M-15: Clinical Trials of Medical Devices: Decisional Trees for Optimizing the Choice of Study Design
Université Claude Bernard - Institut De Pharmacie Industrielle De Lyon France
The goal was to identify the characteristics of diseases and medical devices (MDs) making clinical trials (CT) with a high level of evidence difficult; we aimed to propose decisional trees to describe methodological choices for developing better strategies.
To identify the main methodological issues in MDs clinical evaluation, we analyzed review and meta-analysis data from online databases. We focused on diseases where MDs and drugs CT are available, i.e Parkinson disease, arthritis, benign prostatic hypertrophy, hypertension and myocardial infarction.
Our results show that obstacles to successful randomized controlled trials (RCTs) using MDs include: a low number of patients, delayed endpoints, difficulties in the choice of an adequate comparator, and a lack of study blindness. The rapid evolution of the MDs as well as their costs might also be challenging issues. Based on these methodological limits, major decisional nodes appeared to be: i) whether the MD is implantable or not; ii) the nature of the outcome (reversible or not); iii) the scarcity of the disease; iv) existing comparators (drugs or other MDs); and v) the possibility for blinding. Finally, we also considered the necessity of a learning curve in case of a new surgical technique used for the implantation of the MD. We built three decisional trees to identify situations where it is feasible to perform blinded/open randomized clinical trials, and those for which adapted designs are acceptable.
(1) The first tree deals with the choice of the design regarding the nature of the outcome and the scarcity of the disease. It shows that the randomization is feasible in most cases, except when the MD is still at the feasibility stage or when it is the last chance to save the patient. We also proposed adaptive designs (e.g. tracker trials, n-of-1 trial) when RCTs are not possible due to a few number of patients and/or a delayed outcome.
(2)The second tree describes the choice of control treatments and the possibilities for blinding. We took into account the availability of a standard treatment as a control, with the mandatory use a double placebo if the control is a drug, or the use of a device with a similar appearance if the control is a MD. When blinding seemed difficult to reach, we considered the intervention of an independent assessor or the use of a sham procedure.
(3)Our third tree explains the necessity to learn new surgery techniques when the MD is implantable. The main solution proposed was cluster RCT depending on the site training.
Currently, medical devices are constantly evolving with innovation. Various MDs often gather very heterogeneous and numerous characteristics, which poses challenges in evaluating them. This clearly calls for a need to prove the clinical benefit of such devices after the feasibility stage. They gathered very heterogeneous and numerous characteristics. Thus, their evaluation faces specific challenges. In our work, we showed that we can always find an adaptive design to reach a high level of evidence. The proposed algorithms could constitute a first step for future guidelines regarding the methodology of medical devices clinical trials, and may help reinforce the power and reliability of clinical trials to maximize the clinical relevance of the data while minimizing biases. The next step would be to work in collaboration with institutions and industries specialized in medical devices to test and implement these algorithms on a more practical level.