innoliticsdev

Ask most people what a valid predicate looks like and you get the same checklist: same body region, same modality, same product code. The clearance record says otherwise. One in five AI 510(k)s cites a predicate under a different product code. 85 cleared

Figure 1. Choose the smallest evaluable sample size that reaches the target pass probability. Split that total by the prespecified class ratio and inflate it for missing or unusable cases before enrollment. The short answer There is no single sample size

Figure 1. General workflow for sensitivity and specificity sample size planning. Calculate the condition-positive and condition-negative case targets separately, require both endpoint criteria to pass, and then convert those targets into total enrollment

On August 5, FDA named Cadence the second participant in its TEMPO pilot, for a system called HypertensionOS. When FDA named Dexcom as the first participant on July 22, I wrote about how TEMPO and the CMS ACCESS model are designed to address two problems

I've watched too many device companies do everything right, get FDA-cleared, and then stall out because no one will pay for their device. Evidence and payment are two separate mountains, and most teams don't start on the second one until they've finished

On July 14, 2026, FDA posted a warning letter that should make every software device team re-read their change-control SOP. The agency told Nihon Kohden Digital Health Solutions that a software update it shipped in October 2022, documented internally as a

Introduction PCCPs, or Predetermined Change Control Plans, are a new regulatory tool that allow companies to "pre-clear" certain changes within a 510(k), De Novo, or PMA. While the unique regulatory challenges related to AI are what spurred FDA to develop

On July 16, 2026, CMS and CDC published a Request for Information that asks a question most AI diagnostics companies assumed was settled: when your software helps interpret patient-derived laboratory data, are you still just a medical-device manufacturer,

Can a frontier AI model replace your FDA regulatory consultant? We built a private 1,200-question benchmark, an opinionated distillation of 15 years of consulting judgment, and ran more than two dozen frontier AI configurations against it. Every one faile

An FDA AI model description should explain what each model does, what it receives and produces, how its architecture and features support the intended function, how it was trained and selected, what preprocessing and postprocessing surround it, what param