FDA 510(k) Predicate Counts: Parsing Coverage, Distribution, and Regulatory Meaning
A review of 175,718 records separates the 54,423 parsed summaries from historical and partial-period gaps before interpreting predicate counts.
The captured 510(k) corpus contains 175,718 clearance records, but predicate counts are parsed for 54,423. Among parsed records, 25,051 cite one predicate, 19,038 cite two or three, 4,656 cite four or five, and 2,695 cite six or more. Another 2,983 record zero parsed predicates. The mean among parsed records is 2.07, and the maximum is 83. Parsing coverage exceeds 92 percent for each complete decision year from 2008 through 2025, but earlier years and the partial 2026 period remain incomplete. Predicate count describes references captured in a summary. It does not measure chain depth, equivalence quality, evidence strength, or regulatory drift.
1. Cohort and parsing rules
The source spans decisions from 15 July 1976 through 26 July 2026. Predicate count is populated from captured 510(k) summaries and remains null when no parsed count is available. A null count is not zero. A zero count means that the parsed record yielded no predicate entry under the extraction rule. The analytical distribution uses only the 54,423 records with non-null counts.
2. One predicate is the largest parsed category
One predicate appears on 25,051 parsed records, representing 46.0 percent of the parsed cohort. Two or three appear on 19,038 records, while four or five appear on 4,656. Six or more appear on 2,695. Zero parsed predicates appear on 2,983 records. The mean is 2.07, but the upper tail raises the mean above the single-predicate mode.
3. Parsing coverage defines the usable period
Predicate-count coverage is 92.0 percent in 2008 and remains between 95.1 and 99.1 percent from 2009 through 2025. The 2005–2007 records are almost entirely unparsed, while only 12.3 percent of the partial 2026 records contain counts. Mean predicate count declines from 2.37 in 2012 to 1.92 in 2025, but the historical comparison is limited to years with high parsing coverage.
4. High-count records are concentrated in Class II
Among the 2,695 records citing six or more predicates, 2,504 are Class II and 144 are Class III. Class I accounts for 13, while 34 carry another, mixed, unclassified, or missing representation. The largest record cites 83 predicates and describes a group of spinal systems. Other leading records describe portfolios of hip, spinal, dental, fixation, perfusion, and monitoring systems. Counts can therefore reflect grouped product families rather than one narrow device comparison.
5. Count and chain depth are different measurements
A predicate count records references on one clearance. A lineage chain requires resolving each predicate identifier, selecting current records, following its own predicates, and handling references outside corpus coverage. This article does not perform that recursive graph validation. It therefore makes no claim about chain depth, common ancestors, unresolved identifiers, or the age of terminal predicates.
6. Regulatory meaning requires the submission context
FDA substantial-equivalence review compares intended use and technological characteristics against legally marketed devices. FDA distinguishes a primary predicate from additional predicates and reference devices. A higher count can reflect a broad product portfolio, several technological characteristics, or the structure of the submission. Count alone cannot determine whether the comparison is strong, weak, appropriate, or split across incompatible predicates.
7. Supported interpretation
The record supports three conclusions within its extraction limits. One predicate is the largest parsed category, but more than two fifths of parsed records cite at least two. Predicate-count parsing is sufficiently complete for comparison from 2008 through 2025, not for the full historical corpus or partial 2026. High-count records are predominantly Class II and often describe grouped product families. The data do not support a chain-depth or evidence-quality conclusion without recursive graph and document review.
8. Limitations
Predicate lists are extracted from summary documents and can omit references or include formatting artifacts. Zero and null values have different meanings. Device-class normalization remains incomplete for mixed labels. The corpus includes historical clearances without parsed summaries and partial 2026 records with low coverage. No recursive identifier resolution, amendment selection, reference-device separation, or document-level validation was performed.