Kānuka honey and rosacea: reading the New Zealand trial properly

Published 6 August 2026Zane Hitchcox, editor

Most claims made for a New Zealand botanical rest on a laboratory assay and a leap to a human face. This one does not: there is a registered randomised trial in real patients with a pre-specified primary outcome. That makes it worth reading properly, which means reading the parts that are less flattering than the abstract.

Kānuka is not mānuka, and marketing constantly conflates them

The two plants look alike and are routinely swapped in marketing copy, but sit in different genera: mānuka is Leptospermum scoparium, kānuka is Kunzea ericoides and its relatives. The Department of Conservation gives the field test: mānuka has sharp-tipped leaves, short papery bark and large solitary flowers, while kānuka has soft leaves without the sharp tip, long leathery bark strips and flowers in clusters.¹¹ Kānuka was moved out of Leptospermum in 1983, and a 2014 revision in PhytoKeys split what had been treated as one species into ten New Zealand endemics, seven of them new to science.¹⁰ "Kānuka honey" is therefore a floral description covering a group of related plants, not a single-species product.

Methylglyoxal is the marker compound used to grade and authenticate mānuka honey,¹⁴ but it is not unique to it: a 2012 research letter reported that New Zealand kānuka honey also carries high levels of it.¹⁵ What kānuka lacks is not the compound but the grading scheme. No UMF-style number appears on a kānuka jar, and the markers that separate the two honeys in a laboratory — 4-methoxyphenyllactic acid, methyl syringate, p-anisic acid and lumichrome on the kānuka side — live in analytical chemistry journals, not on labels.¹⁶ The trial below used kānuka honey, so it is not evidence about mānuka; and the far larger body of honey wound-care research — 26 trials and 3,011 participants in Cochrane's 2015 review, which found honey heals partial-thickness burns faster than conventional dressings but declined to draw a general conclusion — is not evidence about faces.¹² The mānuka honey evidence is set out separately, and it points somewhere quite different.

What rosacea is, at a reference level

Rosacea is a persistent inflammatory condition of the central face which DermNet describes as usually presenting after about the age of 30 and becoming more common with age, and which runs a relapsing-remitting course. A 2018 systematic review and meta-analysis in the British Journal of Dermatology, pooling 32 studies across 41 populations and 26.5 million people, put the prevalence at 5.46% of the general population (95% CI 4.91–6.04), concentrated in those aged 45 to 60.⁸ DermNet, New Zealand's own dermatology reference, describes the recognised phenotypes: erythematotelangiectatic (persistent redness and visible vessels), papulopustular (inflammatory papules and pustules), phymatous (tissue thickening, most familiarly of the nose), and ocular, which it puts at more than half of people with rosacea and which can progress to keratitis and lasting damage to sight. The pathogenesis is multifactorial, involving genetic susceptibility, altered skin microbiome, neurocutaneous mechanisms, barrier impairment and immune dysregulation.⁹

Two facts follow. Rosacea fluctuates on its own, so an uncontrolled before-and-after observation is close to worthless and only a randomised comparison means anything. And there are already treatments with strong evidence behind them: Cochrane's 2015 review, covering 106 randomised trials and 13,631 participants, rated topical azelaic acid and topical ivermectin against placebo or vehicle as high-certainty evidence and topical metronidazole as moderate.⁷ Any new candidate is judged against that background, not against nothing.

The trial, line by line

Braithwaite and colleagues at the Medical Research Institute of New Zealand randomised 138 adults aged 16 and over with a doctor-diagnosed facial rosacea and a baseline severity score of at least 2, recruited at five New Zealand outpatient primary-healthcare sites; 137 were in the primary analysis.¹ One arm received Honevo, a preparation of 90% medical-grade kānuka honey with 10% glycerine, the other Cetomacrogol, a liquid paraffin and white soft paraffin emollient. Participants applied the cream twice daily, leaving it on for 30 to 60 minutes per application and rinsing with warm water, for eight weeks. The trial was registered in advance on the Australian and New Zealand Clinical Trials Registry as ACTRN12614000004662.

The primary outcome was the proportion of participants achieving an improvement of at least two points on the seven-point Investigator Global Assessment of Rosacea Severity Score (IGA-RSS) at week 8 relative to baseline — a scale running from 0, clear, to 6, severe, integrating papules and pustules, inflammatory lesions, erythema and telangiectasia. The result: 24 of 68 (34.3%) in the honey arm and 12 of 69 (17.4%) in the control arm, a relative risk of 2.03 (95% CI 1.11 to 3.72), p=0.020.¹

The secondary outcomes are mixed. A subject-rated visual analogue scale of change favoured honey by 12.3 points at week 8 — but that measure comes from participants who knew which arm they were in, so it carries the least weight in the paper. The Dermatology Life Quality Index, a validated instrument, showed no significant difference at week 2 or week 8, and neither did the weekly diary-captured severity scale. Adverse events were comparable: 23 participants reported 31 events on honey, 27 reported 36 on control. Withdrawals ran 7 of 68 (10.3%) with honey against 15 of 69 (21.7%) with the emollient — a greater-than-anticipated dropout rate the authors flag as a limitation, though they also argue it cuts against themselves rather than for them: withdrawals specifically for worsening rosacea ran 12% in the control arm against 4% on honey, and those participants stopped contributing outcome data, which would tend to understate the honey's advantage.¹

Four reasons the headline is weaker than it sounds

The blinding could not work

The paper is direct about it: "Owing to the nature of Honevo, it was not possible to blind the participants and primary investigators to the treatment allocation."¹ Honey looks and smells like honey. What the investigators did instead was sensible: an independent investigator at each site stayed blinded and did the IGA-RSS scoring and nothing else. That protects the primary outcome while leaving every participant-reported outcome exposed, which the authors concede as a detection bias risk in their own limitations.

The comparator was a moisturiser

Cetomacrogol was chosen, the authors write, "as it is a non-ionic moisturising cream, often used as a vehicle for delivery of topical medications."¹ That is defensible for a first trial — it isolates the honey from the act of applying something — but it is not the question a patient has. A patient wants to know whether kānuka honey is as good as azelaic acid, metronidazole or ivermectin, the agents Cochrane rates at moderate to high certainty.⁷ The trial does not address that, and the authors say as much, recommending future trials against topical metronidazole or azelaic cream.

The paper does attempt an indirect comparison, and its own hedging is the useful part. In placebo-controlled trials, it notes, topical metronidazole produced a relative risk of improvement of 1.95 (95% CI 1.5 to 2.6) and azelaic cream 1.52 (1.3 to 1.8), against the 2.03 (1.11 to 3.72) seen here — figures that look flattering to the honey until the authors add that the participants and the outcome measures in those studies were not the same, which makes the comparison uninterpretable as a ranking.¹ Beating an emollient establishes that a preparation does something; it does not place it anywhere in a treatment order.

The estimate is imprecise and the follow-up short

The trial was powered on an assumed 25% response in the control arm against 50% on honey — a 25 percentage-point gap. What it observed was 17.4% against 34.3%, 16.9 points, smaller than the effect it was designed to detect. The confidence interval on the relative risk, 1.11 to 3.72, is consistent with an effect that is barely there and with one that is large. Arithmetic on the published proportions puts the number needed to treat at roughly six for one extra two-point improvement, but that figure inherits the same imprecision and should not be quoted as settled. The most quotable number in the paper is its least reliable: a post hoc analysis found complete resolution in 13% on honey against 3% on control, a relative risk of 4.6 — confidence interval 1.0 to 20.4, and not pre-specified. The authors say themselves it "could be consistent with quite a small effect."¹ Eight weeks is short for a relapsing-remitting condition: nothing here speaks to whether the benefit persists at six months, or what happens on stopping. The authors also note that no validated tool for grading rosacea severity exists — a limitation of the field, not of this trial alone.

The manufacturer paid for it

The funding statement reads: "This study was funded by HoneyLab. HoneyLab provided the Honevo (medical-grade kanuka honey and 10% glycerine) for the study." Competing interests are declared as "None declared."¹ Both facts belong in any honest summary, and the second is worth noticing on its own: a nil competing interests declaration alongside full sponsorship by the product's maker is a narrow reading of the convention. The programme's two pilot papers disclose something this one does not — that HoneyLab's medical director had previously been programme director of complementary medicine at the Medical Research Institute of New Zealand, the institute that ran every trial in the series.³⁴

Industry funding is not, by itself, a reason to discard a randomised trial. The features that protect a result — prospective registration, a pre-specified primary outcome, randomisation, a blinded assessor — are all present here. The stronger argument in the sponsor's favour is downstream: the same funder paid for kānuka honey trials in eczema, psoriasis, acne and cold sores, and every one of those null or equivocal results was published. That is not a programme burying its failures.

The companion trials are the real test, and they are not positive

A single positive trial is easy to over-read; the useful check is what the same group found when it pointed the same product at other skin conditions. The acne trial is the closest companion: Semprini and colleagues randomised 136 people aged 16 to 40 with acne across three New Zealand primary-care localities, where both arms washed with an antibacterial soap twice daily for 12 weeks and one arm additionally applied the same 90% kānuka honey preparation.² The primary outcome — a two-point or greater improvement in Investigator's Global Assessment at week 12 — was reached by 4 of 53 (7.6%) on honey and 1 of 53 (1.9%) on soap alone, an odds ratio of 4.2 (95% CI 0.5 to 39.3), p=0.17. Note the denominators: 136 randomised, 106 analysed, with withdrawals of 22.1% and 20.6%. The authors' conclusion is unambiguous: the trial "did not find evidence that addition of medical-grade kanuka honey in combination with 10% glycerine to standard antibacterial soap treatment is more effective than the use of antibacterial soap alone in the treatment of acne."

Two earlier pilots by Fingleton and colleagues in JRSM Open were smaller still, and differed from the rosacea trial in ways worth stating. Each enrolled 15 adults; each was a within-participant comparison in which plain medical-grade kānuka honey went on a lesion on one limb and aqueous cream BP on the matching lesion on the other, side chosen by coin toss, under an overnight dressing for a fortnight. Not a face, not eight weeks, not the 90%-honey-with-10%-glycerine preparation. In eczema, the blinded change in the lesion component of SCORAD differed by −0.1 (−1.5 to 1.4), p=0.92, and the authors concluded that topical medical-grade kānuka honey "does not appear to be effective in the management of eczema."³ In psoriasis, the blinded PASI intensity component fell 4.2 to 2.9 under honey and 4.0 to 3.2 under aqueous cream — changes of −1.3 and −0.8, a difference of −0.5, p=0.25; the authors said the honey "may have similar efficacy to aqueous cream" and that this required confirmation in a suitably powered study.⁴ At n=15 over two weeks, neither could have detected a modest effect. They are feasibility studies, and should be read as such.

The largest trial in the programme was in cold sores. Semprini and colleagues randomised 952 adults presenting within 72 hours of a herpes simplex labialis episode at 76 New Zealand community pharmacies between September 2015 and December 2017, comparing 5% aciclovir cream against the same 90% preparation, both applied five times daily; 852 were analysed.⁵ Median time to return to normal skin was 8 days (95% CI 8 to 9) for aciclovir and 9 days (8 to 9) for honey; hazard ratio 1.06 (0.92 to 1.22), p=0.56. The conclusion was that there was "no evidence of a difference in efficacy" between the two. It is important not to over-read this in either direction: it was designed as a superiority trial and failed to show superiority, which is not the same as demonstrating equivalence, and there was no untreated arm, so it does not establish that either treatment beat doing nothing.

Has anyone repeated the rosacea result?

Not as far as the published record shows, and the treatment the trial received from the next systematic review is instructive. van Zuuren and colleagues, updating the rosacea evidence in the British Journal of Dermatology in 2019 across 152 randomised trials and 20,944 participants with searches to 6 March 2018, put kānuka honey in a list of interventions — alongside permethrin, dapsone, tranexamic acid and praziquantel ointment — that "were at unclear-to-high risk of bias, inadequately reported or provided very limited data," and handled it in an appendix rather than the summary-of-findings tables. No GRADE certainty rating was assigned to it in the main text.⁶

Searching PubMed for kānuka honey in August 2026 returns no second randomised controlled trial in rosacea. Eleven years after a positive primary outcome in a registered trial, the result stands alone — the fact least likely to appear in a product description.

The proposed mechanism is borrowed, not demonstrated

The discussion offers candidates, not findings. It concedes that "the mechanism of action was not assessed in this study," then cites prior work on anti-inflammatory effects "including inhibition of neutrophil superoxide production, reduction in inflammatory leucocyte infiltration and arachidonic-induced oedema," and raises the proposed role of Bacillus oleronius — a bacterium associated with the Demodex mite — in rosacea inflammation, before stating that "the effect of honey on B. oleronius and the Demodex folliculorum mite requires further investigation."¹ A plausible mechanism has never been a substitute for an outcome.

What the label is allowed to say about it

The gap between "a trial found this" and "this product may say so" is a legal one. Medsafe's guidance on product categorisation states that therapeutic claims are not permitted for products supplied as cosmetics, and that a product used "wholly or principally" for a therapeutic purpose is a medicine; the Medicines Act 1981 defines a medicine at section 3 and therapeutic purpose at section 4, the latter covering "preventing, diagnosing, monitoring, alleviating, treating, curing, or compensating for, a disease, ailment, defect, or injury" as well as "influencing, inhibiting, or modifying a physiological process."¹³ Medsafe adds that the purpose is read off labelling, advertising, websites, testimonials and expected use, not only off the carton.

So the same jar can be sold two ways: as a cosmetic it may speak about appearance and comfort, and the moment it claims to treat rosacea it is in the medicines regime. The rest of that architecture — cosmetics administered by the Environmental Protection Authority under a hazardous-substances statute, Medsafe relevant only once the claim turns therapeutic — is set out in who actually regulates cosmetics here. One note on vocabulary: "medical-grade" is the phrase the trial papers use for the honey supplied to them, describing a manufacturing and sterilisation standard used in wound care. It is not a legally defined category on a cosmetic label, and it belongs with the other label words that carry no statutory definition.

How much weight the result will bear

Set against the rest of the New Zealand botanical literature, this is the strongest thing there is. Nothing in the kawakawa, horopito or harakeke evidence reaches a registered, randomised, primary-outcome-driven trial in patients with a diagnosed condition. Taken on its own terms, the trial supports a narrow statement: over eight weeks, in adults with rosacea, applying a 90% kānuka honey preparation twice daily produced a two-point improvement on a severity score in about twice as many people as applying a paraffin emollient.

It does not support the statement that kānuka honey treats rosacea better than established topical therapy, because it was never compared with any. It does not transfer to acne or cold sores, both tested with the identical preparation and neither positive, nor to the limb lesions of eczema and psoriasis, tested with plain kānuka honey and neither positive. It licenses no conclusion about honey generally, about mānuka honey, or about a product carrying a trace of honey below the preservative in the ingredient list: a trial of a 90% preparation says nothing about a 0.5% one, which is a general limit on what a topical formulation can do. And it does not replace a consultation: rosacea has phenotypes that respond to different treatments, ocular involvement that can threaten sight, and a differential diagnosis requiring quite different management. That is a doctor's work.