Bringing AI Revolution to the KW Region: Formulating the Strategic Blueprint for Next-Generation Healthcare

A data-centric approach to unlock healthcare innovation

By Sirisha Rambhatla, Ph.D.

Assistant Professor, Department of Management Science and Engineering

The business case for artificial intelligence (AI) and machine learning (ML) is compelling — offering organizations faster insights and improved decision-making. However, according to a recent study [1,2], only 32% of AI/ML deployments successfully move from pilot to production, down from a 2022 Gartner survey which reported it to be 54% [3]. Despite these variations, the fact remains that a significant majority of AI/ML projects in companies fail to reach the deployment stage, highlighting the challenges organizations face in implementing AI/ML solutions effectively [4]. This is because ML modelling is never one-and-done, requiring continuous training on evolving data and monitoring for reliable operation. Hence, organizations face challenges at every stage of AI/ML workflow [5]. As a result, despite organizations' eagerness to embrace AI, successful deployment demands mastery of data and engineering ecosystems. This expertise is crucial for identifying use cases, training models, and maintaining AI/ML pipelines — a process known as machine learning operationalization (MLOps).

At the heart of any ML model is the data. Hence, the foundation of a good ML pipeline begins with a good data strategy. However, the data strategy for each organization can look different and needs to be customized. This is because downstream use cases can drive critical decisions in data collection, storage, and retrieval, requiring a tailored approach for a sound data strategy. Efficient data retrieval is thus crucial, as the engineering effort required can grow rapidly, potentially becoming cost prohibitive. Hence, by aligning both long-term and short-term goals, organizations can effectively manage expectations and control expenses.

But all data is not created equal, the quality of data — which encompasses both the degree to which records are complete and their accuracy, and the quantity — the number of such records, both can be critical for the downstream application. In the context of healthcare, collecting demographic information is also key to measuring existing quality of service (QoS) differences between groups, and to ensure that models trained on such data do not reinforce any existing biases. As a result, assessment of data quality and clear documentation of its provenance is essential to assess the org needs to consider and improve, as established by our previous work with Grand River Hospital (GRH) [6].

With the merger of Grand River Hospital (GRH) and St. Mary’s General Hospital, alongside the launch of the CareNext Coalition uniting these hospitals with the University of Waterloo (UW) [7], our UW-GRH team is laying down a robust foundation for the AI revolution in healthcare, ensuring that our region is well-positioned to leverage advanced analytics and ML algorithms for improved patient outcomes and operational efficiency. This partnership transforms local healthcare capabilities while serving as a blueprint for other organizations in the region looking to adopt AI/ML solutions. Thus, creating a thriving ecosystem for technological innovation via a robust pipeline for top AI/ML talent.

Our initial pilot study is supported by generous contributions from the Graham Seed Funding for Transformative Health Technologies [8] and focuses on analyzing the current state of data collection and engaging with stakeholders to listen to their needs and perspectives. This community-driven approach is key to ensuring that any subsequent solutions are well integrated. Scaling AI/ML solutions is also fraught with challenges, and this is often why many promising proof-of-concept AI/ML projects do not get deployed [5]. Successful implementation requires both engineering and AI/ML expertise to transform innovative concepts into practical, real-world applications. As a result, our work also critically impacts other AI/ML initiatives with GRH, ensuring that these healthcare solutions get deployed, improving patient experience and their outcomes.

Our initiative also aligns with UW’s Healthcare Global Futures vision by leveraging real-world AI/ML expertise to address pressing healthcare challenges in KW, Canada, and globally [9]. It represents a significant advancement towards healthcare accessibility, reflecting the university's commitment to inclusive, innovative, interdisciplinary solutions for optimal health and well-being.

References

[1] Eric Siegel, "Survey: Machine Learning Projects Still Routinely Fail to Deploy”, KDNuggets, 2024. https://www.kdnuggets.com/survey-machine-learning-projects-still-routinely-fail-to-deploy 

[2] Karl Rexer and Eric Siegel, "2023 Data Science Survey”, 2023. https://drive.google.com/file/d/1Mz3WmtcvUl-00gaT2XKCxdE5-pqbOOjz/view?pli=1 

[3] New Gartner survey: Only half of AI models make it into production, Sean Michael Kerner, 2022. https://venturebeat.com/ai/new-gartner-survey-only-half-of-ai-models-make-it-into-production/ 

[4] Machine Learning Deployments Suffer High Failure Rates, George V. Hulme, 2024. https://digitalcxo.com/article/machine-learning-deployments-suffer-high-failure-rates/

[5] Paleyes, Andrei, Raoul-Gabriel Urma, and Neil D. Lawrence. "Challenges in deploying machine learning: a survey of case studies." ACM computing surveys 55.6 (2022): 1-29.

[6] Anand Murugan, "Implementing Fairness in Real-World Healthcare ML through Datasheet for Database", Masters Thesis, Supervised by Prof. S. Rambhatla and Prof. A. Wong, University of Waterloo, 2024.

[7] Bringing innovation to the bedside, Media Relations, University of Waterloo, 2024. https://uwaterloo.ca/news/media/bringing-innovation-bedside

[8] Graham Seed Fund for Transformative Health Technologies, University of Waterloo. https://uwaterloo.ca/transformative-health-technologies/graham-seed-fund#partnerfocus

[9] Waterloo at 100: University of Waterloo Strategic Vision, 2024. https://uwaterloo.ca/waterloo-100/sites/default/files/uploads/documents/fp2245_waterlooat100.pdf