Original research · Open data

The Vocabulary Beats the Mandate: Starch Source Disclosure in Paired US and EU Drug Labels

A within-molecule comparison of 86 medicines authorised in both markets. US labels name the starch's botanical source for 88.4% of molecules against 38.4% for EU leaflets; all 43 discordant pairs favour the US label (McNemar exact p = 2.3e-13).

A within-molecule comparison of 86 medicines authorised in both markets

Abstract

The European Union requires that wheat starch, when used as an excipient, be declared on the product information leaflet. The United States has no equivalent requirement; the FDA's guidance on gluten in drug products has remained in draft since 2017. The reasonable expectation is that EU labels are more informative for patients avoiding gluten.

They are not. We compared starch excipient disclosure within 86 molecules authorised in both markets, holding the active ingredient constant. US labels name the botanical source of the starch for 88.4% of these molecules; EU leaflets do so for 38.4%. Of 43 discordant pairs, all 43 favour the US label. There is not one molecule in the sample where the EU leaflet names a starch source and the US label does not (McNemar exact two-sided p = 2.3 × 10⁻¹³).

The mechanism is not regulatory intent but data structure. US Structured Product Labels encode excipients using the FDA Substance Registration System, whose controlled substance names embed the botanical source: SODIUM STARCH GLYCOLATE TYPE A POTATO, STARCH, CORN. EU leaflets carry free text, which for the identical excipient reads simply Sodium starch glycolate.

A mandate to declare one specific ingredient produced less usable information than a controlled vocabulary adopted for unrelated reasons. We suggest this generalises: for allergen avoidance, structured substance identifiers may deliver more patient value than targeted disclosure mandates.

1. The expectation this paper tests

EMA guidance CHMP/704219/2013 requires that where wheat starch is used as an excipient in a centrally-authorised medicinal product, the leaflet declare it, with the product describable as gluten-free only where residual gluten falls below 20 ppm. The FDA's corresponding document — Gluten in Drug Products and Associated Labeling Recommendations — was issued as draft guidance in December 2017 and has not been finalised. Excipient source disclosure in the US is voluntary.

From this, the celiac literature and patient-facing guidance have generally concluded that European labels are the more reliable. Our own earlier census of 1,360 EMA products found no product declaring wheat starch, which is consistent with the mandate working as designed.

That census also found something that complicated the picture: of 218 EU products containing a starch or dextrin excipient, exactly half named no botanical source at all. A patient reading such a leaflet learns that the starch is not wheat only by inference from regulatory compliance they cannot verify.

The obvious next question — is the US actually worse? — cannot be answered by comparing the two corpora directly, because they contain different drugs. The EU central register skews toward biologics and recently-approved agents; DailyMed is dominated by small-molecule generics. Any difference could reflect product mix rather than regime.

This paper removes that confound by pairing on the molecule.

2. Method

2.1 Pair construction

From the 1,360-product EMA corpus we selected products that (a) contain a starch or dextrin excipient and (b) carry a single, clean INN — excluding combination products and multi-word entries that do not resolve to one active ingredient. This yielded 128 candidate molecules.

For each, we queried DailyMed and retrieved up to three current Structured Product Labels, extracting all ingredients with classCode="IACT" (inactive) together with their UNII codes.

A molecule entered the analysis only where a starch or dextrin term appeared in both the US and EU excipient lists, giving a final paired sample of n = 86.

2.2 Disclosure scoring

A label was scored as source-disclosing if any of its starch terms named a botanical origin (maize, corn, potato, wheat, rice, tapioca, pea). Scoring is per molecule per market, and identical on both sides — no different standard was applied to the two regimes.

2.3 Analysis

Because observations are paired, we report the discordant pairs and apply an exact McNemar test. Marginal percentages are reported for readability but the discordance is the substantive result.

3. Results

3.1 The disclosure asymmetry

Paired breakdown across the 86 molecules:

McNemar exact two-sided p = 2.3 × 10⁻¹³.

The complete absence of EU-only disclosures is the striking feature. Across 86 molecules there is no case in which a European patient learns something about starch origin that an American patient does not.

3.2 The mechanism is the substance name

The asymmetry is not caused by US manufacturers volunteering extra detail. It is caused by the naming system.

US SPLs encode excipients through the FDA Substance Registration System, assigning each substance a UNII code and a controlled name. For starch derivatives, that controlled name incorporates the botanical source:

These are, in most cases, the same product from the same manufacturer. The information exists. Only one of the two labelling systems transmits it.

Across US source-naming terms in the sample, potato and corn account for the entire distribution (corn 50 mentions, potato 36). No US label in the paired set names wheat.

The macitentan row is worth dwelling on. The EU leaflet is longer and carries more cross-referenced allergen information — lactose, soya lecithin, sodium — yet omits the one fact the starch itself raises. Free-text leaflets optimise for the allergens someone thought to flag; controlled vocabularies carry whatever the vocabulary encodes, including things nobody was thinking about.

3.3 Sodium starch glycolate, revisited

Our earlier census identified sodium starch glycolate as the largest blind spot in EU labelling: present in 87 products under 17 distinct spellings, with zero naming a botanical source.

The paired data resolves what those 87 products almost certainly contain. Wherever the same molecule appears in the US with sodium starch glycolate, the FDA controlled name specifies potato or corn. Neither is a gluten risk.

So the reassuring reading is available — but note where it comes from. It is inferable only by leaving the European labelling system entirely and consulting an American one. For a European patient with a European prescription and a European leaflet, that inference is not available in any document they possess.

4. Discussion

4.1 Two mechanisms, and the weaker one won

The EU approach is a targeted disclosure mandate: identify the hazardous ingredient, require its declaration. It works on its own terms — no product in our 1,360-product census declares wheat starch, and none should if compliance holds.

The US approach is not an approach at all. Nobody required source disclosure. The FDA adopted a substance registration system for regulatory identification, and that system happens to encode botanical origin because starch derivatives are chemically distinguished by it.

The accidental mechanism outperforms the intentional one by fifty percentage points, and does so without a single counterexample.

The reason is structural. A mandate answers exactly one question: is it wheat? It answers by absence, and absence is a weak signal — indistinguishable from an omission, a parsing failure, or a non-compliant filing. A controlled vocabulary answers a different and better question: what is it? From a positive identification the patient derives the wheat answer and every other source question for free.

4.2 Implication for the earlier census

This finding materially qualifies our own prior paper. We characterised the EU regime as one where absence of a wheat declaration constitutes weak positive evidence, in contrast to the US where label silence means nothing.

That framing was too generous to the EU and too harsh on the US. The correct statement is that US label silence is rare — for 88.4% of these molecules the US label simply says what the starch is — whereas EU silence is the majority case at 61.6%. The regime that provides no guarantees provides more information.

We flag this because the earlier paper is published and should be read with this correction. This is what the corrections policy is for.

4.3 Recommendation

Our earlier recommendation — require botanical source for all starch excipients, not only wheat — stands, but the implementation should be borrowed rather than invented. The EU does not need a new mandate. It needs a controlled excipient vocabulary.

Adopting substance identifiers with source-bearing controlled names, whether the FDA SRS/UNII system or an EU equivalent, would resolve the source-disclosure gap and simultaneously address the 38.1% name fragmentation we measured in EU leaflets. One change fixes both problems, and the reference implementation already exists and is free to copy.

5. Limitations

• Selection on EU starch presence. Molecules entered the candidate pool because their EU leaflet listed a starch. This does not bias the within-pair comparison — both markets are scored identically within the matched set — but the sample is not representative of all co-authorised medicines.
• Up to three US labels per molecule. Where a molecule has many labelers, we sampled. A different sample could alter individual rows, though the 43-to-0 discordance leaves no plausible room for reversal.
• Formulation differences. Products sharing an INN across markets are not guaranteed to share a formulation. Some pairs may genuinely contain different starches. This is a real limitation for any individual row and a weak one for the aggregate, since it would have to operate asymmetrically in one direction 43 times to explain the result.
• Label-derived. This measures disclosure, not gluten content. No claim is made about any product's actual gluten status.
• Point-in-time. US labels harvested 2 August 2026; EU corpus extracted 30 April 2026.
• Not clinically reviewed. The author is an independent developer, not a pharmacist or physician.

6. Reproducibility

python3 us_fetch.py paired_inns.json # harvest US SPL inactive ingredients python3 analysis_paired.py # pairing, scoring, McNemar

paired_rows.json contains one record per paired molecule with verbatim US and EU starch terms and both disclosure flags. Released CC0.

Corrections: https://glutenscreen.org/about.html#corrections

References

European Medicines Agency. Wheat starch containing gluten as an excipient in medicinal products for human use. CHMP/704219/2013.

US Food and Drug Administration. Gluten in Drug Products and Associated Labeling Recommendations. Draft guidance, December 2017.

US Food and Drug Administration. Substance Registration System — Unique Ingredient Identifier (UNII).

Matson KL, et al. Gluten-Free Options for the Top 100 Pediatric Medications. Journal of Pediatric Pharmacology and Therapeutics 2025;30(3):367–371.

Walsch N. The Source-Disclosure Gap: Starch Excipients in 1,360 EU Centrally-Authorised Medicines. Gluten Screen, August 2026.

Gluten Screen is a free, no-signup medication screening tool. No advertising, affiliate links, or manufacturer funding. Editorial policy: https://glutenscreen.org/about.html#editorial-policy