Advancing Healthcare With Data: The Critical Juncture Between Progress And Privacy

As data-driven technologies continue to evolve, healthcare stands at an important crossroads. Private health data plays a crucial role in advancing research and personalized medicine, as it enables researchers to identify patterns and insights that can lead to significant breakthroughs in disease treatment. However, managing this sensitive data requires careful consideration.

In some jurisdictions, researchers can obtain consent for future, unspecified studies, while in others, personal data is de-identified before use. While these approaches protect privacy, they can sometimes limit the depth of insights by removing key context.

The risk of data leakage is a real concern. Sensitive health information can potentially be bought and sold, contributing to the development of models that may expose individuals or cause unintended consequences. The healthcare sector faces unique challenges around privacy, legal compliance, and data security. Balancing innovation with the need for privacy, public trust, and fairness is critical to unlocking the full potential of data-driven healthcare while improving patient outcomes and operational efficiency.

This examines AI adoption in healthcare, as I elicit views from two healthcare startup organizations, who are addressing the risks associated with the most sensitive personal information amid regulatory uncertainty, and lack of data sharing standards. Is it possible to balance the need for data-driven innovation while maintaining patient privacy and trust?

Devin Singh is the Founder and CEO of Hero AI , a health tech startup in Canada that offers advanced clinical automation solutions, giving healthcare systems AI-powered tools to improve patient care and efficiency. Dustin O'Dell is the Cofounder and CEO of SymetryML, a US startup which has created a cutting-edge federated AI platform that enables data collaboration and analysis, without data movement, enhancing how healthcare and life science companies utilize sensitive patient data for research.

Current State Of AI Healthcare: The Potential vs. The Reality

Of the top sectors adopting AI, according to Statista , healthcare ranks third, with a global 6% adoption rate, compared to technology and finance at 63.7% and 10.4% respectively. While this percentage might seem lower compared to some other sectors, according to Vention, healthcare is seeing significant AI integration and impact:

The global AI healthcare market was worth $19.68 billion in 2023

In the US, this year , about 10% of respondents working in healthcare organizations reported their organization was at mid-stage adoption of generative AI while, 14% stated their organization was at early-stage adoption, with a first solution running in production as a customer-facing or mission-critical system.

In Canada, over the coming year, AI software adoption according to Stats Canada plans vary across industries in Canada. The sectors showing the highest likelihood of implementing AI software are: Professional, scientific and technical services (26.6%), Information and cultural industries (24.3%) and Finance and insurance (12.9%).

While there are no recent stats on the state of Canada's health care sector's AI adoption recently, DIGITAL, Canada's Global Innovation Cluster, announced $26M in AI-powered healthcare projects in May 2024, demonstrating a significant commitment to advancing AI-powered solutions in the healthcare sector. These investments have targeted key areas in improving healthcare access particularly in rural areas, accelerating clinical trials, enhancing medical imaging and diagnostic capabilities, optimizing patient care processes, and developing AI-driven platforms for telemedicine and patient contact centers.

Devin Singh from Hero AI, a Canadian startup, explains how regulation has had a dual effect on AI innovation in healthcare, "In one sense, regulation has stifled innovation in AI for Healthcare in that it has limited the ability for hospitals to share data and use it, but in another very real sense innovation has been stifled by a lack of regulation. Healthcare institutions inherently exercise great caution and require well-defined protocols and safeguards to be established before they can integrate any novel technology, pharmaceutical, or therapeutic approach into their standard treatment practices."

Singh continues, "If there are no clear rules to demonstrate that something meets the standard of care then it opens hospitals up to liability. AI models are very nuanced and do not normally generalize well from one institution to the next, particularly when the contextual information is often captured in free text notes." For AI to be widely adopted in healthcare, Singh emphasizes that institutions require clear regulations defining the necessary checks and safeguards to ensure models are safe for broad use as well as within their specific institutions.

EHR Integration Challenges: Privacy Laws, Market Dominance, And Antitrust Concerns

Electronic Health Records (EHRs) were created to make healthcare more efficient and provide useful data for AI solutions. Singh discusses the importance of systems that allow different EHR platforms to work together, like HL7, FHIR, and SMART.

However, Singh points out key challenges that slow progress. "Healthcare leaders are struggling with outdated privacy laws, which leave AI use in a legal grey area and increase perceived risks." He presses on the need for updated laws that clearly explain how AI can be used, so healthcare organizations can innovate while staying compliant.

The integration of EHRs with other systems is also challenging. "Getting data from EHRs is often difficult, not just for third-party solutions but for hospitals too, " Singh adds, highlighting problems like expensive API fees and reoccurring data quality issues caused by EHR software updates, which add significant costs for healthcare organizations trying to improve patient care with AI.

While Singh acknowledges the potential of systems like HL7, FHIR, and SMART on FHIR, he warns that as major EHR companies dominate the market, their influence on AI deployment will face greater scrutiny.

Recently reported by HCI Innovation , P Health, a healthcare data integration platform, filed a federal antitrust lawsuit against Epic Systems Corporation, alleging that Epic has leveraged its dominant position in the electronic health records (EHR) market to stifle competition. The lawsuit claims that Epic controls EHRs for over 75% of the U.S. population and is now attempting to extend its dominance into the emerging payer platforms market. P Health accuses Epic of using "a multi-tentacled approach to try to squash P... Over the past six months, it cut off access to data for P's customers, lobbed now-discredited complaints, and overwhelmed P's support operations by stoking baseless security concerns."

Healthcare Obstacles: Data Silos, Complacency, Established Player Dominance And The Struggle To Innovate

AI solutions frequently face challenges due to inadequate data availability. As Dustin O'Dell of Symetry points out, "Data often exists in silos, and strict privacy laws hinder the collaboration necessary to compile large, high-quality datasets for specific patient populations." Furthermore, many companies struggle with non-standardized and "messy" data. O'Dell emphasizes that "Assessing, cleaning, and preparing data for AI or ML-driven analysis demands significant manual effort."

There is also a reluctance to change according to O'Dell, "Innovation in large healthcare organizations often faces roadblocks due to reluctance to disrupt the status quo, especially when considering AI projects that push boundaries. This is a classic example of the Innovator's Dilemma. "

For AI technology providers, O'Dell identifies several key challenges, "Companies are still trying to figure out how they leverage AI which first requires them to understand their data and business problems, so they move very slowly, if at all." The dominance of established players also poses a significant hurdle, as O'Dell observes that "most companies will first gravitate towards the usual suspects to test and learn with AI," sidelining startups, who struggle to breakthrough and scale.

Information overload is another issue, with O'Dell describing the "AI noise" that makes it "simply overwhelming and hard to sort out who does what and why one solution is different or better than the other." Finally, he cautions against overpromising, noting that AI companies often harm their own prospects by setting unrealistic expectations, which leads to customer disappointment and frequently results in project abandonment or delays.

AI-Driven Healthcare: Enhancing Patient Outcomes While Safeguarding Privacy

Singh says patients will consistently express their desires for shorter wait times, quicker diagnoses, and more precise treatments. However, implementing AI solutions in healthcare requires a careful balance between these demands and privacy concerns, alongside the necessity for personalized care. His company is addressing this challenge by using AI to enhance patient care without compromising privacy. As he explains, "Hero AI empowers hospitals to meet these needs by implementing real-time AI systems that automate certain aspects of patient care delivery while protecting identifiable data."

He illustrates an example from the Hospital for Sick Children (SickKids) in Toronto, Canada—the second-ranked pediatric hospital in the world and a leader in health AI —where the Hero AI platform accelerates care for children experiencing acute mental health crises. By utilizing triage data to identify when psychiatric support is needed, the system automates consultations within minutes of a patient arriving in the emergency department. This innovation has cut wait times for psychiatric care by over 50% for many patients, significantly reducing their stay in the emergency department and freeing up capacity.

Additionally, Hero AI has introduced automated safety alerts for surgical patients facing prolonged wait times, which has led to a 30% reduction in the time required for ordering diagnostic tests for some patients. Singh stresses that access to data is limited to healthcare providers within a patient's circle of care to "ensure that data is never exposed unnecessarily." He adds that all data is securely encrypted and used exclusively for its intended purpose. Importantly, hospitals maintain ownership of any new machine learning models developed through the platform. Singh is committed to ensuring that highly sensitive information remains secure, stating, "This approach respects the lived experiences of providers and patients that generated the training data, while delivering impactful, privacy-preserving clinical automation solutions. All data is owned by the hospital, and it's encrypted in transit and at rest, strictly protected from unauthorized exposure."

Singh also highlights the collaborative efforts between Hero AI and SickKids in the responsible adoption of AI: "Informed consent will be critical when AI begins to autonomously influence care, and transparency is essential. Hero AI and SickKids achieve this through deep patient engagement and co-design." He references Dr. Sasha Litwin, an emergency physician at SickKids and expert design methodologist, who emphasizes the importance of ensuring that "technologies are equitable, accessible, and informed by end users." This is vital for ensuring that patients understand and consent to how AI influences their care.

Addressing the limitations posed by biased models, Singh points out the need for transparency, stating, "Patients and providers need to explicitly understand whether an AI solution has been validated in populations that are demographically and medically representative of our patient base, and if the solution will perform well for them."

Privacy-Preserving Data Sharing Through Federated Learning

Healthcare organizations encounter significant obstacles in sharing and analyzing patient data across borders due to stringent privacy laws. In Canada, Health Information Custodians (HIC) are organizations that have custody or control of personal health information. These can include hospitals, pharmacies, healthcare practices, and nursing homes, all of which are governed by The Personal Health Information Protection Act (PHIPA). This law establishes the framework for when HICs can collect, use, and disclose personal health information (PHI). Generally, physicians may access PHI only with patient consent, and unauthorized access is strictly prohibited. Patients can also request restrictions on who can access their PHI, and custodians must report certain privacy breaches to affected individuals, the Information and Privacy Commissioner, and/or regulatory bodies.

In the United States, the comparable legislation is the Health Insurance Portability and Accountability Act (HIPAA), which mandates the protection of personal health information. Across jurisdictions, strict compliance in accessing, controlling, and disclosing PHI is similar.

Federated Learning (FL) is a decentralized machine learning approach where models are trained across multiple devices or servers without sharing raw data, enhancing privacy and security. O'Dell explains how Symetry ML evolved its FL solution to address these existing gaps, stating, "Our privacy-preserving solution virtually unifies patient data worldwide, making it available for on-demand analysis without the data ever leaving its point of origin." Unlike conventional federated learning, which trains models locally and then shares model weights to central server, O'Dell expresses: "SymetryML has developed a method that creates secure mathematical abstractions of the local data, sharing only those abstractions for robust analysis and inference through a built-in analytics toolkit." This approach enables partners to share their underlying data in a privacy-preserving manner, offering key advantages in data governance, data leakage prevention, scalability, and performance. He adds that their solution complies with HIPAA and GDPR regulations, as verified by third-party audits.

O'Dell illustrates the effectiveness of their solution: We're working with a top-five pharmaceutical company and their research partners to help them connect patient data globally to enhance analysis. Valuable patient data was stored locally across different countries and couldn't be shared or combined due to strict privacy laws. Our solution provided a new way to collaborate, resulting in access to more patient data, which is critical for advancing research."

He argues that the prevailing industry constraint limiting efficient data sharing and analysis at scale, while maintaining privacy, negatively impacts essential research. Symetry addresses this challenge, enabling healthcare organizations to leverage larger datasets for research while upholding strict privacy standards.

O'Dell notes that limitations in traditional federated learning methods could raise concerns about overall model efficacy, particularly if one stakeholder's model accuracy is questioned. He explains that traditional federated learning "often struggles with data overfitting to local models, and optimizing the central model is complex and limited from an experimentation perspective." In contrast, their newer version functions analytically as if all parties have access to the same distribution of data because it shares mathematical abstractions rather than model weights. He adds, "We offer the ability to conduct analysis or build models independently or collaboratively, allowing each party to ensure their models are highly accurate while keeping them private if they choose."

Disruption will not wait for the Healthcare Sector

The intersection of data innovation and privacy in healthcare presents enormous promise but also reveals a sector mired in complacency and a reticence for disruption. Advancements in AI and data sharing can lead to significant improvements in patient care and enabling expanded healthcare capacity. However, this progress must be tempered with a strong commitment to protecting sensitive health information and maintaining patient trust. The ongoing dialogue around regulatory frameworks, informed consent, and equitable access to AI technologies is crucial. Transparency and collaboration are the current priorities to achieve the balance of innovation and compliance but must be in lock-step with the speed of artificial intelligence.