By Anna Mouzenian

Graphic design by Raymond Zhang

When Dr. Muhammad Mamdani reflects on what first drew him to artificial intelligence, he recalls a simple but powerful thought: Wouldn’t it be great if a computer could look at 100,000 patients and based on learnings from these patients, tell me what to do for the one patient sitting in front of me? It’s a question that resonates strongly in the emergency department (ED), where real-time decision-making is essential.

A pharmacist by training with degrees in econometric theory and statistics, Dr. Mamdani has built a career around turning this thought into reality. As Director of the Temerty Centre for Artificial Intelligence Research and Education in Medicine (T-CAIREM) and Clinical Lead – AI for Ontario Health, he is helping to redefine evidence-based medicine for the era of big data and bringing artificial intelligence (AI) out of the lab and into clinical spaces like the ED.

Dr. Mamdani’s journey in AI began when he realized that the evidence supporting conventional clinical decision making often reflects an idealized patient sample rather than the people physicians actually treat, leaving doctors to apply sample-level averages to unique individuals. With the rise of AI, the possibility to draw on large-scale clinical data to support individualised decision-making is within reach.

Few places illustrate the potential of AI more clearly than the fast-paced and complex world of emergency medicine, where even small gains in speed or accuracy can save lives. The pace of the ED, he says, demands systems that operate in real time. “In trauma care, you can’t have an algorithm that runs every hour. It has to run every second,” Dr. Mamdani says. Previously, Dr. Mamdani was the Vice President of Data Science and Advanced Analytics at Unity Health Toronto. St. Michael’s Hospital, part of Unity Health Toronto, provides an ideal setting for these AI solutions, given its commitment to advancing artificial intelligence in healthcare and its status as one of only two Level I trauma centres in Toronto.

Several AI technologies are already reshaping the clinical space. Dr. Mamdani and his multidisciplinary team have developed over 50 AI-driven tools designed to make hospitals safer and more efficient. One of the best-known is CHARTWatch, an AI tool at St. Michael’s Hospital which continuously monitors patient data such as vital signs, lab results, and nursing notes, assessing between 150 and 170 parameters every hour to flag those at risk of unexpected death or transfer to an intensive care unit (ICU).¹ Though currently used in the General Internal Medicine Ward as well as the General Surgery unit, tools like CHARTWatch hold great promise for the ED, where delayed transfer of critically injured patients to the ICU results in higher hospital length of stay and increased mortality.²

Additionally, AI scribes, now being tested in Canada and abroad, generate automated medical notes by listening to clinician–patient conversations. At Michael Garron Hospital, for example, the introduction of an AI transcribing tool enabled emergency physicians to see 10 to 13 percent more patients per shift and spend up to two fewer hours per shift on documentation, contributing to shorter wait times and reduced administrative burden.³ Tools like these could solve massive problems facing Canadian EDs, which are plagued by overcrowding, long wait times, and physician burnout.^4,5

The impact of these tools is also evident in a nurse-assignment system developed at St. Michael’s Hospital. The system reduced the time from 3 hours per day to under 15 minutes, while cutting error rates from over 20 percent to under 5 percent. Another project, ASIST-TBI⁶, which is currently deployed in the St. Michael’s ED, is an AI algorithm that analyzes CT scans in under 2 minutes to identify whether there is need for immediate surgical intervention as opposed to medical management, with accuracy comparable to that of neurosurgeons. In suspected stroke or traumatic brain injury, delays in identifying brain bleeds may mean the difference between recovery and death.⁷ Such rapid and reliable tools represent a transformational shift in care.

Though in the early phases of deployment, technologies like these could elevate emergency care by enabling AI-assisted triage, flagging urgent cases, or ordering imaging before a physician even arrives. These innovations succeed, he says, because clinicians help build them. Instead of delivering ready-made software, his group begins by asking front-line teams, like ED physicians, what problems they most want solved. “Everybody buys in because we built the solution together,” he says.

As AI tools become increasingly prevalent in clinical care, they introduce ethical tensions that are especially acute in the ED, where decisions must be made quickly, and often without complete information. The reach and ability of these tools necessitate responsibility, and Dr. Mamdani is candid about the gray zones that accompany medical AI. One algorithm designed by his team, which identifies structurally vulnerable people who use intravenous drugs and are at high risk of death, has never been deployed despite achieving over 80 percent accuracy. Patient interviews revealed discomfort with being flagged without consent and concerns about the stigma associated with addiction, underscoring the importance of trust and patient autonomy in AI-supported care.

Bias in AI decision-making presents another major challenge. Race data is rarely collected in Canadian hospitals, but omitting it limits the ability to meaningfully address bias. “When we ask for race information, some find it offensive,” Dr. Mamdani explains, adding that “when we don’t [ask for that information], others say it’s unethical. You can’t have it both ways.”

Clinical AI presents a difficult tradeoff between ethical idealism and lives saved. CHARTWatch, which does not include race data and has faced criticism for this limitation has nevertheless been associated with a 26% reduction in unexpected in-hospital mortality.¹ He asks, “If a tool saves lives, is it ethical not to use it while we wait for perfect data?” At Unity Health, ensuring the responsible use of AI is a central priority. Projects undergo ethics review and bias assessment, and then enter a “silent testing” phase, during which models run without influencing patient care. CHARTWatch underwent nine months of silent testing before its official rollout. Rigorous evaluation and ongoing performance monitoring are essential to minimizing bias and ensuring responsible AI use in clinical practice. These frameworks reflect a serious commitment to ethical use of AI in clinical care and enable the safe, responsible integration of AI tools into high-stakes clinical environments like the ED.  

Dr. Mamdani emphasizes that training the next generation of AI researchers is critical for its continued success. For students eager to enter the field, Dr. Mamdani offers practical advice: identify what you enjoy and what you excel at and let that guide your path. “If you love math and you’re good at it, learn to code. If you’re more interested in people and processes, focus on change management or ethics. There’s a role for everyone in AI.” Dr. Mamdani’s vision is both ambitious and humanistic. His work reflects a future where AI strengthens, rather than replaces, the human connection at the heart of medicine. It’s a vision that holds promise for the ED, reshaping care by equipping clinicians with real-time insights when they matter most.

References: 

1. Verma AA, Stukel TA, Colacci M, et al. Clinical evaluation of a machine learning–based early warning system for patient deterioration. CMAJ. 2024 Sept 16;196(30):E1027–37.
2. Gregory CJ, Marcin JP. Golden hours wasted: The human cost of intensive care unit and emergency department inefficiency*. Crit Care Med. 2007 June;35(6):1614.
3. Michael Garron Hospital. MGH’s Stavro Emergency Department adopts AI transcribing tool to reduce patient wait times and address physician burnout [Internet]. Toronto (ON): Michael Garron Hospital; 2024 Nov 6 [cited 2025 Dec 16]. Available from: https://www.tehn.ca/about-us/newsroom/mghs-stavro-emergency-department-adopts-ai-transcribing-tool-reduce-patient-wait
4. de Wit K, Tran A, Clayton N, et al. A Longitudinal Survey on Canadian Emergency Physician Burnout. Ann Emerg Med. 2024 June;83(6):576–84.
5. Emergency Department Overcrowding in Canada: CADTH Health Technology Review Recommendation [Internet]. Ottawa (ON): Canadian Agency for Drugs and Technologies in Health; 2023 [cited 2025 Dec 16]. (CADTH Health Technology Review). Available from: http://www.ncbi.nlm.nih.gov/books/NBK599980/
6. Smith CW, Malhotra AK, Hammill C, et al. Vision Transformer-based Decision Support for Neurosurgical Intervention in Acute Traumatic Brain Injury: Automated Surgical Intervention Support Tool. Radiol Artif Intell. 2024 Mar;6(2):e230088.
7. Kwon H, Kim YJ, Lee JH, et al. Incidence and outcomes of delayed intracranial hemorrhage: a population-based cohort study. Sci Rep. 2024 Aug 22;14(1):19502.