On May 15, 2024, more than 100 IT healthcare experts listened to an expert panel discussion on cloud development and AI adoption. Hosted by Calian IT and cloud development expert Ambrose Mok, the webinar panel included Alicia St.Hill, Executive Director at BORN Ontario, Dr. Lillian Sung, Pediatric Oncologist at The Hospital for Sick Children (SickKids), and Peter Jones, Industry Lead, Canadian Healthcare, Microsoft Canada.

During the one-hour discussion panelists were keen to share their cloud development journeys and the priorities that drove the decision to move from on premises to the cloud. Most notably, St.Hill and Dr. Sung highlighted the following benefits:

• A cloud environment has enabled new services, such as a big data transformation into manageable results-driven insight without software or hardware investments.
• Because healthcare is cost-conscious, moving to the cloud has provided the ability to introduce a managed service model that reduces maintenance overhead and only requires payment for resources when they are being used.
• Being in the cloud enables a more flexible architecture with on-demand capacity, which is needed as we continue ingesting a deluge of data in healthcare.

AI rounded out the discussion focusing on the impacts across the healthcare ecosystem and use cases that are being explored to support patient care and administrative processes.  A few considerations were top of mind:

• Patient-level predictive modelling to improve patient outcomes is at the core of many AI use cases. Large sets of data can be used to understand linkages to health outcomes enabling improved support at the right time and level the patient requires.
• AI can streamline administrative tasks that bog down a healthcare system so that important information is available to researchers and patients faster and with less privacy risk.
• AI can be used to identify data gaps where the quality of data may not broadly represent a cohort or serve everyone equitably.

When we polled our registrants, most identified themselves as being at a stage where they are discussing and developing a list of valuable use cases to implement. And because we are living in a healthcare ecosystem where 97 per cent of the data goes unused, as the industry begins to truly adopt AI, it will influence positive patient outcomes and medical research.

The industry is shifting budget attention towards operating rather than capital expenditures and this will ultimately enable the flexibility for health institutions to stay current and use the latest technology in a secure, compliant way.

Together Microsoft and Calian are here to help you. Consider Azure OpenAI, a service that you could implement, use to build a sandbox in your dev environment, play with and learn from, all in a controlled environment.

Some of the questions asked during the Q&A period included:

What are some of the challenges you came across when transitioning to the cloud?  What issues did you come across that you didn’t foresee?

Ambrose Mok: Healthcare clients who move existing data from on-premises systems to the cloud can be complex. Data formats, compatibility and ensuring data integrity during migration are critical. Organizations can also find themselves resistant to change or have a lack of cloud expertise within, so it is a good idea to enlist a cloud management vendor who can guide and educate teams on how to best adapt to these new processes.

You should continuously evaluate your cloud strategy, the applications that you need and the associated costs. By doing this, you are better able to manage transitional and operational costs, create more efficient and adaptable processes and ultimately keep the data secure and compliant.

---

How did SickKids create a realistic cloud budget, recognizing all the different services offered by a cloud provider, plus the egress fees, plus security, etc.?

Dr. Lillian Sung: I think one of the beautiful things about Microsoft is that it has lots of tools available to project what a particular resource will cost. We monitor all costs carefully and if the costs appear too high, we evaluate whether we actually do need that service or not. In the end our costs have actually been very modest.

Peter Jones: More and more hospitals are shifting to the cloud and accommodating for that. Most organizations are starting to put more OpEx in place to allow for cloud environments. Most CIO’s want to move to the cloud, and it is a matter of managing the expenses associated with it. So, Microsoft has taken to sharing best practices on how to manage these costs and the benefits, like enabling hospitals to stay current and up to date while remaining secure and compliant within their IT infrastructure.

---

How did you address the apprehensiveness of adopting a cloud solution vs. on-prem solution?

Alicia St.Hill: Ask lots of questions and pressure test every assumption. There is no question that you have that should go unanswered. Additionally, healthcare is very collaborative.  Find allies and partner organizations to share and present their journey with your teams.

---

AI systems in medicine need to be trained on medical data, not through public produced content easily available on the web (like chat GPT is trained on). While the cloud is fairly secure from cybersecurity threats, will the info in the cloud be “sold” or used for training future AI tools built specifically for the medical field? While this is great to develop diagnostic tools, what does this mean for the privacy of patient data?

Ambrose Mok: Cloud providers are diligent and focused on safeguarding patient confidentiality and ensuring compliance with privacy regulations. De-identification techniques can help protect privacy. Removing personally identifiable information (PII) while retaining useful features for AI training is essential. At the end of the day, collaboration between healthcare providers, researchers and cloud providers is essential to navigate this complex landscape.

---

With synthetic data, while there are lots of benefits as you had mentioned, if you provide AI ability to generate new “made-up” data, this sometimes can lead to hallucinations in the output. How are hallucinations being mitigated?

Alicia St.Hill: Great question. It starts by putting the synthetic data to the test—testing whether the analysis from the synthetic data gives the same conclusions as past analysis (including both complex and simple) that was performed on the raw data that was used to create the synthetic data.