Clinical research Proposed service Rules checked October 2026

A human study supports a claim only if its question, size and safeguards are fixed before enrollment, and results in cells never substitute for it

How human studies of botanical ingredients are designed, sized, overseen and registered, and what large trials have shown. EUTECNIQ would plan studies with independent partners; it has run none.

1,285interventional trials with a dietary-supplement intervention started in 2025 on ClinicalTrials.gov, up from 26 in 2000

13.5%of the 599 such trials started in 2010 have results posted on the registry (81)

3,069adults aged 75 and older randomized in the GEM trial of Ginkgo biloba; dementia in 277 on ginkgo, 246 on placebo

Status. EUTECNIQ Sciences is in formation, with no clinical data, studies under way, or IRB or CRO agreements. This page describes how studies would be planned and the rules that govern them. Educational, not legal or medical advice.

In brief

Six decisions come before the first participant is enrolled

Each decision constrains the next.

Six decisions in order, each fixed before enrollmentProposed planning sequence

21 CFR Parts 50, 56 and 312; ICH E6(R3); 42 CFR Part 11; ICMJE; CONSORT 2025. Sequence: EUTECNIQ proposal.

Text version of the six decisions

Question and claimWhich population, product, dose, comparator and endpoint; structure or function versus disease
Regulatory pathIND under 21 CFR 312 when the study tests a disease effect; IRB oversight in every case
Design and sizeRandomization, blinding, control, the analysis and a justified sample size
Ethics and consentIRB approval and informed consent (21 CFR 50 and 56)
Conduct and safetyGood Clinical Practice (ICH E6(R3)), adverse-event reporting, monitoring
Registration and reportingClinicalTrials.gov when required by law or for publication; CONSORT reporting

A study designed backward from a desired claim, or sized after the data arrive, cannot support that claim.

The registry record

Registered supplement trials rose from 26 starting in 2000 to 1,285 in 2025, and few post results on ClinicalTrials.gov

Interventional trials with a dietary-supplement intervention, by start year: choose count or shareRegistry data

Show

ClinicalTrials.gov API v2, queries AREA[InterventionType]DIETARY_SUPPLEMENT, AREA[StudyType]INTERVENTIONAL, AREA[HasResults] and AREA[StartDate] by year; retrieved Oct 24, 2026. Counts change as records are added or updated.

Data table, 2000–2025

Interventional studies by start year (ClinicalTrials.gov, Oct 24, 2026)
Start year	With a dietary-supplement intervention	Of which results posted	All interventional	Supplement share
2000	26	3	1,876	1.4%
2001	38	4	2,407	1.6%
2002	58	8	3,497	1.7%
2003	78	11	4,770	1.6%
2004	115	17	6,319	1.8%
2005	129	21	7,729	1.7%
2006	216	34	9,299	2.3%
2007	315	44	10,422	3.0%
2008	453	60	11,900	3.8%
2009	591	69	12,849	4.6%
2010	599	81	13,420	4.5%
2011	634	69	14,217	4.5%
2012	698	70	15,027	4.6%
2013	721	72	15,896	4.5%
2014	784	60	17,600	4.5%
2015	769	55	18,951	4.1%
2016	869	67	20,369	4.3%
2017	887	53	21,265	4.2%
2018	996	71	22,871	4.4%
2019	947	44	23,831	4.0%
2020	843	49	23,601	3.6%
2021	1,063	50	28,342	3.8%
2022	1,118	30	28,118	4.0%
2023	1,213	21	29,490	4.1%
2024	1,242	13	30,689	4.0%
2025	1,285	5	31,021	4.1%

Of the 599 supplement trials that started in 2010, 81 (13.5%) have results on the registry; registering a trial does not make its outcome public.

Explanation, details and sources

The registry's intervention type "Dietary Supplement" is chosen by the sponsor, and a study can list several intervention types. Registration became more common after journal and legal requirements, so early counts partly reflect registration practice. Federal registration and results duties in 42 CFR Part 11 attach to "applicable clinical trials" of FDA-regulated drug, biological and device products; a structure/function supplement study without an IND is generally not one (an inference from the definitions), while ICMJE journals require registration of any trial they publish. Results counts for recent start years are low partly because trials are still running.

Who would do what

EUTECNIQ would design the evidence plan; regulated partners would conduct the study

Explanation

Human studies need independent ethics review, qualified investigators and sites, and often an independent safety committee. Those roles cannot sit with the sponsor.

Seven activities, each with a proposed EUTECNIQ role and a regulated partnerProposed division of work

EUTECNIQ proposal. No IRB, CRO or site agreements exist.

Text version: the division of work as a table

Activity	EUTECNIQ (proposed)	Qualified external partner
Claim and endpoint strategy	Map the intended claim to an endpoint, population and comparator	Regulatory counsel confirms claim category and IND status
Protocol and statistical analysis plan	Draft the design, sample-size justification and analysis plan	CRO biostatistician and principal investigator finalize
Study product	Specification, identity and stability data for the test and placebo products	Manufacturer releases the lots under the applicable cGMP
Ethics review and consent	Supply product and safety information to investigators	Independent IRB reviews, approves and oversees the study
Conduct and data	Sponsor oversight	Clinical sites, monitors and data management under GCP
Safety monitoring	Define stopping rules and reporting lines in the protocol	Investigators report; a data monitoring committee where risk warrants
Registration and publication	Prepare registration and reporting content	Responsible party registers; authors publish

A calculated sample size≠an approved or finalized protocol
IRB approval≠FDA approval of the product or claim
Synergy in cultured cells≠a combined benefit in people
Higher blood levels from a partner≠synergy at a biological target
A registered trial≠a trial reviewed or endorsed by FDA
A biomarker change≠clinical benefit, unless the biomarker is validated

Technical detail

The rules, designs, statistics and evidence standards behind a defensible human study

Explanation

Calculators are educational tools, each with a methods panel and checked against independent reference software.

A supplement study needs an IND when it tests a disease effect; a study of structure or function does not

Explanation

Under FDA's 2013 guidance, the intent of the study decides, not the product category.

Two regulatory paths for human researchRegulations and guidance

21 CFR Parts 50, 56 and 312 (eCFR, current as of October 20, 2026); FDA, Investigational New Drug Applications (INDs): Determining Whether Human Research Studies Can Be Conducted Without an IND (final guidance, September 2013), section VI.D.1; FTC, Health Products Compliance Guidance (December 20, 2022).

IND, IRB and registration: what each study needs, and whyRegulation FDA guidance

A December 2022 proposed rule (87 FR 75536) would have exempted certain studies from IND requirements; as of October 22, 2026 it had not been finalized.FDA, INDs: Determining Whether Human Research Studies Can Be Conducted Without an IND (final guidance, September 2013), section VI.D.1; 80 FR 66907 (2015); FDA, NDI identity and safety guidance (October 2026), footnote 20; 21 CFR Parts 50, 56 and 312; 42 CFR 11.10 and 11.22; ICMJE, Clinical Trial Registration.

Text version: the guidance quoted, the 2015 stay, safety-only studies and the 2022 proposal

FDA's 2013 guidance on studies that can proceed without an IND states that if a clinical investigation "is intended only to evaluate the dietary supplement's effect on the structure or function of the body, an IND is not required", but one intended to evaluate its ability to "diagnose, cure, mitigate, treat, or prevent a disease" requires an IND under 21 CFR 312.

FDA guidance, final 2013

The supplement subsection still applies

A 2015 notice (80 FR 66907) stayed parts of the guidance's conventional-food and health-claim subsections, VI.D.2 and VI.D.3. The dietary-supplement subsection, VI.D.1, was not stayed.

FDA guidance, final October 2026

Safety-only studies

FDA's October 2026 NDI identity and safety guidance (footnote 20) says a study of a supplement or ingredient "generally does not require an IND if the only endpoints measured are safety endpoints", and that such studies must comply with 21 CFR Parts 50 and 56.

Proposed rule, never finalized

A proposed IND exemption

A December 2022 proposed rule (87 FR 75536) would have exempted certain studies of a drug use of a lawfully marketed food, supplement or cosmetic from IND requirements. It was never finalized: as of October 22, 2026, no final rule had been published, so the 2013 guidance still describes the current position.

Oversight rules: consent, ethics review, Good Clinical Practice, safety reporting and registration

Explanation

Whether Parts 50 and 56 legally reach a structure or function study never submitted to FDA is unconfirmed; EUTECNIQ would treat IRB review and consent as the baseline.

Show the table (9 rows)

Rule	Status	What it requires or recommends	Applies to
21 CFR Part 50 Protection of Human Subjects	Final regulation	Informed consent with the required elements; added safeguards for children	Clinical investigations under an IND, and those supporting applications for FDA research or marketing permits, including for foods and dietary supplements that bear a nutrient content or health claim
21 CFR Part 56 Institutional Review Boards	Final regulation	IRB review, approval and continuing review; written procedures for prompt reporting of "any unanticipated problems involving risks to human subjects or others" (§ 56.108(b))	The same investigations
21 CFR Part 312 Investigational New Drug Application	Final regulation	An IND in effect before a drug is studied in people, unless exempt; IND safety reports (§ 312.32)	Drugs, and supplement studies that evaluate a disease effect (FDA guidance, 2013)
ICH E6(R3) Good Clinical Practice	Final guidance	Risk-proportionate quality management, participant protection and reliable results	Principles and Annex 1 adopted by ICH on January 6, 2025 and issued by FDA as final guidance in September 2025; Annex 2 adopted by ICH on June 3, 2026
Data monitoring committees	Final guidance Draft revision	When an independent committee should review accumulating safety and efficacy data	FDA guidance of March 2006; a February 2024 draft would supersede it once finalized
42 CFR Part 11 FDAAA § 801 registration and results	Final regulation	Register on ClinicalTrials.gov and submit results	"Applicable clinical trials": for trials started on or after January 18, 2017, interventional, not phase 1, studying an FDA-regulated drug product, with a U.S. site, a product made in and exported from the U.S., or an IND
ICMJE registration policy	Journal policy	Registration "at or before the time of first patient consent for enrollment" in ClinicalTrials.gov or a WHO primary registry	Any trial submitted to a member journal; interventions include "dietary interventions"
FTC Health Products Compliance Guidance	Final guidance	"Competent and reliable scientific evidence"; health benefit claims generally need randomized, controlled human clinical testing	Advertising for supplements, foods and other health products (December 20, 2022)
CONSORT 2025	Reporting guideline	30-item checklist and participant flow diagram	Publication of randomized trials

eCFR 21 CFR 50.1, 56.101, 56.108, 312.2 and 312.32, and 42 CFR 11.10 and 11.22 (current as of October 20–21, 2026); FDA, E6(R3) Good Clinical Practice (final guidance, September 2025; docket FDA-2023-D-1955); Swissmedic, introduction of ICH E6(R3) (January 6, 2025 adoption); EMA, ICH E6 Good clinical practice scientific guideline page (Annex 2 adopted by ICH June 3, 2026); FDA, Establishment and Operation of Clinical Trial Data Monitoring Committees (March 2006) and Use of Data Monitoring Committees in Clinical Trials (draft, February 2024); ICMJE, Clinical Trial Registration; FTC (December 20, 2022); Hopewell S et al. BMJ. 2025;389:e081123.

Registration of supplement studies. The legal duty in 42 CFR Part 11 attaches to applicable clinical trials of drugs and devices. Our reading of the definitions, not confirmed by an NIH or ClinicalTrials.gov statement, is that a structure or function study of a supplement without an IND is generally not an applicable clinical trial. EUTECNIQ would register such studies anyway, before enrollment, because ICMJE journals require it.

The design follows the question: parallel groups for most claims, crossover for stable conditions, factorial designs to show what each ingredient contributes

Four randomized designs, what each suits, and its main riskSchematic, not to scale

FDA, Adaptive Design Clinical Trials for Drugs and Biologics (final guidance, December 2019); 21 CFR 300.50; ICH E9(R1) Addendum on estimands and sensitivity analysis (FDA final guidance, May 2021).

Text version: the four designs as a table

Design	How it works	Best suited to	Main risk
Parallel groups	Each participant receives one product, assigned at random	Most efficacy questions and conditions that change over time	Needs more participants than a crossover; chance imbalance at baseline in small trials
Crossover	Each participant receives both products in random order, separated by a washout	Stable conditions and short-acting, reversible effects; each person is their own control	Carryover between periods; a dropout loses both periods; unsuitable for lasting effects
Factorial 2 × 2	A alone, B alone, A + B and placebo	Showing what each component adds, as the fixed-combination drug rule requires (21 CFR 300.50)	A trial sized for the main effects usually has low power to detect an interaction
Adaptive	A prespecified interim analysis may stop the trial or re-size it	Uncertain effect sizes	Inflated false-positive rate unless every adaptation and its decision rule is fixed in advance

The primary endpoint carries the claim and sets the sample size; secondary and exploratory endpoints have a different status

Explanation

FDA's multiple-endpoints guidance is written for drug trials; its hierarchy is the design reference EUTECNIQ would use.

Endpoint families by role in the analysis, and endpoint types by distance from clinical benefitFDA guidance and glossary

FDA, Multiple Endpoints in Clinical Trials (final guidance, October 2022), section III.A and appendix section 5; FDA–NIH BEST (Biomarkers, EndpointS, and other Tools) Resource, glossary (updated January 16, 2025). Read October 23, 2026.

Sample size follows from the smallest difference worth detecting: halving that difference roughly quadruples the trial

Explanation

Per-group size grows with (σ/Δ)². An underpowered trial can miss a real effect, and a significant small trial tends to overstate it (Button et al. 2013). Δ and σ are justified before enrollment.

Two parallel groups, continuous endpoint, exact t testn = smallest integer with power ≥ 1 − β
power = P(|T| > t_{1−α′/2, 2n−2}), δ = d√(n/2)
d = Δ/σ α′ = α/m

Difference to detect, Δ (endpoint units) Standard deviation, σ Two-sided α (family-wise) Power, 1 − β Primary comparisons, m (Bonferroni) Baseline–outcome correlation ρ (ANCOVA) Expected dropout

Standardized effect d

α per comparison

Evaluable per group

Randomize in total

Assumptions

Assumptions

Two parallel arms, 1:1 allocation, normally distributed endpoint, equal variances
ANCOVA reduction (1 − ρ²) applies when the baseline value is measured and used in the analysis
Dropout inflates enrollment as n/(1 − dropout) per arm; it does not repair bias from unequal losses
Crossover, cluster, non-inferiority and equivalence designs need other formulas

Participants per group versus standardized effect sizeModel output

Exact power of the two-sided two-sample t test (noncentral t; Lenth 1989, algorithm AS 243). Guenther WC. Am Stat. 1981;35(4):243–244. Julious SA. Stat Med. 2004;23(12):1921–1986. Borm GF et al. J Clin Epidemiol. 2007;60(12):1234–1238. Button KS et al. Nat Rev Neurosci. 2013;14(5):365–376.

Method, assumptions and limits: continuous endpoint

What it calculates: The number of evaluable participants per group that gives the chosen power for a two-sided, two-sample t test at a family-wise significance level α split across m primary comparisons, optionally reduced for an analysis of covariance on the baseline value, and the number to randomize after dropout.
Equations: Power(n) = 1 − F(t_c; ν, δ) + F(−t_c; ν, δ), with F the noncentral t distribution, ν = 2n − 2, δ = d√(n/2), t_c the 1 − α′/2 quantile of the central t with ν degrees of freedom, and α′ = α/m. The search starts from the closed form 2(z_1−α′/2 + z_1−β)²/d² + z²_1−α′/2/4 (Guenther 1981; Julious 2004) and moves to the smallest integer whose exact power reaches the target. ANCOVA: n_ANCOVA = ⌈(1 − ρ²)n⌉ (Borm et al. 2007). Randomized per arm = ⌈n/(1 − dropout)⌉; total = twice that.
Variables and units: Δ and σ in the endpoint's units (only their ratio d matters); α ∈ {0.01, 0.025, 0.05, 0.10}; power ∈ {0.80, 0.85, 0.90, 0.95}; m = 1–5; ρ = 0–0.8; dropout 0–40%.
Assumptions: Independent observations; normal endpoint with a common variance; one primary analysis per comparison; Bonferroni control of the family-wise error, which is conservative when endpoints are correlated. If all of several co-primary endpoints must succeed, α is not split but overall power falls, which this calculator does not model. The ANCOVA factor is a large-sample approximation.
Valid range: Standardized effects from about 0.03 to 15 over the slider range. Effects above about 1 imply very small trials, where the normality assumption and chance baseline imbalance matter more; a warning appears.
Does not establish: An approved or finalized protocol. A real sample-size justification also covers the estimand (ICH E9(R1)), the source of the assumed difference and variance, the handling of missing data, interim analyses, and review by the study statistician and IRB.
References: Lenth RV. Algorithm AS 243. Appl Stat. 1989;38(1):185–189. Guenther 1981; Julious 2004; Borm et al. 2007; FDA, Multiple Endpoints in Clinical Trials (final guidance, October 2022); ICH E9(R1) (FDA, May 2021).
Verification: Unit tests match scipy's noncentral t to an absolute error below 10⁻¹⁴ and statsmodels' TTestIndPower power to 10⁻¹⁰; exact sample sizes equal statsmodels' in every test case. Across 9,472 combinations of d, α and power, the closed form alone was exact in 88% and one participant short in 12%, which is why the calculator uses the exact search. Browser readouts match a Python reference built on statsmodels.

For a responder endpoint, the sample size depends on both response rates, not only on their difference

Explanation

For a binary outcome the variance depends on the rates: the same 15-point difference needs more participants between 40% and 55% than between 5% and 20%.

Two proportions, normal approximation (Fleiss, Levin & Paik 2003)n = [z_1−α′/2√(2p̄q̄) + z_1−β√(p₁q₁ + p₂q₂)]²(p₁ − p₂)²
n_cc = n4[1 + √(1 + 4/(n|p₁ − p₂|))]²

Response rate, control group Response rate, product group Two-sided α (family-wise) Power, 1 − β Primary comparisons, m (Bonferroni) Expected dropout

Difference in response rate

Per group, no continuity correction

Per group, with continuity correction

Randomize in total (corrected)

Assumptions

Assumptions

Two independent groups, 1:1 allocation, analysis by a two-sided test of two proportions
The responder definition is fixed before the study; dichotomizing a continuous endpoint loses power
Normal approximation; small expected counts need exact methods

Participants per group versus the product group's response rateModel output

Fleiss JL, Levin B, Paik MC. Statistical Methods for Rates and Proportions, 3rd ed. Wiley; 2003, chapter 4. Continuity correction: Fleiss JL, Tytun A, Ury HK. Biometrics. 1980;36(2):343–346.

Method, assumptions and limits: binary endpoint

What it calculates: The number of evaluable participants per group needed to detect the difference between two response rates with the chosen power, with and without a continuity correction, and the number to randomize after dropout.
Equations: n = [z_1−α′/2√(2p̄(1 − p̄)) + z_1−β√(p₁(1 − p₁) + p₂(1 − p₂))]²/(p₁ − p₂)², p̄ = (p₁ + p₂)/2, α′ = α/m; corrected n_cc = (n/4)[1 + √(1 + 4/(n|p₁ − p₂|))]². Both are rounded up; randomized per arm = ⌈n_cc/(1 − dropout)⌉.
Variables and units: p₁, p₂ response proportions, 5–95%; α, power, m and dropout as in the continuous calculator.
Assumptions: Independent binary outcomes; the two-proportion test (equivalently a 2 × 2 chi-square test) with pooled variance under the null hypothesis; Bonferroni-split α across m primary comparisons.
Valid range: Normal approximation; when n p or n(1 − p) is below about 5 in either group, exact or simulation-based methods are needed, and a warning appears. Equal rates have no finite sample size.
Does not establish: An approved protocol, the clinical relevance of the responder threshold, or the expected rates themselves, which must come from prior data.
References: Fleiss, Levin & Paik 2003, chapter 4; Fleiss, Tytun & Ury 1980; FDA, Multiple Endpoints in Clinical Trials (October 2022).
Verification: Unit tests compare both formulas with an independent Python evaluation to 10⁻¹². Browser readouts match the Python reference exactly after rounding up.

Every participant is accounted for from screening to analysis

Explanation

CONSORT 2025 is a 30-item checklist with a flow diagram of everyone assessed, randomized, lost and analyzed. Unequal losses can bias the result, so the flow is read before the effect size.

Counts invented for illustration

Hopewell S et al. CONSORT 2025 statement. BMJ. 2025;389:e081123. Schulz KF, Altman DG, Moher D. CONSORT 2010 statement. BMJ. 2010;340:c332.

Timeline of a 12-week, two-arm supplement trialIllustrative durations

Durations are invented to show the sequence; real timelines depend on the endpoint, recruitment rate, sites and review cycles. Study product would be released against its specification before the first participant is dosed.

Each step from target to people answers a narrower, more relevant question, and only human studies can support a claim about people

Explanation

At each step, ask what a positive result does and does not show.

From target to people: relevance rises, throughput fallsQualitative schematic

Show the table (6 rows)

Level	Typical methods	A positive result can show	It cannot show
1. Biochemical assays	Enzyme inhibition (IC50), receptor binding (Ki), in chemico reactivity	Direct interaction with a defined target at a known concentration	That the compound reaches the target in the body, or that the interaction matters clinically
2. Cell-based assays	Reporter genes, signaling, viability, cytokine release	Activity in a living cell; potency; early cytotoxicity	Absorption, metabolism and distribution; most cell lines are not the target tissue
3. Ex vivo and tissue models	Isolated tissues, primary cells, organoids, skin explants	Effects in more realistic tissue architecture	Whole-body exposure and long-term effects
4. Animal models	Mechanism and disease models; pharmacokinetics in animals	Integrated physiology, exposure–response, dose ranging	Translation to people: species differ in metabolism and disease biology
5. Human mechanistic studies	Biomarker, pharmacokinetic and tolerability studies	Absorption, exposure and short-term biological effects in people	Clinical benefit: a biomarker change is a surrogate unless validated
6. Randomized controlled trials	Parallel or crossover trials with clinical or validated endpoints	Benefit in the population, dose and duration studied	Benefit in other populations, doses or products; safety beyond the trial

Dose–response: potency (EC50) and steepness (Hill coefficient) are separate properties, and a steep curve narrows the useful dose range

Explanation

The Hill equation describes most concentration–effect curves; inverted, it gives the IC50. EC50 shifts the curve sideways; the Hill coefficient sets the fold-rise in concentration from 10% to 90% of the maximal effect.

Hill (sigmoid Emax) equationE = E_max · CⁿEC₅₀ⁿ + Cⁿ
EC₉₀ / EC₁₀ = 81^1/n

EC50 (arbitrary units) Hill coefficient n Maximal effect E_max

EC10

EC90

EC90 / EC10

Assumptions

Assumptions

Concentrations in arbitrary units; the curve is a model, not data
Effect measured at equilibrium in a single system; E_max below 100% describes a partial effect
An in vitro EC50 is not a human dose: binding, metabolism and distribution intervene

Concentration–effect curve (log concentration)Model output

Goutelle S et al. The Hill equation: a review of its capabilities in pharmacological modelling. Fundam Clin Pharmacol. 2008;22(6):633–648.

Method, assumptions and limits: Hill equation

What it calculates: The concentration–effect curve of the sigmoid Emax (Hill) model and the concentrations giving 10% and 90% of the maximal effect.
Equations: E = E_maxCⁿ/(EC₅₀ⁿ + Cⁿ); EC_x = EC₅₀(x/(100 − x))^1/n, so EC₁₀ = EC₅₀·9^−1/n, EC₉₀ = EC₅₀·9^1/n and their ratio is 81^1/n. EC values are fractions of E_max, so they do not change with E_max.
Variables and units: EC₅₀ 0.01–1,000 arbitrary units (logarithmic slider); n 0.4–4.0; E_max 20–100%.
Assumptions and limits: Single-site, equilibrium response with no tolerance, receptor reserve or time dependence. Fitted EC50 values depend on the assay system and exposure time.
Does not establish: A human dose, efficacy or safety.
Verification: Unit tests confirm that the model returns exactly 10%, 50% and 90% at the computed concentrations and the 81^1/n span. Browser readouts match the Python reference.

Cards: Test-tube antioxidant capacity is not an effect in the body; The evidence standard rises with the claim

Antioxidant assays

Test-tube antioxidant capacity is not an effect in the body

In 2012 the USDA Nutrient Data Laboratory removed its ORAC database, citing evidence that the values had no relevance to the health effects of specific bioactive compounds. ORAC and DPPH rank radical-scavenging chemistry, not absorption, effect or benefit.

SupplySide Supplement Journal, "USDA Says ORAC Tests Useless, Removes Database for Selected Foods," June 12, 2012 (trade-press report of the USDA statement).

Claims

The evidence standard rises with the claim

FTC guidance (December 2022) expects competent and reliable scientific evidence, generally randomized, controlled human testing, for health benefit claims. Drug approval rests on adequate and well-controlled investigations (21 CFR 314.126).

FTC, Health Products Compliance Guidance (December 20, 2022); 21 CFR 314.126.

"Synergy" means more than expected, so it only has meaning against a stated reference model of what to expect

Explanation

Loewe additivity treats the partners as dilutions of one another; Bliss independence treats them as acting through separate mechanisms; the highest-single-agent rule only asks whether the pair beats its better member. A result can be synergistic under one and additive under another.

Cards: Dose additivity; Probabilistic independence; Better than the better partner

Loewe additivity · 1953

Dose additivity

d_A/D_A,x + d_B/D_B,x = 1

A drug combined with itself is additive; combinations reaching effect x at doses below this line are synergistic. The combination index (CI) is this sum.

Bliss independence · 1939

Probabilistic independence

E_AB = E_A + E_B − E_AE_B

Effects as fractions from 0 to 1; an observed effect above the expectation is Bliss synergy.

Highest single agent

Better than the better partner

E_AB > max(E_A, E_B)

The weakest test, close to the drug-combination rule's question of whether each component contributes; it does not establish synergy.

Loewe S. Arzneimittelforschung. 1953;3:285–290. Bliss CI. Ann Appl Biol. 1939;26:585–615. Berenbaum MC. Pharmacol Rev. 1989;41:93–141. Greco WR, Bravo G, Parsons JC. Pharmacol Rev. 1995;47:331–385. Tang J, Wennerberg K, Aittokallio T. Front Pharmacol. 2015;6:181.

Synergy evidence has levels too: a combination index in cells is a hypothesis about people, not a finding

Explanation

Only the last level can show that a combination works better in people than its parts.

Show the table (4 rows)

Evidence	What it can show	What it cannot show
Cell or enzyme assay (CI, Bliss excess)	Greater-than-expected effect in that system, at that ratio, effect level and exposure time	That the ratio is reached in the target tissue in people, at safe doses
Animal study of the combination	Interaction in an integrated organism, with exposure data	Human metabolism, disease biology and doses
Human pharmacokinetic study	Whether one partner changes the other's blood levels (pharmacokinetic enhancement)	Synergy at a target, or benefit
Human factorial trial (A, B, A + B, placebo)	Each component's contribution and whether their joint effect exceeds the sum	Effects in other populations, doses or products

Documented botanical, vitamin and mineral interactions, each with its evidence level, and the sample-size cost of testing an interaction in a factorial trial: Combination science and synergy.

Combination index: a mixture tested at a fixed ratio can be synergistic at one effect level and antagonistic at another

Explanation

The Chou–Talalay method fits the median-effect equation to each agent and the mixture, then computes the combination index at every fraction affected (fa). Change the mixture's potency and slope and watch the points move.

Median-effect equation · combination indexf_af_u = (DD_m)^m CI = d_AD_A,x + d_BD_B,x

Share of A in the mixture (by dose) Mixture potency versus dose additivity at fa 0.5 Mixture slope m

CI at fa 0.50

CI at fa 0.75

CI at fa 0.90

Assumptions

Assumptions

Single agents: A with D_m = 10, m = 1.0; B with D_m = 50, m = 1.4 (arbitrary units, invented)
CI < 0.9 synergism, 0.9–1.1 nearly additive, > 1.1 antagonism (Chou 2006); real reports give CI with its confidence interval from replicated data
An in vitro CI is a property of that assay system only

Normalized isobologram and Fa–CI plotModel output

Chou TC, Talalay P. Adv Enzyme Regul. 1984;22:27–55. Chou TC. Theoretical basis, experimental design, and computerized simulation of synergism and antagonism in drug combination studies. Pharmacol Rev. 2006;58(3):621–681.

Method, assumptions and limits: combination index

What it calculates: The Chou–Talalay combination index of an invented fixed-ratio mixture at three effect levels, the corresponding isobologram points, and the Fa–CI curve.
Equations: Median-effect dose D_x = D_m(f_a/(1 − f_a))^1/m. The dose-additive mixture at f_a = 0.5 is 1/(p/D_A,50 + (1 − p)/D_B,50); the mixture's own D_m is that dose divided by the potency factor k. CI = p D_mix/D_A,x + (1 − p)D_mix/D_B,x, which equals 1/k at f_a = 0.5.
Variables: p share of A by dose, 5–95%; k 0.33–3 (logarithmic slider); mixture slope m 0.5–3.0. All doses in arbitrary units.
Assumptions and limits: Each agent and the mixture follow the median-effect equation; the mixture ratio is constant; effects are measured in one system. CI values without replicate-based confidence intervals are not interpretable as evidence.
Does not establish: Synergy in people, a dose, or that the ratio can be achieved in tissue.
Verification: Unit tests confirm that a mixture following dose additivity gives CI = 1 at every effect level and ratio, and that CI = 1/k at f_a = 0.5. Browser readouts match the Python reference.

Bliss independence on a dose matrix: where the combination beats the expectation, and where it falls short

Explanation

Each cell's excess is the observed effect minus the effect expected if the two acted independently; single-agent rows carry no excess by definition.

Excess over BlissΔE = E_obs − (E_A + E_B − E_AE_B)

Interaction built into the invented data

Largest excess

At doses (A, B)

Mean excess

Assumptions

Assumptions

Single agents follow Hill curves (A: EC50 3, n 1.2; B: EC50 20, n 1.0; arbitrary units)
The "observed" matrix is generated by shifting the Bliss expectation on the logit scale by g · 4 E_AE_B; it is not experimental data
Real screens need replicates; single wells cannot separate synergy from noise

Excess over the Bliss expectation, percentage pointsModel output

Bliss CI. The toxicity of poisons applied jointly. Ann Appl Biol. 1939;26:585–615. Tang J et al. Front Pharmacol. 2015;6:181.

Cards: Piperine raises curcumin's blood levels; that is not synergy at a target; A factorial trial isolates each partner; the drug rule dema

Pharmacokinetic enhancement

Piperine raises curcumin's blood levels; that is not synergy at a target

Shoba et al. (1998): 20 mg of piperine with 2 g of curcumin increased curcumin bioavailability by 2000% in volunteers, from serum levels undetectable or very low without it; in rats, by 154%. Higher exposure needs a new safety assessment, and does not show joint action at a target.

Peer-reviewedShoba G et al. Planta Med. 1998;64(4):353–356.

Proving each component contributes

A factorial trial isolates each partner; the drug rule demands it

21 CFR 300.50 allows a fixed-combination prescription drug when each component contributes to the claimed effects. A factorial design (A, B, A + B, placebo) shows contribution directly. Botanical extracts add many constituents that may act together or against each other.

Regulation21 CFR 300.50; Wagner H, Ulrich-Merzenich G. Phytomedicine. 2009;16:97–110; Caesar LK, Cech NB. Nat Prod Rep. 2019;36:869–888.

In vitro synergy is a hypothesis, not a benefit. A combination index below 1 in cells holds for that cell line, ratio and exposure time. Whether the same ratio reaches the same tissue in people, at safe doses, is a separate question that only pharmacokinetic and clinical data can answer. EUTECNIQ would report synergy findings with the reference model, the ratio, the effect level and the uncertainty, and would not present an in vitro result as a human effect.

What large trials show

Large botanical trials show why design decides the answer: 3,069 people for six years in one, a failed active comparator in another

GEM trial: 3,069 participants randomized, followed for a median 6.1 years, with dementia in 277 on ginkgo and 246 on placeboPeer-reviewed trial

DeKosky ST, Williamson JD, Fitzpatrick AL, et al. Ginkgo biloba for prevention of dementia: a randomized controlled trial. JAMA 2008;300(19):2253–2262 (NCT00010803). Hazard ratio 1.12 (95% CI 0.94–1.33).

Data table

GEM flow as reported
Group	Randomized	Dementia	No dementia (computed)
Ginkgo, 120 mg twice daily	1,545	277	1,268
Placebo	1,524	246	1,278
Total	3,069	523	2,546

Dropout and loss to follow-up 6.3%; dementia rate 3.3 (ginkgo) and 2.9 (placebo) per 100 person-years.

A three-arm depression trial: neither St John's wort nor sertraline separated from placebo, so the trial could not judge eitherPeer-reviewed trial

Hypericum Depression Trial Study Group. Effect of Hypericum perforatum (St John's wort) in major depressive disorder. JAMA 2002;287(14):1807–1814.

Data table

Primary outcomes at week 8
Arm	HAM-D change (95% CI)	P vs placebo	Full response
Placebo	−9.20 (−10.51 to −7.89)	–	31.9%
St John's wort (LI-160), 900–1,500 mg/day	−8.68 (−10.01 to −7.35)	.59	23.9%
Sertraline, 50–100 mg/day	−10.53 (−11.94 to −9.12)	.18	24.8%

An active comparator tests assay sensitivity: when a licensed drug also fails to beat placebo, the trial cannot show that the botanical does not work.

Explanation, details and sources

GEM enrolled community volunteers aged 75 or older with normal cognition (2,587) or mild cognitive impairment (482); 92% of dementia cases were classified as possible or probable Alzheimer disease or Alzheimer disease with vascular disease. The trial measured dementia incidence; this page states its result and makes no health claim for any botanical. In the depression trial, sertraline did better than placebo only on a secondary measure (Clinical Global Impressions improvement, P = .02). More trials of widely used botanicals are summarized on Lessons learned.

Planning human evidence for an ingredient?

Start with the claim you need to support. EUTECNIQ proposes to map it to a design, a sample-size justification and the regulated partners who would run the study.

Discuss a study plan Nonclinical safety