Software Components in FDA 510(k) Clearances: Denominators, Device Class, and Review Timing

FDA 510(k) records contain 18,756 clearance decisions from 2020 through 2025. The software-component field identifies software in 8,593 records and remains blank in 3,582. Software accounts for 45.8 percent of all clearances but 56.6 percent of records that answer the field. The annual all-record share crossed 50 percent in 2025, while the answered-record share remained above 50 percent throughout the study window. Field completeness therefore changes the apparent timing of the majority threshold. Among identified software records, 98.0 percent are Class II. Review-panel data remain absent on 2,617 software records decided since 2023.

1. Cohort and field rules

The cohort includes 510(k) records with decision dates from 1 January 2020 through 31 December 2025. The source now contains 175,718 records through 26 July 2026, but 89.6 percent of 2026 records leave the software field blank. The analysis excludes 2026 because that field is not sufficiently populated for comparison. A true value means that the record identifies a software component. A false value means that the record answers negatively. A blank value remains unknown and is not equivalent to no software.

2. Software prevalence uses two denominators

Software appears in 45.8 percent of all cohort clearances. Restricting the denominator to records that answer the software field raises the share to 56.6 percent. The two measures answer different questions. The all-record share is a lower bound on documented software presence and moves with missingness. The answered-record share describes documented records but assumes that answered and blank records are comparable.

3. Missingness changes the trend

Blank software fields declined from 23.4 percent of clearances in 2020 to 15.7 percent in 2025. The all-record software share rose by 8.4 percentage points across the same interval. The answered-record share rose by 4.7 points. Improved field completion explains part of the observed increase, while the answered records also show a smaller increase. The record supports both effects and does not support attributing the full change to either one.

4. Annual clearance composition

Annual clearance volume remained between 2,890 and 3,335 records. Identified software records increased from 1,261 in 2020 to 1,671 in 2025. Answered records without identified software changed from 952 to 1,038, while blank records declined from 677 to 503. The software count increased within a stable annual clearance denominator.

5. Identified software is concentrated in Class II

Class II devices account for 8,422 of 8,593 software records, or 98.0 percent. Class I accounts for 107 records, while 59 are unclassified. Five records contain another blank or null class representation. The result describes the regulatory class attached to software-containing clearances. It does not measure software risk or autonomy.

6. Review-panel records are incomplete

Among software clearances decided from 2023 through 2025, 2,617 records leave the review panel blank. Radiology leads the populated panel values with 519 records. General and Plastic Surgery follows with 264 after two equivalent labels are consolidated. Cardiovascular records 144, Physical Medicine records 104, and Neurology records 95. These counts describe documented panel assignments and exclude no software record from the underlying cohort.

7. Median recorded decision time remained stable

Median days from receipt to decision remained between 31 and 33 across all six years. The ninetieth percentile increased from 90 days in 2020 to 152 days in 2024 before declining to 118 days in 2025. A stable median therefore coexists with a longer upper tail. These dates describe fields in the clearance record and do not include development, presubmission work, or time before formal receipt.

8. Supported interpretation

The record supports four conclusions within its measurement limits. Software appears in a majority of clearances that answer the software field throughout 2020–2025. The all-record share crossed 50 percent in 2025 because both documented software counts increased and blank fields declined. Identified software clearances are concentrated in Class II devices. The captured software field cannot distinguish artificial intelligence, standalone software, embedded firmware, or clinical importance.

9. Limitations

Blank software and panel fields can represent absent documentation, extraction limits, or a negative value that was not captured. The analysis cannot separate those mechanisms. Class and panel values require consolidation because equivalent labels appear in several forms. Receipt-to-decision dates omit work outside the formal interval. The 2026 records are excluded from analytical figures because software-field completeness falls to 10.4 percent. Later refreshes can change historical counts when source records or parsing improve.