DDReg Pharma

DDReg Pharma
Artificial Intelligence

Regulatory Guidelines for Software and Artificial Intelligence as a Medical Device

Technological disruptors are gradually revolutionizing various aspects of life sciences and healthcare. Software is slowly becoming an important part of products and are being integrated into digital platforms for medical and non-medical purposes. There are 3 types of software-related medical devices: 1) Software as a medical device (SaMD), 2) software in a medical device, and 3) software used for manufacturing and/or maintaining a medical device [1].

Software as a Medical Device (SaMD)

There are many unique features that are associated with SaMD which makes the device relatively non-conventional compared to a typical medical device. Thus, regulators across the globe understand the need to converge on common principles and framework that allow various stakeholders, including regulatory agencies, to continue safeguarding public health while promoting innovation. For example, as recent as 26th April 2023, the TGA Australia released regulations for software based medical devices [2].
As this technology further develops, artificial intelligence is also becoming a key component of medical devices and, in parallel, the need for more harmonized guidelines is becoming more crucial.

Artificial Intelligence and Machine Learning in Healthcare

The global artificial intelligence (AI) in healthcare market size was valued at USD 15.4 billion in 2022 and is expected to grow at a CAGR of 37.5% from 2023 to 2030. AI is rapidly integrating into the medical field, with medical devices incorporating AI capabilities such as enhanced imaging systems, wearable technology, smart robots, simulation platforms, and AI-based data analysis [3].
Machine learning (ML), a subset of artificial intelligence, helps doctors make more accurate patient diagnoses and treat them more effectively by analyzing vast volumes of clinical data to spot patterns and predict outcomes more precisely than ever before. It is used in tasks such as disease diagnosis, drug discovery, and precision medicine. It allows healthcare providers to uncover correlations in healthcare data between diseases and detect subtle changes in vital signs that may indicate potential health issues. Precision medicine is the most common use of machine learning in healthcare, predicting successful treatment procedures based on the patient’s makeup and the treatment framework.
AI algorithms offer a level of precision not achievable with traditional approaches to analytics and clinical decision-making, providing unprecedented insights into care processes, diagnostics, and patient outcomes. Despite its potential, it is important to approach the integration of AI in healthcare with caution and ensure that it is properly regulated to ensure patient safety and ethical considerations are met.

UK MHRA: Guidance on AI and Medical Device Software

In April 2023, the UK Medicines and Healthcare Products Regulatory Agency (MHRA) released a guidance document on Software and Artificial Intelligence as a Medical Device [4]. The level of regulation depends on the level of risk associated with the software, with higher-risk devices being subject to more stringent regulations. The guidance also provides specific instructions on using AI and MI in medical devices, emphasizing the importance of ensuring the algorithms are validated, transparent, and accountable. The top priority is to ensure that medical device software is safe, effective, and reliable for patients.
The guidance from the UK MHRA provides additional information on the risk classification of medical device software, with the highest-risk devices being classified as Class III and the lowest-risk devices as Class I. Factors such as the intended use of the software, potential harm to patients if the software malfunctions, and the ability to mitigate those risks all determine the level of risk classification.
Some of the guiding principles cover :
  • Ensuring multi-disciplinary expertise is utilised throughout the whole product life cycle.
  • Implementing security practices and good software engineering.
  • Ensuring that the data sets and participants in clinical studies reflect the target patient population.
  • Ensuring that test and training data sets are independent.
  • Using specific reference datasets and the most advanced techniques.
  • Model design should be adjusted to reflect the device’s intended use and the matching data.
  • Concentrating on the output of the human-AI team
  • Demonstrating device functionality under therapeutically relevant circumstances
  • Providing users with clear, necessary information
  • Monitoring the use of models for risk and performance management

US FDA: Approach to Artificial Intelligence and Machine Learning Medical Devices

The US FDA’s premarket approval pathway for medical devices include the 510(k), De Novo, and Premarket Approval (PMA); however, these do not cover AI and ML technologies. In 2019, the FDA published a discussion paper which described the approach to premarket review for AI and ML technologies in medical devices [5]. In 2021, in response to this discussion paper, the FDA subsequently released an AI/ML SaMD Action Plan which outlines the actions that the FDA is to take and its intention on developing an update to the discussion paper which includes issuing a draft guidance on pre-determine change control plan [6].
Collectively, these guidance documents provides recommendations for the development and validation of medical devices that incorporate AI algorithms. They emphasize the need for transparency and explainability of AI algorithms and robust performance evaluations to ensure their safety and effectiveness. It also covers topics such as data management, algorithm training and optimization, and cybersecurity considerations. Additionally, the guidance highlights the importance of continuous monitoring and updating of AI algorithms to maintain their performance over time.
The FDA’s Action Plan outlines recommendations for regulating software designed to diagnose, treat, mitigate, or prevent diseases without being a part of a traditional hardware-based medical device. The guidelines include principles for clinical evaluation, risk management, level of concern, continuous monitoring, and cybersecurity. SaMD manufacturers must provide evidence to the FDA to demonstrate the safety and effectiveness of their products. The document emphasizes the importance of a risk-based approach to ensure that SaMD is safe and effective for its intended use.
In April 2023, following the Action Plan, the FDA issued a draft guidance on ‘Marketing Submission Recommendations for Predetermined Change Control Plan for Artificial Intelligence/Machine Learning (AI/ML)-Enabled Device Software Functions’ that is intended to provide a “forward-thinking” approach to promote the development of safe and effective medical devices that use ML models trained by ML algorithms [7]. The guidance document demonstrates the FDA commitment towards developing approaches for better and more robust software-device regulation. In October 2022, the FDA approved 178 new machine learning (ML) and artificial intelligence (AI) applications. Currently, the number of AI/ML-enabled devices have surpassed 520 in the United States [8].
AI has now outperformed humans in various medical areas, particularly in medical imaging. Consequently, medical device manufacturers are now integrating AI and ML applications to enhance diagnostic accuracy and improve patient care.
AI is becoming more prevalent in the medical and life sciences field with devices such as enhanced imaging systems, smart robots, wearable technology, AI-based data analysis, and simulation platforms emerging as ‘show-stoppers’.
The integration of AI in healthcare has the potential to revolutionize how medical data is handled, diagnoses are made, and treatments are created, but it is important to regulate AI to ensure patient safety and ethical considerations. The UK’s MHRA and the US FDA provide guidelines on the development, validation, and regulation of medical devices that incorporate AI algorithms, emphasizing transparency, explainability, and robust performance evaluations.
References and Further Reading