P19 · Chapter 4
After the Balm: Reading Experience, Labels and Research
A public post describes a balm infused with lavender, chamomile, mugwort and passionflower in olive oil, lists many other things already tried, and asks for more ideas [P19Q01]. The honest response is not “that balm works” or “that balm does not work”. It is a set of questions: what would it mean for it to work, what was actually in it, what else changed, and what kind of evidence could answer the question at all?
1. What the experience record establishes
The post reports trying several products, including a multi-herb olive-oil balm containing mugwort. It documents the author’s account and request for further ideas, rather than independently verifying the person or product. The described balm was a blend of lavender, chamomile, mugwort and passionflower in olive oil; the report does not isolate mugwort [P19Q01].
It does not establish that the balm failed, that it helped, that mugwort was responsible for anything, or that the person’s condition is comparable to anyone else’s. The poster’s identity and diagnosis are not verified, the product’s exact formula and strength are not stated, and there is no follow-up describing what happened next. These are not reasons to dismiss the record. They are reasons to classify it correctly: it is evidence about a question, not evidence about an effect.
The distinction that matters. A user report answers “does this concern exist, and what does it feel like?” It does not answer “did this product cause that outcome?” Those are different questions, and a guide that blurs them is not being generous to the user — it is borrowing certainty the user never claimed.
2. Four questions to separate before judging
When someone says a product helped or did not help, four questions should be separated before any conclusion is drawn.
Question one: what exactly was used?
Product name, code, variant, batch if available, and full formula. In the record, the balm is described by its herbs and carrier but not by brand, strength or preparation method [P19Q01]. Without that, even a perfect outcome could not be attributed to a specific article.
Question two: what was the outcome, and how was it measured?
“It helped” could mean less pain at rest, less pain on movement, better sleep, less stiffness, or a pleasant sensation during use. Those are different outcomes. A trial uses defined scales; a forum post uses ordinary language. Both are legitimate, but only one can be compared across people.
Question three: what else changed?
Time, rest, other products, activity, weather, stress and concurrent treatment all change how a person feels. If several things changed at once, the product is one candidate among many.
Question four: how would we know if it were not the product?
This is the counterfactual question, and it is the one anecdotes cannot answer. A trial answers it with a comparison group. An individual report answers it with memory and expectation, which are unreliable.
3. The evidence ladder, without the marketing
Evidence types are often presented as a hierarchy in which everything above a certain line is “real” and everything below is worthless. That is too crude. Each type answers a different question, and the useful skill is matching the type to the question.
| Type | Can answer | Cannot answer |
|---|---|---|
| Label or listing claim | What the seller says the product is for | Whether the claim is true |
| Individual user report | That a concern exists; what an experience felt like | Whether the product caused the experience |
| Case series | That a pattern was observed in a clinic | Whether the pattern would occur without treatment |
| Observational study | Association in a defined population | Causation; control of unmeasured factors |
| Randomised controlled trial | Effect of an intervention versus a comparator, in that population and protocol | Effect of a different product, species, dose or population |
| Systematic review | What the body of trials shows, with heterogeneity | Anything the included trials did not study |
The ladder is not a licence to skip the lower rungs. A trial with the wrong product does not beat a well-documented user report for the question “what does this product feel like?”. It beats it only for the question the trial was designed to answer.
4. A real trial, read closely
The one controlled study connected to this collection is a randomised double-blind trial of topical Artemisia absinthium formulations in knee osteoarthritis, published in 2017 in the Iranian Journal of Medical Sciences [K03]. For this revision the full text was read through PubMed Central (PMC5684373), not only the abstract. The summary below follows the methods, results and discussion sections; the limitations are the authors’ and the reader’s, and they are stated separately from the findings.
Design
The authors describe a randomized, double-blind, parallel-group non-inferiority trial in primary knee osteoarthritis at a clinic in Hamadan, Iran, conducted in 2012–2013 [K03]. The enrolment, completion and table denominators are distinguished in section11 below.
What was applied
The extract came from plants grown at a research station in Hamadan, harvested in August 2012 and authenticated at the station’s herbarium. The dried, powdered plant material was soaked in 8% acetic acid, filtered and dried by rotary evaporation. The ointment contained 3% w/w of the extract in an ointment base; the liniment contained 3% w/w in olive oil. A pharmacist prepared both, and they were packaged alike and marked only A, B and C [K03].
Patients were told to rub the assigned formulation on the affected knee — front, back and sides — three times a day for 28 days, and to return at baseline, week 4 and week 6. In bilateral cases both knees were treated but only the more painful knee was evaluated. Patients were also told to stop and return to the clinic if redness, itching or burning appeared at the application site [K03].
Outcomes and analysis
The primary outcome was reduction in mean VAS pain on a scale from 0 to 10. Secondary outcomes used the WOMAC index: pain, stiffness and physical function, with 5, 2 and 17 items respectively, using the five-point Likert version [K03]. The report’s analysis definition and non-inferiority wording are examined in section11.
Results as reported
Baseline scores were similar across groups: mean VAS about 6.0 and mean total WOMAC about 42–50. In the ointment group, total WOMAC, WOMAC pain, WOMAC physical function and VAS changed significantly across the three time points (P<0.001); the stiffness score did not. In the liniment group, VAS and WOMAC pain changed significantly, but physical function and stiffness did not. In the piroxicam group, all scores changed significantly [K03].
When the three groups were compared with a three-way repeated-measures ANOVA, the analysis found no difference between groups in the trend of pain severity or physical function. The authors also report that the ointment group showed no recurrence two weeks after treatment ended, whereas the liniment’s effect on pain attenuated by week 6. Their conclusion is that topical ointment containing the extract “may improve clinical manifestation” and that further work is needed on the optimal dose [K03].
What the full text adds to the abstract
Three details change how the result should be read. First, the reported absence of a statistically significant between-group difference is not demonstrated equivalence. Second, there was no placebo arm; the authors also compared each group with its own baseline. Third, stiffness results must be read by treatment arm: the ointment did not show a significant change, whereas the Results and Table 2 report a change in the piroxicam group. The Discussion’s broad stiffness wording should not replace those arm-specific results [K03].
What the authors did not say
The trial did not evaluate a mugwort patch, A. argyi, A. vulgaris or the reference consumer products. It did not measure a biological mechanism in these participants. Its discussion of other experimental literature does not turn the trial into a mechanism study for the products in this collection.
5. Why the trial cannot be transferred
Transferring this trial to a mugwort patch requires at least six bridges, and none of them is established by the trial.
Species
The trial used Artemisia absinthium. The reference listings do not establish a Latin species. Without that identity, the study and the offered product cannot be assumed to contain the same botanical material. The product needs its own species and preparation record before a meaningful comparison [K03].
Preparation
The trial used a 3% ointment and a 3% liniment. The reference patches are described with paste or adhesive layers; the report supplies no bridging comparison of their release with the trial preparations. The balm in the discussion record was described as a multi-herb olive-oil infusion, with no concentration established [K03] [P19Q01].
Dose and regimen
Three applications daily for four weeks, at a defined 3% concentration. No reference listing states a concentration, and none states a three-times-daily regimen. A patch worn for six to eight hours [4] is a different exposure pattern.
Population and condition
Adults 30–70 with primary knee osteoarthritis. The experience record is not a knee osteoarthritis case, and the reference listings are marketed for the neck, shoulders, waist, knees and legs generally. A population with a different condition is a different question.
Comparator
The comparator was piroxicam gel. There was no placebo arm. The reported absence of a significant between-group difference does not by itself establish superiority, equivalence or non-inferiority; section 11 examines the actual analysis and its limits [K03].
Size and setting
The study enrolled 90 participants and reports 80 completing follow-up, at one centre in a defined period. Keep those denominators separate from the n=24, n=28 and n=28 column labels in its outcome table, as discussed in section 11. None establishes an outcome for a different product [K03].
| Bridge | Trial | Reference products | Status |
|---|---|---|---|
| Species | A. absinthium | Not stated | Not established |
| Preparation | 3% ointment / liniment | Balm oils, paste, hot-melt adhesive | Not established |
| Dose | 3× daily, 4 weeks | “Suitable amount”; 6–8 h wear | Not established |
| Population | Knee OA, 30–70 | General musculoskeletal use | Not established |
| Comparator | Piroxicam gel | None | Not applicable |
| Size | 90 enrolled; 80 completed; single centre | No trial | Not established |
6. Reading outcome measures without a statistics degree
Two measures appear in the trial, and both are easy to misread.
Visual analogue scale (VAS)
A line, usually 10 cm, anchored at “no pain” and “worst pain imaginable”. The patient marks a point; the distance is the score. Whether a one-point change matters depends on the baseline and on the minimal clinically important difference for that condition, which the trial report does not give in the text reviewed.
WOMAC
A questionnaire with three subscales: pain, stiffness and physical function. Higher scores mean worse symptoms. The trial reports total pain, total physical function and total stiffness scores [K03]. When a result says “improved on all factors except stiffness”, it means the stiffness subscale did not show the same change — not that stiffness was unaffected in every patient.
Statistical significance versus importance
Statistical significance does not state the size or clinical importance of a change. The source’s means and comparisons are examined in section 11. A reader should retain the outcome, comparison and uncertainty rather than translate “significant” into “large” or “no significant difference” into “equivalent”.
8. What product-specific evidence would look like
If a buyer wanted evidence about a specific product rather than about a species, the minimum would be:
- A characterised product. Species, plant part, preparation, concentration, full formula and batch.
- A defined use. Site, amount, frequency, duration and population.
- A defined outcome. A validated scale, a measurement time, and a pre-specified analysis.
- A comparison. Placebo, vehicle, or an active comparator, with blinding where feasible.
- A report. Enough detail to reproduce the method, with conflicts of interest declared.
Anything less can still be useful — a skin-penetration study, a stability study, a user study — but it answers a narrower question. The error is not using weak evidence; it is using weak evidence for a strong claim.
9. Writing an evidence note
An evidence note is a short record that keeps the claim and the support together. The following template can be completed for any product in this category.
| Field | What to write |
|---|---|
| Claim being assessed | One sentence, no marketing language |
| Product identity | Code, variant, species, preparation, concentration |
| Source type | Label, user report, trial, review |
| What the source studied | Population, product, dose, duration, comparator |
| What the source found | Outcome and direction, with the authors’ wording |
| What it does not cover | The gaps that prevent transfer |
| Read state | Full text, abstract, or metadata only |
| Conclusion | Supported, partially supported, or not supported |
Filling this in for the Artemisia absinthium trial produces a conclusion of “partially supported for a 3% ointment in knee osteoarthritis; not supported for a mugwort patch”. That is a useful sentence. “Mugwort is proven” and “mugwort is unproven” are both less useful, because neither says what was studied.
11. Reading the trial’s actual numbers
The study summary above should be checked against the full report, not used in place of it. Table 2 of the Basiri study presents outcome means at baseline, the end of four weeks of treatment, and two weeks after treatment stopped. The following table extracts only its VAS pain means. It does not reproduce the complete statistical analysis or provide a treatment recommendation. Basiri et al., Table 2.
| Study arm | Baseline | Week 4 | Week 6 | Week 4 minus baseline | Week 6 minus week 4 |
|---|---|---|---|---|---|
| A. absinthium ointment | 5.99 | 5.07 | 4.93 | −0.92 | −0.14 |
| A. absinthium liniment | 6.11 | 4.79 | 5.51 | −1.32 | +0.72 |
| Piroxicam gel | 6.05 | 4.12 | 4.60 | −1.93 | +0.48 |
The first three numeric columns are reported group means. The last two columns are our subtraction of those published means. They are not participant-level analyses, confidence intervals, adjusted effects or results of new significance tests. The source table also reports standard deviations, which are omitted from this compact display; consult it before interpreting the variability of the observations.
A visible change and a statistical test answer different questions
The ointment mean is lower at week 4 than at baseline. The liniment mean is also lower at week 4, then rises at week 6. The piroxicam mean rises numerically between weeks 4 and 6 as well. The presence of that last increase illustrates why the words “no recurrence” should not be read as “every recorded number stayed unchanged”.
In Table 2, the reported P value for the week-6 versus week-4 VAS comparison is 0.472 for ointment, 0.01 for liniment and 0.056 for piroxicam. At the article’s chosen threshold, the liniment change is marked significant, while the other two are not. Thus a numeric increase can appear in a group whose reported test does not reach the chosen significance threshold. The appropriate description retains both the direction of the mean and the actual result of the comparison.
These comparisons are within study arms over time. They do not themselves estimate the difference in benefit between ointment and piroxicam. Taking the difference between two simple changes in our display would still not supply the missing uncertainty of that between-arm estimate. Individual data, variation and the specified analysis matter; a subtraction of rounded published means is descriptive arithmetic.
What “non-inferiority” adds to the reading task
The authors describe the design as a non-inferiority trial. The statistical-methods section includes a sample-size calculation and refers to a non-inferiority limit of −8. This means it would be inaccurate to say that the report never mentions a margin. It also does not mean that a reader can take any nonsignificant comparison in the paper as proof of non-inferiority.
The main results table presents means, standard deviations and within-group comparisons; the narrative reports no detected difference in trends across the three groups using repeated-measures ANOVA. That is the analysis result to report as written. A product claim of equivalence or non-inferiority would require a clear connection among the relevant endpoint, the margin, the estimated between-treatment difference and its uncertainty. The compact table above does not supply that connection, and we do not reconstruct it from a P value.
The buyer’s conclusion should therefore be narrower than “as effective as piroxicam”. The paper reports changes under its studied conditions and a particular comparison analysis. It is a source to examine, not a ready-made comparative claim for a retail product with another species, preparation, carrier, application regimen or population.
Keep the enrolment and table denominators visible
The report states that 90 participants were enrolled, with 30 allocated to each arm, and that 80 completed. Table 2 labels the ointment, liniment and piroxicam arms with n=24, n=28 and n=28 respectively. Those labels sum to 80, so they should not be described as three complete groups of 30 when presenting the outcome table.
The methods call the analysis intention-to-treat and define its population as randomized participants who took medication and supplied baseline and post-baseline efficacy assessments. That is the authors’ stated definition. Preserve it alongside the completion and table counts; do not silently replace it with an assertion that outcome data were available for every randomized participant.
The results say discontinuation was higher in the ointment group. That observation is relevant to the denominator, but does not establish why each person left. We do not infer a numerical adverse-event rate, a customer-satisfaction rate or a formula defect from an unexplained dropout count. Nor do we convert the 80 completers into 80 people who benefited.
The measured knee is part of the study object
The protocol treated both knees when participants had bilateral involvement, but evaluated the knee with more severe pain. The participant and the evaluated knee should consequently remain distinguished when summarizing the study. A reader should not double the sample because some participants had two treated knees, or assume that every knee treated contributed an independent outcome.
The same principle applies to later product evidence. Count the unit actually measured: people, knees, applications, patches or laboratory samples. A box containing twelve patches does not create twelve independent clinical participants. Multiple observations from one person likewise need their actual study relationship rather than a larger marketing denominator.
Population exclusions limit the question answered
The trial concerned primary knee osteoarthritis in adults aged 30–70 at a single clinic. Its exclusions included several other musculoskeletal or medical conditions, pregnancy and nursing, specified recent medicine use, and severe pain. These details explain the population studied; they are not general instructions to stop medicines or criteria for a person to select treatment.
They also prevent broad transfers. A public post about chronic pain from an unverified cause is not the same population. A report about difficulty applying a product to the back is a usability question, not evidence that this knee trial answers it. Separate those questions even when the search results place the posts and the study next to one another.
The useful final evidence note can now name the species and preparation already discussed, identify the actual population and measurements, retain all relevant denominators, and report the findings with their analytical limits. This is a more reproducible use of the trial than repeating its conclusion sentence without the table that qualifies it.
Technical evidence records
12. Similar composition, similar structure and clinical equivalence are different conclusions
A product’s list of ingredients, physical profile and clinical results describe different evidence. Matching one does not silently establish the others. This distinction matters when a seller describes a balm as equivalent to a studied product, or when a buyer compares a patch with a topical preparation mentioned in research.
The FDA’s final March 2026 Q3 guidance, sections I, III and IV, gives a defined pharmaceutical framework for comparing physical profiles of topical products submitted in ANDAs. That framework is not a determination that any reference botanical product is bioequivalent, approved or eligible for a particular regulatory route.
Description comes before comparison
Q3 characterisation describes the physicochemical and structural attributes of an individual formulation. It can exist without a comparison. To compare products, the reader needs to know which test product and which reference standard were characterised, which attributes were relevant, and how the profiles were compared.
The guidance distinguishes a basic description useful for identifying dosage form from a comprehensive profile of attributes potentially important to performance. The latter is selected for the nature and complexity of the reference standard. It is not a universal panel in which a buyer can tick a few familiar terms and conclude that two botanical products are equivalent.
In the ANDA setting, the FDA recommends considering relevant product-specific and other guidance alongside this document. A comprehensive Q3 comparison can support a demonstration of bioequivalence with the appropriate evidence. The guidance should not be rewritten as “a similar viscosity proves equal clinical effect” or “the same ingredient names prove the same physical structure”.
Variation has to be represented in its actual context
The source recommends relevant comparative Q3 characterisation using at least three batches of the test product and three batches, as available, of the reference standard. That is a recommendation within this defined drug-development programme. It is not a general retail sample-count rule or the number of participants needed in a clinical trial.
For an actual report, retain the batch information and comparison conditions rather than reducing the result to one selected sample. A test-product attribute falling outside the characterised reference variability raises a different question from a difference between two isolated numbers measured without a common basis. The report and the applicable assessment determine the significance.
Physical differences can identify questions about performance
The FDA describes potential consequences of Q3 differences, including changes in solubility or stability, phase states, diffusion and partitioning, changes after application, and the active ingredient’s activity profile. These can matter to availability and bioequivalence in the pharmaceutical context. They are possible mechanisms to investigate, not measured findings about the mugwort reference set.
Conversely, a physical profile that appears comparable does not supply a missing clinical study or resolve a different botanical species, preparation or formula. The trial’s actual group results and the identity chapter remain the appropriate records for those questions.
The useful evidence statement names the tested products, the kind of comparison performed and its actual conclusion. Keep an ingredient comparison as an ingredient comparison, a structural comparison as a structural comparison, and a clinical conclusion within the study that supports it. The cited guidance is nonbinding except where it refers to specific legal requirements, and it expressly does not address Q3 characterisation for product quality control.
13. Read the adhesion data before accepting the headline
A headline saying “good adhesion” does not identify whether a report contains area observations, a mean score, a comparison with another product or a force measurement. The FDA’s final July 2026 ANDA adhesion guidance, sections III–V, provides a detailed example of how these records are kept and analysed. The following is a reading guide to that example, not a test result or acceptance criterion for a mugwort patch.
Observed scores and analysis scores are separate records
The source calls for the actual adhesion observation at each scheduled assessment, with successive assessments made independently. It also specifies a convention for the comparative noninferiority analysis: after baseline, the highest observed detachment score is carried forward until a higher score occurs. Complete detachment is assigned the completely detached category for the remaining scheduled assessments.
The source’s example makes the distinction concrete. Observations of 1, 2, 1, 3 become analysis values of 1, 2, 2, 3. The third actual observation remains 1 in the raw record. Its analysis value is 2 because of the stated convention. Replacing the raw observation with the imputed value would lose information about what was seen.
With equally spaced post-baseline observations, the arithmetic mean of the raw example is 1.75, while the mean of the analysis values is 2. These calculations explain the convention; they are not results from a study of the reference products. The guidance calls for submitting both raw and analysis data when imputation is applied.
Time points determine what the mean represents
The guidance recommends equally spaced assessments throughout the labelled wear period. It explains that collecting disproportionately many observations early, when attachment may be better, can make a simple mean unrepresentative of the whole period. A report’s time schedule therefore belongs alongside its mean score.
Where observations are unequally spaced, the source describes interval-length weighting. Its example uses post-baseline assessments at 2, 4, 8, 12 and 24 hours over a 24-hour period. The intervals represented are 2, 2, 4, 4 and 12 hours, respectively. The final observation represents a different interval from the first; an unweighted average of these observations answers a different question.
This example does not prescribe a wear period for a herbal patch. The relevant medicinal reference label and study purpose determine the study setting. For any offered product, preserve the actual assessment schedule rather than borrow a familiar schedule from another report.
Count people and repeated observations separately
The guidance includes a hypothetical frequency table with 100 subjects per treatment and five assessment times. Summing the five rows gives 500 assessment records per treatment; it does not create 500 distinct participants. A buyer should therefore read the denominator before repeating a large number in a product claim.
For the hypothetical test treatment, the combined rows contain 443 score-0 observations and 57 score-1 observations. The calculation 57 divided by 500 gives 0.114, reported as a mean score of 0.11 after rounding. For the hypothetical reference treatment, the combined frequencies give a mean score of 0.436, displayed as 0.44. These are illustrative values in an official document, not measured performance of a named commercial patch.
Neither 0.11 nor 0.44 is a percentage adhered, a percentage detached or a pain-response rate. The categories describe different ranges of adhered area, so a mean category score cannot simply be multiplied by a constant to recover a precise mean area. A report should retain the frequency distribution and relevant individual events alongside the summary.
A comparative conclusion needs its specified analysis
In this ANDA framework, the FDA specifies a noninferiority comparison of treatment mean adhesion scores with a margin of 0.15 on the defined five-point scale. The source states that this margin does not transfer to another adhesion scale, transformed scores such as logarithms, or median scores. A document cannot keep the familiar margin while changing the quantity being compared.
The analysis population is also specified. The guidance discusses reasons for exclusions, early removal and dropout, and sensitivity analyses for potentially informative or unbalanced missing data. A result based only on the patches still present at the last visit can mean something different from the prespecified comparison across the relevant population and wear period.
For a combined adhesion and pharmacokinetic bioequivalence study, the source treats the purposes and analysis populations separately. Adhesion observations do not become blood-concentration results, and a pharmacokinetic result does not supply an absent adhesion record. A single study can collect both kinds of evidence without making them the same endpoint.
For a buying decision, ask for the actual product and comparator, scale, observations, time schedule, population and analysis conclusion. Keep this statistical example within its medicinal regulatory context. It explains what a report can contain; it does not approve the construction, label, wear time or effectiveness of the reference mugwort patches.
14. A safety-study average does not replace adverse-event records
A small mean response can coexist with a more substantial reaction in a few participants. To understand a skin-study headline, read the distribution, the reasons people stopped and the reported complaints, not only the summary score.
The FDA’s July 2026 revised draft for ANDA TDS irritation and sensitization studies discusses this problem explicitly. It remains a draft medicinal research framework. The distinctions below explain how to read such evidence without declaring a mugwort product safe or assigning its study requirements to every herbal patch.
A score needs its own scale and endpoint
The draft separates a dermal-response scale from a scale for other effects, with specified handling for analysis. A score from that framework is not the five-point adhesion score described in the preceding section. Similar-looking numbers can describe different observations.
The source also says that a mean irritation score alone can be difficult to interpret clinically: a small number of higher responses and a larger number of lower responses can produce a similar mean. It therefore considers the proportion with excessive irritation and when it occurred, as well as the comparative mean. A headline reporting only the mean leaves those questions unresolved.
Read complaints and adverse events alongside scored observations
The draft asks for an overall adverse-event assessment including everyone who received at least one dose. It addresses all adverse events, including systemic events, and separate documentation of application-site complaints such as dryness, itching, burning, pain or soreness. Those complaints are not erased because a visual score is low.
For a report, distinguish the number treated, the population in a particular endpoint analysis, the people who discontinued and the events recorded. A statement that no event occurred in one selected analysis group should not be expanded to everyone originally exposed without checking the records.
The source also calls for the actual site and relevant treatment relationship to be recorded. That preserves the difference between an event that happened during a study and an event assessed as related to a particular treatment. Neither category should silently be substituted for the other.
A control is defined by what it contains and why it was used
The draft discusses vehicle and positive-control delivery systems where safety concerns preclude the usual comparison. A vehicle is intended to match the inactive formulation without the active ingredient; differences need explanation and supporting information. A generic “blank patch” is not automatically that defined vehicle.
It also discusses selecting a suitable positive control in advance, rather than trying several and choosing one after seeing the results. That is a safeguard against changing the comparison to fit an observed outcome. A buying dossier should retain the actual comparator and prespecified purpose when summarising the report.
Separate a complete result from a convenient excerpt
Observer training, consistency between scorers, randomisation and evaluator blinding where possible are part of the draft’s study discussion. They concern how observations are made and potential bias is managed. An attractive photograph alone does not establish those features.
When requesting evidence for a selected patch, ask for the exact product version, study purpose, population, endpoint definitions, relevant results and adverse-event information. A summary can then be written at its actual scope. It should not turn a comparison under a controlled protocol into a guarantee for every user, site, frequency or wear period.
15. Clinical evidence belongs to the preparation and batches studied
A study of a botanical product does not describe only a plant name. It also involves particular raw materials, preparation and manufacturing conditions, finished formulation and batches supplied to participants. Understanding that connection helps explain why a study of one Artemisia preparation cannot be assigned to every product carrying a related common name.
The FDA’s final 2016 Botanical Drug Development guidance, sections III.B, VI.C and VI.F, discusses therapeutic consistency for botanical drugs developed through investigational and new drug applications. The scientific discussion is used here to read evidence. It is not a determination of the regulatory status or therapeutic consistency of a reference balm or patch.
Chemical and clinical consistency are connected questions
The guidance describes a totality-of-evidence approach that can include raw-material controls, chemical testing, manufacturing controls, biological assays and clinical information. This responds to the complexity of botanical mixtures and possible uncertainty about the active constituents. It does not mean that any collection of certificates automatically proves consistent clinical effect.
A chemical profile can provide information about composition, while clinical outcomes concern people receiving the product under the study conditions. The connection between the two needs evidence. A passing chemical test should retain its actual conclusion rather than become an unqualified therapeutic guarantee.
The source also discusses keeping clinical-study material and the proposed marketed material consistent through a robust manufacturing process. If raw materials or processes change, the relationship to earlier evidence may need bridging information. That is a reason to identify which version was studied, rather than assume that a familiar brand or plant name establishes continuity.
Multiple batches can reveal a question hidden by pooled results
The guidance considers batch-effect analyses when variations between product batches could affect clinical outcomes. These analyses are additional to a primary comparison that pools treated participants across batches. An overall treatment result can therefore coexist with an unanswered question about consistency among the particular batches.
The batches selected for such work should represent the intended marketed batches, rather than be so homogeneous that relevant variation is excluded. The guidance discusses representative raw-material sources in late-phase development as part of that problem. These recommendations concern a defined botanical-drug programme, not a universal rule for the number of commercial suppliers or samples a buyer must obtain.
For a research report, useful information includes which batches participants received, the basis for their selection and whether batch effects were analysed. If that information is not reported, record the limit. Its absence does not establish that the product was inconsistent; it establishes that the report has not answered that particular question.
Site and batch must not be confused
The FDA gives a specific example of confounding: if each clinical site receives only one batch, differences in outcomes between batches can also reflect differences between sites. A comparison cannot simply assign every observed difference to the manufacturing batch.
The guidance therefore recommends planning appropriate batch assignment and analyses in advance where such work is warranted. It discusses tables or forest plots with estimates and confidence intervals by batch, relevant baseline characteristics and assessment of possible confounders. The figure or table needs its actual design context before it can support a consistency statement.
It also distinguishes receiving one batch throughout a study from changing batches during follow-up. Where participants receive a sequence of batches, an analysis may concern that sequence rather than a single isolated batch effect. A report that compresses both situations into “three batches tested” discards part of the evidence.
A dose-response finding needs its own controlled result
The guidance describes another possible source of evidence: a controlled study can investigate whether clinical effects are sensitive to dose over the studied range while also establishing an appropriate comparison with control. A similar effect across several studied doses may reduce certain concerns about variation, under the conditions described.
This is not a rule that the amount of a botanical ingredient never matters or that buyers may adjust application amounts freely. The actual controlled results, studied range and formulation are essential. An ingredient list without quantities cannot supply such a dose-response finding.
Use this framework to state the evidence gap precisely
For the mugwort-related product set, the relevant comparison still begins with species, plant part, preparation, complete formula and actual instructions. The Artemisia absinthium trial discussed earlier remains evidence about its own preparation, participants and methods. This consistency framework does not change its species or turn it into a study of the reference patches.
A useful research summary therefore identifies the exact studied product and reports its result without silently extending it to unstudied preparations or production histories. A buyer can ask for the missing link between a claim and the offered configuration while keeping the published study’s original contribution intact.
16. Biological activity, marker content and clinical effect are separate records
A report may quantify a chemical marker, measure activity in a biological assay or describe clinical outcomes. All can contribute information, but their units and conclusions differ. The buyer needs to know which kind of result supports the statement being made about a topical product.
In section VII.B of its 2016 Botanical Drug Development guidance, the FDA discusses biological assays where chemical testing alone may be insufficient to ensure quality and therapeutic consistency of a botanical drug. It prefers an assay reflecting the known or intended mechanism of action. This is a medicinal development framework, not a biological result for any reference mugwort product.
A chemical amount does not state an activity value
A marker assay reports the selected chemical quantity under its method. A biological assay measures a response in its defined system. One result should not be relabelled as the other simply because both appear in the same supplier file.
The guidance discusses measuring batch potency or activity relative to a suitable reference, with activity expressed in units calibrated against that reference. Those units are not automatically milligrams of plant, grams of extract or an amount to apply to skin. A conversion needs an established relationship, not an assumption based on a product name.
For a report, retain the assay’s actual purpose, system, endpoint, reference and reporting unit. If two laboratories use different assays or references, their numbers do not become directly comparable merely because both are called activity units. The relationship between the methods needs to be established.
Assay controls matter to interpretation
The source notes that biological-assay results are inherently more variable than most chemical-assay results. It calls for system suitability and quality controls so the assay performs reproducibly and predictably. A report’s activity number should therefore be read with the supporting method information rather than as an isolated potency score.
The guidance discusses validation addressing accuracy, precision, specificity, linearity and range. These are attributes of the analytical procedure, not claims that a finished patch is clinically effective. A document saying a method was validated should identify the method and relevant scope; it does not approve every use of the resulting number.
The FDA also discusses the possibility that different intended indications may require different biological-assay considerations. That reinforces the need to identify what response was evaluated. A result in one test system should not be expanded into an unrestricted list of product functions.
A laboratory response is not a patient outcome
Clinical evidence concerns the people, product, exposure and outcomes studied. A biological assay can support a defined quality or consistency question without itself being a clinical study. It does not establish how much a person absorbs from a patch, how a mixed balm will feel or whether an individual will experience relief.
For the current reference set, a useful evidence request is therefore specific: provide the exact formulation or preparation represented, the type of assay and its conclusion, and the link between that record and the proposed claim. If a document concerns only an upstream extract, preserve that specimen rather than call it a test of the complete patch.
Read the evidence together without merging the endpoints
The FDA’s totality-of-evidence approach can bring raw-material controls, chemistry, manufacture, biological assays and clinical information into one assessment. The approach depends on the relevance and quality of those records. It is not a method of adding unrelated tests until an efficacy claim appears justified.
A clear summary names the contribution of each record: the chemical result identifies or quantifies a defined constituent; the biological result concerns its stated response system; the clinical result concerns the studied population and product. Keeping those roles explicit lets the evidence support a precise statement without manufacturing a therapeutic conclusion.
17. Read prior human experience as a documented record
A statement that a botanical preparation has been used before can contribute context, but it leaves several questions open. Which preparation was used, how was it used, who was exposed, what was recorded and how does it relate to the product now being discussed? A long history attached to a plant name does not answer those questions on its own.
The FDA’s 2016 Botanical Drug Development guidance, sections V.A, V.B and V.C, describes prior human experience in a defined investigational-drug setting. Its discussion is useful for reading an evidence file. It is not an exemption or market authorisation for a reference topical product, and this chapter does not summarise the current legal route for selling one.
Identify the record rather than repeat its headline
The guidance discusses earlier clinical studies, marketing history, published literature and relevant compendia. For clinical work, it calls for critical review of the data quality and relevance to the proposed use. These sources are not interchangeable: a clinical report, a sales record and a book describing a traditional preparation contain different information.
For a reader, keep the source type visible. A description in a traditional reference can establish what that source describes. A marketed-product record can document a particular supply history. A controlled study can report its outcomes under its methods. None should silently acquire the strengths of the others when condensed into a product paragraph.
The direct experience record used by this collection remains an individual public discussion. Its contribution is the documented question and reported experience at that scope. It is not converted into a clinical cohort, a verified purchasing history or a measured market merely because it is considered alongside formal research.
Compare the preparation and proposed use
The FDA discusses bridging information between earlier experience and a proposed investigation. Examples include comparing the preparation’s identity, the relevant amount of raw material or traditional preparation, and the clinical setting. That connection is part of the evidence, not an optional inference left to the reader.
For a mixed topical product, name the actual preparation and other relevant formulation information. An earlier account of a botanical material does not automatically describe the balm base, adhesive system or other ingredients in a later product. The identity chapter supplies the vocabulary needed for that comparison without replacing the missing records.
The proposed use also needs its own description. Differences in route, exposure amount or duration can change the relevance of earlier experience. The guidance specifically considers additional assessment where proposed exposure exceeds prior use or differs in route. This does not provide an application amount or a longer wear period for the reference products.
Sales volume and exposed people have different denominators
The guidance’s marketing-history discussion includes annual sales volume, an estimate of the exposed population and adverse-effect rates. These are separate fields. A unit count is not automatically a count of distinct users, and a count of comments is not a count of all exposed people.
The shared-jar question makes the distinction concrete. One container could be used across several clients, while one person could use more than one container. These are explanatory possibilities, not measured usage patterns in this collection. They show why the denominator must be documented rather than supplied by dividing or multiplying a pack count without evidence.
An adverse-effect rate also needs a defined numerator and population. If the record does not state how reactions were collected or who was followed, do not invent those fields. “No adverse effects reported in this source” and “no adverse effects occurred among everyone exposed” are different statements.
Read reliability and relevance separately
A record can be accurately copied while remaining insufficiently relevant to the proposed product. A translation can preserve a claim faithfully without establishing that the claim is true. The FDA discussion asks for justification of the reliability and relevance of foreign-market information in the proposed clinical setting.
The source also calls for literature submitted in that programme to include English and the original language where applicable. The underlying lesson for an evidence file is to retain the source and distinguish it from its translation or summary. A changed plant name, omitted preparation qualifier or shortened warning can change the meaning even when the prose remains fluent.
When comparing an account with the present product, preserve the original observation and any unresolved identity fields. Do not fill a missing formulation from a later advertisement or treat a later formula as the one necessarily used in an old account. Version and timing are part of the relationship.
Experience can guide a question without closing every evidence gap
The guidance considers the relevance of earlier experience case by case and relates the information needed to the development stage and product circumstances. It does not support a universal rule that any traditionally used plant needs no further assessment, nor a rule that every old account has no value.
A useful summary therefore states what the experience records establish, what comparison to the present preparation has been made and what remains unknown. It can identify a question worth investigating while preserving the difference between a historical account, a laboratory result and a controlled clinical outcome.
For the reference balms, patches and liquid formats, that means keeping the current product identity and instructions central. Earlier experience may help frame a document request or explain why someone asks about the product. It does not supply the missing formula, certify an organic claim or establish relief for an individual buyer.
Continue: Chapter 5 — Warmth, cooling or burning