Helping healthcare use the power of AI⚕️

With a childhood filled with hospital visits, I knew I would be a part of improving healthcare one day.

From seeing the fragile human life, to how much the treatment will influence the future.

My dad's car accident marked my life, not physically, but mentally.

As a witness to my dad's struggle, I saw the realities of healthcare. Its triumphs and its failures.

My glimmer of hope for healthcare lies in innovation, specifically the transformative power of AI.

The hope has evolved to become my passion.

A passion that makes sure:
Patients get the best care out there
Clinicians can focus on their patients
Peace of mind for caregivers

At the bottom line, is dignity for every individual.

In order to do so, healthcare as is needs to change.

Technology is the only feasible way to truly change healthcare as is.

A future where AI in healthcare is not just a possibility.

It's a necessity.

Knowing what to change though, can be difficult.

Should you be following the old ways of:
Prioritizing aesthetics over functionality
Making small tweaks ignoring core issues
Inefficient exchange of information

AI can create true value and impact in healthcare.

Interested in working towards a clear vision of AI for your organization?

Contact me to plan a call (sigrid@eirhealth.ai)

Looking for a framework on how to integrate AI in healthcare organizations?
I’ve got you covered. 

Here are 10 frameworks and guidelines where to start, from implementation to evaluation:

ESMO guidance on the use of Large Language Models in Clinical Practice (ELCAP) 
https://lnkd.in/ecb4cnAC

QUEST Framework for human evaluation of LLM in healthcare
https://lnkd.in/eKxJ3n53

FUTURE-AI: Best practices for trustworthy AI in medicine
https://lnkd.in/ebvysJRq

Evaluation framework to guide implementation of AI systems into healthcare settings (TEHAI)
https://lnkd.in/eAusWgtU

Reporting guideline for the early stage clinical evaluation of decision support systems driven by artificial intelligence: DECIDE-AI 
https://lnkd.in/ekqFGQkx

Critical activities for successful implementation and adoption of AI in healthcare: towards a process framework for healthcare organizations
https://lnkd.in/eQHyk3-e

Implementation frameworks for end-to-end clinical AI: derivation of the SALIENT framework
https://lnkd.in/eDCd7C6Z

AI evidence pathway for operationalising trustworthy AI in health
https://lnkd.in/euRsdN2z

Standing on FURM Ground: A Framework for Evaluating Fair, Useful, and Reliable AI Models in Health Care Systems
https://lnkd.in/e2eRnkhj

AI for IMPACTS Framework for Evaluating the Long-Term Real-World Impacts of AI-Powered Clinician Tools: Systematic Review and Narrative Synthesis
https://lnkd.in/eFhYt_FP

What framework would you include?

Sigrid Berge van Rooijen

With a childhood filled with hospital visits, I knew I would be a part of improving healthcare one day.

From seeing the fragile human life, to how much the treatment will influence the future.

My dad's car accident marked my life, not physically, but mentally.

As a witness to my dad's struggle, I saw the realities of healthcare. Its triumphs and its failures.

My glimmer of hope for healthcare lies in innovation, specifically the transformative power of AI.

The hope has evolved to become my passion.

A passion that makes sure:
Patients get the best care out there
Clinicians can focus on their patients
Peace of mind for caregivers

At the bottom line, is dignity for every individual.

In order to do so, healthcare as is needs to change.

Technology is the only feasible way to truly change healthcare as is.

A future where AI in healthcare is not just a possibility.

It's a necessity.

Knowing what to change though, can be difficult.

Should you be following the old ways of:
Prioritizing aesthetics over functionality
Making small tweaks ignoring core issues
Inefficient exchange of information

AI can create true value and impact in healthcare.

Interested in working towards a clear vision of AI for your product and processes?

Contact me to plan a call (sigrid@eirhealth.ai)

Ethics when implementing AI in healthcare is 𝗻𝗼𝘁 𝗼𝗽𝘁𝗶𝗼𝗻𝗮𝗹

If you’re not prioritizing ethics in AI, you’re risking lives.

Most AI systems lack ethical oversight.

But you can do it differently, by avoiding these common mistakes:

1. Ignoring patient consent forms
2. Overlooking data privacy laws
3. Disregarding bias in algorithms
4. Failing to report errors
5. Neglecting transparency standards

Instead, focus on:

1. Implementing human agency and oversight
2. Ensuring technical robustness and safety
3. Train staff of data protection and ethical AI usage
4. Assess impact on patient outcomes
5. Clearly communicate to patients when AI is being used

So remember, ethics is not optional.

How are you ensuring ethical use of AI in your organization?

AI will never revolutionize healthcare
If we don’t look at the structural issues. 

Only when these issues have been resolved can we truly benefit from the AI solutions. 

𝐅𝐢𝐫𝐬𝐭𝐥𝐲, 𝐭𝐡𝐞𝐫𝐞 𝐢𝐬 𝐚 𝐡𝐮𝐠𝐞 𝐝𝐚𝐭𝐚 𝐩𝐫𝐨𝐛𝐥𝐞𝐦.
Data in healthcare is siloed, but also full of mistakes. 
With inconsistent formats. Making it difficult to get a clear patient picture. 

Even with data standards, there is still a lack of integration. Within existing systems and workflows, but also across healthcare providers. 

But we also should not forget bias in the data. For instance, what is the impact when people from a different background receive lower-quality care? Or when females are considered to be exaggerating their symptoms?

For AI to work well, we need to fix the data problems. Making sure we record the data with similar terminologies, similar measurements, and remove errors. 

Then we can work on connecting the data to a larger infrastructure. And accessible for AI systems. 

𝐒𝐞𝐜𝐨𝐧𝐝𝐥𝐲, 𝐡𝐞𝐚𝐥𝐭𝐡𝐜𝐚𝐫𝐞 𝐩𝐫𝐨𝐟𝐞𝐬𝐬𝐢𝐨𝐧𝐚𝐥𝐬 𝐚𝐧𝐝 𝐭𝐡𝐞 𝐮𝐬𝐞𝐫 𝐢𝐬 𝐨𝐟𝐭𝐞𝐧 𝐢𝐠𝐧𝐨𝐫𝐞𝐝.
Innovations should be straightforward to use. 

Today, they often don’t integrate with current systems, and if they do, additional steps need to be taken to access the tools. 

The more steps and clicks a healthcare professional needs to take, the lower the adoption. 

Workflow analyses before implementing AI solutions need to be conducted. With clinician feedback. Feedback on the tool design, but also on workflow integration. 
Clinicians are the best to identify inefficiencies in the tool and integration thereof, and to help improve user experience. 

If the added value and user experience are high, the adoption will follow. 

𝐓𝐡𝐢𝐫𝐝𝐥𝐲, 𝐧𝐨𝐛𝐨𝐝𝐲 𝐢𝐬 𝐰𝐢𝐥𝐥𝐢𝐧𝐠 𝐭𝐨 𝐩𝐚𝐲.

Data preparation, AI deployment, and skill training require investments. But the ROI is still unclear. 

Where will the money come from, and who will pay? What about a hospital that uses a tool to reduce the need for patient care? Since the hospital gets reimbursed for the delivered service, what incentive would they have to spend money on a tool eventually reduce income? 

The investments and reimbursements are not clear, and before the ROI is fully clear, this will still be a big challenge to overcome. 

𝐋𝐚𝐬𝐭𝐥𝐲, 𝐞𝐭𝐡𝐢𝐜𝐚𝐥 𝐜𝐨𝐧𝐬𝐢𝐝𝐞𝐫𝐚𝐭𝐢𝐨𝐧𝐬 𝐚𝐫𝐞 𝐧𝐞𝐠𝐥𝐞𝐜𝐭𝐞𝐝.

The will to innovate is higher than acknowledging the risks. From an early stage, organizations need to start thinking ethics. Starting by establishing an ethics committee to oversee AI implementation.

Understanding the patient impact and mitigating negative impact. 
Educate clinicians and patients on how AI is being used. And give patients a chance to object. 

It all comes back to trust. 

If we are not willing to look at the larger picture and how the structural challenges impact the implementation of AI, we will not benefit from AI. 

What other challenges do you see?

Val Fraser

Why 90% of healthcare AI projects fail (and how to be in the 10%)

Are you thinking of integrating AI into healthcare?

Unfortunately, many healthcare organizations are struggling.

Here is a list of key steps that can help you be successful.

These steps can guide you through challenges, from system integration to data privacy.

Here's a look at what you can focus on:
1. Integration Plan: 
Carefully blend AI into current processes with a thoughtful strategy that combines new tech with existing systems.

2. Healthcare Vision: 
How do you want to impact healthcare in the future, and how can you disrupt current processes?

3. Navigate Regulations: 
The regulatory framework can be complex. Proceed wisely to introduce innovation.

4. Gain Clinical Acceptance: 
To bring the team on board with AI, focus on demonstrating real-world benefits rather than using hype.

5. Protect Data: 
Don't compromise on patient data privacy. Make sure AI integration aligns with high standards of data protection.

6. Develop Expertise: 
Lack of in-house AI skills is a major obstacle. It requires a shift in the skills matrix.

Every significant breakthrough begins with overcoming challenges.

Now is the opportunity to reshape the healthcare landscape. 
AI should not be pushed without a plan. 
And AI should not be seen an obstacle, but a catalyst for a transformation.

How have you overcome challenges in integrating AI in healthcare?

A proven way of evaluating AI solutions could reduce adverse events in healthcare. 

But too many tools show unanticipated biases. 
And performance varies too much depending on local data characteristics. 

Having a regulatory approval doesn’t necessarily mean the tool will work in the local practices. 
So healthcare organizations need setting-specific assessments that consider their unique populations. 

Many AI frameworks can be difficult to translate into practical assessments. 
Luckily, the OPTICA tool has been developed and tested on several AI solutions. 

The early results show that OPTICA identifies critical issues before deployment. 
Even having some solutions stopped when found unsuitable. 

And this is what healthcare needs. 
A framework to evaluate AI solutions before they do harm. 

And hearing the perspectives of different stakeholders. 
The clinical experts
The AI solution developers
The organizational data leads
The MLOps experts
The Organizational AI leads

With seven different stages, it defines who does what and when. 
Stage A: Clinical expert defines needs and explores alternatives
Stage B: Developer describes solution details, training data, and methods
Stage C: Clinical expert assesses developer responses from clinical perspective
Stage D: Data lead evaluates organizational data availability and compatibility
Stage E: MLOps expert addresses technical deployment aspects
Stage F: Clinical expert provides final assessment and post-deployment plans
Stage G: AI lead offers overall strategic assessment

And when a solutions has been approved and implemented, one should always plan for monitoring. 
Monitor for stability, output consistence, and ongoing performance. 

What do you think about the OPTICA tool? Does it have added value compared to other frameworks out there? 

__________________________

Interested in the tool? Find it here:
https://lnkd.in/e-3nePdg

Your AI strategy will fail if clinicians are not involved from day one. 

There will be no use for AI if they are not involved.

Most organizations tend to miss this key part, 
involving key players early on, 
and it's hurting your adoption rates.

From healthcare professionals to patients, 
administrative staff to tech providers, 
each stakeholder plays a crucial role in successful implementation.

Here's how to engage them effectively:

Healthcare Professionals (Doctors, Nurses, Specialists)

1. Engage early in the decision-making process
2. Gather input on practical requirements and potential challenges
3. Involve in pilot programs to assess usability and integration challenges
4. Provide comprehensive training on new technologies

Patients

1. Educate about new technologies and their benefits
2. Screen for digital literacy to identify those who may need extra support
3. Choose user-friendly technologies that don't require logins or downloads
4. Explain how new tools will save time or improve health outcomes

Administrative Staff

1. Include in needs assessment to identify inefficiencies in workflows
2. Provide training on new systems and processes
3. Gather feedback on technology effectiveness and areas for improvement

Technology Providers

1. Involve in stakeholder discussions to understand healthcare-specific needs
2. Collaborate on pilot programs and validation of technologies
3. Ensure technology is effective for healthcare professionals and interoperable with existing healthcare infrastructure
4. Ensure intuitive navigation in healthcare technology systems to facilitate adoption

Organizational Leadership

1. Conduct thorough needs assessments to align technology with organizational goals
2. Develop a strategic plan with SMART goals for digital transformation
3. Establish key success metrics to evaluate technology effectiveness
4. Create a common forum for stakeholder discussions

Including the different stakeholders can lead to:
1) Shared vision
2) Trust building
3) Addressing (and avoiding) conflicting interests
4) Improved compatibility
5) Ethical considerations

What learnings do you have from implementing new technical tools in your organization?

The push to innovate is starting to worry me. 

I know, I know, I am often talking about the possibilities of AI. 
And, I strongly believe that AI can add a great benefit to healthcare. 

But, the push to innovate, and to implement innovation without a clear goal still worries me. 

Ignoring the risks of the innovations means that the patients can get ignored. 

I wouldn't want to see that. 

In the race for innovative solutions, it's essential to be aware of the ethical dilemmas that could undermine our progress.

So, how do we ensure the ethical deployment of AI?

Here are 6 key aspects to get you going.
1️⃣ Start Ethical: 
Integrate ethical considerations into AI medical device development from day one, prioritizing patient well-being and ethical standards.

2️⃣ Bias Awareness: 
Understand and address data and algorithmic biases to prevent skewed outcomes and safeguard patient care.

3️⃣ Guidelines for Ethical Data: 
Establish clear guidelines for ethical data collection, conduct regular audits to maintain integrity.

4️⃣ Transparency Matters: 
Ensure transparency and explainability of algorithms to build trust among stakeholders and foster accountability.

5️⃣ Diverse Teams: 
Build diverse and ethically aware AI development teams to mitigate oversight in ethical decision-making.

6️⃣ Continuous Monitoring: 
Regularly monitor and update AI medical devices ethically throughout their lifecycle.

What are you doing to ensure ethical development and deployment of AI in healthcare?

Ethical AI in healthcare

Is your organization's AI putting patients at risk? 

Here are 6 critical checks to do to detect the risks before it’s too late.

More and more healthcare organizations are implementing AI.

But are they doing it responsibly?

Before you trust AI with patient care, run through these checks to protect both your patients and your organization. 

1️⃣ Assess Fairness

AI shouldn't discriminate. Regularly audit your systems for bias against protected groups. Crucial for patient trust.

2️⃣ Ensure Transparency

Can you explain how your AI makes decisions? If not, it might be time to invest in explainable AI. Clinicians need to be critical to decisions and understand why a decision is made.

3️⃣ Prioritize Privacy

With 112 million patients impacted by data breaches in 2023, data protection is non-negotiable. How robust are your AI's privacy safeguards?

4️⃣ Implement Accountability

Who's responsible when AI makes a mistake? Establish clear oversight mechanisms. Remember, AI should assist, not replace human judgment.

5️⃣ Test for Safety and Robustness

Is your AI resilient against unexpected inputs? Conduct rigorous testing to ensure reliability and security. A robust system not only protects data but also safeguards patient outcomes.

6️⃣ Evaluate Ethical Impact

Track your AI's real-world effects. 
Is it improving patient outcomes across all demographics? 
Are you seeing these benefits equitably distributed?

We should be talking more about responsible AI in healthcare, not just leaving AI to coincidences. By measuring these aspects, we can ensure AI benefits all patients equitably and safely.

What's your biggest challenge in implementing responsible AI in healthcare?

AI in healthcare isn’t as neutral as you think. 

AI could harm the very patients it’s meant to help. 

Without addressing the bias, we will never be able to benefit from the good. 

Here’s how we can fix it. 

1. 𝗜𝗺𝗽𝗿𝗼𝘃𝗲 𝗗𝗮𝘁𝗮 𝗤𝘂𝗮𝗹𝗶𝘁𝘆
AI models are only as good as the data they are trained on. 
Unfortunately, many datasets lack diversity, 
often overrepresenting patients from certain regions or demographics. 

Making sure datasets are inclusive of all populations is key to reducing bias.

2. 𝗥𝗶𝗴𝗼𝗿𝗼𝘂𝘀 𝗩𝗮𝗹𝗶𝗱𝗮𝘁𝗶𝗼𝗻
AI tools must be tested across diverse populations before deployment. 
Studies have highlighted how biased algorithms can worsen health disparities,
at every stage of development. 

Rigorous validation makes sure that these tools perform equitably for all patients.

3. 𝗧𝗿𝗮𝗻𝘀𝗽𝗮𝗿𝗲𝗻𝗰𝘆 𝗮𝗻𝗱 𝗘𝘅𝗽𝗹𝗮𝗶𝗻𝗮𝗯𝗶𝗹𝗶𝘁𝘆
Healthcare professionals need to understand how AI models make decisions. 
Lack of transparency can lead to mistrust and misuse. 

Explainable AI not only builds trust but also helps identify and correct biases in the system.

4. 𝗠𝘂𝗹𝘁𝗶-𝗦𝘁𝗮𝗸𝗲𝗵𝗼𝗹𝗱𝗲𝗿 𝗔𝗽𝗽𝗿𝗼𝗮𝗰𝗵
Bias mitigation requires collaboration between AI developers, clinicians, policy makers, and patient advocates. 

Diverse perspectives help identify blind spots and create solutions that work for everyone.

5. 𝗢𝗻𝗴𝗼𝗶𝗻𝗴 𝗠𝗼𝗻𝗶𝘁𝗼𝗿𝗶𝗻𝗴
Bias doesn’t stop at deployment. 
Continuous monitoring is needed to ensure AI tools adapt to new data and evolving healthcare needs. 

For instance, algorithms trained on outdated or incomplete data may maintain errors over time.

We will not see benefits of AI in healthcare if we don't address these areas. 
Focus is needed on reducing errors, aiding diagnoses, and personalizing treatments for all.

What steps are your organization taking to ensure fairness in AI healthcare tools?

Bias in AI for Health

Healthcare organizations are skipping this important step with AI

The step that will make or break the success of your AI project

If you want successful AI initiatives, you need to prepare. 

Healthcare needs more than just good algorithms to transform healthcare. 

Readiness ensures:
- Motivated and trained staff
- Infrastructure supporting AI
- Culture that embraces change
- Avoids costly failed investments

There are three pillars of readiness:
Psychological readiness: Are people willing and motivated to change?
Structural readiness: Are resources, data, and IT systems in place?
Cultural readiness: Does organization’s culture support innovation and learning?

These need to be assessed and common barriers identified, before implementation. 

Common barriers are:
- Lack of clear vision or strategy for AI
- Insufficient IT infrastructure and data quality
- Staff resistance or fear of job loss
- Regulatory and ethical uncertainties
- Limited funding and expertise

Organizational readiness can be built, by creating a positive environment where people feel safe and supported to embrace AI. 

Open discussions around fears should be encouraged. 
And leadership should encourage change while supporting staff.
And, of course, IT systems and data should be ready before any implementation. 

Guide your organization step-by-step. 
Define goals
Build multidisciplinary teams
Align strategy with regulatory requirements
Plan testing, training, and gradual rollout

Do you need help to assess AI readiness in your organization? Send me a DM.

The ugly truth about AI in healthcare: It's not the tech that's failing us.

It's the systemic barriers holding us back from improving healthcare.

Even with a lot of developments in AI in healthcare, adoption is still not where it could be.

Here are some observations of current challenges in the healthcare sector:

1️⃣ 𝗔𝗳𝗳𝗼𝗿𝗱𝗮𝗯𝗶𝗹𝗶𝘁𝘆 𝗦𝘁𝗿𝘂𝗴𝗴𝗹𝗲:
Healthcare settings face budget constraints limiting AI accessibility, and healthcare is focusing on short-term return on investment.

2️⃣ 𝗜𝗻𝘁𝗲𝗿𝗼𝗽𝗲𝗿𝗮𝗯𝗶𝗹𝗶𝘁𝘆 𝗣𝘂𝘇𝘇𝗹𝗲:
EHR systems lack standardization, hindering seamless data and AI integration.

3️⃣ 𝗗𝗮𝘁𝗮 𝗣𝗿𝗶𝘃𝗮𝗰𝘆 𝗖𝗼𝗻𝗰𝗲𝗿𝗻𝘀:
Sensitive medical data vulnerability raises patient confidentiality and trust issues, both with patients and clinicians alike.

4️⃣ 𝗟𝗮𝗰𝗸 𝗼𝗳 𝘁𝗿𝘂𝘀𝘁:
Lack of trust in the accuracy and reliability of AI algorithms, especially for high-stakes medical decisions hinders AI implementation and usage.

5️⃣ 𝗥𝗲𝗴𝘂𝗹𝗮𝘁𝗼𝗿𝘆 𝗟𝗮𝗴:
AI advancements outpace regulations, creating uncertainty and outdated models.

Instead, healthcare should start doing this:
1️⃣ 𝗔𝗳𝗳𝗼𝗿𝗱𝗮𝗯𝗶𝗹𝗶𝘁𝘆 𝗦𝗼𝗹𝘂𝘁𝗶𝗼𝗻:
Government incentives, partnerships, and targeted investments for financial accessibility, especially for lower-income countries. For more high-income countries, the long-term value of AI tools should be considered.

2️⃣ 𝗜𝗻𝘁𝗲𝗿𝗼𝗽𝗲𝗿𝗮𝗯𝗶𝗹𝗶𝘁𝘆 𝗮𝗻𝗱 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻:
Standardizing data formats and fostering collaboration between tech and healthcare, ensuring quick and simple integration of tools, reducing the time healthcare needs to spend on implementing new tools.

3️⃣ 𝗗𝗮𝘁𝗮 𝗣𝗿𝗶𝘃𝗮𝗰𝘆 𝗦𝗼𝗹𝘂𝘁𝗶𝗼𝗻:
Robust encryption to protect sensitive patient data and adherence to data protection regulations. Protecting this data will increase public trust in these systems.

4️⃣ 𝗟𝗶𝗺𝗶𝘁𝗲𝗱 𝗧𝗿𝘂𝘀𝘁 𝗮𝗻𝗱 𝗔𝗱𝗼𝗽𝘁𝗶𝗼𝗻 𝗦𝗼𝗹𝘂𝘁𝗶𝗼𝗻:
Clinicians and patients need training on how to use AI and its benefits. It should also be integrated into healthcare education so that the new workforce is aware of its value. Increasing education will improve adoption.

5️⃣ 𝗥𝗲𝗴𝘂𝗹𝗮𝘁𝗼𝗿𝘆 𝗦𝗼𝗹𝘂𝘁𝗶𝗼𝗻:
Regulation should reflect the need for regular updates and continuous monitoring of AI in healthcare. It should protect patient safety and, at the same time, support impactful innovation.

We need to recognize solutions that will get us to the future we want, not only acknowledging the challenges.

What would you include to reduce the gaps of AI in healthcare?

What if your AI-driven diagnosis is built on data that’s fundamentally flawed?

Even the most advanced AI can only be as good as the data it learns from. 

The biased data it has leads to dangerous errors.

Bias in medical research is becoming more essential for AI in Health.

So, I compiled top 5 points on bias in clinical use:

1) Bias distorts the data AI depends on, leading to misdiagnosis and ineffective treatments that could affect millions globally.

2) Nearly 40% of clinical trials remain unpublished, creating a skewed evidence base that AI systems unknowingly inherit.

3) AI trained on biased datasets can reinforce existing health inequities, disproportionately harming vulnerable groups.

4) Fixing bias requires more than algorithms. It needs transparency, ethical oversight, and cross-disciplinary collaboration.

5) Healthcare organizations need to prioritize identifying and mitigating bias to unlock AI’s full potential.

I am sure I am missing important points, but these were my top. 

What are you doing to reduce bias in your organization?

Interested in more on ethics and bias of clinical AI? You can read more here: https://lnkd.in/eJQRK945

Most AI tools in healthcare fail governance checks

Risking compliance failures and loss of trust

Companies that balance innovation with ethical accountability build stronger trust.

Here are 10 steps to start with AI Governance:

1. Establish AI Governance Structure
- Define AI governance roles: appoint an AI governance leader and form an oversight board.
- Formalize AI policies aligned with corporate strategy and regulations.
- Clarify definitions and scope of AI within the organization.

2. Inventory and Prioritize AI Use Cases
- Catalog all AI applications in clinical and operational settings.
- Identify high-risk, high-impact use cases for focused governance.
- Prioritize based on clinical value, regulatory impact, and ethical concerns.

3. Assess Regulatory and Legal Risks
- Map applicable regulations (FDA, HIPAA, EU AI Act, etc.) and compliance requirements.
- Conduct risk assessments to identify legal, privacy, and intellectual property risks.
- Embed automated compliance checks and audit trails in AI lifecycle management.

4. Ensure Data Quality and Integrity
- Implement strong data governance: validation, classification, and metadata management.
- Use diverse, representative datasets to reduce bias and improve fairness.
- Maintain data privacy and consent management protocols rigorously.

5. Mitigate Bias and Promote Fairness
- Regularly audit AI models for bias using fairness metrics.
- Adjust training data and algorithms to correct detected biases.
- Ensure AI tools are accessible and equitable across patient demographics.

6. Promote Transparency and Explainability
- Develop explainable AI models and clear documentation of AI decision logic.
- Communicate AI capabilities and limitations to clinicians and patients.
- Maintain version control and detailed model documentation.

7. Monitor AI Performance and Safety Continuously
- Set up ongoing monitoring for accuracy, errors, and model drift.
- Use “red teams” or independent testing groups to stress-test AI applications.
- Establish incident response protocols for AI-related adverse events.

8. Embed Ethical Principles and Accountability
- Create ethical guidelines involving multidisciplinary stakeholders.
- Define clear accountability for AI decisions and outcomes.
- Document governance processes and decisions for transparency.

9. Balance Innovation with Risk Management
- Use risk frameworks to evaluate new AI projects before deployment.
- Encourage innovation while enforcing safety and compliance guardrails.
- Adapt governance policies as AI technologies evolve.

10. Engage Stakeholders and Build Trust
- Educate clinicians, patients, and partners on AI governance practices.
- Report governance activities and AI outcomes transparently.
- Collaborate with regulators, industry groups, and standards bodies.

What are you doing to make sure AI governance is in place in your organization?

Organizations are removing AI tools from their tool stack. 

After a flood of AI tools, healthcare-related or not, 
Organizations felt pressure to implement innovations. 

But many organizations quickly realized that the tools didn't add enough value.

I think this stems from AI tools being implemented without having a clear problem or objective in mind. 

You see, many (healthcare) organizations feel the pressure to start implementing AI. 

So they quickly implement without a clear goal in mind. 
Nor do they know where the tool should be implemented. 
Or what value they should expect.

Leading to many AI projects failing. 

Here’s how you can do better:
- What process do you want to improve?
- What is your organizational problem or what process do you want to improve?
- Specific measurable objectives for clinical/operational objectives
- How does the tool support broader institutional goals or strategies?
- What are the risks or benefits on patient safety, efficiency, cost, savings, etc?
- How will success be measured?
- Which stakeholders should be involved?

Rushing to implement AI will eventually be a risk for the whole project. 

Not only can the project fail, but patient safety, data security, and overall trust are being gambled on.

If we want to get the true value from AI, we need to start with the basics first. 

What has your organization done to define its objective for AI projects?

Unpopular opinion: Your doctor shouldn't be the gatekeeper of your health data.

What would healthcare look like if it were more patient-centered?

This shift in power dynamics could change healthcare.

The European Health Data Space (EHDS) has the potential to improve healthcare in Europe in several ways:
- Improved quality
- Improved safety
- Improved access
- Decreased costs

With patients having better access to their health data, both the patients and the healthcare professionals can make informed decisions about their health and treatments.

Layer digital health and AI on top, and patients can access a broader scope of healthcare, which can improve the effectiveness and accessibility of the treatment.

The patients will be more empowered, but there are some concerns.

How do we fully ensure the safety and privacy of the health data?
How do we make sure second opinions are not negatively influenced by previously recorded information?
And how do we make sure all the data will not overload the end-user, that being the patient or the healthcare professionals?

Next to that, estimated numbers show low digital and health literacy across citizens.

The EHDS should benefit all, not only those who can currently manage their health.

I am optimistic that the EHDS can benefit both patients and healthcare professionals, and can offer a substantial benefit to innovation and research.

What do you think about the EHDS? Will the patients be further empowered now that the EHDS is coming?

If clinicians don’t speak up, how will HealthTech ever deliver real solutions?

If HealthTech doesn’t know what challenges clinicians face, 
How can they provide useful solutions?

Wondering about why HealthTech solutions miss the mark in healthcare?

When providers don’t fully understand hospital challenges, innovation is interrupted. 

True innovation starts when clinicians, 
Or even better, their organization openly communicates their needs.
Anything less is wasted potential.

Here are some observations:

1. Clinicians are taught not to question why clinical workflows are the way they are.
2. HealthTech organizations often lack insight into healthcare needs.
3. Fragmented communication hinders understanding of clinical workflow challenges.
4. Implementation inefficiencies arise from poorly tailored tech solutions.
5. Existing solutions remain underutilized due to a lack of shared challenges.

Instead, we need to:

1. An open dialogue between healthcare organization and providers.
2. Conduct thorough needs assessments with healthcare stakeholders.
3. Prioritize interoperable, scalable technology solutions.
4. Collaborate on co-creating tools addressing real healthcare pain points.
5. Share challenges transparently to discover existing innovations.

Bridging the gap between healthcare realities and technology has the potential to unlock efficiency and improve patient care. 

We cannot wait for the solutions to solve itself. 

What are you doing to share your challenges or understand healthcare challenges?

The regulatory environment in Europe sucks.

Especially if you can bring added value from your product.

One says it is for patient safety, and the other that it blocks innovation.

I am somewhere in the middle.

So what regulations do you need to be aware of with your AI tool?

Here is an overview of 12 important regulations to look into.

EU AI Act:
Aims to ensure ethical AI development while protecting health, safety, and fundamental rights. Non-compliance could result in legal issues and market access restrictions. NB. pending changes. 

GDPR:
The legal framework for personal data protection in the EU. It governs data protection, access, and (re)use. GDPR violations can lead to significant fines and reputational damage.

MDR:
Medical Device Regulation ensures safety and performance requirements. Non-compliance can prevent market access and pose risks to patient safety.

IVDR:
The In Vitro Diagnostic Regulation establishes specific requirements for AI-based diagnostic tools. Non-compliance can prevent market access and pose risks to patient safety.

Data Governance Act:
Increases data availability and facilitates data sharing. Specifies conditions for re-use of certain protected data, including data protected by intellectual property rights. It also provides processes and structures to facilitate voluntary data sharing.

Data Act:
Encourages fair access to and use of data generated by products or services. It specifies requirements for accessibility and transfer of generated data, which is crucial for AI development and use in healthcare.

Interoperable Europe Act:
Framework for cross-border interoperability of public services, including health systems. Ensures interoperability standards for public health systems across EU member states.

Open Data Directive:
Promotes the use of open data and stimulates innovation in products and services. Sets rules for re-use of public sector information, which can be valuable for AI development in healthcare.

NIS2 Directive:
Aimed at improving cybersecurity across critical sectors by imposing stricter security requirements and incident reporting obligations. Protection of sensitive data and ensuring the resilience of healthcare systems against threats.

GPSR:
General Product Safety Regulation is a broad framework for the safety of products placed on the EU market. It covers AI technology embedded in other products not covered by sector-specific regulations.

EHDS:
Facilitates the sharing of health data across EU borders for better patient care and research. Impacts how health data is managed and shared. Will create new opportunities and requirements for health data processing and sharing.

Product Liability Directive:
Modernizes existing product liability rules to include AI systems, software, and digital health products. Covers claims for personal injury, property damage, destruction or corruption of data, and other damages caused by defective products.

What would you add or remove?

20 ways to ensure digital health technologies reduce health disparities

Diverse data representation
→ Ensure algorithms use data from all population groups served

Comprehensive data capture
→ Include broad demographic, social, and clinical features in datasets

Identify high-impact areas
→ Target solutions to specific minority health needs and challenges

Co-design with stakeholders
→ Involve clinicians, patients, and public in AI application development

Integrate diverse partners
→ Collaborate with varied stakeholders for wider perspective and outreach

Recruit end-users
→ Involve patients and care teams throughout problem-solving process

Develop diverse workforce
→ Build teams equipped to identify and mitigate bias

Create ethical frameworks
→ Establish governance to protect rights and ensure transparency

Implement participatory design
→ Engage local communities in technology design and implementation

Assess algorithm fairness
→ Regularly test for bias across different population groups

Improve digital literacy
→ Provide training to enhance tech skills in underserved communities

Ensure accessibility
→ Design interfaces usable by people with various abilities

Address language barriers
→ Offer multilingual options in digital health tools and services

Culturally sensitive design
→ Incorporate cultural nuances in user experience and content

Promote equitable access
→ Develop strategies to provide technology to underserved areas

Monitor health outcomes
→ Track disparities reduction through continuous data analysis

Collaborate with policymakers
→ Advocate for regulations supporting equitable digital health access

Invest in infrastructure
→ Improve internet and device access in disadvantaged communities

Conduct impact assessments
→ Regularly evaluate how technologies affect different population groups

Foster public trust
→ Maintain transparency in data use and algorithm decision-making

What are you doing to reduce health disparities?

Only 61% of hospitals validated models locally before deployment. 

And only 44% evaluated models for bias. 

A study from the US looked at how hospitals use and evaluate AI predictive models. 
2425 hospitals participated. 

65% reported using AI or predictive models integrated in their EHR. 
But only 61% of the hospitals evaluated models for accuracy with local data, before deployment. 
56% of the hospitals did not systematically protect patients from potential biased or unfair AI. 

Most hospitals (92%) used models to predict health trajectories or risks for inpatients. 
79% to identify high-risk outpatients for follow-up.

If 946 of the hospitals surveyed do not validate the models used, 
imagine the impact it will have on clinical outcomes. 
Or time savings. 

And more so, if most hospitals use tools to predict inpatient health trajectories or identify high-risk patients,
How will biased tools impact patient safety and care? 

If these biases are not evaluated, the tools can propagate bias in access to care. 

Biased models trained on incomplete data will increase disparities, 
Where high-risk predictions lead to reduced care, slower recovery, and worse outcomes for marginalized patients. 

This shows me that there is an urgent need for intervention, to make sure that every patient population benefit from the AI tools. 

Healthcare systems need to be more critical on model bias. 

If this continues, not only will patient safety suffer, 
Treatment quality reduce, 
But overall trust in AI in healthcare will also decline. 

Though these numbers don’t reflect the reality in Europe, I don’t think the reality is very different. 

We should not keep validating models as optional. 
It is crucial for patient safety, treatment quality, and trust. 

How many tools in your organization can you say have been evaluated before use?

_________

Read the study here: https://lnkd.in/eD4dJccf

I have seen it firsthand. 
Ethical being forgotten. 
Innovation pushed forward. 

But, without considering patients and the risk of innovations, 
Trust in tools, both from healthcare professionals and patients alike, will fail. 

The first thing organizations need to do, is to start thinking ethics. 
Not just after something is implemented, but also during the exploration phase. 

Start with creating a multidisciplinary ethics board. Emphasizing the multidisciplinary. 

Then identify risks, whether a tool is already implemented, but preferably before. 

Don’t forget policies on data collection and informed consent. Focusing on the patient confidentiality. 

Make sure healthcare professionals can understand the AI outputs. 

But also, remember to have methods to identify, mitigate and monitor bias. 

There should be clear accountability for AI-supported decisions, but make sure healthcare professionals are in charge. 

And at last, continue to evaluate how the tools impact the patients - effectiveness, safety and fairness. 

These are steps to take when creating an ethical AI decision-making framework. 

But it should not end with a framework. 
The framework must become alive with the healthcare professionals working with the tools. 

What have you done in your organizations to give ethics a place in innovation?

taplio

Sigrid Berge van Rooijen's Linkedin Analytics