# Seven Years of Drift: Measuring the Decay of the Reference List Celiac Patients Are Told to Use **A comparison of glutenfreedrugs.com against current US drug labels** Ned Walsch · Gluten Screen · August 2026 **Data and code:** [`plogsted-drift-2026.csv`](https://glutenscreen.org/data/) (CC0) · `analysis_plogsted.py` · `dm_fetch.py` · `results_plogsted.json` --- ## Abstract For two decades, patients with celiac disease asking whether a medication contains gluten have been directed to a single free resource: **glutenfreedrugs.com**, compiled by pharmacist Steve Plogsted. It remains the most widely recommended answer to the question, cited by advocacy organisations, dietitians, and pharmacists. **Its structured list was last updated 23 April 2019. Its alphabetical list was last updated 28 April 2017.** We compared all 739 structured entries (304 distinct drug names) against current US Structured Product Labels retrieved from DailyMed on 2 August 2026. **31.6% of drug names fail a direct lookup against current labels.** This decomposes into two distinct failure modes: **16.1% have no current US label under any name variant**, and **15.5% cannot be found as written because of typographical or formatting errors in the list itself.** Separately, among the 213 drug–manufacturer pairs where a complete current labeler set could be retrieved, **the manufacturer named by the list is no longer among the current labelers for 67.6%.** We report this not as a criticism of an unpaid community resource maintained for decades, but as a measurement of a gap the celiac community has not acknowledged: **the answer patients are routinely given is, for roughly a third of its entries, no longer actionable, and there is no maintained successor.** --- ## 1. Background The clinical advice given to celiac patients about medications is consistent across guidelines: check whether the product contains gluten, and when the label is unclear, contact the manufacturer. The practical execution of that advice has, for most of the last twenty years, run through one resource. glutenfreedrugs.com is a hand-curated list compiled by Steve Plogsted, a pharmacist at Nationwide Children's Hospital. It records, per product, whether the manufacturer confirmed gluten-free status, and flags the presence of corn starch, potato starch, maltodextrin, lactose, and soy. It is free, has never carried advertising, and represents an enormous amount of uncompensated labour. It is also, by the evidence of its own published timestamps, no longer being updated. This paper measures the consequence. We want to be explicit about intent up front, because the framing matters. The finding here is not that the list is bad. It is that **a community of patients with a serious autoimmune condition is relying on a static snapshot of a market that churns**, and no institution has stepped in to maintain a replacement. That is a structural failure, not a personal one. --- ## 2. Method ### 2.1 Sources - **Reference list:** glutenfreedrugs.com structured list, 739 entries, 304 distinct drug names, `source_updated` field 2019-04-23. - **Comparator:** DailyMed SPL API (`spls.json`), the FDA's repository of current Structured Product Labels. Harvested 2 August 2026; DailyMed database publication date 31 July 2026. ### 2.2 Name resolution Each distinct drug name was queried against DailyMed. A name "resolves" if at least one current SPL matches. Names returning zero results were re-queried under normalised variants: correcting observed typographical errors in the source list, stripping parenthetical annotations and strength qualifiers (`all strengths`, `all types`), converting slash- and hyphen-separated combination names to space-separated, and falling back to the leading brand token. This second pass is essential for honesty. Without it, a list entry reading `Buproprion HCl` — a misspelling of bupropion present in the source — would be scored as a discontinued product rather than a data-quality error. The two failure modes have completely different implications and are reported separately throughout. ### 2.3 Attribution drift For each entry, the manufacturer recorded by the list was compared against the labelers named in current SPL titles. Both were canonicalised by stripping corporate suffixes (Inc, LLC, Pharmaceuticals, Laboratories, and similar) before matching on substring containment in either direction. **Critical constraint:** the DailyMed query returns a maximum of 25 SPLs. For a widely-genericised molecule, that set is truncated, and a manufacturer's absence from a truncated page is not evidence of absence from the market. Drift was therefore computed **only** over entries where fewer than 25 SPLs were returned, guaranteeing a complete labeler set. Removing this constraint inflates the reported drift figure by approximately ten percentage points; we flag this because it is the sort of error that would otherwise be invisible in a summary table. --- ## 3. Results ### 3.1 A third of the list no longer resolves | Outcome | Names | Share | |---|---:|---:| | Resolves against DailyMed as written | 208 | 68.4% | | Resolves **only** after correcting the list | 47 | 15.5% | | No current US label under any variant | 49 | 16.1% | | **Fails a direct lookup** | **96** | **31.6%** | ### 3.2 The list contains errors that block lookup 47 entries (15.5%) cannot be found as written but resolve once corrected. These fall into recognisable categories: - **Typographical errors in drug names.** `Buproprion HCl`, `Buproprion HCl SR`, `Buproprion HCl XL` (bupropion); `Ipratroprium Nasal` (ipratropium); `Dilaudid tabslets`. - **Manufacturer misspellings.** Abilify is attributed to `Otysuka` (Otsuka). - **Formatting artifacts treated as part of the name.** `Amlodipine Besylate -Benazepril Hydrochloride-all strengths`, `Amlodipine Besylate -Atorvastatin-all strengths`. - **Category entries rather than products.** `Insulin all types`. - **Truncation.** `Quillivant XR Powder for` — the entry ends mid-phrase. These matter more than they might appear. A patient searching the list for their medication, or software attempting to query it programmatically, performs an exact or near-exact match. An entry that requires a human to infer the intended drug is, functionally, not an entry. ### 3.3 Products that no longer have a US label 49 names (16.1%) return nothing under any variant. **23 of these — 47% — are oral contraceptives**: Brevicon, Cesia, Cyclessa, Desogen, Estrostep Fe, Generess Fe, Jolivette, Leena, Levlen, Levlite, Lo Minastrin Fe, Lo/Ovral, Lybrel, Mircette, Modicon, Nordette, Norinyl, Nortrell, Ovcon, Quasense, Solia, and others. This concentration is not coincidental. Oral contraceptive brands turn over unusually fast through authorised-generic launches, brand retirements, and portfolio divestitures. Any static list ages fastest in this category — and it is a category where consistent daily adherence matters and where patients switch products frequently. The remaining 26 span cardiovascular (Accupril, Accuretic, Altace, Altoprev, Amturnide, Liptruzet), respiratory (Aerospan, Duoneb, Zetonna), psychiatric and sleep (Brintellix, Edular, Halcion, Intermezzo, Zolpmist, Zyban), thyroid (Levothyroid, Thyrolar), and analgesia (Hycodan, Oxecta, Tussigon). ### 3.4 Manufacturer attribution has substantially drifted | Measure | Value | |---|---:| | Drug–manufacturer pairs with a complete labeler set | 213 | | Listed manufacturer still among current labelers | 69 (32.4%) | | Listed manufacturer no longer among current labelers | **144 (67.6%)** | The drift reflects genuine corporate history since 2019. Adderall XR is listed under Shire, now Takeda. Aranelle under Barr, now Teva. Actonel under Warner Chilcott, now Allergan. Amethia under Watson, now Mayne Pharma. Aceon under Abbott, now XOMA. This is the finding with the most direct patient consequence. The list's core clinical value is the `gf_status` field — "confirmed gluten-free by the manufacturer." **That confirmation is an attribute of a specific manufacturer's specific formulation.** When the manufacturer changes, the confirmation does not transfer. A patient reading "confirmed gluten-free by the manufacturer" for a product now made by a different company is reading a statement that was true, about a product they are not holding. --- ## 4. Discussion ### 4.1 The compounding problem The three findings are not independent — they stack against the same patient. Consider someone prescribed a generic whose brand entry is on the list. They may find the entry resolves (68% likely). They then read a gluten-free confirmation attributed to a manufacturer who, on these numbers, has a roughly two-in-three chance of no longer being among the current labelers. They are then advised to call that manufacturer to confirm — a company that may no longer make the product, or may no longer exist under that name. At no point does the resource signal that any of this has happened. A static list has no mechanism for expressing its own staleness. ### 4.2 Why nothing replaced it The obvious question is why a maintained successor does not exist. The likely answer is that the work is unrewarding in every conventional sense: it requires pharmaceutical knowledge, it must be redone continuously, it produces no publications, and the output is a public good with no revenue model. Plogsted did it anyway, for years. What has changed since 2019 is that the underlying inputs are now machine-readable. DailyMed publishes Structured Product Labels through an open API with per-product inactive-ingredient lists. The European Medicines Agency publishes product information leaflets. The curation task that once required a pharmacist reading labels one at a time is now substantially automatable — not fully, because excipient source ambiguity is irreducible from labels alone (see companion paper on EU source disclosure), but enough that the maintenance burden is no longer the barrier it was. ### 4.3 What a successor requires From these results, four properties are non-negotiable: 1. **Continuous revalidation against primary sources**, not periodic manual review. 2. **Manufacturer-scoped answers.** A verdict must attach to a specific labeler and NDC, because that is the granularity at which excipients actually vary. 3. **Visible staleness.** Every answer should carry the date of the label it was read from. A resource that cannot express its own age will eventually mislead. 4. **Explicit uncertainty.** Where a label does not disclose a starch source, the correct output is "unresolved," not a pass. We note the obvious conflict of interest: the author maintains a tool built on these principles. Readers should weight §4.3 accordingly. The measurements in §3 are independently reproducible from the published code and do not depend on that judgement. --- ## 5. Limitations - **Name-based matching.** Comparison keys on drug names, not NDCs, because the source list records names. Brand-name reuse and reformulation under an existing name are not detected. - **Repackagers count as labelers.** DailyMed labelers include repackagers such as Aphena Pharma Solutions and Physicians Total Care. These are legitimate labelers a patient will encounter, but they are not original manufacturers. The drift figure therefore measures **attribution staleness** — whether the contact the list names is still associated with the current label set — rather than corporate ownership. Where a repackager occupies the returned SPLs, an original manufacturer may still market the product. - **Absence from DailyMed is not proof of discontinuation.** It indicates no current SPL under that name. Products may persist under renamed brands or as unbranded generics. - **US only.** DailyMed covers the US market. Products discontinued in the US may remain available elsewhere. - **Structured list only.** The 1,048-entry alphabetical list (last updated 2017) was not assessed and is two years staler; figures here are likely optimistic relative to the full resource. - **Point-in-time.** Harvested 2 August 2026 against DailyMed's 31 July 2026 publication. --- ## 6. Reproducibility ``` python3 dm_fetch.py plogsted_names.json # harvest current SPLs (cached) python3 analysis_plogsted.py > results_plogsted.json ``` `plogsted-drift-2026.csv` contains one row per drifted drug–manufacturer pair (n=144) with the listed manufacturer, current labelers, and SPL count. Released **CC0**. Corrections: --- ## Acknowledgement Steve Plogsted maintained glutenfreedrugs.com as a free public service for the celiac community over many years, without funding or institutional support. This paper measures the cost of that work not being continued by anyone. It should be read as an argument for institutional investment in a successor, not as a criticism of the person who carried it alone for two decades. --- ## References Plogsted S. *Gluten-free drug list.* glutenfreedrugs.com. Structured list last updated 23 April 2019. Matson KL, et al. Gluten-Free Options for the Top 100 Pediatric Medications. *Journal of Pediatric Pharmacology and Therapeutics* 2025;30(3):367–371. doi:10.5863/JPPT-23-00084. US National Library of Medicine. *DailyMed Structured Product Labeling Resources.* dailymed.nlm.nih.gov. 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. The author is not a clinician. Editorial policy: *