Healthcare systems across the world have invested billions in clinical digitalization. Electronic health records (EHRs), clinical data warehouses, integrated hospital platforms, and now artificial intelligence systems have been introduced under the promise of efficiency, transparency, interoperability, and improved patient outcomes.
Yet, after more than two decades of implementation, the recurring pattern remains strikingly consistent: resistance from clinicians, workflow friction, hidden cognitive overload, interoperability failures, and, in many cases, silent abandonment.
The problem is rarely technological immaturity. Nor is it simply “resistance to change”.
The deeper issue is structural: clinical digitalization often fails because it attempts to impose the logic of information systems onto a domain—medicine—that operates according to fundamentally different epistemological principles.
This is not a story of bad software. It is a story of a category error.
Key takeaways (in 60 seconds)
- EHR failures are often modeling failures: clinical work is non-linear, contextual, and narrative.
- Digitalization changes visibility and power, so adoption is also a governance problem.
- “Interoperability” fails when architectures remain closed and vendor-optimized.
- Many systems increase cognitive load by externalizing clerical work to clinicians.
- AI helps when it reduces clerical friction and mediates semantics, not when it is layered on top of broken workflows.
1. The Illusion of Rational Digitalization
At the outset, digitalization projects appear rational and unavoidable. They promise:
- Standardization
- Traceability
- Efficiency
- Data accessibility
- Decision support
From a managerial perspective, these objectives are legitimate. Healthcare is complex, costly, and increasingly accountable. Digital systems promise control over that complexity.
However, what appears rational at the level of governance often becomes frictional at the bedside.
Clinicians experience rigidity where administrators see structure.
They perceive surveillance where policymakers see transparency.
They feel increased workload where vendors promise efficiency.
This divergence is not accidental. It reveals a structural mismatch between two different ways of organizing reality.
2. Data Are Not Neutral: Visibility, Power, and Ownership
One of the least discussed dimensions of digitalization is that clinical data are not merely informational assets. They are also instruments of power.
When data become universally visible:
- Clinical decisions become retrospectively examinable.
- Performance becomes measurable.
- Activity becomes comparable.
Digitalization modifies the balance of visibility within institutions.
Questions inevitably arise:
- Who can access my clinical data?
- Why can others audit my activity while I cannot audit theirs?
- Is this system designed to support care—or to monitor me?
These are not irrational fears. They reflect a shift in governance architecture.
Digital systems do not simply store information. They redistribute authority.
Any digitalization project that ignores this political dimension is likely to encounter resistance—not because clinicians oppose technology, but because clinicians understand its institutional implications.
3. The Ontological Error: Treating Patients as Inventory
Software engineering thrives on structured entities:
- Defined states
- Clear transitions
- Deterministic workflows
- Stable categories
This logic works remarkably well in administrative domains. Billing, scheduling, inventory management, procurement—these processes are structured, rule-based, and relatively stable.
Clinical medicine is not.
The patient is not a static entity with predefined states. The patient is a dynamic, context-dependent system characterized by uncertainty, incomplete information, evolving trajectories, and narrative complexity.
When software systems model patients as if they were items in a warehouse—admitted, processed, discharged—they inevitably simplify the ontological richness of clinical reality.
This explains a persistent observation: administrative software often works better in hospitals than clinical software.
Administration resembles structured logistics.
Clinical care resembles complex adaptive reasoning.
The failure lies not in programming skill, but in modeling assumptions.
4. The Semantic Problem: Medicine Is Narrative
Medical language is inherently variable.
Consider a simple postoperative complication:
- “FA in POD2”
- “Episode of paroxysmal atrial fibrillation occurring on postoperative day two”
Both refer to the same phenomenon. Yet clinical expression varies according to habit, context, training, and communicative intent.
Information systems, however, require:
- Stable ontologies
- Consistent syntax
- Standardized coding
- Structured data fields
The attempt to compress medical narrative into fixed data-entry schemas inevitably generates friction.
Clinicians compensate by:
- Using free-text fields
- Entering partial information
- Circumventing rigid workflows
Standardization initiatives (ICD, SNOMED, HL7, FHIR) are essential and valuable. Yet they cannot eliminate the fundamental variability of clinical language, because medicine is not merely descriptive. It is interpretive.
Clinical reasoning is contextual, provisional, and narrative-driven. A purely tabular representation cannot fully capture this dimension.
A further complication is that clinical documentation is not only “data”. It is also a medico-legal artifact and a medium of human-to-human communication. Any system that treats notes as mere structured fields will collide with this reality.
5. Cognitive Load and the Hidden Cost of Rigid Systems
Digitalization frequently externalizes cognitive costs onto clinicians.
Common patterns include:
- Multi-layer authentication
- Sequential, non-adaptive workflows
- Mandatory structured data entry
- Alert fatigue
- Redundant documentation
Instead of reducing workload, poorly designed systems increase task fragmentation and mental switching.
The result is not only frustration, but measurable cognitive burden. Documentation time expands. Direct patient interaction contracts.
What is rarely measured in digitalization projects is the cognitive cost per clinical action.
Systems are evaluated on deployment success (“go-live”), not on net cognitive efficiency.
When a digital platform adds friction without reducing risk or saving time, clinicians do not reject technology. Clinicians reject inefficiency.
6. Interoperability: The Persistent Illusion
Another recurring failure lies in integration.
Healthcare institutions often deploy multiple systems:
- EHR
- Laboratory software
- Imaging platforms
- Administrative modules
- Pharmacy systems
In theory, interoperability standards exist. In practice, integration remains partial, fragile, or vendor-dependent.
Data silos persist.
Interfaces are brittle.
APIs are limited.
As a result, clinicians must navigate multiple platforms, duplicating entries and reconciling inconsistencies.
Fragmented architectures generate friction not because integration is technically impossible, but because systems are often built as closed environments optimized for internal coherence rather than modular interoperability.
Without architectural openness, digital ecosystems become digital labyrinths.
A non-technical driver matters here: procurement and incentives. When purchasing decisions reward feature checklists over usability, and lock-in over openness, architectures predictably become closed and brittle.
7. Why Artificial Intelligence Will Not Magically Fix This
The current wave of artificial intelligence in healthcare promises to solve many of these problems.
However, AI is unlikely to correct structural dysfunctions on its own. It will not, by default:
- Fix flawed governance structures
- Resolve institutional distrust
- Repair closed architectures
- Eliminate incentive misalignment
- Transform rigid workflows into adaptive ones
If deployed on top of structurally misaligned systems, AI risks amplifying existing dysfunctions.
Adding prediction to chaos does not generate coherence.
So where can AI actually help?
8. Where AI Can Actually Help
Despite these limitations, AI does offer meaningful opportunities—if used appropriately.
1. Semantic Mediation
Large language models and NLP systems can translate narrative text into structured representations, reducing the tension between clinical variability and system rigidity.
They can:
- Normalize expressions
- Map narrative to ontologies
- Extract structured variables from free text
This does not eliminate variability, but it can mediate it.
2. Reduction of Clerical Burden
Speech-to-text systems, intelligent pre-filling, and automated documentation support can reduce repetitive administrative tasks.
When AI reduces documentation time rather than increasing it, adoption becomes organic rather than enforced.
3. Adaptive Workflow Orchestration
AI can help design event-driven, non-linear systems that adapt to clinical trajectories rather than forcing clinicians into rigid sequences.
Such systems must remain clinician-in-the-loop, transparent, and auditable.
AI’s value lies not in replacing reasoning, but in absorbing clerical friction.
9. A Minimal Framework for Non-Failing Digital Systems
If digitalization is to succeed, certain structural principles must guide implementation:
- Workflow-first design Systems must model real clinical processes before enforcing data structures.
- Semantic flexibility with structured back-end mapping Allow narrative input, then translate into structured data through mediation layers.
- Measurement of cognitive cost Evaluate systems based on net reduction of clinician time and cognitive burden.
- True interoperability Modular architectures, open APIs, and standardized exchange protocols.
- Transparent data governance Clarify access rights, audit structures, and visibility symmetry.
- Clinician-centered evaluation metrics Success must be measured in time saved, errors reduced, and communication improved—not merely in deployment completion.
To make this concrete, consider a common clinical reality: trajectories branch. A postoperative course can shift quickly from “stable recovery” to “arrhythmia”, “bleeding”, “infection”, “AKI”, or “delirium”, each altering priorities, documentation needs, and team communication. A linear software funnel will always feel wrong against this event-driven logic.
Conclusion: Digitalization Must Adapt to Medicine
Clinical digitalization fails when it attempts to constrain medicine within the deterministic logic of software architecture.
It succeeds only when digital systems acknowledge that medicine is:
- Dynamic
- Narrative
- Uncertain
- Context-dependent
- Relational
The patient is not an inventory unit.
Clinical reasoning is not a linear transaction.
Healthcare is not a warehouse workflow.
Artificial intelligence may become a powerful mediator between structured systems and narrative medicine—but only if we first correct the structural mismatch at the heart of digital healthcare.
Technology must adapt to clinical epistemology—not the other way around.
FAQ
Isn’t clinician resistance the real problem?
Often, “resistance” is a signal of design debt: misfit between tools and real work, plus legitimate concerns about governance and accountability.
Do standards like FHIR solve interoperability?
Standards help, but interoperability also depends on incentives and architecture. Closed systems can “support standards” and still prevent meaningful modular integration.
Can AI reduce burnout?
Yes, if it removes clerical burden and improves navigation of narrative data. No, if it adds alerts, extra steps, or opaque recommendations to already fragile workflows.