بررسی عوامل مهم در توسعه دستگاه های پزشکی از طریق شبکه های بیزی

In this paper, we investigate the impact of product, company context and regulatory environment factors for their potential impact on medical device development (MDD). The presented work investigates the impact of these factors on the Food and Drug Administration’s (FDA) decision time for submissions that request clearance, or approval to launch a medical device in the market. Our overall goal is to identify critical factors using historical data and rigorous techniques so that an expert system can be built to guide product developers to improve the efficiency of the MDD process, and thereby reduce associated costs. We employ a Bayesian network (BN) approach, a well-known machine learning method, to examine what the critical factors in the MDD context are. This analysis is performed using the data from 2400 FDA approved orthopedic devices that represent products from 474 different companies. Presented inferences are to be used as the backbone of an expert system specific to MDD.

Although advances in technology has provided many indispensable medical products to improve human health and sustain it, the development cost of medical devices burdens the healthcare systems as the industry is more technology-centric than ever before. Accordingly, the identification of critical success factors for medical device development (MDD) has become increasingly important. These critical factors should be identified so that device development can be managed to minimize the adverse effects of these factors. Many factors are related to the likelihood of success of devices in the market; and based on a company’s ability to react to them, these factors are considered to be either internal or external (Medina, Okudan Kremer, & Wysk, 2012). Internal factors mostly focus on the organizational context within which design is executed, and among others these factors include organization’s composition in terms of the level of experience in design teams (Lucke, Mickelson, & Anderson, 2009) and effective communication of the development priorities (Brown, Dixon, Eatock, Meenan, & Young, 2008). Likewise, several publications (Brown et al., 2008, Millson and Wilemon, 1998 and Rochford and Rudelius, 1997) report that the execution of a complete development process that includes preliminary market analyses, financial analyses, and customer involvement is critical for the further commercial success of medical devices. On the other hand, external factors are mostly related to costs and profits, research and development (R&D), clinical research and insurance companies’ reimbursement (Advanced Medical Technology Association, 2003). Intellectual property protections and overseas market opportunities are also among these external factors. More importantly, the Food and Drug Administration (FDA), regulatory agency of medical devices marketed in the Unites States, has been suggested as the primary external factor influencing the development priorities (Advanced Medical Technology Association, 2003). Success factors in product development have been examined in various ways thus far. However, existing methods used for the identification of critical success factors have a number of shortcomings; among these are the subjectivity of survey-based studies and the complexity to comprehensively and rigorously address both internal and external factors (Medina et al., 2012). This paper discusses these shortcomings and proposes to overcome them with the application of a Bayesian network (BN) approach to examine the impact of product characteristics, company context and regulatory environment related factors on MDD. BN is a well-known method used for machine learning. Furthermore, it has been cited in the literature as a preferred method to address the limitations of other analysis methodologies (e.g., Kim and Park, 2008 and Venter and Van Waveren, 2007). The BN approach allows for a scientifically objective analysis with the ability to simultaneously consider quantitative and qualitative data (Chiang and Che, 2010 and Venter and Van Waveren, 2007). In the paper, we use the BN approach to investigate the critical factors of MDD. BN analysis is performed using data from 2400 FDA approved orthopedic devices. In the remaining sections of the paper, we first provide a summary of the reviewed literature to identify potential factors with implications on MDD; then, we introduce the methodology. Details about the data set and results follow before we provide conclusions.

This paper presents our inference framework along with initial observations in an effort to solidify the relationships of the MDD setting in an evidence-based fashion in order to result in a useful expert system. Causal relationships identified in the network structure will be used as the significant predictors within the expert system. Although we have used only 2400 actual data points, we are able to show the legitimacy of the chosen methods in yielding useful inferences. In the subsequent expert system development, not only will the data set be enhanced but also cases from other regulatory settings will be considered. A valid network is developed and analyzed in detail based on product, company and regulatory environment variables in relation with FDA decision time. Some of the relevant variables included the type of submission, year of decision and historical reference. Based on these results, further research should include the application of supervised learning along with further study of other aspects of MDD. Of importance are the potential impacts of this planned expert system with implications on innovation/technology development with a focus on MDD. The variables included, relating to product, company and regulatory setting, show the most pertinent factors with which product launch time can be reduced. Companies armed with this knowledge can preplan their knowledge and monetary capital in seeing through the projects to their completion. This paper also extends the literature on MDD and the identification of critical factors with the implementation of a BN approach with unsupervised learning. Some of the future research directions may include investigating the importance of historical reference and innovation further, along with the interrelations between different variables.