AI’s growing influence on the medical device company’s value chain (MedTechDive)
With the rise of interest around AI since the release of large language models (LLMs), there is pressure on MedTech executives to get ahead of industry peers regarding AI, not only within devices but in the conception, design, manufacturing, approval, distribution and post-market management of devices.
The medical device industry is unique across the value chain and life stages. It spans research, design, quality and product engineering, clinical trials, validation, pre-market approval, production, post-market support and product obsolescence. Adopting the digital thread throughout the value chain is essential for scaling organizations in response to industry growth. It enables companies to proactively address or leverage disruptive forces more effectively.
Medical Device Companies and the Digital Thread
Medical device companies must see digital thread investment powered by artificial intelligence (AI) models as their number one strategic imperative or risk falling behind. The digital thread refers to the seamless flow of data and information across different stages of a product’s life cycle. This includes design, manufacturing, use and service, creating an integrated and connected data ecosystem.
Benefits of the digital thread include:
- Faster development cycle time
- New product-driven growth
- Enablement of business scaling
- Streamlined regulatory compliance
- Reduced business and clinical risk
- Simplified product registrations
- Improved asset utilization and reduced downtime
- Reduced scrap and waste in prototyping and manufacturing
- Better energy / reduced environmental footprint
- Reduced manufacturing and field service costs
- Closed-loop quality management
- Improved insights from connected products
Those who can get ahead of the curve and carefully leverage AL/ML in this context will be most competitive in the coming years. It’s essential to have an effective digital thread strategy that considers practical and realistic uses of AI to supercharge processes throughout the value chain.
Let’s look at examples of how your organization can use AI across the product lifecycle including:
- Research and development
- Quality
- Regulatory
- Manufacturing engineering
- Manufacturing
- Supply chain
- Post market & field
R&D oriented
- Generative design and incomplete-design-finishing: Generate new designs for new products such as medical devices, allowing for faster and more efficient design exploration
- Simulation: CAD simulation tools are used for simulating part design performance such as finite element analysis, computational fluid dynamics and electromagnetic simulation
- Labeling/instructions for use (IFU): Develop various IFU content for readability and accuracy with LLMs. An LLM can also produce clear, comprehensive instructions that are easy to understand for end-users. This can streamline the process of document creation and ensure consistency across documents
- Translation services: Translate content from one language to another. While we would not yet advocate for full human replacement, there is potential to reduce costs in this space
- Medical device software development: Use AI in the development of Software as a Medical Device (SAMD) with AI. Many large tech companies such as Microsoft, IBM and Google are now providing integrated AI-based frameworks and toolkits for custom purposes for developers of healthcare applications
- Product lifecycle intelligence (data analytics of product data): Leverage analytics to describe and diagnose the state of the process as well as predict the future and prescribe actions. An example may be predicting the speed of engineering changes and the likelihood of rejection at any given stage
Quality oriented
- Synthetic test data: Manufacture synthetic data for simulating the performance of medical devices or components of devices, for testing and validation
- Personalization: Manufacture customized medical devices for individual patients based on their medical imaging and other data with Generative AI. This has been used by companies like UNYQ
- Analytics: Describe the state of a process, diagnosing issues as well as predicting and prescribing outcomes
- Computer vision (inspections): Use cameras and algorithms to capture, process and analyze images of objects to gather information, make decisions, or perform inspections. While setting up and optimizing this process is not without challenges, it can greatly speed up the inspection process vs. using more traditional, slower and expensive tools
Regulatory oriented
- Submission package review: Review and draft individual artifacts or subsections of submission packages and seek potential gaps and make recommendations to increase the likelihood of success for a particular regulation
- Product Registration Management: Understand what can or cannot be shipped to a certain country and within what date-span to remain compliant. This includes translating data from certificates into structured data that can be managed by Regulatory Information Management Systems (RIMS). This information may otherwise not be well tracked as the data cleansing, enrichment and migration obstacle is too high to get reliable information into RIMS
- Analytics: Apply analytics to the regulatory process, including predicting the likelihood a submission would be accepted and how long it may take
Manufacturing engineering oriented
- Co-piloting of PLCs: Streamline Product Lifecycle Approach (PLC) development. This can result in better code, which can improve production efficiency. Additionally, given the current skilled labor shortage, making PLC developers more efficient through these types of AI-empowered tools may be increasingly crucial
- Digital twin, simulation and emulation: Create virtual modeling of shop floor processes with simulation tools, enabling the identification and rectification of bottlenecks. You can also test different scenarios without disrupting actual operations. Emulation, on the other hand, involves creating a digital twin of the production environment, allowing for real-time analysis and optimization of workflows. We are seeing increasing demand for digital twinning of plants, especially as manufacturers with high-demand products seek to scale up operations
- Cost and risk analysis: Assess the cost and risks associated with different process designs by analyzing and predicting raw material and component costs, labor and potential reliability. Simulating these variables in a production design could improve overall production throughput
Manufacturing oriented
- Material flow optimization: Optimize flow on the shop floor using a variety of tools including simulation and emulation which was discussed previously. Automated Mobile Robots (AMRs) can significantly boost flow and improve yield. AMRs are versatile, intelligent machines capable of navigating and performing tasks autonomously within the shop floor environment. They can transport materials and products efficiently, reducing the need for manual labor and minimizing human errors
- Shop floor analytics: Combine these various automated systems, AMRs and IoT devices to provide your organization with real-time data and analytics, further enhancing decision-making processes
- Model predictive control: Leverage trained AI models to predict the future behavior of manufacturing equipment and optimize control actions. It repeatedly solves an optimization problem at each control step to find the optimal control inputs and make time-based predictions. The model will try to close the gap continuously between desired and predicted outputs
- Worker support: Support many personas throughout the supply chain and in the plant. There is a premium on understanding and training. For newer workers, LLMs are proving useful in simplifying complex work instructions and providing an interactive knowledge engine. For more advanced use cases or experienced workers, information can be provided in real time and contextualized, enabling skilled workers to make faster decision
Supply chain oriented
- Supply chain stability: Have access to good quality internal data, integrated with external data sources. This is key to the start of having a resilient supply chain. Internal data sources can include items such as inventory on-hand and demand forecasts. External data feeds may include predictions for supplier and shipment disruption, regulatory changes and supplier performance monitoring.
Post market and field oriented
- Predictive maintenance: Predict failures for manufacturers and set up new service business models that can result in improved asset utilization by hospitals
- Real world data contextualization: Contextualize Real World Data (RWD) along with other inputs from Research and Development (R&D), such as the Design and Manufacturing Failure Effects Analysis (DMFEA), to understand failures and provide early warning signals if issues arise. The FDA requires medical device companies to collect Real World Data (RWD) and Real World Evidence (RWE), which is a beneficial practice overall
- Sentiment analysis: Understand market sentiment to products by taking data feeds from external sources such as social media. This is especially useful for consumer medical devices. There is a general expectation from regulatory authorities that if the information is publicly available, MedTech companies need to pay attention and consider the impact of evolved understanding. Traditionally this has been quite challenging with conventional tools, certainly without considerable human intervention
We have seen many remarkable examples of excellent business results such as 18% improvement in production yield, 35% reduction in inventory and between 10-20% reductions in machine down time. However, it’s important to note that not all forms of AI can be hands-off, especially in a validated environment.
When implementing AI throughout the value chain, a risk-based approach is warranted to determine the level of validation needed and consider how much a human must be in the decision-making process. In general, the FDA’s stance remains progressive on the use of AI in the value chain, especially in more recent years where quality is prioritized over simple compliance.
FDA’s Good Machine Learning Principles are a good example of how the FDA is supportive. However, heavy use of validation across a variety of scenarios remains critical, whether a human is involved or not. Organizations need to ensure that decisions are made about the use of any technology by the right persons, with the right skill level and this is all verifiable.
Relevance to the digital thread
The digital thread seeks to automate and integrate the flow of data across the value chain and through automation super-charge this flow. All use cases cited above are dotted throughout the digital thread, will leverage core enterprise systems such as:
- Systems engineering/application lifecycle management
- Product lifecycle management
- Enterprise resource planning
- Manufacturing execution systems
- Supply chain management systems
- Customer resource management systems
These can often feed upon each other. The more mature these base systems are, especially if the data models are harmonized or at least well translated, the more likely AI will be useful to the manufacturer. It’s essential to have strong programs focused on core enterprise systems to be in the best position to take advantage of these exciting capabilities.