FDA PMA Approvals: The Federal Record of How High-Risk Medical Devices Reach Market

A pacemaker that paces a failing heart, a mechanical valve that opens and closes a hundred thousand times a day, an implantable defibrillator that decides whether to shock—these are the devices that cannot simply borrow their way onto the market by resembling something already sold. They must prove, on their own evidence, that they are safe and that they work. The federal record of that proof is the Premarket Approval database: roughly 56,000 PMA approvals and supplements, each keyed to a PMA number, tracing how the highest-risk medical devices in America earn—and keep—the FDA's most demanding license to market.

This article covers what the PMA dataset is and how the Federal Food, Drug, and Cosmetic Act frames it; the device-classification scheme that funnels the highest-risk products into the PMA pathway; how PMA differs fundamentally from the far more common 510(k) clearance route; the structure of the data—original approvals versus the supplements that track a device's changes over its commercial life; the advisory-committee specialties (cardiovascular, orthopedic, ophthalmic, and the rest) that organize the review; the review timeline and what an approval actually grants; how the PMA table joins to the device-classification and device-recall datasets through the product code and PMA number to reconstruct a device's full regulatory life; a Python workflow that pulls PMAs from the openFDA device/pma endpoint, separates originals from supplements, and tallies approvals by year and specialty; and the caveats—the supplement-versus-device counting trap, the originals-only question, and what a PMA approval does and does not certify—that any analyst must internalize before drawing conclusions.

What the dataset is

The Premarket Approval (PMA) database is the FDA's public record of every device that has cleared the agency's most stringent review pathway. Maintained by the Center for Devices and Radiological Health (CDRH), it documents the approvals granted to Class III medical devices—the highest-risk category—along with the supplements that track changes to those devices after they reach the market. Each record represents a decision: the FDA reviewed an application and, on a particular date, approved a particular device or a particular change to one. Surfaced through the openFDA device API and the PMA database on fda.gov, the record comprises roughly 56,000 approvals and supplements.

In our database this record is stored as the table fda_device_pma, keyed by the PMA number. The grain is not one row per device but one row per approval decision: an original PMA is the first approval, and each later supplement—a new manufacturing site, a labeling change, an expanded indication—is its own row. A single long-lived device can therefore accumulate dozens of supplement records over its lifetime, all sharing the same base PMA number. The columns capture which device was approved, who makes it, when the decision was reached, what kind of approval it was, and which review division handled it:

pma_number                   -- the PMA application number (the join key)
supplement_number            -- "0" / blank for an original; non-zero for a supplement
supplement_type              -- the kind of change a supplement records
supplement_reason            -- why the supplement was filed (new site, labeling, etc.)
applicant                    -- the manufacturer / sponsor that holds the approval
trade_name                   -- the marketed name of the device
generic_name                 -- the device type in plain terms
product_code                 -- three-letter code; joins to device classification
advisory_committee           -- the reviewing medical-specialty panel (CV, OR, OP...)
advisory_committee_description -- the specialty in words (Cardiovascular, etc.)
decision_code                -- the disposition of the application (e.g. APPR)
decision_date                -- the date the approval decision was reached
date_received                -- the date FDA received the application

The pma_number and supplement_number together identify a row. The base PMA number ties every supplement back to the original approval, so the full history of a device's regulatory changes is recoverable by grouping on the base number. The product_code is the second load-bearing column: this three-letter code is the same identifier used throughout the device-classification system, which is what lets a PMA be joined to the device's classification record (its regulation number, its device class, its medical specialty) and, through that, to recalls of the same device type. The advisory_committee field records which medical-specialty review panel handled the application—cardiovascular, orthopedic, ophthalmic, and so on—which is the natural axis for analyzing where high-risk device innovation concentrates. And decision_date stamps each approval in time, turning the dataset into a longitudinal record of how the high-risk device market has grown and how individual devices have evolved.

What it is and the statutory frame

For most of the twentieth century, medical devices reached the US market with almost no premarket federal oversight. The Federal Food, Drug, and Cosmetic Act of 1938 gave the FDA authority over devices, but only to act against products that were already adulterated or misbranded—after the fact. There was no requirement to demonstrate, before sale, that a device was safe or that it worked. A series of injuries from poorly understood devices— most prominently the Dalkon Shield intrauterine device—made the gap intolerable.

The answer was the Medical Device Amendments of 1976, which built the modern device-regulation system and created the Premarket Approval pathway. The Amendments did three things at once: they established a risk-based classification of all devices into three classes; they required the highest-risk class to obtain premarket approval demonstrating safety and effectiveness before marketing; and they grandfathered the devices already on the market in 1976 (the “preamendment” devices), a decision whose consequences ripple through the device system to this day. Later statutes refined the framework—the Safe Medical Devices Act of 1990 strengthened postmarket surveillance, and the Medical Device User Fee program established the fees and performance goals that now govern review timelines—but the 1976 Amendments are the statutory bedrock on which the PMA pathway stands.

The legal standard a PMA must meet is exacting and specific. The applicant must provide valid scientific evidence—in practice, typically data from one or more clinical investigations—establishing a reasonable assurance of safety and effectiveness for the device's intended use. “Safety” here is not the absence of risk but a favorable balance: the probable benefits to health outweigh the probable risks, judged across the population of intended users. “Effectiveness” means that, in a significant portion of that population, the device will provide clinically significant results under the conditions of its intended use. An approved PMA is, in effect, a private license: it authorizes the marketing of that specific device, by that specific manufacturer, for that specific intended use, on the strength of that specific body of evidence. It is not a general permission that competitors can lean on—a rival who wants to sell a similar high-risk device must, in general, file and win a PMA of its own.

Device classification and how products reach the PMA pathway

The PMA pathway is the destination for one slice of the device universe, and understanding which slice requires understanding the FDA's three-class, risk-based classification. The 1976 Amendments sorted every device type into a class according to the degree of control necessary to provide reasonable assurance of its safety and effectiveness. Class I devices—tongue depressors, elastic bandages, examination gloves—pose the lowest risk and are subject only to “general controls” (registration, good manufacturing practice, labeling), with most exempt even from premarket submission. Class IIdevices—infusion pumps, surgical drapes, many diagnostic instruments—pose moderate risk and require, in addition to general controls, “special controls” such as performance standards and, usually, a 510(k) premarket notification.

Class III is the apex of the scheme and the source of the PMA database. It is reserved for devices that support or sustain human life, are of substantial importance in preventing impairment of health, or present a potential unreasonable risk of illness or injury— and for which general and special controls alone are judged insufficient to assure safety and effectiveness. These are the implantable and life-sustaining devices: pacemakers and their leads, mechanical and tissue heart valves, implantable cardioverter-defibrillators, coronary stents, deep-brain stimulators, many intraocular lenses, and a long tail of permanent implants and high-risk diagnostics. For most Class III devices, premarket approval—a PMA—is the required route to market. Each device type carries a product code and a regulation number in the classification database, and it is that classification record which establishes a device type as Class III and PMA-required in the first place. The PMA, in other words, is the marketing authorization that a Class III classification demands; the classification dataset says which device types need a PMA, and the PMA dataset records the approvals that actually cleared.

PMA versus 510(k): the two roads to market

No fact about the device-approval system matters more for interpreting the PMA database than the contrast between PMA and the 510(k) premarket notification, the pathway that the great majority of devices actually travel. They embody two completely different evidentiary philosophies, and confusing them is the most common error in device-data analysis.

The 510(k) pathway rests on substantial equivalence. A manufacturer seeking 510(k) clearance for a moderate-risk device does not have to prove the device is safe and effective from first principles. It has to demonstrate that its device is substantially equivalent to a “predicate” device already legally on the market—that it has the same intended use and the same technological characteristics, or that any differences do not raise new questions of safety and effectiveness. The reasoning is comparative: because the predicate is presumed safe and effective, a device shown to be equivalent to it inherits that presumption. Clearance can often be obtained on bench testing and comparison alone, without a clinical trial, which is why the 510(k) route is faster, cheaper, and far more heavily used—the source of several thousand clearances a year.

The PMA pathway rests on independent proof. There is no predicate to lean on and no equivalence to claim. The applicant must build the case for the device's safety and effectiveness on its own evidence, which for a genuinely novel, life-sustaining device almost always means one or more clinical trials enrolling real patients, with prespecified endpoints, statistical analysis plans, and follow-up periods that can stretch for years. The application is correspondingly larger, the review longer and more involved—often including a referral to an outside advisory panel—and the bar higher. The terminological distinction follows the evidentiary one and should be observed precisely: a 510(k) device is cleared; a PMA device is approved. The word “approved” is reserved, by the agency, for the products that earned it under the more demanding standard. An analyst who treats the PMA and 510(k) datasets as interchangeable counts of “FDA-authorized devices” erases exactly the distinction the two systems exist to draw.

Originals and supplements: the device lifecycle in data

The single most important structural feature of the PMA dataset—and the one most likely to trip up a naive analysis—is the distinction between an original PMA and a PMA supplement. An original PMA is the first approval of a device: the moment a new Class III product clears review and is licensed to market. In the data it carries a supplement number of zero (or blank). It is, conceptually, the birth certificate of a high-risk device.

But a device's regulatory life does not end at approval. Almost any meaningful change to an approved device—and to the conditions under which it is made and sold—requires FDA review, and each such change is filed as a supplementto the original PMA, carrying the same base PMA number with a non-zero supplement number. The reasons span the device's whole commercial existence: a new or modified manufacturing site; a change to the device's design or materials; a revision to its labeling or instructions for use; a change of ownership; and—most consequential clinically—an expanded indication, in which the manufacturer seeks approval to market the device for an additional patient population or condition, often on the strength of a fresh clinical study. The FDA recognizes several supplement types calibrated to the change's risk, from full-review “panel-track” supplements for the most significant changes (such as a new indication) down to lighter-touch notifications for minor modifications. Over a long-lived device's life—a workhorse pacemaker platform, a durable heart-valve line—the supplements can number in the dozens, each one a row in fda_device_pma sharing the original's PMA number.

This structure is what makes the dataset rich and what makes it treacherous to count. The roughly 56,000 records are not 56,000 distinct devices; they are originals plus supplements, and the supplements vastly outnumber the originals. The number of original PMAs is far smaller—the true measure of how many novel high-risk devices have been approved. Any analysis of “how many devices the FDA has approved” must filter to originals (supplement number zero); any analysis of “how much regulatory activity a device generated” should count its supplements. Reading the supplement number correctly is the difference between counting devices and counting paperwork. At the same time, the supplement record is itself a valuable signal: a device that accumulates many indication-expansion supplements over time is a commercial and clinical success story, while the supplement stream also reveals how manufacturing footprints shift and how a device's labeling evolves in response to real-world experience.

Advisory committees, review timelines, and what approval grants

Each PMA is handled within a medical-specialty review division, and the dataset records the corresponding advisory committee—the panel that, for the most significant or novel applications, may be convened to review the evidence in a public meeting and advise the agency. The committee field is the cleanest available proxy for the device's clinical domain: Cardiovascular (the largest by volume, reflecting the dominance of cardiac implants—pacemakers, defibrillators, stents, heart valves), Orthopedic (joint implants and spinal hardware), Ophthalmic (intraocular lenses, surgical lasers), along with neurology, general and plastic surgery, clinical chemistry, and the rest. Tallying approvals by advisory-committee specialty is the standard way to map where high-risk device innovation concentrates and how the center of gravity shifts over time.

The dataset also carries the dates needed to measure review time: a date the application was received and a decision date. The interval between them—the FDA review clock, which excludes the time an application spends back with the manufacturer awaiting additional data—is a long-studied performance metric for the device program, and the user-fee performance goals are written in its terms. Original PMAs, with their clinical evidence and frequent panel review, take far longer to decide than minor supplements; computing review-time distributions therefore demands the original-versus-supplement split before any averaging. It is worth stressing what the dates do and do not bound: the decision date marks when the FDA decided, not when the device reached patients, and the review interval reflects only the portion of elapsed time the application sat on the FDA's side.

What an approval grants is narrow and specific, and the narrowness is the point. A PMA approval authorizes the named applicant to market the named device for its approved indication for use, manufactured under the conditions and to the specifications the application described. It does not bless a category of devices, it does not transfer to a competitor, and it does not end the manufacturer's obligations: an approved device remains subject to postmarket conditions—often including required postapproval studies—to adverse-event reporting, to manufacturing inspection, and to the recall authority that the device-recall dataset records. The PMA is the entry ticket to the market, not a permanent guarantee of good standing.

Joining to classification and recalls: a device's full regulatory life

The PMA table is most powerful not in isolation but as one chapter in an integrated device record, and two join keys—the product code and the PMA number—stitch the chapters together.

The first join is to the device-classification databasethrough the product code. The classification record is what establishes a device type's class, its governing regulation number, its medical specialty, and whether it is PMA-required in the first place. Joining fda_device_pma to the classification table by product code is what lets an analyst confirm that a given approval really is for a Class III device, read the device type's formal definition, group approvals by regulation rather than by the free-text trade name, and reconcile the advisory-committee specialty against the canonical medical-specialty taxonomy. Without the classification join, a PMA is an approval floating free of the regulatory category that required it; with it, every approval is anchored to a defined, classified device type.

The second join is to the device-recall database. When a marketed device is found to violate FDA law or to pose a risk to health, it can be recalled, and the recall record identifies the affected product—frequently by the same PMA number and product code. Joining approvals to recalls is what closes the loop on a device's regulatory life: it lets an analyst ask whether devices that cleared the most stringent premarket pathway nonetheless went on to be recalled, how long after approval the recall came, which advisory-committee specialties show the highest post-approval recall burden, and whether devices that accumulated many design-change supplements were more or less recall-prone than those that stayed unchanged. The 510(k) clearance dataset adds a third perspective—the far larger universe of moderate-risk devices—so that, taken together, the classification, PMA, 510(k), and recall tables reconstruct the entire arc of a medical device: how it is categorized, how it reached the market, how it changed while there, and how it left if it ever did.

Analytical uses

A national, device-resolved, date-stamped record of high-risk device approvals supports a distinctive set of analyses that no other device dataset can.

Innovation tracking by specialty and over time is the most immediate use. Filtering to original PMAs and tallying them by advisory-committee specialty and decision year reveals where genuinely new high-risk devices are emerging— the cardiovascular dominance, the rise and fall of particular implant categories, the years a new technology cluster appears. Because the data spans decades, it makes the long arc of high-risk device innovation legible in a way anecdote cannot.

Manufacturer and market-structure analysis exploits the applicant field. Ranking the holders of original PMAs and of total approvals shows which companies dominate the high-risk device market and in which specialties—and the supplement stream reveals how those positions are maintained, as established players file indication-expansion and design-change supplements to extend the life of their approved platforms. Device-evolution analysis follows a single device through its supplements: grouping on the base PMA number reconstructs the full history of changes—the indication expansions, the manufacturing moves, the labeling revisions—turning a static approval into a longitudinal biography of a device.

Finally, linking approval to outcome is the analytic payoff of the joins: pairing the PMA record with the recall record by PMA number and product code measures how often the most rigorously reviewed devices are later recalled, how the lag from approval to recall is distributed, and whether review time, supplement count, or specialty predicts post-approval problems—the closest the public data comes to asking whether the front-end rigor of the PMA pathway actually buys back-end safety.

Python workflow: PMAs from the openFDA device API

The script below pulls PMA records for a medical specialty from the FDA's openFDA device/pma endpoint, separates original approvals from supplements using the supplement number, and computes two of the core metrics: approvals by calendar year (from the decision date) and the count of distinct device families among the originals. A second helper tallies approvals across the whole database by advisory-committee specialty using the server-side count parameter. No API key is required for public data within the default rate limits. Because openFDA caps skip+limit paging at 26,000 hits, a full national pull must be sharded—here by advisory committee—and concatenated; any production use should also be validated against the current openFDA device/pma field reference, which is the authoritative source for the field names.

import requests
import pandas as pd
from collections import Counter

# ---------------------------------------------------------------
# openFDA Premarket Approval (PMA) API
# Endpoint: https://api.fda.gov/device/pma.json
# No API key required for <= 240 requests/min and 1,000/day.
# Register at https://open.fda.gov/apis/authentication/ for a
# higher daily ceiling.
#
# This script:
#   1. Separates original PMAs from supplements
#   2. Tallies approvals by calendar year
#   3. Counts approvals by reviewing advisory-committee specialty
# ---------------------------------------------------------------

BASE = "https://api.fda.gov/device/pma.json"


def count_field(field, search=None):
    """Server-side term frequencies via the openFDA count parameter.

    Returns aggregate counts without transferring individual records --
    the fast way to size a slice before downloading rows.
    """
    params = {"count": field}
    if search:
        params["search"] = search
    resp = requests.get(BASE, params=params, timeout=30)
    resp.raise_for_status()
    results = resp.json().get("results", [])
    return {row["term"]: row["count"] for row in results}


def fetch_records(search=None, page_size=1000):
    """Page through PMA records for an optional search expression.

    openFDA lets skip/limit page through at most 26,000 hits, so a
    full pull must be sharded (here, by advisory_committee) and
    concatenated.
    """
    records = []
    skip = 0
    while True:
        params = {"limit": page_size, "skip": skip}
        if search:
            params["search"] = search
        resp = requests.get(BASE, params=params, timeout=60)
        if resp.status_code == 404:   # openFDA returns 404 on empty results
            break
        resp.raise_for_status()
        batch = resp.json().get("results", [])
        if not batch:
            break
        records.extend(batch)
        if len(batch) < page_size or skip + page_size >= 26000:
            break
        skip += page_size
    return records


def is_original(rec):
    # An original PMA carries a blank supplement_number (or "0");
    # any non-zero supplement number is a later change to the device.
    sn = (rec.get("supplement_number") or "").strip()
    return sn in ("", "0", "000")


def analyze(specialty="Cardiovascular"):
    recs = fetch_records(search=f’advisory_committee_description:"{specialty}"')
    if not recs:
        print(f"No PMA records returned for {specialty}.")
        return

    originals = [r for r in recs if is_original(r)]
    supplements = [r for r in recs if not is_original(r)]
    print(f"{specialty}: {len(recs):,} records "
          f"({len(originals):,} originals, {len(supplements):,} supplements)")

    # --- Approvals by year -----------------------------------------
    df = pd.DataFrame(recs)
    df["year"] = pd.to_datetime(df["decision_date"], errors="coerce").dt.year
    by_year = df.dropna(subset=["year"]).groupby(df["year"].astype("Int64")).size()
    print("  Recent approval years:")
    for yr, n in by_year.tail(5).items():
        print(f"    {yr}: {n:,}")

    # --- Originals per device family -------------------------------
    families = Counter(r.get("product_code") for r in originals)
    print(f"  Distinct product codes among originals: {len(families):,}")
    return df


# Approvals by advisory-committee specialty across the whole database.
def by_specialty():
    return count_field("advisory_committee_description.exact")


analyze("Cardiovascular")
# print(by_specialty())

Two practical notes apply. First, the original-versus-supplement split done in the script is the single most important step and the one most often skipped: every count that purports to measure “devices” rather than “decisions” must filter to supplement number zero, and every review-time or approval-volume comparison must hold the original/supplement distinction fixed, because supplements are far more numerous and far faster to decide than originals. Second, for national-scale work—ranking every manufacturer, building the full classification-joined and recall-joined device biography, or computing review-time distributions across decades—the openFDA bulk download files for the device endpoints are far more efficient than thousands of paginated API calls and ship with the version-stamped field definitions for the release, so they should be preferred over the live API for anything beyond an exploratory slice.

Limitations and analytical caveats

The PMA database is the authoritative public record of high-risk device approvals, but it carries structural limitations that an analyst must internalize before drawing conclusions.

Records are approvals, not devices. The roughly 56,000 rows are originals plus supplements, and the supplements heavily outnumber the originals. Counting rows answers “how much approval activity occurred,” not “how many devices exist.” The questions diverge sharply, and the only correct way to count devices is to filter to original PMAs by supplement number. Treating the raw record count as a device count overstates the number of distinct high-risk products on the market by a wide margin— the most common and most consequential error this dataset invites.

The PMA universe is not the whole device market. PMA covers only the highest-risk Class III devices that require premarket approval—a small fraction of devices on the market. The vast majority of devices reach market through the 510(k) clearance pathway or are exempt entirely, and they do not appear here. Any statement about “FDA-authorized medical devices” drawn from the PMA dataset alone describes only the apex of the risk pyramid; it must be read alongside the 510(k) and classification datasets to characterize the device market as a whole, and it must never be presented as a comprehensive count of authorized devices.

Approval is a point-in-time decision on the evidence then available. A PMA records that the FDA found a reasonable assurance of safety and effectiveness on the basis of the data submitted at the time, for the device as then designed and indicated. It is not a permanent certificate of safety, and it does not capture what was learned afterward. Devices that cleared the most stringent review have nonetheless been recalled, restricted, or withdrawn; postapproval studies have sometimes revised the understanding of a device's risks. The approval record must therefore be read together with the recall and adverse-event records to assess a device's real-world performance, never as a standalone verdict that a device is safe.

The coded fields summarize a far larger record. The structured columns—product code, supplement type, decision code, advisory committee—compress applications that can run to tens of thousands of pages of clinical data, engineering analysis, and labeling into a handful of codes. The substantive content of an approval—the trial design, the endpoints met and missed, the conditions and postapproval-study requirements attached—lives in the summary of safety and effectiveness data and the approval order, not in the tabular extract. The dataset is excellent for counting, cohorting, trending, and joining; it is not a substitute for reading the underlying decision documents when the question turns on what an approval actually established.

Held with these caveats in mind, the fda_device_pma table is a uniquely valuable resource: a device-resolved, date-stamped, specialty-coded record of the approvals that stand at the most demanding gate in American device regulation—the gate the highest-risk implants and life-sustaining machines must pass, and keep passing through their supplements, to reach and remain on the market that the device-recall record watches over from the other side.