Heath Data Ethics, Inclusivity and Safeguards

Introduction

This section will explore health data in depth. We will first define health data, then look at “different from the typical” types of health data through two case studies.

Before we get going, what is health data?

Health Data

We saw in previous sections that health data may encompass things as general as a person’s post-code or as granular as the sequence of their DNA. If you were to try to group data about “you”, it is difficult to imagine many aspects of these data that could not be tied to your health. 

Because of this, health data is inherently messy, big and incomplete. Health data does not represent a nicely organised .csv file with clearly labelled rows and columns. With the “four ingredients of modern AI” (discussed in Introduction to AI, Week 1) modern systems can however learn from these data in a way that was previously unimaginable.

Remember, the power of these modern AI tools comes from this “extraordinarty pattern recognition”. For image processing, this may be picking out subtle biomarkers from sense images (by pixel size). For other applications, though, the recognised patterns may be something tied to other aspects of your data. 

We will adopt the UK’s 2018 definition of health data as our working definition, however you are encouraged to critique this:

‘Data concerning health’ [is] personal data relating to the physical or mental health of an individual, including the provision of health care services, which reveals information about their health status – [UK Data Protection Act, 2018]

To get an idea of what sort of data may typically be used in healthcare, take a look at the Institute for Health Metrics and Evaluation data sources section. Here you can browse all the data sources used, organised by country.

TASK: What do you notice about the distribution of these data? Do data sources cover all countries equally?

The Less Obvious Faces of Health Data

In this section, we will look at one of the less obvious faces of health data. 

TASK: Explore different, less typical types of health data. A good starting point could be the idea of “digital phenotyping“. What safeguards or governance would you want in place for this type of data?

Case Study 1: Your Health Data in the Private Sector

When watching the following video, “Health data’s journey in the intelligent health ecosystem” by EY Global and think critically about the types of data being used, where these data come from and where they go. Also, make note of the producer of this video and think critically about interests that they may have other than the user’s personal health.

TASK: Identify a specific element of the video you might want to be clarified before handing over your data to a private sector company.

Case Study 2: Applying AI Systems to Mental Health

The first chatbot ELIZA was introduced as a therapist. The UK 2025 Parliament Post report by Hannah Gardiner and Natasha Mutebi, “AI and mental healthcare: opportunities and delivery considerations” demonstrates that applications of AI to the healthcare sector are also being explored for current mental health applications within the NHS.

Image from AI and mental healthcare: opportunities and delivery considerations by Hannah Gardiner and Natasha Mutebi

The application of AI to mental health highlights some of the different faces of health data we need to be prepared to critically evaluate:

• As identified by the UK Parliament in the above report, mental healthcare provision in the UK faces rising demand because of the increase in mental ill-health. 
• There have been multiple ways that AI and machine learning techniques have been highlighted as methods to support mental healthcare delivery.

The following video, “AI-powered mental health chatbots developed as a therapy support tool” by 60 Minutes, showcases the application of a mental health chatbot for therapy.

TASK: Think about what additional safeguards are required when we consider applying AI tools to mental health?

Ethical Issues and Health Data

How, then, are we to maximise the mental health support offered by AI-driven methods, while ensuring we do not dispense with ethical considerations?

In this section, we will take a deep look at the challenges, barriers and future directions of some of these more creative applications of Health Data.

TASK: Consider the following image from the 2025 nature article, “The comprehensive clinical benefits of digital phenotyping: from broad adoption to full impact” by Zhang et al. 

Choose one topic from “Challenges and Barriers” and one topic from “Future Directions”. Create a project proposal for how you could address these within your current area of study.

The next video “How is data kept secure?” from Research Data Scotland, addresses how your right to your data is balanced by researcher requirements of health data, specific to Scotland. 

TASK: If AI systems can identify patterns within seas of data that no human could, are we as humans capable of defining what “identifying details” are?

Knowing how AI systems work is crucial to making use of these tools, while ensuring that the outputs of these systems do not negatively impact our society or specific groups within our population. This balance will be addressed in video interviews with Dr Clara Aranda Jan in the next section.