secondary and explicitly exploratory.

The trial opened in April 2009 and reached primary completion in August 2012. Enrollment stopped at 27 participants: 10 on 500 mg, 7 on 1000 mg, and 10 on placebo. The registry states that it stopped early for funding limitations, and that the resulting sample was insufficient for meaningful exploration of functional outcomes. Niacin is ChEMBL CHEMBL573, at maximum clinical phase 4 and marketed for decades as a lipid therapy. New here was the indication and the dosing window.

The biological hypothesis

Niacin binds HCAR2 (Ensembl ENSG00000182782), the high-affinity receptor for nicotinic acid and beta-hydroxybutyrate, which signals through Gi-mediated inhibition of adenylyl cyclase. Open Targets scores the HCAR2 association with ischemic stroke at 0.69197, with genetic association at 0.8482 and animal model evidence at 0.6831.

The rationale came from the institution that ran the trial. Chen and colleagues reported in 2007 that Niaspan increased angiogenesis and improved functional recovery in a rat stroke model, and Shehadah and colleagues reported in 2009 that it promoted arteriogenesis. The hypothesis was restorative rather than protective: dose days after the infarct and improve the repair phase. It rode on a second assumption, that raising HDL cholesterol was itself the useful lever. In 2009 that was mainstream.

What actually happened

Baseline NIHSS was 8.8 with a standard deviation of 4.4 on 500 mg, 9.7 with 3.0 on 1000 mg, and 10.4 with 6.8 on placebo. At 24 weeks the posted values were 2.1 with 2.4, 3.7 with 3.0, and 4.3 with 5.4. Every arm improved substantially, which is what subacute stroke cohorts do untreated.

On the primary safety outcome, 2 of 10 participants on 500 mg, 1 of 7 on 1000 mg and 1 of 10 on placebo had a serious adverse event. One death occurred in the 1000 mg arm and one in placebo, and the two serious events in the 500 mg arm were coded as stroke.

The investigators presented the trial at the International Stroke Conference in 2018. That abstract reports an HDL cholesterol increase of 12.5 mg/dL with p equal to 0.035, NIHSS improvement of 6.5 points on niacin against 6.1 on placebo with p equal to 0.845, and modified Rankin Scale of 2.1 against 2.0 with p equal to 0.860. The drug did what it was known to do to lipids, and nothing detectable to function, in a trial the sponsor says was too small to answer the functional question.

Failure mechanism, best guess

The archetype is sponsor_decision. The registry reason is funding, and the sponsor states that the sample could not support the functional analysis. Nothing in the posted data licenses a claim that HCAR2 agonism fails in stroke recovery, because the trial was never in a position to test it.

What the record does establish is that the pharmacology worked. HDL rose significantly on drug, so this was not a failure to dose or to engage the lipid axis. That narrows the explanations to two: either the rat effect does not transfer at these doses in these patients, or it does and 27 people could not see it.

Context matters here. AIM-HIGH reported in 2011 that adding niacin to intensive statin therapy produced no cardiovascular benefit, and HPS2-THRIVE reported in 2014 that extended-release niacin with laropiprant produced no benefit and meaningful harm in 25,673 high-risk patients. The HDL-raising rationale that made niacin attractive collapsed during and just after this trial's enrollment window, and a programme whose premise had become unfashionable was a poor candidate for renewed grant support.

How to prevent this next time

The posted data is enough to size a successor honestly. Pooling the three posted 24-week NIHSS standard deviations of 2.4, 3.0 and 5.4 across arms of 10, 7 and 10 gives a pooled standard deviation of 3.92 on 24 degrees of freedom. For a two-arm comparison at 80% power and two-sided alpha of 0.05, detecting a 2-point NIHSS difference requires about 61 patients per arm, or 122 in total. The trial randomised 27 across three arms, so the shortfall was structural from the first patient rather than a consequence of stopping early.

Two design levers follow. A restorative trial should carry a pharmacodynamic gate that is not the clinical endpoint, and a marker closer to the mechanism, such as a circulating angiogenic readout, would have separated a mechanism that did not engage from one that engaged and did not help. Base-rate adjustment is the second lever and stays qualitative here, because no sourced prior exists for HCAR2 in stroke recovery. A programme built on rodent data from its own laboratory should treat internal replication as weaker than external validation.

The single highest leverage change would have been sizing and funding the trial to the 122-patient target implied by its own variance before enrolling anyone, rather than opening a three-arm study powered for nothing and discovering the shortfall at termination.

What this means for similar programs

HCAR2 is uncrowded. Open Targets lists a single clinical candidate against it, SCH-900271, which reached Phase 2. Little accumulated failure evidence means a well-powered trial would still be informative. The Claidex mechanism risk score for HCAR2 after this entry is 20 and sits in the green band, with recency contributing only 0.69.

Interest in the receptor has moved past lipids. A 2022 report described HCAR2 modulating microglial response in a mouse Alzheimer model, and a 2025 report described niacin enhancing hematoma clearance and neurological recovery through an HCAR2, SIRT1 and Nrf2 pathway. Programmes reviving niacin or a selective HCAR2 agonist for a nervous system indication should be explicit that they are not relying on the lipid rationale, which was tested at scale and failed. For any team reading an old terminated entry as a negative result, this record is a caution. The numbers look like a null. The sponsor says they are not one.

Open questions

Which pharmacodynamic marker would have distinguished restorative engagement from lipid engagement, and was any such marker collected but not posted?

Would a selective HCAR2 agonist be a better probe of the neurorestorative hypothesis than repurposing Niaspan?

Sources

1. ClinicalTrials.gov record and posted results, NCT00796887, retrieved 31 August 2026. https://clinicaltrials.gov/study/NCT00796887 Russman AN, et al. Abstract WP156, Phase IIa study of extended-release niacin for stroke recovery. Stroke.

2. Chen J, et al. Ann Neurol.

3. Shehadah A, et al. J Cereb Blood Flow Metab.

4. AIM-HIGH Investigators. N Engl J Med.

5. HPS2-THRIVE Collaborative Group. N Engl J Med.

6. Moutinho M, et al. Sci Transl Med.

7. Redox Biol.

8. Open Targets Platform, HCAR2 and ischemic stroke, retrieved 31 August 2026. https://platform.opentargets.org/target/ENSG00000182782 ChEMBL CHEMBL573, retrieved 31 August 2026. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL573/ openFDA FAERS drug event endpoint, niacin, queried 31 August 2026. https://open.fda.gov/apis/drug/event/.