Only 3 of 1,357 FDA-Cleared AI Medical Devices Have Tested Patient Outcomes, Study Finds

Only 3 of 1,357 FDA-cleared AI medical devices have ever been evaluated for patient-centered outcomes such as mortality, morbidity, or readmissions, according to a systematic analysis published in PLOS Digital Health. For builders deploying AI in clinical settings, this number is a stark reminder that regulatory clearance is not a proxy for clinical effectiveness.

The evidence thins quickly after FDA authorization

Researchers from the University of Toronto linked all FDA-cleared AI and machine-learning-enabled medical devices (through December 5, 2025) to registries on ClinicalTrials.gov and PubMed. They found that of 1,357 cleared devices, only 2.5% were linked to registered prospective trials, 0.9% (12 studies) posted results, and 0.9% had peer-reviewed publications. The sharpest drop was in patient-centered outcomes: a mere 0.2% - three studies - evaluated endpoints that actually matter to patients.

Among the studies that did exist, 62% used observational designs with small, homogeneous cohorts, limited subgroup analyses, and frequent exclusion of vulnerable populations. This pattern makes it difficult to generalize results across diverse patient groups.

Why the gap exists

The study identifies several structural barriers that discourage rigorous clinical validation. Misaligned financial incentives, reliance on predicate-based regulatory pathways (such as the 510(k) clearance process), and the logistical challenges of multicenter trials all contribute to what the authors call "evidence attrition." In a predicate-based pathway, a new device can be cleared by citing similarity to an already-marketed device, which may itself lack clinical outcome data. The result is a cascade of devices with thin evidence.

What this means for builders shipping AI in healthcare

If you are building an AI medical device or deploying one within a hospital system, regulatory clearance does not guarantee improved patient outcomes. The crowded market of over 1,350 cleared devices may give false confidence to purchasers and clinicians. For product teams, investing in prospective trials or real-world evidence studies early can become a competitive differentiator. For internal tool builders, the findings suggest that institutional adoption decisions should demand more than a clearance letter.

Study limitations and the bigger picture

The analysis relied on publicly available trial registrations and publications, which may undercount unpublished or non-public trials. It also excluded software updates not labeled as recalls and devices cleared after 2022 where follow-up time was limited. The landscape is evolving rapidly: as of mid-2026 the count of cleared AI devices likely exceeds 1,500. Still, the direction of the findings is clear - regulatory authorization and clinical validation are not the same thing, and builders should plan accordingly.