How is the clinical evaluation of software as a medical device conducted?

Clinical Evidence Begins with the Intended Purpose

Article 61 and Annex XIV Part A of Regulation (EU) 2017/745 require a clinical evaluation for medical devices. It is not a one-time literature chapter at the end of the project, but a systematic and ongoing process. The starting point includes the medical intended purpose, target population, intended users, environment of use, contraindications, and any specific claims regarding safety, performance, or clinical benefit.

A Clinical Evaluation Plan specifies in advance which fundamental safety and performance requirements need clinical data, which endpoints and acceptance criteria apply, how the state of the art is determined, and where evidence gaps exist. Results are assessed in a traceable manner in the Clinical Evaluation Report; both favorable and unfavorable data must be included.

Three Evidence Components for Medical Software

MDCG 2020-1 structures the clinical evidence for Medical Device Software along three interconnected components:

1. Valid Clinical Association: It must be scientifically justified that the parameter processed by the software or the output generated is related to the intended clinical condition or benefit.
2. Technical Performance: The software must reliably and reproducibly convert inputs into the specified output. This includes aspects such as accuracy, robustness, data quality, and edge cases.
3. Clinical Performance: In the intended context of use, the output must achieve the claimed clinical purpose. The evidence must be appropriate for the target population, user group, and significance of the information.

The evidence may come from scientific literature, existing clinical data, performance studies, comparative data, and proprietary clinical trials, depending on the product. A clinical trial is not automatically required for every software; however, if existing data is insufficient, new clinical data must be generated. The scope and quality must correspond to the risk class, novelty of the algorithm, and significance of the clinical claim.

Evaluation Continues After Market Access

Post-Market Surveillance and, if applicable, Post-Market Clinical Follow-up provide real-world data on misuse, rare events, changed populations, and the state of the art. This information is fed back into the clinical evaluation report, risk file, and product measures. Annex XIV requires that the clinical evaluation is updated throughout the entire product lifecycle with clinical data from the market phase.

The crucial factor is not the quantity of publications but their suitability for the claimed function. Data on a similar algorithm, a different population, or another clinical workflow is only transferable if this transferability is methodologically justified. Therefore, the specific evidence strategy should be established early with regulatory, clinical, and statistical expertise.