Healthcare Has a Design Problem
In 2018, a review in Science Translational Medicine noted in passing that the average American adult visits a health care provider fewer than four times a year. The authors moved on. It is worth stopping there.
Four appointments is perhaps three hours. A year contains about eight thousand waking ones. Whatever is happening to a person’s health is happening in the other seven thousand nine hundred and ninety-seven, and the health system has no view of any of it.
That blind spot is not neglect. It is culture, set into structure. Medicine was built to identify sickness and treat it, and by that standard it is one of the most successful things our species has ever organised. But an institution whose entire culture forms around the moment something goes wrong will struggle with a person going wrong slowly, and has nothing at all to say to a person who is currently fine.
Consumer digital health was supposed to fill that space. Fifteen years in, it has mostly added to the confusion rather than resolved it. This essay is about why, and about what the job actually is.
A system designed around sickness
Call it a design and the word sounds accidental. It was not. Every layer of the system independently rewards the same thing.
The clearest evidence is where the money sits. In the United States, 90% of the country’s $5.3 trillion in annual health expenditure goes to people with chronic and mental health conditions, by the CDC’s own accounting. That is not damning on its own. People with chronic conditions need care and should get it. The question is about the other end of the timeline, the twenty or thirty years during which those conditions were forming, were cheaper to influence, and were invisible to everyone holding a budget.
Canfell and colleagues describe the result as a break-fix model standing in for a predict-prevent one, and identify why it persists: acute activity is easy to measure, and therefore easy to fund (Canfell et al., 2022). Four mechanisms hold it in place, each reinforcing the others.
| Layer | What it rewards | Effect |
|---|---|---|
| Payment | Treatment episodes, which can be coded and billed | Prevention produces no billable event |
| Data | Whatever is captured inside a clinical encounter | The well person generates no record at all |
| Measurement | What instruments in buildings can read | The ~70% of outcome variance outside them goes uncounted |
| Culture | Intervention as the definition of care | Staying well is nobody’s clinical responsibility |
The measurement row is the one to sit with. Social, environmental and behavioural determinants explain roughly 70% of the variance in health outcomes, and there is no contemporaneous, population-scale way to capture any of them (Canfell et al., 2022). Treat that figure as an order of magnitude rather than a precise quantity, since estimates in this literature run from about two-thirds to about 90% depending on method. The direction is not disputed. Most of what determines health happens where nothing is recorded.
A system funds what it can count, and it can only count the part that happens after.
Where these figures come from, and where they do not apply
Both numbers above are American, from a country with insurance and third-party reimbursement. The mechanism differs elsewhere. The conclusion does not.
In Nigeria, 72% of current health expenditure was paid directly by households in 2023, according to the WHO Global Health Expenditure Database. Where care is bought at the point of use, the reimbursement argument has little purchase, and something harsher takes its place. A visit is not a copayment but a real sum weighed against school fees, rent and food. The rational response to a vague symptom is to wait and see. People go less often, and they go later.
So the distance between what medicine knows and what reaches a life is not a peculiarly American problem. It widens as the financial cushion thins, which means it is widest exactly where the chronic-disease burden is climbing fastest.
How the burden changed shape
The reactive design was, for most of history, the only design available.
Before modern microbiology, societies handled outbreaks through observation, quarantine and sanitation, without understanding pathogens mechanistically. Prevention was practised, and it worked, but it was empirical rather than explained. It also operated on cities rather than on people. You cannot offer drainage to one individual.
The nineteenth century changed the method. Sanitary reform and germ theory moved disease control away from broad mandates toward precise biomedical intervention. Once a cause could be named it could be targeted, and medicine spent the following century becoming very good at targeting. That is the lineage the modern hospital comes from, and it is why a hospital is organised the way it is: an arriving problem, a diagnosis, an intervention, a discharge.
Progress was never purely technical, which is the part usually left out. Epidemics reshaped governance and institutions, and biomedical advances landed unevenly depending on infrastructure and political will. Public health reform and behaviour change drove the decline of infectious disease as much as clinical medicine did.
Then the burden underneath the system changed shape. Across the twentieth century, chronic disease became the organising problem, and it was shaped by politics and administration as much as by biology. Epidemiological transition theory named the pattern: societies moving from infectious-disease dominance toward chronic disease. The framework has been revised and criticised repeatedly for oversimplifying, and the honest contemporary picture is messier still. Many countries now carry a double burden, chronic disease rising while infectious threats persist. Nepal is one of the clean illustrations, shifting from tuberculosis, diarrhoea and malaria toward cardiovascular disease, cancer and diabetes, tracking an ageing population and diets higher in fat, sugar and salt. Nigeria, meanwhile, continues to grapple with longstanding infectious diseases such as malaria, cholera while also facing a growing burden of cancer.
This is the mismatch at the centre of everything else. The conditions that now end or shorten most lives do not arrive in an afternoon:
- Cardiovascular disease accumulates across decades of blood pressure, lipids and inactivity.
- Type 2 diabetes is preceded by years of measurable, reversible metabolic drift.
- Most cancers develop over a long period before anything is detectable, let alone symptomatic.
A hospital is organised around an event. A chronic condition is not an event. It is a gradient, and there is no moment on a gradient at which somebody presents to reception.
What moves that burden is not a mystery. Nutrition, sleep, movement and mental wellbeing carry most of it, and they interact rather than acting separately: sleep changes appetite, appetite changes activity, activity improves sleep. The effects are not small. In a UK Biobank cohort of 59,078 adults, the most favourable combination of sleep, activity and diet was associated with about nine more years of life, and nine more lived free of disease, than the least favourable (Koemel et al., 2026). Read that as odds across a population rather than a promise to anyone, given a median age of 64 and volunteers healthier than the country they came from. The more useful number is at the other end of the scale: fifteen minutes more sleep, under two minutes more activity and half a serving of vegetables a day tracked with 10% lower all-cause mortality (Stamatakis et al., 2025), and the evidence for movement and sleep runs to mental health as much as to physical (Firth et al., 2020).
The part nobody watches
Between well and ill there is a long, quiet interval that no one observes.
The same 2018 review contains the most useful sentence in it: disease often begins with a period of subclinical decline before progressing to symptoms that lead a person to seek medical care (Gambhir et al., 2018).
Read that as a design statement rather than a clinical one. The window in which a problem is most tractable is the window in which nobody knows there is a problem. You cannot book an appointment for early-stage nothing-in-particular. There is no diagnostic code for it, no specialty that owns it, and no reason for anyone to look. The years in which small changes still compound pass unobserved, and by the time the system has a name for what is happening, the cheap options have closed.
Consumer digital health was supposed to close this
It is the obvious candidate. The sensors are already in people’s pockets. The gap is continuous and personal, and software is continuous and personal. On paper the fit is exact.
The premise was that visibility would be enough: show people their own numbers and they will act. A great many people now have a great many numbers, and the premise has not held.
The reasons are fairly legible once stated.
A metric is not a decision. A resting heart rate of 64 does not tell anyone what to do this afternoon. The translation from measurement into action is the difficult part of the problem, and it was handed to the user. Gambhir and colleagues said as much in the same 2018 review: monitoring technologies must prioritise the collection of actionable data and long-term engagement. That priority is still largely unmet.
Statistical risk resists personal interpretation. Being placed in an elevated-risk group is a true statement about a population and a nearly unusable one about a Tuesday. When a number contradicts how somebody feels, the number usually loses, and dismissing it is not irrational so much as a reasonable response to information that cannot be acted on.
Generic advice was already freely available. Nobody is unaware that sleep and exercise matter. Awareness was never the binding constraint, so a product whose output is awareness changes nothing.
More information is not less ambiguity. This is the failure that matters most, because it is the opposite of the intended effect. A person who previously had a vague sense that they should sleep more now has a score, a trend line, a percentile and no idea which of them is important. Offloading interpretation onto the user does not resolve uncertainty. It manufactures it.
The field’s own assessment is appropriately modest. Reviewing consumer health informatics for precision prevention, Canfell and colleagues report mixed evidence for whether these tools actually improve prevention-related health outcomes (Canfell et al., 2025). A 2026 review of precision prevention in BMC Medicine reaches a compatible conclusion from another direction: the remaining barriers are not principally technological, and the models for discovery and implementation have not kept pace with the tools (Tsakiroglou et al., 2026).
The sensors got good. The layer that decides what to do with them did not get built.
What it should do instead
The useful move is to treat each gap as a specification rather than a complaint.
| The gap | What the design has to do |
|---|---|
| Data centred on disease | Keep the well person visible between visits, continuously rather than at points |
| Much data, little meaning | Produce direction, not more metrics: one specific thing, with a time attached |
| Weak personalisation | Fit the actual person, because general advice has already been tried and has already failed |
| Motivation without structure | Build for repetition in a stable context until an action stops requiring a decision |
| Engagement without support | Combine monitoring with guidance, structure and, where it matters, a human |
| Equity and trust as afterthoughts | Design for the people furthest from care first, and treat privacy as a precondition |
Two of those deserve expanding, because they are where most products go wrong.
Personalisation is the mechanism, not a feature. Generic health messaging fails to change behaviour with a consistency that ought to have settled the argument. A recommendation that does not fit the life it is given to will not survive contact with a Tuesday afternoon. The finding that very small combined changes track with measurable benefit is what makes this tractable: the correct output is not a programme, it is one small finishable thing that fits.
The goal is participation, not compliance. A meta-analysis of 138 studies in the Journal of Clinical Nursing found a moderate association between health literacy and self-care behaviours, r = 0.29 (Magi et al., 2026). The same paper reports that 62 of those studies carried a high risk of bias and that certainty of evidence ranged from very low to moderate, which is a fair description of this whole field and a reason to state the finding at exactly its strength. The direction still points somewhere useful. People who understand their own health act on it more, which means the job is building understanding rather than delivering data.
And the last row is not decoration. If the distance between medical knowledge and an ordinary life grows as out-of-pocket spending grows, then a design assuming a wearable, an uncapped data plan and a doctor to escalate to has solved the easiest version of the problem and left the rest.
The actual job
Healthcare became scientifically powerful without becoming structurally preventive. That is a statement about culture before it is one about budgets: a system whose idea of care is intervention will keep producing intervention, and will keep treating the well person as somebody with no clinical business being there.
Consumer digital health inherited the space that leaves. Its strongest role is not to add another layer of ambiguity to a picture that is already confusing, which is broadly what the last fifteen years produced. It is to make prevention continuous, personal and actionable: to keep somebody visible to themselves between the appointments nobody is booking, and to turn what that reveals into one small thing worth doing today.