Ontario Audit Finds AI Doctor Note Tools Miss Key Facts
Auditors from Canada's Office of the Auditor General of Ontario examined 20 AI systems approved for use by healthcare providers in the province. These tools, known as AI Scribes, aim to generate patient notes from doctor-patient interactions. The review revealed frequent errors, including missed details, incorrect data, and invented content not present in the original recordings.
The audit forms part of a broader examination of AI use in Ontario's public services. It focuses on the AI Scribe program, started by the Ontario Ministry of Health for physicians, nurse practitioners, and other health professionals. Officials tested the systems during procurement with simulated doctor-patient audio. Medical experts then compared the recordings to the AI-produced notes for accuracy checks.
Widespread Errors in AI-Generated Notes
Reviewers found serious issues across the systems. Nine out of 20 AI tools created information and suggested treatment plans not covered in the recordings. Examples included claims of no masses detected or patient anxiety, none of which appeared in the simulated consultations.
Twelve of the 20 systems added wrong drug details to notes, affecting 60 percent of those tested. Seventeen systems failed to capture key mental health details from the discussions. The report specifies that six systems omitted mental health issues entirely or partially, or skipped vital information.
OntarioMD, which supports doctors in adopting technologies and helped with AI Scribe procurement, advises manual reviews of AI notes. However, none of the approved systems require doctors to confirm accuracy through a mandatory feature.
Flaws in the Evaluation Process
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
The audit points to problems in how officials scored the AI systems. Domestic presence in Ontario accounted for 30 percent of the total score. Medical note accuracy weighed only 4 percent.
Bias controls made up 2 percent, while threat, risk, and privacy assessments added another 2 percent. SOC 2 Type 2 compliance contributed 4 percent. Overall, factors like accuracy, bias, security, and privacy formed a minor part of the scoring.
The report warns that such scoring could lead to choosing vendors with tools that generate wrong or biased records, or fail to protect health data properly.
Context and Official Response
Errors in AI medical tools surprise few, given past reports on consumer AI providing faulty health advice. Studies show large language models struggle with diagnoses in about 80 percent of cases. These AI Scribes target professionals, so the performance raises questions.
The Office of the Auditor General of Ontario conducts independent audits of provincial government operations to ensure accountability. OntarioMD, part of the Ontario Medical Association, assists over 40,000 physicians with electronic records and tech integration since 2002. The AI Scribe program seeks to ease administrative burdens amid doctor shortages.
The Register contacted the Ontario Health Ministry for comment on the audit and plans to follow its suggestions. No immediate reply came. A ministry spokesperson told CBC that more than 5,000 physicians use the program, with no reported patient harms linked to the technology.