Reading an FDA device review summary starts with the decision pathway. For a 510(k), the public record typically includes the K number entry, the decision letter, and a summary the submitter files under 21 CFR 807.92. For a PMA, the comparable record includes the approval order and the summary of safety and effectiveness data. The useful reading strategy is to treat these documents as regulatory evidence boundaries, not product reviews [1]. The summary describes equivalence to a predicate device, not proof of better patient outcomes. Checking the indications for use against the company's public statements is the single most reliable verification step, because that line defines what the agency actually cleared [3][4]. A practical read therefore moves from the database entry to the summary sections, then to what is absent. This guide explains that sequence, the key fields, and a concrete evaluation example for a digital health device.

Start with the database entry and product code

Every 510(k) submission receives a number beginning with K followed by six digits, where the first two digits show the year FDA received the submission. A number such as K253055 therefore indicates a 2025 receipt year, and searching by that number, the company name, or the device name returns the public entry [3]. The entry shows the decision date, the decision, the review panel, the product code, and whether a third party performed the review [3]. This is the factual spine of the record.

The product code deserves a click because it links to the device classification, the governing regulation, and the risk class. Most software cleared through a 510(k) is Class II, a moderate-risk class with special controls such as testing requirements [3]. FDA assigns every cleared device a three-letter product code based on its classification, which supports post-market tracking of similar devices and identification of safety signals [2]. Treat the product code as the route to understanding which regulatory controls apply, not as a performance rating.

Read the summary as an equivalence argument

Many submitters file a 510(k) Summary instead of a statement certifying that safety information will be provided on request. The summary must include the submitter identity, device names, the legally marketed predicate, a description of significant physical and performance characteristics, and an intended use statement [4]. FDA's content guidance also requires that the summary contain only information covered in the body of the 510(k) and no puffery or unsubstantiated labeling claims [1].

The summary is structured around substantial equivalence. For a new device, the document names one or more predicate devices, often with their own K numbers, and describes where the new device matches and where it differs [3]. A reader should follow the predicate chain backward to understand the regulatory lineage. The comparison of technological characteristics is where the company argues why any differences do not raise new questions of safety and effectiveness [3]. That argument, not the marketing announcement, is what the agency reviewed.

A suggested workflow for evaluating one digital health clearance

Begin with the K number in the 510(k) database. Record the decision date, product code, and whether a third party reviewed the file. Open the summary and extract the exact indications for use sentence, then copy the predicate device name and its K number if shown [3]. Compare the indication sentence against the vendor's press release or product page. If the announcement claims diagnosis but the indication describes triage or flagging, that is a consequential discrepancy [3].

Next, read the device description and technological comparison for the claimed differences from the predicate [3]. Check the performance section for the test population and reference standard; note what is reported as standalone versus reader study or clinical data. Finally, verify whether the summary cites a predicate chain or a newer decision summary that includes real-world data. FDA's RWE examples show that such data can support clearances, but the public file may provide limited detail on study design [7]. This workflow identifies the evidence boundary without inventing conclusions.

The indications for use are the boundary

The statement of intended use within the summary describes the diseases or conditions the device will diagnose, treat, prevent, cure, or address, including the patient population where appropriate [4]. For a digital health tool, this field states the exact patients, settings, and purpose the clearance covers. A tool cleared to flag possible findings for a radiologist to review is not cleared to make the diagnosis [3]. The distinction matters because company announcements tend to describe a clearance in its broadest terms [3].

When indications differ from those of the predicate, the change may create a new intended use that leads to a not substantially equivalent determination [2]. This means readers should compare the new device's indication line against the predicate's stated use, not the vendor's headline. The FDA 510(k) database includes the indications for use form, the summary, and the decision letter for many submissions [2]. Those documents, not press releases, define the regulatory perimeter.

What the performance section shows and omits

For AI software, the performance section in a 510(k) summary often reports standalone results, such as sensitivity or specificity against a labeled reference set [3]. Preclinical and bench testing may follow recognized consensus standards when applicable, though conformance is voluntary unless a standard is incorporated by reference into regulation [6]. FDA's content guidance asks submitters to present test data and results with logical presentation, scientific soundness, and relevance to the device and intended use [1].

The record does not show price, hospital adoption, or post-deployment performance. Researchers tracing AI devices through the public database found that most algorithm updates came with no reported clinical benefit [3]. A clearance rests on equivalence to a predicate, not on demonstrated better outcomes for patients [3]. Where real-world evidence did support a device decision, FDA's fiscal years 2020–2025 report describes 73 examples, but the underlying public documentation can vary in detail and may not confirm whether a clinical evidence component was real-world evidence [7]. Readers should distinguish evidence submitted from evidence summarized.

PMA summaries read differently

A premarket approval application follows a four-step review: acceptance and filing review, substantive review by FDA personnel, advisory panel review when applicable, and final decision documentation [8]. The approval package includes the approval order, the summary of safety and effectiveness data, and final draft labeling [8]. Unlike a 510(k) summary, the PMA record is built around an order approving the application rather than a finding of substantial equivalence [8].

FDA publishes a Federal Register notice of approval, which includes information about any adverse health effects and provides an opportunity for administrative review [8]. Conditions of approval attach to the approval letter and require adverse event reporting, annual reports, and supplements for certain changes [8]. For a reader, the PMA file therefore offers a different evidence scope: clinical studies and nonclinical studies support an approval decision, while a 510(k) summary supports equivalence to an existing device.