Clinical Evidence Supporting FDA Clearance of First-of-a-Kind Therapeutic Devices via the De Novo Pathway Between 2011 and 2019

Abstract

Importance: In recent years, the US Food and Drug Administration (FDA) and manufacturers have increasingly sought to expedite patient access to first-of-a-kind devices via the De Novo premarket review pathway. Understanding the strength of clinical evidence supporting FDA clearance through this pathway can help guide clinical adoption of novel devices and ongoing regulatory development of the postmarket surveillance infrastructure. Objective: Our primary objective was to characterize the strength of clinical evidence supporting FDA clearance of therapeutic De Novo devices. Key secondary objectives were 1) characterization of FDA post-marketing requirements for De Novo devices and 2) study of the use of these devices as the basis for devices subsequently cleared via the 510(k) process. Design: Retrospective cross-sectional analysis Setting: Publicly available online FDA databases, including the De Novo database, the 510(k) clearance database, the 522 Post Market Surveillance database, and the Recalls of Medical Devices database Participants: All moderate-risk therapeutic devices cleared via the De Novo pathway between January 1, 2011, and December 31, 2019. Main Outcome Measures: (1) proportion of De Novo devices cleared based on evidence from a pivotal clinical study, (2) proportion of pivotal study primary effectiveness endpoints that were met, (3) proportion of De Novo devices subject to FDA-required postmarket studies, and (4) proportion of De Novo devices serving as the basis for at least one subsequently cleared 510(k) device (i.e., new models or competitor products). Results: There were 63 (of 65; 96.9%) moderate-risk therapeutic devices cleared by FDA via the De Novo pathway between 2011 and 2019 for which decision summary documentation was publicly available. Of the 63 devices, 51 (81.0%) were supported by pivotal clinical studies (n=54 studies); the remainder (n=12; 19.0%) were not supported by a pivotal clinical study. The majority of pivotal studies were randomized (57.4%), multi-armed (61.1%), and used an active (25.9%) or sham (35.2%) comparator arm; 17 (31.5%) failed to meet at least one primary effectiveness endpoint. Among the 63 devices cleared via the De Novo pathway, one (1.6%) was subject to an FDA-required posttmarket study and 32 (47.8%) served as a predicate device for new models or competitor devices subsequently cleared through the 510(k) process. Conclusions: Between 2011 and 2019, the FDA cleared the majority of first-of-a-kind moderate-risk therapeutic devices via the De Novo pathway based on premarket evidence from pivotal clinical studies. However, 43% of devices were cleared without clinical evidence from pivotal studies or based on pivotal studies that failed to meet at least one primary effectiveness endpoint. The FDA rarely required postmarket studies of these devices, which often served as the basis for new models and competitor products subsequently cleared via the 510(k) process.

Supervised machine learning in psychiatry

“…Thus, before a medical predictive model can be safely applied in clinical practice, it is crucial to test it not only in a single but in multiple datasets that are independent both to each other and to the data used during the development of the algorithm. As a matter of facts, the majority of medical device filings to regulatory bodies such as the US Food and Drugs Administration are based on multi-center clinical studies (Johnston, Dhruva et al 2020), and multi-centric testing seems to have progressively become more and more used in the recent literature of ML for medical applications (Abraham, Milham et al 2017, Meyer, Mueller et al 2017, Gabr, Coronado et al 2019. However, previous studies using ML to predict clinical course and treatment response prediction in OCD patients are mostly based on data recruited in a single center (Salomoni, Grassi et al 2009, Hoexter, Miguel et al 2013, Yun, Jang et al 2015, Mas, Gasso et al 2016, Lenhard, Sauer et al 2018, Reggente, Moody et al 2018, Metin, Balli Altuglu et al 2020.…”

Affecting OCD

“…Thus, before a medical predictive model can be safely applied in clinical practice, it is crucial to test it not only in a single but in multiple datasets that are independent both to each other and to the data used during the development of the algorithm. As a matter of facts, the majority of medical device filings to regulatory bodies such as the US Food and Drugs Administration are based on multi-center clinical studies (Johnston et al, 2020), and multi-centric testing seems to have progressively become more and more used in the recent literature of ML for medical applications (Abraham et al, 2017;Gabr et al, 2019;Meyer et al, 2017).…”