During my father’s cancer treatment, I spent months taking him to chemotherapy, radiation, and medical appointments whenever I could return to Canada from overseas.
What I witnessed was organized chaos.
Doctors gave verbal instructions to a man with dementia. Those instructions would be forgotten. Written explanations were rarely provided during the encounters I attended, and we were not given a consistent, usable way to follow his care.
He underwent tests for lung cancer and other conditions, yet the results were not routinely placed in our hands. Finding the information and ensuring the next person had it became another responsibility for a family already dealing with serious illness.
At one hospital in Longueuil, Quebec, I was handed a folder of his medical documents to carry from one doctor to another on a different floor.
I had become part of the hospital’s information infrastructure.
Patients and families were expected to connect the pieces precisely when they were least equipped to do so.
Canada has an opportunity to change this. It could build healthcare around a continuous, accessible patient record, supported by AI that helps patients, families, and professionals act on the information it contains. If we can demonstrate that this improves care at home, we could also create an important Canadian export.
The medical record should follow the patient
It would be inaccurate to say Quebec has no digital health infrastructure.
The Dossier santé Québec, or DSQ, allows authorized professionals to access shared health information. It shares laboratory results, medical imaging results, and certain medication information. But Quebec’s own explanation explains that only some information from local records is automatically shared. The provincial record is not necessarily a complete account of a patient’s care.
Quebec is also modernizing. Santé Québec describes its Dossier santé numérique and associated patient service, Mon dossier santé, as bringing together information, results, appointments, and communication. Santé Québec currently lists the Dossier santé numérique as a pilot in the Nord-de-l’Île-de-Montréal and Mauricie–Centre-du-Québec regions, where Mon dossier santé is also offered.
A family should not have to know which database contains a result. Relevant information should reach the authorized people responsible for care, wherever that care happens.
During my father’s treatment, teams in different hospitals repeatedly appeared unaware of one another’s work. I could not tell whether information was missing, inaccessible, incompatible, or simply unread.
From the patient’s perspective, the result was the same: continuity depended on someone manually reconstructing it.
Estonia offers a useful reference. Its national health information system has operated since 2008, enabling providers to exchange information across institutions. Through its Health Portal, patients can review submitted medical documents, prescriptions, and referrals, designate representatives, and check who has accessed their information. Digital identification provides a secure means of access.
Estonia has not solved every clinical problem. But it demonstrates a principle Canada should treat as basic infrastructure: care should remain connected across providers.
Every patient should have access to a dedicated AI assistant
An AI assistant could make that record useful in daily life. Consider the questions a family caring for someone with dementia needs answered:
When is the next appointment? Where is it? What preparation is required? What changed at the last consultation? Has a referral been accepted? Who is responsible for following up?
With my father’s permission, or the appropriate authority to act on his behalf, I should have been able to ask those questions from overseas and receive answers drawn from his current care plan.
Such an assistant should retrieve authorized records as needed, identify the source and date of its answers, and acknowledge missing or conflicting information.
It could send reminders by text, email, or voice call, notify an authorized caregiver when an appointment remains unconfirmed, and help reschedule before a slot is lost.
After consultations, it could help prepare plain-language summaries for clinical review and share them with the patient and authorized family. Printed copies, telephone support, and caregiver access would remain essential. The system must adapt to the patient’s abilities.
In my father’s case, appointments were missed. A process that relies on someone remembering instructions must account for whether that person can reliably remember them.
The financial case can be measured
A Cochrane review of eight randomized trials published through August 2012 found moderate-quality evidence that text reminders improved attendance compared with no reminders. AI would need to demonstrate additional value through conversation, coordination, and caregiver follow-up.
For illustration, suppose a service schedules one million appointments annually. If an intervention reduces missed appointments from 10% to 7%, that represents 30,000 additional attended appointments. At an assumed recoverable capacity value of C$150 each, the benefit would be C$4.5 million annually before implementation and operating costs.
These are hypothetical inputs, not measured Canadian savings. Recovered capacity may mean more patients treated or shorter waits rather than money returned to the treasury.
Hospitals could similarly measure time saved locating records and clarifying requests, accounting for implementation, errors, and oversight.
A call button should communicate more than a light
When I returned after receiving the call in late September that my father was terminally ill and in hospital, the failures became more intimate.
He shared a room with another seriously ill patient. This patient was frequently distressed and shouted throughout the day. From what I witnessed, he struggled to understand his condition and what was happening around him.
Nurses tried to explain. Their time was limited, and other patients needed them.
My father also asked questions repeatedly. An answer could soothe him, but only briefly before the question returned.
There was another detail that seemed remarkably ordinary to everyone around us: the call button.
A patient pressed it, a light appeared outside the room, and an audible alert sounded at the other end. I watched staff enter and ask which patient had called. They then had to find out why, and sometimes leave to locate the appropriate colleague or equipment.
The request carried almost no useful context.
A better system could identify the bed immediately and allow the patient to communicate a need through speech or simple choices: help getting to the washroom, discomfort, a question, or urgent assistance.
Where appropriate, AI could turn a spoken request into a concise message for staff. The system could record whether someone acknowledged it and escalate an unanswered request according to clinical rules.
The existing emergency call function would remain available. A patient should never have to complete a conversation with a chatbot to summon help, and an algorithm should never independently dismiss a report of pain.
Bed identification is straightforward engineering; interpreting a spoken request is where AI could help. Both should save staff from having to discover information that could have accompanied the call.
Reassurance should be available when the question returns
A carefully designed conversational assistant might repeat an explanation approved by the care team, help a patient contact family, or relay a concern to a nurse. It could respond patiently when the same question returns, as it did so often with my father.
For patients with dementia, careful testing is essential: false reassurance could deepen confusion, and distress must reach a human.
I cannot claim such a tool would have comforted my father. I wish we had had a safe, well-tested option to try.
Patients should be able to decline it. Making support more available must never become a justification for reducing human contact.
Monitoring should connect the numbers to the person
A patient cannot always explain what is wrong. AI could also help care teams recognize changes the patient cannot communicate.
Where monitoring is clinically appropriate, software could combine heart rate, breathing rate, oxygen saturation, and blood pressure with laboratory results and relevant medical history. Trends could be assessed alongside the person’s usual readings, diagnoses, recent treatment, and medications. This requires a clinical system validated for its intended use; connecting a chatbot to a monitor is not enough.
Rising breathing rates alongside falling oxygen levels could prompt an assessment. An alert should show what changed and when; identifying the cause would still require clinical judgment and bedside observations.
Canada already has experience here. At Unity Health Toronto, CHARTwatch uses information such as test results and vital signs to help predict the level of support a patient will need.
Evidence from trials is promising but mixed. A 2024 randomized trial involving 15,965 patients tested AI analysis of electrocardiograms with alerts to physicians about high-risk patients. Mortality within 90 days was 3.6% with the intervention versus 4.3% with usual care. The finding applies to that ECG-based intervention, not all AI monitoring.
A cluster-randomized trial published in 2026, involving 10,422 inpatient visits tested passive risk displays combining continuous cardiorespiratory monitoring with other clinical measurements, without active alerts or a mandated response. It found no significant improvement in the primary outcome of time free from clinical deterioration. Patient movement between groups complicated interpretation. A prediction on a screen does not guarantee better care.
Alerts must reach an accountable clinician, with systems tested for false alarms and missed deterioration across patient groups. An assistant could explain the clinician-reviewed assessment to authorized family; diagnosis and treatment remain with the care team.
What I wanted for my father was a system that noticed meaningful changes, brought his history to the bedside, and helped someone respond. Near the end of life, the purpose of that response must be guided by comfort and the person’s wishes.
Dignity at the end of life must be part of the design
On my second day back in Quebec, beside my dad in what would become his final hospital bed, I was told he would be added to a palliative care waiting list. He had been given no more than three months to live, and the focus had shifted to end-of-life care.
More than two weeks later, I was told he had been accepted. Then I was told he was not in enough pain to be moved.
He was not connected to monitoring equipment. That unsettled me, although I understand that assessing comfort at the end of life does not necessarily require machines. My concern was more fundamental: how was his suffering being assessed when dementia made it difficult for him to explain what he was experiencing?
I needed someone to explain the decision: what had been assessed, why he did not qualify for a transfer, what alternatives existed, and who would take responsibility for the next step.
Instead, the explanations felt like excuses. From his bedside, it seemed that moving him had become more trouble than the system was willing to accept. I was left convinced that there was too little palliative care capacity, and that families like ours were being left to bear the consequences.
My father entered the hospital on September 22, 2026. Only on October 12, 2026, after endless back-and-forth about finding a palliative care home, was I offered the option of moving him back to his residence. By then, he was unresponsive and unable to move. I felt it was too late to put him through another transfer. I declined.
My father died in hospital two days later, on October 14, 2026. The phone call I had been expecting finally came. His suffering was over.
Palliative care encompasses physical, psychological, social, and spiritual suffering. It is broader than pain intensity or admission to a particular room. A 2023 report from the Canadian Institute for Health Information also identifies persistent barriers to access that vary with age, location, and diagnosis.
More measurement is not automatically better care. UK NICE guidance on care in the last days of life recommends avoiding investigations unlikely to change care and using validated behavioural pain assessments when someone cannot explain their pain, including because of dementia. AI should support that assessment and the agreed care plan.
Technology cannot create a staffed palliative bed. But it could make referrals, reassessments, and alternatives visible, exposing unmet need and giving families a clear account of what was actually arranged. What I do know is that I spent weeks passing along assurances to my father and his sibling that better care was coming. Those assurances became promises I could not keep. A family should not be put in that position.
Being better than the United States is an insufficient ambition
Canadians often defend our healthcare system by comparing it with the United States.
The life-expectancy advantage is real. For 2024, Statistics Canada’s preliminary estimates put life expectancy at birth at 80.03 years for males and 84.29 years for females. The corresponding U.S. figures were 76.5 and 81.4 years—Canadian advantages of approximately 3.5 and 2.9 years.
Life expectancy reflects many influences beyond healthcare delivery. It cannot tell us whether a confused patient understood an instruction, whether a family could locate a test result, or whether someone died with sufficient comfort and privacy.
A favourable comparison should strengthen our ambition, not become an excuse to accept preventable dysfunction. The American system has serious shortcomings of its own. Preliminary U.S. government estimates for 2025 put the number of people uninsured at the time of interview at 28 million, or 8.3% of the population.
Yet we can learn from specific American approaches to more personalized, coordinated care. The Comprehensive Primary Care Plus model emphasized continuity, care management, and patient and caregiver involvement. A five-year evaluation found a roughly 2% greater mean annual reduction in emergency department visits relative to comparison practices over five years. It did not produce overall Medicare savings once the extra program payments were included.
We can also learn from hospitals in Scandinavia that treat food as part of care. In a 2026 Danish hospital study, expanding breakfast options with familiar, energy- and protein-rich foods increased daily intake by an average of 198 calories and 6.8 grams of protein in the main analysis of 71 patients. That small study measured intake, not recovery or the quality of food across Scandinavia. It still offers a practical lesson in listening to patients.
The food my father received felt like an afterthought. Much of it was, in my view, embarrassing to serve. A meal is one of the most ordinary ways to show someone they are being cared for. It should not be treated as incidental.
We should be willing to learn from anyone who has found a better way.
Think from first principles, then build for compounding progress
Canada needs a larger ambition than digitizing the processes that failed my father.
First-principles thinking starts with what the patient needs and asks which existing arrangements actually serve that need. My father needed people to know his history, understand his condition, explain what was happening, and respond when he needed help. None of those needs required his family to act as a courier, a memory aid, or a switchboard.
If we designed that journey today, would we make a man with dementia responsible for remembering the next step? Those are design choices we can change.
AI could connect these functions into personalized support across the entire journey, including the long intervals between appointments. That is a much larger ambition than adding a chatbot to each hospital’s website.
This is where exponential thinking becomes useful. A reliable record could improve an assistant’s answers. Better follow-up could keep the care plan current. Clearer requests could help staff respond sooner. Each improvement could make the others more useful, while a validated software improvement could be deployed across participating hospitals without every institution rebuilding it independently.
The law of accelerating returns and debates about the exponential trajectory of AI development encourage us to think beyond today’s capabilities. They do not guarantee exponential improvements in health. But they give us a reason to build infrastructure that can incorporate better tools as they prove themselves.
For me, the moonshot ambition is concrete: every patient has continuity of information and support, and every care team can act on it. Pursuing that ambition requires leadership willing to redesign workflows, fund implementation, and give clinicians, patients, and Canadian builders the authority to solve problems together.
Canada could build something worth exporting
Canada’s AI ambitions and its relationship with Europe create an opening.
Canada–EU trade in goods and services reached C$178.6 billion in 2025, and bilateral merchandise trade increased by more than 77% between 2016 and 2025, according to Global Affairs Canada. The relationship also includes cooperation on AI and digital infrastructure.
Canada’s 2026 national AI strategy explicitly proposes using government purchasing to help domestic companies establish the revenue and validation needed to export. It also announces an AI Missions Program beginning with C$200 million toward improving health outcomes. These are policy commitments, rather than evidence of results already achieved.
Canadian companies could build and validate a package combining connected records, patient assistants, caregiver access, appointment coordination, staff communication, and clinically validated monitoring tools.
A turnkey offering would still need to adapt to local languages, clinical practices, laws, and existing infrastructure. Countries would retain control over their data and public services.
Europe already has substantial expertise and its own agenda. The European Health Data Space regulation entered into force in March 2025, with major requirements applying in stages from 2029 and 2031. Canadian products would need to fit that direction and earn their place through demonstrated value.
The export would combine proven software, implementation expertise, training, and evidence that care improves.
Start with one complete patient journey
Canada should begin with a bounded but meaningful challenge: connect the full journey of a patient receiving cancer treatment across a participating network.
Include referrals, diagnostics, treatment appointments, instructions, caregiver permissions, symptom reporting, clinically appropriate monitoring, and transitions into palliative care.
Build it with patients, nurses, physicians, support staff, and families. Give Canadian startups a realistic route into procurement. Require participating products to exchange information and allow providers to change suppliers.
Measure missed appointments, time spent chasing records, response times, patient understanding, caregiver burden, safety, and total costs. Publish the results, including failures.
Then expand what works. Shared standards would let each network build on the last, turning one patient journey into a route to national transformation.
The conservative-versus-liberal debate matters, but I believe its usual framing is too narrow: conservatives emphasize spending restraint and efficiency; liberals emphasize public investment and expanded services. Those are broad political instincts, not a description of every government. My concern is the cycle they can produce: cuts that strain care, followed by spending that restores capacity without fixing the underlying processes.
We should demand more from both. Conservatives should have to explain how efficiency will improve care, not simply reduce budgets. Liberals should have to explain how additional spending will change outcomes, not simply fund more of the same. AI-enabled redesign offers a shared ambition: more effective care from the resources we commit.
We can preserve public healthcare while demanding a much better operating system.
I watched nurses and doctors work hard inside arrangements that made their work harder. I watched my father depend on explanations he could not retain and on a family that could not always be physically present.
I cannot know how much technology would have changed his final weeks. I know what I wanted: information that followed him, instructions we could revisit, requests that reached the right person, and decisions we could understand.
Canada should be capable of providing that. AI gives us tools to pursue it at scale. First-principles thinking helps us redesign the system around that promise. We need leaders willing to carry it through.
If we build it well enough for our own families, we may also build something the world wants.
My father cannot benefit from that future. Other families still can.
Credit: Source link
