# Data-Ethical Consultation about Sanquin

1. The organization: Sanquin

The company for which I will be doing a short ethical consultation is Sanquin, the Dutch blood bank. I found that this company articulated a new strategy in 2021 that stated that they would like to prevent health problems before they occur. To achieve this, they wanted to use longitudinal health data that is available to them, data science, and artificial intelligence technology. They also wanted to establish active collaborations with other stakeholders in healthcare, science, and the business sector. Their website can be found via this link: https://www.sanquin.org/.

2. The AI technologies Employed

I did not find any AI technologies that this company currently uses, but I would like to explore the possibilities and think about ethical concerns that could be relevant in the future. First of all, machine learning algorithms can be trained on longitudinal health data to identify patterns between health variables. These algorithms can predict the likelihood of developing health conditions or diseases. By combining these algorithms with other prediction techniques, Sanquin can forecast the probability of specific health problems. This can help identify individuals who may be at a higher risk of developing certain diseases or conditions and require early interventions or preventive measures.

With pattern recognition, AI algorithms can detect patterns in the data that may indicate the onset or progression of certain health conditions. By continuously monitoring and analyzing data, Sanquin can identify abnormal patterns and alert healthcare professionals for intervention. Clinical decision support systems (intended to improve healthcare by enhancing medical decisions with specific clinical knowledge, patient information, and other health information) can help Sanquin identify preventive measures, suggest appropriate screenings to healthcare providers, or recommend lifestyle modifications to individuals based on individual health profiles.

Another AI technique that could aid Sanquin is data visualization. Data visualizing tools can help Sanquin, their donors, employees and stakeholders to understand health data better: interactive dashboards and visual representations of trends and patterns can facilitate decision-making, but can also aid in transparency and trust in these systems.

3. Ethical concerns

At this moment, Sanquin already has and processes a lot of personal donor data. They state on their website that they require valid reasons in order to be allowed to process personal data: consent, performance of a contract, a legal obligation, and the legitimate interest of Sanquin or a third party. For special personal data that concerns health and genetic data they use explicit consent, the necessity for purposes of preventive medicine, medical diagnoses or the provision of health care or treatments, the necessity for reasons of general interest in the field of public health or ensuring high standards of quality and safety of health care and medicinal products, and the necessity with regard to scientific research.

If Sanquin processes personal data for the purpose of legitimate interests or those of third parties, the processing will be preceded by carefully weighing up these interests against the donor's right to privacy (as stated on their website). They take precautionary measures to protect privacy and, where necessary, to prevent the donor's interests from being harmed.

The application of AI techniques in analyzing longitudinal health data for preventive purposes raises several related ethical concerns.

privacy and Data Security

Health data contains highly sensitive and personal information that can provide insights into individuals' medical conditions, treatments, genetic predispositions, and lifestyle choices. This data is very valuable, but it must be handled with care to protect privacy. Safeguarding privacy ensures that individuals maintain control over who can access their health data, empowering them to make informed decisions about how their information is used. Privacy and data security also play an important role in the trust that patients have in their healthcare providers. Patients need to feel confident that their health information will remain confidential and secure. When patients trust that their data will be protected, they are more likely to openly share accurate and complete information with their healthcare providers, which will make it easier for Sanquin to make accurate AI models, develop treatment plans, and provide better care, leading to improved healthcare.

Maintaining patient privacy and ensuring data security is also crucial for preventing unauthorized access, breaches, or misuse of health data. Unauthorized access to health data can result in harm, such as identity theft, fraud, or the potential for discriminatory practices. Techniques such as de-identification, encryption, and strict access controls help ensure that only authorized individuals can access the data, reducing the likelihood of breaches and maintaining individual privacy.

If Sanquin wants to combine different datasets for a more detailed view of individual or population health, this aggregation must be done cautiously. Techniques such as data anonymization, aggregation, and statistical safeguards can be employed to protect individual identities, while maintaining the important details for the analyses.

Bias and Discrimination

AI models trained on health data have the potential to inherit biases present in the data, which can result in biased predictions and discriminatory outcomes. These biases can affect certain demographics or marginalized groups, leading to healthcare disparities. It is crucial to address these biases and ensure fairness in both the data and the algorithms. Additionally, when multiple datasets are combined, the risk of revealing sensitive information that could lead to discrimination increases. For example, the combination of genetic data with other demographic information could be used to discriminate against individuals based on their predisposition to certain diseases. Such practices can increase stigmatization and potentially deny some individuals good healthcare opportunities.

It is also important to recognize that historical biases and underrepresentation in healthcare data can cause unfairness. When certain groups have been frequently underrepresented in the collection and analysis of health data, AI models may be less accurate when applied to these groups. For example, if a specific group has been underrepresented in medical studies or has less access to healthcare, the AI models trained on such data may not capture the details and specific healthcare needs of that group. Biases can also emerge over time. Monitoring AI models and their predictions can help identify and address these biases. Lastly, fairness metrics should be established to assess the performance of AI models across various demographic groups. This will enable the detection and mitigation of disparities in healthcare outcomes.

informed consent and transparency

Informed consent and transparency relates to the first ethical consideration (privacy), these concepts play an important role in safeguarding individuals' privacy rights. Informed consent is important because it allows individuals to be fully informed about the purpose, risks, and potential outcomes of using their health data for predictive analysis. By obtaining informed consent, Sanquin (and the other organizations that will share their data) will respect individuals' autonomy and privacy rights. This ensures that individuals have control over their health information. Furthermore, obtaining informed consent and being transparent about practices align with legal and regulatory requirements, such as the General Data Protection Regulation in the European Union. Compliance with these regulations protects individuals' rights and holds organizations accountable for their actions.

Transparency in data collection, storage, and usage practices is needed for maintaining trust with patients. Individuals need to be aware of how their data will be used, who will have access to it, and what potential risks or benefits may arise from AI analysis. Moreover, transparency about AI practices (data sources, model development, and validation processes) helps identify and address biases that may be present in AI models trained on health data. Informed consent provides individuals then with the opportunity to be informed about potential biases and how they may impact the analysis and subsequent decisions. By respecting privacy, obtaining informed consent, and being transparent about practices, Sanquin can ensure that individuals are fully informed, minimize surprises or unintended consequences, address biases, and maintain trust in Sanquin.

4. Recommendations

Based on the three ethical considerations above, there are some concrete steps that Sanquin can take:

With respect to privacy and data security, Sanquin should implement robust security measures to protect patient data from unauthorized access, breaches, or misuse. The data should be anonymized and de-identified to minimize the risk of re-identification. Data practices should comply with relevant data protection regulations and standards, ensuring data encryption, access controls, and secure storage.

To address bias and discrimination, Sanquin should ensure diverse and representative datasets, accounting for demographic, socioeconomic, and other factors. They should also regularly evaluate and audit the AI algorithms for potential biases or unfair outcomes. Lastly, they can implement fairness measures, such as bias detection and correction techniques, to minimize biased predictions.

With respect to informed consent and transparency, Sanquin should obtain informed consent from individuals before using their data for predictive analysis. They should clearly communicate the purpose, potential risks, and benefits of using their data, ensuring transparency about data collection, storage, and usage practices. With this, they should also provide individuals with the option to opt in or opt out of data analysis while respecting their autonomy.

There are other ethical concerns that should be considered: interpretation and accountability (the people that interpret the models and outcomes should be skilled), data governance and ownership (determining who owns the data), psychological impact and autonomy (providing individuals with predictive health information can have psychological impact), and lastly, unintended consequences (e.g. false positives and false negatives have different consequences). Addressing these ethical concerns requires a multidisciplinary approach together with policymakers, healthcare professionals, data scientists, ethicists, and patients. Ensuring transparent, responsible, and accountable use of these AI technologies can help maximize the benefits while minimizing potential ethical challenges.