Mobility Atlas research: lightweight-wheelchair-guide
Data and method for the article at mobilitatlas.com/blog/lightweight-wheelchair-guide/
License: CC BY 4.0 (credit "Mobility Atlas")
Prepared: 10 October 2026

WHAT WAS READ
FDA MAUDE medical device reports under product code IOR (Wheelchair, Mechanical), received 1 January 2023 to 31 August 2026 (the latest data in openFDA on 10 October 2026), downloaded through openFDA (api.fda.gov) on 10 October 2026. 3,501 reports downloaded (441 for 2023 to 2025, 3,060 for 2026); 2,041 of the 2026 reports carry no text; 1,460 coded, 1,457 describe a problem or event.
FDA recall records for the same product code, downloaded through openFDA the same day.

HOW IT WAS CODED
1. A codebook (codebook.txt) was written before any report was coded. Its SHA-256 fingerprint, recorded at 2026-10-10T05:55:31Z, is bd0696fd4682c1179d070177915e119bcdd7785fcc14d7cadae8edeb66eb194b. Any later change to the codebook would change this fingerprint.
2. The reports were split into chunks. Two coders (AI language models working separately, each without access to the other's answers) coded every report in every chunk against the codebook.
3. Agreement between the two coders: chair 98.5% (kappa 0.967); category 97.6% (kappa 0.968); injury 99.9% (kappa 0.997); fall 99.9% (kappa 0.992); arrival 99.7% (kappa 0.989); retrospective filing 99.4% (kappa 0.988).
4. Every report where the coders differed was decided by the lead analyst under written rules (below). 72 disagreements decided under rules M1 to M9; a consistency pass on the retrospective flag changed 3 more.
5. Counts in the article were produced by script from the final table, not by hand. Some sub-counts in the article (for example parts named, places named) are keyword counts over the report text, and the article says so where it matters.

ADJUDICATION RULES
M1: CHAIR: a model named anywhere in the report counts, including an owner-manual title quoted in the maker narrative (Quickie 2 = A, Quickie Iris = P); when two different models are quoted the chair is U.
M2: CHAIR: a Medline item number beginning K1, K3 or K4 (for example K1186N13S) names a Medline K1/K3/K4 chair and is S; "Medline wheelchair" with no model is U; "reclining wheelchair" is P; a Drive Silver Sport model number is S; "Focus" alone is U.
M3: CATEGORY: recline or tilt mechanisms, pistons, struts, recline locks and handles are attached parts (R), as is a part named by its function when it is not frame, wheel or wheel lock (legrest, footplate, armrest, backrest pin or button, anti-tipper).
M4: CATEGORY: a broken weld is coded by the part it joins: a weld on a legrest, footrest or armrest bracket is R; a weld on the frame, cross brace, seat rail or back post is F.
M5: CATEGORY: a fall, slide-out or tip with no failed chair part named is T, including when the maker says the chair was not at fault or a non-chair item (a cushion from another maker, a belt) is involved.
M6: CATEGORY: harm from contact with an intact part (a cut, a stubbed toe) with no fall and no failure is X; a vague complaint with no failure named and no fall is U.
M7: ARRIVAL = y only when the report says the problem was found on delivery, unboxing, assembly or before first use; a maker's suspicion of shipping damage alone is n.
M8: RETRO = y exactly when the report text contains the word "retrospective" (applied to every report in the consistency pass).
M9: INJURY and FALL follow the same rules as the power study (R2, R8): pain or medical care = injury y; the user leaving the seat (including throwing himself out) = fall y.

QUALITY NOTES (disclosed in full)
- During coding, some coder runs took shortcuts: a few used scripts on near-identical templated reports instead of reading each one, and two runs briefly shared a scratch folder. Those chunks were re-run with stricter instructions (read every report, no keyword scripts, a private scratch folder). The agreement figures above are for the final runs.
- The article then went through three independent fact-check passes (a full pass, a second pass with a reader review, and a control pass over the changes), each recomputing numbers from the files here and reading quotes against their sources; every error found was corrected before publication.

AMOUNTS
Every report text was searched for money figures. None of the reports gives a money figure. Prices in the article are list prices from makers', dealers' and retailers' pages and the Mobility Atlas catalog, listed in lightweight-wheelchair-guide-documents.csv and the article sources.

LIMITS
- Reports are not rates. Nobody publishes how many devices of each model are in use.
- A report is an allegation or a description; the FDA says the submission of a report "itself does not necessarily demonstrate that the device caused or contributed to the adverse outcome or event."
- Counts follow paperwork: some makers filed many reports at once from reviews of old complaints (see the retro column).
- Reports with no text could not be coded.

FILES
codebook.txt        the codebook as fingerprinted
coded_reports.csv   one row per coded report: both coders' answers, the final code, and how a disagreement was resolved (resolution, rule_applied)
report_pool.csv     the reports in the pool with product code, brand, date received and event type
The full report texts are public: search MAUDE or openFDA by report key.
lightweight-wheelchair-guide-documents.csv  the documents and pages read, with links