Seeing the Forest and Missing the Trees

Before this moment, Bryan had already seen three versions of how the same problem had been approached, each shaped more by circumstance than design. First, he had encountered a world where records were held firmly within institutions, stored with care but rarely moving with the patient, creating a system where information existed yet could be frustratingly distant from the point it was needed. He then witnessed a quiet shift, where records began to travel, not as part of a grand plan but because it made practical sense; maternity notes, personal child health records, and health passports simply followed the person, accumulating a continuous story across different settings.

Finally, he observed how this approach placed continuity quite literally into the hands of patients themselves, who carried their own histories from one encounter to the next. It was not a perfect solution, but it had a certain coherence. Information, for a time, moved with care because it recognised where care actually happened.

The next time jump for Bryan was less like travel and more like someone had quietly replaced the universe while nobody was looking directly at it, which, in fairness, is still the most reliable way most large information systems evolve when everyone is busy assuring everyone else that nothing fundamental is changing.

He arrived in what was recognisably the NHS again, although “recognisably” now felt like the kind of recognition you get when you return to a house you once lived in and discover that all the furniture has been replaced by dashboards explaining where the furniture used to be. The buildings were familiar. The people were familiar. The biscuits were still reassuringly indifferent to technological progress. But the information had developed a new habit of arriving in pieces that only made sense if you already knew where the rest of it had gone.

Screens had become architectural features. Not installed so much as grown, like a new layer of institutional coral. They displayed information with calm assurance, each one convinced it had the most accurate version of reality, or at least the most recently updated one that could be made to fit on a spreadsheet without complaint.

Bryan recognised the familiar names. GP systems. Hospital systems. Radiology systems. Pathology systems. Pharmacy systems. Each one competent, each one internally consistent, each one quietly behaving as though it was the main custodian of truth.

The difficulty, as he would soon discover, was that truth had been politely divided into departments.

A GP moved through a consultation, opening records with the speed of long practice, pausing occasionally at entries that looked correct but slightly unfamiliar, like handwriting from a previous life. A hospital clinician did the same, but in a different system, with different assumptions about what mattered. A pharmacist checked medication histories that were precise, detailed, and occasionally unaware that the patient had recently been somewhere else entirely.

Nothing appeared broken. This was, in fact, the problem.

The first hint of difficulty arrived in the form of a simple question.

“What medication are you currently taking?”

It was the kind of question that once lived comfortably in the space between memory and paper. Now it arrived in a world where memory had been distributed and paper had been digitised into several competing interpretations.

The patient answered. The answer matched one record, partially matched another, and politely disagreed with a third, depending on how one treated timing, dosage amendments, and the occasional enthusiastic attempt at reconciliation by a previous clinician who had clicked “update” with optimism rather than certainty.

Bryan began to see the shape of something that had not so much gone wrong as gone sideways with good intentions.

Bryan recognised the pattern gradually. In the earlier systems he had observed, the information had followed the patient as a single, travelling account. It was incomplete at times, occasionally untidy, but fundamentally unified.

What had changed, though not all at once, was what happened when those records began to be computerised. As information entered digital systems, it quietly stopped travelling. It became anchored to the place in which it was recorded, accessible in principle but no longer physically accompanying the patient in the same way. At the same time, the paper records that remained began to change their behaviour too. Instead of moving as they once had, they became increasingly static, as if everyone had collectively decided that their movement was only temporary and that, at some point in the near future, they would finally be absorbed into a digital world where travel would no longer be necessary.

Still, each place knew its own patients, and if you wanted to understand what was happening nationally you asked people to count things. This produced what were known as aggregate returns, which had the distinct advantage of not requiring computers that agreed with each other.

Among the earliest and most influential of these were the Korner returns. Named after a committee rather than a person who might have been able to escape the responsibility, they represented a time when healthcare information was deliberately summarised into categories that could be collected, compared, and placed into national reports without anyone having to open the metaphorical box and examine the contents too closely.

Hospitals reported activity in structured totals. Admissions. Discharges. Lengths of stay. Diagnoses grouped into tidy classifications that behaved well when printed and even better when filed. The system worked because it did not attempt to describe patients in any detail that might cause disagreement between observers.

It was not that nuance did not exist. It was simply not invited to the reporting process.

These returns were useful, in the way that weather summaries are useful if your only question is whether you should bring a coat, not whether you should build a house in a particular place.

So here, the information had settled into places. It had been distributed across organisations, each maintaining its own version of events.

This had not happened carelessly. It had happened because the system had learned to ask a new kind of question.

Instead of focusing only on individual encounters, it had begun to consider populations. How many patients were treated, how quickly, and with what outcomes. Whether services were improving. Where variation existed. What patterns could be identified if enough information could be brought together and examined.

These were important questions, and answering them required something the earlier systems had struggled to provide. Information needed to stay still long enough to be counted, structured, and analysed. It needed to be consistent. It needed to belong somewhere.

So it did.

Information was recorded where care occurred, and then it remained there. It was updated locally, maintained locally, and interpreted within each system’s own logic. Over time, this produced something that worked remarkably well for organisations and for the system as a whole.

It also produced duplication.

Not duplication in the sense of careless copying, but in the sense of multiple valid versions existing simultaneously. A diagnosis existed in primary care. It also existed in secondary care. Medication histories were present in several places, each reflecting the moment at which they had last been updated. Test results were available where they had been processed, and sometimes elsewhere, depending on how successfully systems had communicated.

Each version was correct when recorded.

Each version gradually became slightly out of step with the others.

The result was not confusion but something quieter and more persistent. It was a situation in which no single place contained the complete, current narrative at any given moment. Instead, the full picture emerged only when these parts were brought together.

Into this arrangement stepped the patient, who had acquired a role without being formally assigned one. They provided continuity. They answered questions, clarified changes, and resolved differences between what systems displayed and what had actually happened. In doing so, they became a point of connection between structured but separate accounts.

This was not inefficient. It worked reliably in practice. It depended, however, on the patient being present, able to recall events accurately, and able to translate their experience into something each part of the system could recognise.

As computing entered the NHS in more earnest fashion, the desire grew to move beyond aggregates and into something more granular. Something that could be queried rather than merely counted. Something that could, in theory, be reused for more than one purpose.

This was the rise of secondary use data. Information was no longer solely for the consultation. It contributed to planning, evaluation, and improvement across the entire system. It allowed healthcare to learn from itself in ways that had previously been impossible.

Yet there was an underlying tension.

The same information was expected to support immediate care and broader analysis. For one purpose it needed to be available, up to date, and unified in the moment. For the other it needed to be structured, standardised, and consistent across time and place.

These goals were aligned but not identical.

In meeting both, the system had achieved something significant. It could now see itself clearly. It could understand populations, identify change, and monitor performance with a level of detail that would have seemed extraordinary in earlier times.

But this clarity came at a cost that was not immediately visible.

The information had become easier to see collectively and slightly harder to experience as a single, unified narrative at the point of care. It had gained structure and perspective, but lost some of its natural tendency to gather in one place at one time.

That was where primary care immediately developed what can only be described as a deeply principled reluctance.

General practice data lived in systems that were local, varied, and historically resistant to outside interpretation. Not because anyone intended obstruction, but because the data had evolved inside consulting rooms where meaning was negotiated between clinician and patient, not between database schemas.

Every attempt to extract information nationally ran into the same difficulty. The data existed, but it did not exist in a standardised way that survived contact with other systems.

So when national reporting began to require more detail from primary care, something had to be invented to bridge the gap between clinical reality and statistical ambition.

This led to MIQUEST, which sounded like a polite request and behaved like a slightly more determined one.

MIQUEST allowed queries to be run against GP systems in a structured way, extracting data according to predefined definitions. It was, in principle, an elegant solution. In practice, it required every system to interpret clinical codes in exactly the same way, which turned out to be a bit like asking every local dialect of English to agree on the precise meaning of “fine”.

Still, it represented progress. Data could be extracted, provided one was willing to define precisely what one meant by data, and also willing to accept that what was extracted might not resemble what clinicians thought they had recorded.

Over time MIQUEST gave way to more standardised extraction methods, each one promising slightly better consistency and slightly fewer philosophical disagreements about what a diagnosis actually was when viewed from a different database.

Meanwhile, general practice continued to do what it had always done, which was to record patient information primarily for the purpose of caring for patients, with secondary consideration given to whether it could later survive national interrogation without collapsing into ambiguity.

And still, access remained difficult. Not because the data was absent, but because it was protected by a combination of clinical workflow, technical variation, contractual caution, and a long institutional memory of what happens when someone turns up asking for “just a simple extract”.

Even when access improved, what emerged tended to be carefully shaped.

This was where aggregation reasserted itself.

Despite all the progress in electronic records, a large proportion of national understanding still came from aggregated returns. Counts. Summaries. Activity totals. Carefully constructed views that could be compared across organisations without requiring them to agree on the underlying detail.

Korner had not disappeared so much as evolved into a mindset. Information was safest, and therefore most shareable, when it had been reduced to something that could not embarrass anyone by being interpreted too literally.

And then, somewhere in this evolution, primary care found itself in a peculiar arrangement.

In order to improve recording and standardisation, incentives were introduced. The Quality and Outcomes Framework meant that general practice was, in effect, paid to record specific pieces of information in specific ways so that they could be reliably counted, compared, and extracted.

The irony was not subtle. The system had discovered that if it wanted clean, structured, shareable data, it needed to fund the act of producing clean, structured, shareable data in the first place. It was a bit like paying someone to label their own drawers so that a national audit could confirm that socks were still being stored where socks were expected to be stored.

It worked, broadly speaking. But it also reinforced a curious separation between data recorded for clinical care and data recorded for system reporting. The same consultation could generate both a living clinical narrative and a parallel version optimised for measurement, each valid in its own environment, each slightly surprised to meet the other.

Bryan watched this separation with growing familiarity. The more the system improved its ability to measure itself, the more it seemed to rely on translations of clinical reality into formats that could survive being moved around without context.

He noticed something else too. The more data was prepared for secondary use, the more it was cleaned, coded, standardised, and abstracted, the less it resembled anything a clinician would recognise when looking at a patient.

A diagnosis that had once been a sentence became a code. A symptom that had once been described in human terms became a selectable field. A story that had once unfolded in a consultation became a dataset that behaved beautifully in analysis tools and slightly strangely in real life.

Clinicians sometimes opened records and paused, not because anything was missing, but because what was present felt like a summary written by someone who had never actually met the patient but had been given very good instructions about how to pretend they had.

At the same time, the system had become extraordinarily good at seeing patterns. It could aggregate across populations, identify trends, monitor outcomes, and produce insights that no single organisation could have generated alone. The view from above had become remarkably clear.

The view from inside the consultation room had become more dependent on interpretation.

Bryan stood for a moment, watching information move between systems, sometimes directly, often indirectly, frequently through carefully structured summaries that had once been rich detail and were now mathematically reliable approximations.

He watched the screens a little longer, the way one might watch a very efficient machine quietly insist it understands everything that is happening in a forest.

It did, in fact, understand the forest. Remarkably well. It could tell you how many trees there were, what kinds they were, how fast they were growing, and how often they were visited by other slightly smaller systems carrying clipboards. It could even predict changes in canopy coverage with an accuracy that would have impressed anyone who had never tried to recognise an individual tree.

“We’ve become very good at seeing the forest,” Bryan thought, watching another beautifully aggregated view of reality assemble itself in reassuring blocks of colour and statistical confidence.

A pause, as a clinician somewhere nearby tried to reconcile that view with a patient who was, inconveniently, still a single human being with a story that did not naturally compress into bar charts.

It occurred to him that somewhere along the way the system had learned to recognise patterns of woodland at planetary scale, while occasionally struggling to notice that the trees were still standing there, mildly confused, waiting to be asked what type of tree they were or what was actually wrong with them.

“We can see everything,” he added, “as long as it is sufficiently far away to look like a forest”

He looked once more at the layers of structured certainty, each one behaving as though proximity was a minor technical detail rather than the entire point.

“And up close,” he thought, “it turns out we’re still rather dependent on asking the trees to explain themselves."