Iterative Health

For SKOUT the model is the device. The K213686 clearance describes an artificial intelligence based algorithm that detects potential colorectal polyps in live colonoscopy video and marks them on screen, and without it there is only the endoscope's picture.

The company's larger business is a clinical research site network, which it describes as using machine learning to pre-screen patients for trial eligibility. That network is a services business whose value does not rest on a model in the same way. SKOUT is the product the company's description leads with, and the one that places it among imaging vendors.

A detection aid during a live procedure, with the gastroenterologist deciding. The K213686 indication is a computer aided detection tool that gives polyp location information to assist trained gastroenterologists during colonoscopy in adults having screening or surveillance, for white light colonoscopy only. It places blue boxes on the live video. The labeling states that the gastroenterologist confirms or rejects each suspected polyp on their own judgment, that the device is not for primary interpretation, diagnosis or treatment recommendations, and that a status indicator shows a green square when running and a red X on error.

What is not published is how the system behaves outside the conditions it was tested in, such as poor bowel preparation, or what guidance an endoscopist gets for a mark they disagree with. Concurrent marking during a live procedure is the configuration most exposed to automation bias, because the operator cannot afterward separate what they saw from what they were shown.

The approach is described only as an artificial intelligence based algorithm. The K213686 summary gives no architecture, training data description or data provenance, though it does publish the device's outputs, its standalone detection and false mark rates, and results by polyp size and histology. Six clearances from 2022 to 2025 show the device is versioned through FDA filings, but what changed in each version is not published.

On the research side, the eligibility screening models have no published description or validation, and the network's activation and enrollment figures measure operations rather than the models.

A short chain for the device. SKOUT is the company's own cleared model: K213686 and its five successors are filed by Iterative Scopes, the company's legal name, so the model cannot be swapped without a filing, and no outside model provider sits between the video and the marks.

The research side is less clear. The privacy policy lists categories of service providers, including platform hosting and security monitoring, without naming them, and nothing identifies who hosts screening or trial data.

A multicenter, prospective randomized controlled trial with the method available to read. The K213686 clearance summary reports 1,359 patients in the modified intention to treat population, 682 colonoscopies with SKOUT and 677 without, no adverse events or complications, and, as a second primary endpoint, positive percent agreement of 70.9 percent with the device against 75.7 percent without. The published trial, in Gastroenterology, reports a 27 percent relative increase in adenomas per colonoscopy with the device. Standalone testing on 79 videos and 94 polyps sits behind it.

Adenomas per colonoscopy is a surrogate for cancers prevented rather than an outcome in itself. No evidence is published for the research network's screening models; the network's activation and enrollment figures describe operations.

General assurances only. The company's privacy policy, updated 1 September 2026, covers personal information from business customers and website visitors, says information is kept as long as necessary to provide the services and protected by industry standard measures, and lists categories of service providers, including platform hosting and security monitoring, without naming them. It does not address patient data, endoscopic video, clinical records or trial data, and states no retention period, training use position or de-identification standard for any of them. The K213686 summary does not say whether video is stored or leaves the procedure room.

The company holds more than one data estate: procedure video from SKOUT, clinical records screened across partner practices to find trial candidates, and trial data held for sponsors. It trains a detection model on endoscopic video and also operates research sites that generate such video, and nothing published says whether video or records from those sites inform model development or whether patients are told.

No statement of HIPAA status is published, and no privacy document reaches the product. The privacy policy covers business customer and website visitor information and does not mention HIPAA, protected health information, business associate agreements or patients, so it does not cover SKOUT's procedure video or the research network's patient records.

Two regimes are in play and neither is addressed. As a device vendor selling to provider organizations, ordinary business associate analysis applies. As an operator of research sites that screens patient records for trial eligibility before anyone has consented, the governing instruments are institutional review board oversight and a research authorization or waiver, and nothing says under whose authority that screening runs. The policy names an EU data protection officer and representative for customer and visitor data only.

Nothing independent is published. There is no security page or trust center; the footer carries terms, cookies, privacy and two California notices; and the privacy policy says only that industry standard measures are used, with no certification or attestation named. The K213686 summary carries no cybersecurity statement.

The gap matters because of what the company holds: live procedure video in provider endoscopy suites, and trial data for pharmaceutical, biotech and device sponsors whose vendor assurance programs commonly require an attestation before a supplier handles patient level data.

SKOUT holds six FDA 510(k) clearances under the gastrointestinal lesion detection software classification (product code QNP, 21 CFR 876.1520). The original, K213686 on 12 August 2022, was a traditional 510(k) with predicate DEN200055. Special 510(k)s for later versions followed: K230658 (April 2023), K240781 (April 2024), K241508 (July 2024), K251126 (May 2025) and K253664 (December 2025), the last two filed under the Iterative Health name.

The indication is computer aided detection of potential colorectal polyps during white light colonoscopy in adults having screening or surveillance, assisting trained gastroenterologists, and not a replacement for a full patient evaluation. The company presents its research network as services rather than devices.

The K213686 clearance summary publishes detection results by subgroup, each with its confidence interval. In standalone testing on 79 high definition videos containing 94 polyps, the per polyp detection rate was 97.8 percent for men and 97.9 percent for women, with 25.7 and 20.2 false marks per 15 minutes; 98.4 percent for patients aged 40 to 64 and 97.0 percent at 65 and over; 98.1 percent for polyps under 5 mm and for those of 5 to 9 mm, and 100 percent at 10 mm and over; 100 percent for adenomas and sessile serrated lesions and 96.4 percent for hyperplastic polyps; and 97.5 percent for White patients and 100 percent for Asian and Black patients. So where performance varies across clinical and demographic groups is visible.

Three limits apply. The Asian and Black results are reported as 100 percent with intervals of 100 to 100, which gives no measure of their uncertainty. Bowel preparation quality, which tracks health literacy, support and comorbidity and so varies across populations, is not among the groups reported. And nothing is published on how the research network's patient screening performs across the populations it screens, where under identification would reduce access to trials directly.

The company's own clearance filing publishes an error rate with its method and denominator. The K213686 summary reports standalone testing on a predefined set of 79 high definition colonoscopy videos containing 94 polyps: SKOUT detected 92, a per polyp detection rate of 97.87 percent (95 percent confidence interval 94.96 to 100), with 22.55 false marks per 15 minute interval (18.95 to 26.15). At the frame level it marked 55.7 percent of frames showing a polyp and 2.3 percent of frames without one. These are test set results for white light colonoscopy in adults having screening or surveillance, not performance in use. The false mark rate is the figure that decides whether endoscopists keep a detection aid switched on.

No warranty, indemnity or remediation commitment is published, and there is no route for reporting a missed lesion.

Little integration is required and none is documented. SKOUT sits between the endoscope's video processor and the display, so adoption needs no record system project. Its distribution through Provation, a procedure documentation company, is a sales channel; no published material describes SKOUT results being written into a Provation report or any record system.

On the research side, the company embeds research in partner provider organizations, and how its screening reaches their records, and through what interface, is not described.

A device in the video path, with where it processes left implied. The K213686 summary describes SKOUT as hardware and software that overlays marks on the endoscopic video, adding 56 to 62 milliseconds of video delay depending on the connection used, which places the system in the procedure room. The summary does not state where inference runs, whether any video is kept after the procedure, or whether the unit connects to a network for updates or data transfer.

The research network raises a separate question. It spans North America, Europe, India and Australia, so trial and screening data can cross borders, and no residency commitment or regional arrangement is published.

No price is published. SKOUT is sold in the United States through an exclusive distribution partnership with Provation, a gastroenterology documentation company, so for many buyers the contracting counterparty is the distributor rather than Iterative Health. Research network services are contracted separately with trial sponsors and provider organizations on a different basis.

How each business sells is public; the unit of charge for either is not.

Two unrelated scopes, one validated. Iterative Health appears among clinical trials AI vendors because its research network uses machine learning to prescreen patients for trial eligibility; its cleared product is SKOUT, a colonoscopy device that reads live endoscopic video to find polyps. SKOUT covers one procedure in one specialty, white light colonoscopy in adults having colorectal cancer screening or surveillance, validated by a multicenter randomized trial in that population; performance by setting, such as hospital units against ambulatory centers, is not published.

The research network is broad: the company reports more than one hundred sites across North America, Europe, India and Australia in gastroenterology, hepatology, obesity and cardiology, with no published validation of its screening models in any of them. The device and the network serve different buyers and do not depend on each other.