What this research found
Ten thousand FDA adverse-event reports for implantable cardioverter-defibrillators filed between April and July 2020 were sorted into failure modes using keyword rules plus unsupervised topic modelling, then compared across 36 manufacturers. Failure profiles differ markedly by manufacturer (chi-square 7,075.88, Cramér's V = 0.268), device malfunction is the largest single category at 37.3% of events, and pairwise odds ratios range from 9.5-fold more malfunction for one manufacturer to 64-fold less battery depletion for another.
- Failure mode distributions differ significantly across manufacturers: chi-square = 7,075.88 on 72 degrees of freedom, p < 0.0001, with a Cramér's V of 0.268 reported as a medium-to-large effect.
- Malfunction accounts for 3,728 events (37.3%), followed by battery depletion at 2,257 (22.6%), inappropriate shock at 1,887 (18.9%), infection at 819 (8.2%), recall-related reports at 433 (4.3%) and lead dislodgement at 43 (0.4%).
- Reporting is concentrated: five manufacturers produce 73% of events — ZOLL Manufacturing (23.2%), MPRI (15.6%), Philips Medical Systems (13.3%), Boston Scientific (13.0%) and St. Jude Medical Sylmar (8.5%).
- Pairwise tests with false-discovery correction show large manufacturer-specific skews. ZOLL had 9.52 times the odds of malfunction of St. Jude Sylmar, MPRI had 64 times lower odds of battery depletion than ZOLL (odds ratio 0.016) and 42.77 times the odds of lead fracture compared with Philips, and Philips reported essentially no inappropriate shocks against an 18.9% overall average.
- Topic modelling surfaced themes the keyword rules missed: software and firmware flag malfunctions across 1,371 events, electrode belt cable failures appearing 2,288 times and concentrated in ZOLL wearable devices, and recurring skin irritation and biocompatibility complaints.
- Reports cluster in time, with 66.3% of the four-month total falling in May and June 2020 — a pattern the analysis attributes to a possible recall or pandemic-related effects without establishing a cause.
How it was done
Ten thousand reports were retrieved through the openFDA interface to the Manufacturer and User Facility Device Experience database, covering Class III defibrillators reported between April and July 2020 and drawn from 367,269 available records spanning 36 manufacturers. Report narratives were matched against keyword patterns for eight predefined failure modes, which categorised 6,758 events (67.6%) and left 3,242 for unsupervised discovery. Latent Dirichlet allocation and non-negative matrix factorisation, each fitted with 12 topics over TF-IDF features including bigrams, were then run across the text. A chi-square test with Cramér's V quantified the overall manufacturer-failure association, and 40 Fisher's exact tests with Benjamini-Hochberg correction at a 5% false discovery rate handled the pairwise manufacturer comparisons. The output was a 34-page report with a bipartite manufacturer-failure network, a heatmap of manufacturer failure profiles, and 25 verified citations.
Data sources
- FDA MAUDE via the openFDA API — 10,000 Class III defibrillator adverse event reports, April to July 2020, sampled from 367,269 available records across 36 manufacturers
- 25 peer-reviewed publications on ICD complication rates, used to benchmark the observed distributions
Limitations
The window covers only four months of 2020 and the sample is 10,000 of 367,269 available records (2.7%), so the temporal clustering may be an artefact of that slice. With no market-share denominator, true failure rates cannot be calculated, and because MAUDE is passive surveillance the manufacturer differences may reflect reporting practice and product mix — several categories trace to wearable rather than implanted hardware — rather than device reliability.
Figures from this analysis
How this research was produced
K-Dense Web planned and ran this medical devices investigation end to end — gathering the sources, carrying out the analysis, producing the figures, and drafting the report. The full session transcript, including every intermediate step, is available to view.
