Oral saffron stigma extract,
does it really help with Cognition in mild-to-moderate Alzheimer's disease?
research showsOne small placebo-controlled trial reported better cognitive scores at 16 weeks. Independent replication under the same conditions and long-term clinical benefit remain unconfirmed. This does not establish a treatment for Alzheimer's disease or a replacement for prescribed medication.
ads claimThis assessment does not support claims that saffron tea prevents dementia, stops Alzheimer's progression, is equivalent to standard medication, or can replace prescribed treatment.
Four separate assessment dimensions
| Effect direction and size | A favorable 16-week ADAS-cog change was reported versus placebo. The difference between published mean changes is −7.77 points (editorial calculation). Clinical importance and durable functional benefit are not established. |
|---|---|
| Evidence certainty | Provisionally limited: one small sample, LOCF imputation, unverified prospective registration/SAP, and some internally inconsistent reported statistics. This is not a formal GRADE assessment. |
| Applicability | Restricted to the studied stigma extract at 30 mg/day for 16 weeks in mild-to-moderate Alzheimer's disease. Not transferable to tea, other formulations, isolated constituents, prevention, or prescription replacement. |
| Safety | Caution. Short-term events were recorded, but rare or long-term harm, multiple co-medications, and vulnerable populations remain inadequately assessed. |
C·46 is a retained legacy value under methodological consistency review, not a 46% effect, formal GRADE rating, or journal/expert certification. Correcting S to P does not establish a new final grade. Read the four assessment dimensions and unresolved items separately.
Useful facts when choosing a product
- The studied regimen was an ethanolic stigma extract in 15 mg capsules twice daily for 16 weeks. This describes a trial, not a personal dosing instruction.
- Do not transfer the result to tea, food, different extracts, isolated crocin or safranal, or combination products.
- The core trial does not represent all routine-care patients continuing antidementia medication. It provides no basis for stopping or replacing prescriptions.
Chamgap Semantic Classification Code
Permanent code issued
S.saffron-stigma-ethanolic-extract.oral.alzheimer-adas-cog.reduce.placeboSupplements and nutraceuticals > Saffron stigma ethanolic extract > Oral > Mild-to-moderate Alzheimer ADAS-cog > Score-reduction claim > Placebo
Bound to ADAS-cog change with 15 mg twice daily for 16 weeks in mild-to-moderate disease under the trial's medication restrictions. Issuance identifies scope, not confirmed efficacy or clinical certification. Drug add-on regimens, tea, isolated crocin, and other outcomes remain distinct. The code identifies the intervention and claim; it never contains the evidence grade, score, or safety result.
Exact Claim Classification
These independent facets prevent evidence from different forms, routes, populations, effects, and comparators from being mixed.
| Intervention class | S · Supplement or nutraceutical |
|---|---|
| Canonical ingredient or intervention | Saffron stigma extract |
| Source or part used | Dried stigmas of Crocus sativus |
| Formulation or processing | Capsules of dried 80% ethanolic extract; reported crocin/safranal standardization |
| Route | Oral |
| Dose | 15 mg twice daily, total 30 mg/day; trial regimen |
| Duration | 16 weeks |
| Population | Age over 55, probable mild-to-moderate Alzheimer's disease, MMSE 15–26, subject to the core trial's antidementia-medication restrictions |
| Effect or condition | Improvement in Alzheimer cognition |
| Primary endpoint | Page-selected: between-arm difference in 16-week ADAS-cog change; lower is better; exact version and prospective prespecification unverified |
| Comparator | Visually identical placebo capsules |
| Duplicate-detection key | saffron-stigma-ethanol-extract|oral-capsule|30-mg-day|16-weeks|mild-moderate-alzheimer-no-concurrent-antidementia-treatment|adas-cog-change|placebo |
What the research actually shows
The core trial randomized 46 participants, 23 per arm. Lower ADAS-cog scores indicate improvement. Changes at 16 weeks were −3.69±1.69 versus +4.08±1.34 points (mean±SD). The simple difference in these changes, −7.77 points, is calculated from the published means; it is neither an author-reported adjusted estimate nor an individual's probability of improvement. The paper reported P<.0001 for the change-score comparison. The abstract's P=.04 instead refers to the repeated-measures group effect. An author-reported 95% confidence interval was not verified. Source: S1, author-uploaded full-text transcription, p4, Results/ADAS-cog.
In S3, 60 people receiving donepezil were assigned additional saffron or placebo for 12 weeks; the between-group MMSE comparison was P=.303. Non-significance does not prove zero effect. Improvements in inflammatory markers are not cognitive benefits, and this add-on study is not counted as independent replication of S1. The active donepezil comparator in S2 does not automatically establish placebo-relative efficacy or noninferiority.
Why this is classified as C (46)
The legacy C·46 rating is under review. ADAS-cog directly measures the cognitive function targeted by this claim; its endpoint code is corrected from S (surrogate) to P (symptom or function itself), rather than calling it a surrogate merely because it is a test. Assigning EX solely because a clinical-importance threshold is unverified conflicts with the existing boundary rules. Numerical effect reconstruction and unverified clinical importance are separate issues. The remaining axes and final grade are not newly adjudicated in this correction.
Counterpoint. The positive randomized signal should neither be erased nor inflated into clinical confirmation. These data cannot settle long-term daily functioning, caregiver burden, or disease progression.
Rejudgment record. Editorial assessment using ChatGPT research and Codex source comparison — Existing Chamgap scoring kept separate from four clinical assessment dimensions
| Endpoint | P | Symptom or function itself is the target - including patient reports and performance tests Case application: ADAS-cog is the cognitive-function endpoint itself for this claim, not a biomarker or surrogate. |
| Replication | R1 | Single confirmatory trial Case application: One direct placebo trial; prospective confirmatory status and matching independent replication unverified |
| Independence | I1 | Mixed funding sources |
| Effect size | EX | The clinical size of the effect could not be judged |
| Precision | CX | No pooled confidence interval could be confirmed Case application: Numerical reconstruction is possible, but clinical precision remains unclassified without a verified applicable importance threshold |
Stored derived and displayed grades match; this is not a current recalculation or validity check (C).
Review performed and remaining limitations
Pilot AI-assisted review. No external expert review, blinded independent review, formal GRADE, complete RoB 2 signalling-question assessment, or public comprehension testing was performed. Some full texts, registrations, and analysis plans were unavailable.
Preliminary RoB 2-informed review — not a complete formal assessment
| Randomization | Some concerns: a computerized sequence and blinded codes are described, but operational allocation concealment and separation of recruitment/allocation staff remain unverified. This is not proof of improper allocation. S1, Intervention. |
|---|---|
| Deviations from assigned interventions | Provisionally toward low risk: double blinding, matched appearance, caregiver supervision, and analysis by assigned group are described. Taste/odor matching, quantitative adherence, and co-treatment deviations remain unverified. S1, Participants/Intervention/Statistics. |
| Missing outcome data | Some concerns: 1/23 versus 3/23 dropped out, totaling 4/46 (8.7%, calculated). LOCF was used and a placebo participant died. Other missingness reasons and sensitivity analyses remain unverified; LOCF does not remove missing-data uncertainty. S1, Statistics/Results/Safety. |
| Outcome measurement | Provisionally toward low risk, with documentation gaps: both arms used the same ADAS-cog schedule and double blinding, but exact version, translation, assessor training, and blinding success remain unverified. S1, Measurements/Intervention. |
| Selection of the reported result | Some concerns: the paper specifies a 16-week change-score test, but multiple scales/analyses are reported and a time-stamped prospective registry/SAP was not retrieved. Possible selective reporting is not evidence of manipulation. S1, Statistics/Results. |
Reasons for the certainty judgment — not formal GRADE
| Risk of bias | Uncertainty remains about allocation concealment, missing-data handling, and prospective result selection. Small size or company support alone was not converted into a formal high-risk RoB 2 judgment. |
|---|---|
| Inconsistency | Between-study inconsistency is difficult to assess with one direct trial under matching conditions. S3 differs in co-treatment and scale; its non-significant result is not used to invent heterogeneity statistics or consistent replication. |
| Indirectness | Direct for the narrow question, but not transferable beyond the extract, selected population, medication restrictions, and 16 weeks to prevention, durable daily function, tea, or prescription replacement. |
| Imprecision | An independent-arm equal-variance t calculation using published change SDs and 23 participants per arm gives a difference of −7.77 points and a conditional 95% CI of −8.68 to −6.86 (df44). This is an editorial calculation, not an author-reported CI, and assumes the rounded summaries and LOCF analysis. It does not resolve missingness/reporting bias or establish clinical importance. A formal optimal-information-size assessment was not performed. |
| Publication bias | No funnel plot or Egger test was performed for one direct trial. Registries were not exhaustively searched for unpublished studies, so publication bias cannot be excluded. |
Search scope and limitations. Cutoff: 2026-09-14. ChatGPT supplied 56 web queries; Codex checked core primary sources, arithmetic, and contrary findings. Queries combined saffron/Crocus, Alzheimer, placebo, ADAS-cog, registration, correction/retraction, and recent years. PubMed/registry domain-restricted searches were included, but this was not an exhaustive native database search or systematic review. Recent reviews were trial locators, not additional independent trials. Queries and exclusions are retained in the editorial review record.
Evidence Table
| Study | Design | Sample | Funding | Endpoint | Result | Weight |
|---|---|---|---|---|---|---|
| Akhondzadeh et al. 2010 — 16-week saffron–placebo trial (PMID 20831681) | Three-site randomized double-blind placebo trial; paper-defined ITT with LOCF | 46 randomized, 23/23 in paper-defined ITT analysis; 1/3 dropouts; 22/20 completers calculated by subtraction; analysis df44 | Joint university/Green Plants of Life grant; company donated material. Cash shares, employee authorship, patents, and separate COI declaration unverified | 16-week ADAS-cog change; selected for this page; prospective registry prespecification unverified | Changes −3.69±1.69 vs +4.08±1.34 points (SD); simple difference −7.77 (calculated). Full-text p4, change-score test P<.0001 | One direct trial. Small sample, LOCF, unavailable prospective plan, and statistical reporting inconsistencies limit confirmation |
| Akhondzadeh et al. 2010 — 22-week saffron–donepezil trial (PMID 19838862) | Randomized double-blind active-comparator trial, not placebo-controlled | 54 participants; completion/analysis details not established in this review | Funding, product support, employee authorship, and patent details unverified in this review | 22-week ADAS-cog/CDR-SB, with donepezil as comparator | The abstract reports no significant between-group difference; this is not converted into equivalence, noninferiority, or placebo-relative efficacy | Context under a different comparator; not counted as replication of the direct placebo trial |
| Rasi Marzabadi et al. 2022 — saffron added to donepezil (DOI 10.1016/j.hermed.2022.100574) | Randomized double-blind saffron/placebo add-on trial during donepezil treatment | 60 participants, 30/30 randomized; publisher Results excerpt reports 27/27 analyzed | Tabriz University Neurosciences Research Center grant 62590; authors declared no conflicts. Product provision, employee authorship, and patents not separately established | 12-week MMSE; saffron 15 mg twice daily as add-on | Between-group MMSE P=.303; mean difference/95% CI unverified. Inflammatory-marker findings are not cognitive effects | Important non-significant context. Different co-treatment and scale preclude pooling with S1 or counting it as consistent replication |
Receipt — 3 References
Retained evidence-access date: 2026-09-14. This correction checked record consistency; it did not reverify every original source or recertify the grade.
This correction by: Codex · Evidence date: 2026-09-14 · Corrections: 1
Correction log — 1
Corrections applied to this verdict, in chronological order. Changes are logged, not erased.
- 2026-09-14 · Editorial correction and legacy-grade review notice — Corrected the ADAS-cog endpoint code from S to P and disclosed unresolved EX and score issues. Grade and score are retained, not certified as a new final assessment. (grade C unchanged)
Cite the review status too
Legacy grade under review. New grade-card distribution is on hold.
[Chamgap] Saffron Extract for Alzheimer Cognition—A Short-Term Signal, Not Confirmation — legacy C · 46, under review. https://chamgap.com/en/verdicts/cognition/saffron-extract-alzheimer-adas-cog-placebo/ · CC BY 4.0What this document does and does not do
Chamgap is an information source. It reports what research has and has not confirmed; it does not tell readers what to take or buy. That decision belongs to readers and, when needed, medical or legal professionals. This verdict reflects literature available up to the search date and may change as new research appears. Nothing here is medical advice.