Approaches to identify scenarios for data science implementations within healthcare settings: recommendations based on experiences at multiple academic institutions

Lillian Sung, Michael Brudno, Michael C. W. Caesar, Amol A. Verma, Brad Buchsbaum, Ravi Retnakaran, Vasily Giannakeas, Azadeh Kushki, Gary D. Bader, Helen Lasthiotakis, Muhammad Mamdani, Lisa StrugFrontiers Digit. Health 2025 · DOI

Introduction

While the potential for artificial intelligence (AI) and machine learning in healthcare is vast, many initiatives fail to transition from concept to clinical reality. A primary bottleneck is scenario identification: the process of selecting a specific clinical or operational problem where data science can provide measurable value. This paper synthesizes the collective experiences of seven major academic healthcare institutions to provide a strategic roadmap for identifying and greenlighting the right AI use cases.

Methodology

Moving beyond theoretical frameworks, this study employed a qualitative, experience-based approach. Representatives from seven diverse Toronto-based healthcare institutions—including pediatric, adult, and rehabilitation centers—participated in a one-day workshop co-hosted by the Data Sciences Institute (DSI) and T-CAIREM at the University of Toronto.

Participating institutions included:

  • Baycrest Centre
  • Holland Bloorview Kids Rehabilitation Hospital
  • Mount Sinai Hospital
  • The Hospital for Sick Children (SickKids)
  • Unity Health Toronto
  • University Health Network (UHN)
  • Women’s College Hospital

Each institution presented one successful and one unsuccessful attempt at scenario identification. Researchers performed a content analysis on these case studies to identify recurring patterns, which were categorized into four strategic pillars: Idea Generation, Prioritization, Approval, and Champions.

Key Findings: The Four Pillars of Success

1. Idea Generation and Value Proposition

Successful institutions distinguish between a broad "clinical problem" (e.g., sepsis) and a specific "scenario" (e.g., predicting sepsis during admission to a general ward). The study recommends a portfolio approach, where a wide range of ideas are initially promoted. This allows for a "fail fast" mentality, ensuring that only scenarios with the clearest value proposition and data feasibility proceed to the resource-intensive stages of development.

2. Prioritization and Strategic Alignment

A common failure mode identified was the "silo trap," where research and operations teams work on the same scenario in isolation, leading to redundant efforts. Effective prioritization requires aligning research energy with problems already identified as priorities by frontline clinicians and institutional leadership. The authors emphasize that prioritization should be strictest at the point of clinical implementation rather than during the initial exploration phase.

3. Streamlined Approval Processes

Large executive committees often lack the technical or clinical nuance required to judge the feasibility of a data science solution. The findings suggest that a small group of knowledgeable experts is more effective at adjudicating projects. This group should be capable of evaluating both the technical return on investment (ROI) and the practical clinical impact.

4. The Role of Champions

Projects frequently stall when there is no clear "owner." A successful scenario requires two types of champions:

  • Senior-level champions: To clear administrative hurdles and secure institutional buy-in.
  • Implementation champions: To manage day-to-day integration and ensure the tool meets the needs of the target users.

Recommendations for Healthcare Leaders

Based on these institutional "war stories," the authors offer four concrete recommendations:

  1. Endorsement: Require endorsement from both clinical and operational leadership during idea generation.
  2. Alignment: Ensure data science projects solve problems already prioritized by frontline staff.
  3. Expert Adjudication: Utilize small, expert-led boards to judge project feasibility and ROI.
  4. Accountability: Ensure every project has an individual or group accountable for its success, with heavy representation from end-users.

Limitations and Future Directions

This study is limited by its geographic focus on Toronto-based academic institutions. The factors influencing success in this specific ecosystem may differ in non-academic or rural settings. Future research should address the subsequent hurdles of data access, governance, and the long-term sustainability of AI tools after deployment.

Summary updated

BibTeX
@article{sung2025approaches,
  author = {Lillian Sung and Michael Brudno and Michael C. W. Caesar and Amol A. Verma and Brad Buchsbaum and Ravi Retnakaran and Vasily Giannakeas and Azadeh Kushki and Gary D. Bader and Helen Lasthiotakis and Muhammad Mamdani and Lisa Strug},
  title = {{Approaches to identify scenarios for data science implementations within healthcare settings: recommendations based on experiences at multiple academic institutions}},
  journal = {Frontiers Digit. Health},
  year = {2025},
  doi = {10.3389/fdgth.2025.1511943},
}
Figure