How common is patient harm in Canadian hospitals?
In 2025 – 2026, 1 in 17 acute care hospital stays in Canada involved at least one harmful event: about 155,000 of 2.65 million stays (CIHI). Many of these events are preventable, and the cost is high in lives and in dollars.
| Measure | Canadian figure | Source |
|---|---|---|
| Hospital stays with harm, 2025–2026 | 1 in 17 (about 155,000) | CIHI |
| Average cost of a stay with harm vs without | $49,887 vs $11,157 | CIHI |
| Adverse events per 100 admissions | 7.5 | Canadian Adverse Events Study |
| Adverse events judged preventable | 36.9% | Canadian Adverse Events Study |
| Patients with an adverse event who died | 20.8% | Canadian Adverse Events Study |
| Preventable deaths from adverse events, estimated per year | 9,250 to 23,750 | Canadian Adverse Events Study |
Where harm happens in Canadian hospitals
The six most common harmful events in Canadian hospitals are electrolyte and fluid imbalance, urinary tract infections, delirium, pneumonia, aspiration pneumonitis and post-procedure infections (CIHI). One in 4 hospital stays with harm involved more than one harmful event. Most of these events build over hours or days, which leaves a window to catch them early.
How AI can help reduce harm in Canadian hospitals
AI is already lowering harm in hospitals by spotting risk earlier than people can on their own. The strongest Canadian evidence comes from Toronto.
| Use | What the AI does | Evidence |
|---|---|---|
| Early warning of deterioration | Reads vital signs and lab results every hour and alerts the care team when a patient is at risk of dying or needing ICU care | CHARTwatch at St. Michael’s Hospital, Toronto: 26% fewer unexpected deaths on the internal medicine ward (CMAJ, 2024) |
| Medication safety | Flags prescription orders that look unusual for that patient, for a pharmacist to review | A machine learning tool caught 74% of orders that needed pharmacist intervention in a single-hospital study (JAMIA, 2020) |
| Less documentation burden | AI scribes draft clinical notes from the visit conversation | In an OntarioMD evaluation of 152 family physicians and NPs: 70% less time on documentation, up to 4 hours a week freed |
Fewer missed warning signs and more time with patients both target the harms in the CIHI list above.
How can hospitals use AI safely?
AI supports clinical judgment; it does not replace it. Each tool needs testing on local patients, clear rules for who responds to an alert, and ongoing checks for missed cases and bias. Clinicians stay responsible for reviewing everyAI-drafted note or flagged order before it reaches the chart.
References
- Baker GR, Norton PG, Flintoft V, et al. The Canadian Adverse Events Study: the incidence of adverse events among hospital patients in Canada. CMAJ. 2004;170(11):1678–1686. doi:10.1503/cmaj.1040498
- Canadian Institute for Health Information. Patient harm in Canadian hospitals? It does happen. 2026. cihi.ca
- Healthcare IT News. Machine learning-based CDS can accurately intercept Rx errors, study finds. October 1, 2020. Reporting on a study in the Journal of the American Medical Informatics Association. healthcareitnews.com
- OntarioMD. OntarioMD highlights AI scribes. Canadian Healthcare Technology, November 1, 2024. ontariomd.ca
- University of Toronto. AI tool reduces risk of unexpected hospital deaths by 26 per cent: study. September 2024. Reporting on Verma AA, et al., CMAJ, 2024. utoronto.ca