AI adoption continues to lag in the healthcare sector. The complexity and sensitivity of patient data, how they’re governed and communicated within systems and between institutions requires significant time and investment. The last decade saw health information data slowly becoming digitized however strict privacy regulations, security requirements, and liability concerns–have all made AI challenging to fully integrate into the industry.
The resistance to change is emblematic in a sector that has idealized existing practices and workflows, and the legacy systems that have been working for decades.
The use of patient information to train high performing AI algorithms needs to consider, not only the existing data privacy and consent laws, but also whether there is adequate access to rich data sets that can produce reliable outcomes. In addition, the black box nature of most algorithms does not encourage confidence within a medical profession, which needs to fully understand how these results are generated.
The integration of artificial intelligence (AI) in healthcare holds immense potential for transforming patient care and outcomes. Navigating the ethical landscape of healthcare data usage is a crucial consideration. I met with Dr. Devin Singh, the founder and CEO of Hero AI, an early-stage health tech company, to delve into the challenges surrounding healthcare data usage, the need for advocacy as he pushes for significant change within a highly regulated sector.
A Catalyst for Systemic Change
Dr. Devin Singh’s journey into the world of AI began with a profound experience while working as a junior doctor many years ago at The Hospital for Sick Children (SickKids). Witnessing the tragic death of a young patient due to systemic challenges ignited a fire within him. Fueled by the determination to revolutionize healthcare, Singh began his new journey towards computer science.
“Many years ago, when I was a junior doctor, we had a young patient die, a preventable death, partially because they were waiting too long… I just remember being so angry that this could happen. And it drove me to think through how do we really think in a dramatically different way in solving meaningful problems in healthcare? That’s what led me to go from doctor to computer scientist. So, I did a master’s in computer science and artificial intelligence, and we started to realize that we needed to find a way to empower hospitals–to do more with less–and to liberate their electronic health record (EHR) data so they could solve problems like this in a really agile way.”
The Dichotomy: Agile and Healthcare
At Hero AI, their process involves taking real-time streaming health record data and inputting it into decision logic or machine learning algorithms. The output is then directed to highly customizable front-end user interfaces, including web apps, patient apps, and provider apps. The goal has been to streamline this process, allowing for customization and deployment within a matter of hours rather than weeks or months. This innovation has enabled practitioners to address real clinical problems in hospitals swiftly. As per Singh;
“We can identify an issue during a meeting, rapidly develop solutions, and proceed to testing and deployment at a speed that’s really never been seen before in healthcare”.
Respecting Privacy and Compliance
Respecting patient privacy and complying with stringent regulations surrounding healthcare data are paramount. Singh acknowledges this and underscores the importance of trust between health systems, physicians, and patients. Hero AI adopts an ethical approach by engaging in extensive stakeholder interviews, including patients and their families, to gather insights into their data usage preferences and concerns.
“We’ve done obsessive stakeholder engagement to interview, not only parents and families but also the kids themselves… understanding the user well and trying to gather what they want to see happen to their data and what their concerns are”.
Singh recognized the challenge of working within existing legislation and privacy laws that were not initially designed to address the complexities of big data. Harmonizing patient opinions with legal boundaries is a crucial task.
Navigating the Regulatory Landscape
Singh admits the process of achieving harmonization has not been easy and required years of authentic engagement with lawmakers, Health Canada, patients, privacy lawyers, and experts.
“Our objective was to find authentic ways to merge patient perspectives, legal boundaries, and data security measures… Is there a single clear path right now in Canada to go from creating an AI innovation like what we’ve done to then getting it deployed safely and in a way that respects privacy? There isn’t because the laws weren’t written for AI specifically, however that’s part of the advocacy that our group does as well… So, I have my AI researcher hat at SickKids as well, and part of the research I do is around regulatory reform. How do we think through creating streamlined pathways that can ensure we can deliver AI solutions to patients, but do it in a way that is obsessed with privacy?”
Leveraging AI to Increase Efficiencies
Addressing healthcare disparities and promoting equitable care for all patients lies at the core of Hero AI’s mission. Rather than attempting to absorb all available data, Hero AI focuses on specific clinical challenges faced by healthcare facilities. By leveraging precise datasets tailored to particular problems, such as reducing wait times for abdominal ultrasounds, AI models can provide accurate diagnoses and expedite critical care.
“We strive to identify a very specific clinical problem that an emergency department might be struggling with. Wait times for getting an abdominal ultrasound is an issue. Let’s say you might have appendicitis. For you to get that surgery, you’re going to arrive at the emergency department. You’re going to get that ultrasound done which will give the diagnosis to get that surgical intervention you really need. So, we want that time to be short. So we identify a problem in healthcare, and then we figure out the specific data that we need to build an AI model that will be likely to diagnose that you have an appendicitis, and while you’re waiting those four hours, why don’t we just order the ultrasound? That’s research that we do. We are thinking through how we can use AI and automated ordering potentially in a precise way to advance the speed of care”.
Towards a More Equitable Healthcare
The panacea of AI is to effectively treat patients based on their unique indications, ethnicity, history and geography. What has not been addressed are the systemic challenges that have kept everyone from receiving equal access and equal care. This may imply having the ability to aggregate electronic health record (EHR) data from almost anywhere across health care organizations and jurisdictions to surface insights about diseases with input across genders, ethnicities and geographies.
The opportunities are immense; however, the privacy constraints must be taken into account. Singh acknowledged that data from clinical trials, triage notes, patient symptoms, etc. can be included to build high performing models, however, in order truly declare the model is truly equitable, we need ethnicity data to validate the model plus continuous algorithmic auditing to ensure sustained efficacy.
“And so that’s a key point of advocacy here is that if we just ignore the collection of that data and let’s assume care is equal for everyone. Your initial premise was that care isn’t necessarily equal for everyone; we need that data to demonstrate where those deficiencies might lie for certain groups of people. And we also need that data to challenge the model and demonstrate that it is fair, or it isn’t fair. That transparency is needed. I advocate for creating a regulatory landscape that promotes access to data to make these crucial decisions. The goal is to do this from the patient perspective; they need to be at the center of this”.
Singh emphasizes the importance of ensuring fairness and equity in AI deployment within healthcare. While their AI models have shown promising results in perspective trial data, they are vigilant about potential biases.
Singh points out, “How do we know that that model is going to perform equally depending on your language preference, your gender or your race?”
He stresses the need to assure the public that the solutions they deploy, like faster ultrasound diagnostics, will be fair, irrespective of language preference, gender, or race. Transparency is paramount, and they are committed to addressing any differences in performance among different groups with honesty and openness.
Overcoming Hurdles and Driving Impact
Hero AI has successfully overcome significant hurdles related to privacy, regulation, and data access, establishing a robust foundation. The organization has engaged extensively with stakeholders, adhered to privacy regulations, and prioritized patient outcomes. Singh aims to deploy AI models that prioritize high-risk patients, automate mental health advocacy, and leverage precision medicine frameworks to deliver personalized care.
This founder shares his optimism
“All the uphill battles that you were talking about around privacy, exploring what the regulatory boundaries are, access to the data… we’ve got really great foundations in place. Now, this year is all about deployment and impact”.
With the healthcare industry facing mounting challenges, AI-driven solutions offer immense potential for positive change. Devin Singh and Hero AI strive to reshape the healthcare landscape, ensuring that AI innovations close health gaps rather than widen them. Through collaboration between technology innovators, regulatory bodies, and healthcare providers, a future of equitable, efficient, and patient-centered healthcare is hopefully within reach.
“The real motivation is when that case [tragic death of a young patient] happened, a fire ignited in me, and I said it can’t be how healthcare in Canada is right now. It just can’t. So, I am running up that hill really fast, and we’re making the change to see these things happen”.