World Health Organization Guidelines on Artificial Intelligence

Artificial intelligence is gaining momentum in many industries, where it helps businesses achieve various milestones. In the pharma industry, is also being used to achieve various business related measures, ensuring safety and supplying high quality products.

For the pharma regulators, it is also being accepted as a tool to provide better output and considerably reduce different errors. Additionally, regulatory bodies are also devising regulations to better utilize AI  in the pharmaceutical industry

World Health Organization is an international reputable body that provides best practices for the pharma industry worldwide, and is independent of any local regulatory bodies worldwide. For the AI, it also provide guidelines for using it in medical devices and clinical practice. These guidelines aim to highlight the risks that this new technology poses, and their relevant best practices

Before explaining this guideline, let's discuss how professionals implement the AI and what the position of AI is in the pharmaceutical industry.

Examples of AI in the Pharma industry

Some area that AI is being implemented in pharma industry include the following

Manufacturing

AI can be used in manufacturing to improve the processes by utilizing it as a historical trend to design optimum process values.

Application of AI also helps to reduce waste during manufacturing, by detecting and predicting errors in the process, and defining optimum machine values.

Drug Discovery

AI can be used in drug Discovery for analyzing human specific problems, and using the information obtained to predict optimum chemical composition and therapies. It also helps to improve existing therapies by analyzing the historical pattern of effectiveness and detect anomalies and apply improvements

Research

AI can be used to automate data collection by researchers, and translating them into meaningful information. The data in pharmaceutical research is enormous, and can become difficult and inefficient to translate it into the information through manual process.

The AI can solve this problem, by automating the process and integrating historical  data with current data to analyze and predict accurate information

What regulatory body think about the potential of AI for pharma industry?

In this section, we discussed what regulatory body thinks about artificial intelligence

AI has the potential

Many regulatory bodies considered AI having a great potential to  transform healthcare. They help to provide an insight into current manufacturing processes, and provides a wealth of information about a specific field area, or application.

Regulatory bodies also understand that AI, and technology like these can help achieve the objectives of accurate, safe and patient centric products

Helps to streamline regulatory process

Regulatory bodies are of the view that AI can streamline compliance process for an organization by providing the necessary data, proving the implementation of the required processes and following recommended practices. AI can also help pharma manufacturers in achieving compliance by fulfilling requirements in less time that could take a lot more time without using AI

For the manufacturing organization, AI can streamline the entire elements of manufacturing by utilizing historical data and producing the ongoing process in an efficient manner. For a clinical research organization, AI can automate the data analysis, and can use this historical data to predict effective drug improvements in an existing drug and reducing the time consumed by traditional research practices.

All the four mention circumstances helps an organization to complete the regulatory body regulations in an efficient manner

Regulatory bodies proposing and updating their regulatory guidelines

Regulatory bodies are accepting the use of AI in pharma industry and proposing new guidelines. They continuously update themselves to provide guidance for better application of this technology. Regulatory bodies analyses the application of AI deeply to find out any risks that can cause harm to pharma industry.

Every regulatory authority worldwide has published dedicated guidelines on using AI in pharma industry,  but the focus of this article is on World Health Organization.

World health organization guidelines on artificial intelligence

The World Health Organization’s guideline on AI is called a “regulatory concentration on artificial intelligence for health”. Published in October 2023, it provides guidance for its stakeholders, in using AI in healthcare. It includes, Medical device regulation and healthcare practitioners.

Unlike compliance guidance, it is not an enforcement, but instead provides reference  information for implementing AI in their respective businesses.

The WHO describes six key area of application that are critical in the health Care industry and includes the following

· Documentation and transparency

· Risk management and AI systems development lifecycle approach

· Intended use and analytical and clinical validation

· Data Quality

· Privacy and Data Protection

· Engagement and Collaboration

Let’s discuss each of the above points briefly.

Documentation and transparency

Documentation and transparency are key requirements for assessing the impact of AI. It is critical for developing trust on AI system by different stakeholders such as developers, manufacturers and end user.

It can be implemented by taking a product life cycle approach by pre-specifying documentation processes, methods, resources, and decisions in all the elements of life cycle such as initial conception, development, training, deployment and validation

The benefit of documentation and transparency is that it establishes trust in tracking, recording and retaining records It should be seen as an opportunity to improve the applicability of science based tool.

Some consideration for documentation and transparency includes the following

· Documentation must be designed and implemented across the entire product life cycle

· The purpose of implementing AI should be documented such as what is the reason for implementing AI, procedure and possible outcomes

· Documentation must be designed for both Deployment and post Deployment stages

· Documentation and Record keeping must be implemented by risk proportional approach

Risk management and AI systems development life cycle approach

For implementing AI throughout the product life cycle, a holistic risk management approach should be used for addressing and solving common problems with the AI, such as Cyber security and system vulnerabilities. These principles are particular important to medical devices, and the pre  and post market deployment.

Some consideration of this principle is as follows

· The developers should devise practices to enable responsible AI development, and all activities should be performed under a Quality Management System

· The development stage should include, appropriate practices for AI to facilitate continuous AI learning and product improvement.

· For mitigating risks related to deployment of AI, such as Software, its associated components and cybersecurity, risks should be considered throughout the development stage and holistic risk management should be deployed for mitigating. Additionally, risk management for both pre and post market should be implemented. 

Intended use and analytical and clinical validation

For implementing AI in pharma based application, the developer must also keep in view the safety consideration, and answer safety specific questions such as whether the product is safe, to make sure it will not harm its user, and has enough justification to be accepted into the market. Evaluation of these answers and assessment can be performed as per the following

· Clinical evaluation from development to analytical and clinical validation and to post- market surveillance should be implemented for AI.

· Using use cases, the AI can be assessed for individual requirements, and informing individual users of its application. It can also be used as a proof to regulators, that analytical and clinical validation have been performed appropriately.

· The developers of AI should be able to present datasets used in system training, validation, and testing.

· For lower risk devices, novel tools to collect real world performance data must be used. For high risk devices, analytical validation must be used for clinical evaluation

· Manufacturers or developers should monitor post –market performance, and report any incidents & findings for continuous monitoring and improvement to regulatory bodies.

· AI devices, that are subject to change such as user interface, code or training data, should go through risk based evaluation

Data Quality

Data is the basis of AI, and successful implementation of AI depends on accurate, relevant , good and quality data. During implementation and using, there can be variety of qualities of data, which must be monitored by developers, and decide whether it can be used or not.  A bad quality data could affect the performance of AI and will not serve its intended purpose.  

Some attributes that could affect the data quality,  includes the following

· Dataset management

· Data inconsistency

· Irrelevant dataset selection and curation

· Data usability

· Data integrity

· Model training

· Data labeling

· Documentation and transparency

Privacy and Data Protection

The WHO classifies the health related data as a sensitive data that requires high degree of safety and security. The developers are responsible for devising this feature and the user is responsible for verifying and implementing it. For AI, the big data should be safe by deploying appropriate methods. Additionally, appropriate laws should also be developed for protecting the health data.

Engagement and Collaboration

Engagement and collaboration between stakeholders such as developers, drug manufacturers, patients, advocates and policy makers can be beneficial for the success of AI system. It can be used to communicate the purpose, benefits and applications of AI in healthcare system. It can also be used as a mechanism to receive feedback on existing devices for process and technological improvement.

About the Author:

Muhammad Asim Niazi

Muhammad Asim Niazi has a vast experience of about 11 years in a Pharmaceutical company. During his tenure he worked in their different departments and had been part of many initiatives within the company. He now uses his experience and skill to write interested content for audiences at PharmaSources.com.