Executive Summary

Artificial Intelligence is no longer a peripheral technology. It has the potential to become a next driver of industrial revolution. Using AI is rapidly shifting from “nice to have” to “must have”: it will increasingly differentiate competitive products and services from ineffective ones. Therefore applying AI thoughtfully to government work is essential — just as it is essential for the economy and other critical domains.

Earlier work established the broader strategic frame. In Artificial Intelligence in Education and Management the central thesis was formulated: AI has already ceased to be merely an auxiliary tool and is becoming a factor of productivity, competitiveness, organisational and governmental management, and labour-market structure. The same analysis underlined the systemic risks that accompany uncontrolled adoption and the necessity of controlled, human-centred deployment.

Subsequent research in manufacturing revealed a concrete missing piece. The analysis Production Knowledge as the Foundation for AI in Industry showed that the majority of recurring operational problems are not caused by the absence of modern AI models. They are caused by the absence of an effective system for capturing, storing, searching and reusing production knowledge — the digital memory of the enterprise. Without structured accumulated experience, AI has less reliable organisational knowledge to work with.

Healthcare exhibits a strikingly similar pattern, amplified by additional domain-specific blockers.

Russian healthcare already stores vast volumes of medical information. EGISZ, REMD, regional systems and local MIS successfully answer one important question:

“What data exists about this patient?”

They can also contain information about clinical decisions, treatments and subsequent events. However, the existence of these records does not automatically turn them into reusable institutional clinical experience.

This leads to a second, equally important question:

“What was done in similar documented cases — and what happened afterwards?”

Today, the answer to this question is often difficult to obtain systematically. Clinical experience remains distributed across institutions, systems and documents. Similar cases may be rediscovered repeatedly. Valuable experience accumulated in one medical institution may remain difficult to reuse elsewhere. When an experienced specialist leaves, part of the institution’s practical memory may also become difficult to recover and transfer.

The missing capability is therefore not simply another repository of medical data.

It is the ability to turn accumulated medical records and documented clinical decisions into reusable institutional knowledge .

These three lines of work converge on a single practical proposal for the government healthcare system: create a missing infrastructure layer — a Digital Memory of Medical Institutions.

It sits on top of the existing digital infrastructure. It does not replace EGISZ, clinical guidelines, MIS or the physician. Data standardisation, integration and aggregation from existing systems are necessary technical prerequisites, but they are not the primary goal.

The primary goal is to transform documented clinical practice into a structured, reusable and traceable body of institutional clinical experience .

The core idea is simple:

Existing medical documents should become raw material for institutional clinical memory, not the final product of the system.

Such a system should not merely provide a recommendation such as:

“In similar cases, X was recommended.”

It should make it possible to trace that recommendation back to the documented cases, clinical decisions and outcomes from which the institutional knowledge was derived.

This is an important distinction from a conventional generative AI assistant. The value is not only in producing a plausible answer, but in making its institutional basis inspectable.

The system should not ask a physician simply to trust an AI-generated conclusion. It should allow the physician to understand and verify why that conclusion was reached.

Recent practical work on a university BI system demonstrated how such decision-support infrastructure can be built with explicit traceability, alternative interpretations, confidence levels and known limitations. In Lightweight BI system for university, part I the architecture deliberately separates observations from conclusions and keeps the human expert as the final decision maker — principles that transfer directly to clinical settings.

A Digital Memory of Medical Institutions could therefore provide healthcare with a capability that existing information systems do not fully provide: the ability to learn systematically from its own documented clinical practice .

This could reduce dependence on the physical location of expertise, preserve institutional knowledge, reduce repeated rediscovery of similar clinical approaches, and create a durable knowledge asset from experience that is already being accumulated by the healthcare system.

All of the above opens significant opportunities as well as new risks. Clinical data is highly sensitive, historical experience may contain errors or local biases, and inappropriate generalisation from previous cases can be dangerous. Privacy, security, access control, clinical validation, legal responsibility and governance must therefore be considered integral parts of such a system.

This is a conceptual proposal and only the beginning of a much larger path towards implementing such an innovation. Many technical, clinical, organisational, legal, ethical and security challenges remain to be discovered, understood and overcome. The purpose of this proposal is not to claim that all of these questions have already been solved, but to define a potentially valuable direction for further work.

This proposal has been prepared for the Ministry of Health of the Russian Federation as a conceptual starting point for discussion and further investigation.

The Problem: Data Without Memory

Healthcare systems excel at recording events. They are far weaker at turning those events into transferable institutional knowledge.

Today clinical experience is fragmented across:

  • individual physicians,
  • separate institutions,
  • incompatible MIS,
  • paper notes and informal conversations.

Consequences are concrete:

  • A primary-care doctor rarely sees the full documented decision path and subsequent outcomes from a specialised centre.
  • Similar complex cases are solved from scratch instead of building on previous documented results.
  • Data exchanged between organisations often arrives in a form that is hard to analyse or reuse.
  • Every new MIS integration is custom and expensive.
  • When a key specialist retires or moves, a significant part of hard-won practical knowledge may become difficult to recover and transfer.

The system records clinical events and documents. It does not systematically transform those records into a reusable memory of what happened after particular clinical decisions in comparable cases .

Proposed Solution: Three Linked Functions

1. Capture experience as structured clinical cases

Every case is represented as a linked chain:

Observations → Decision → Treatment → Outcome

Context, data limitations and confidence level are preserved. The result is not another free-text note, but a reusable clinical case that can be compared, searched and explained.

An important distinction must be preserved throughout the process: documented clinical facts are not the same as reconstructed clinical reasoning. Where the reasoning behind a decision is inferred from available documentation, the system should explicitly identify it as reconstructed rather than presenting it as an observed fact.

2. Make experience available to the treating physician

When a doctor works with a new patient, the system surfaces similar historical cases and shows:

  • decisions that were made,
  • treatment options that were applied,
  • observed outcomes,
  • degree of similarity and the factors contributing to it,
  • alternative hypotheses that were considered,
  • known limitations of the recommendation.

Similarity itself is not a guarantee of clinical equivalence. The system should therefore make clear why historical cases were considered similar, what relevant information is missing, and how strong the resulting evidence is.

This is not a black-box answer. It is traceable decision support grounded in documented real practice, with explicit boundaries on what can and cannot be inferred from that practice.

In particular, an observed association between a treatment and a favourable outcome in historical cases should not automatically be interpreted as proof that the treatment caused the outcome. Differences in patient characteristics, disease severity, selection of treatment, physician expertise and other factors may influence observed results.

3. Create a single integration hub

MIS no longer need to build point-to-point integrations with every other system. They connect once to a common interface and use a shared clinical-case format. The Digital Memory becomes both a working tool for physicians and a practical standardisation point for the exchange of clinical experience.

Such an interface requires not only technical interoperability but also semantic standardisation: common definitions, terminology, identifiers, units, classifications and rules for representing the meaning and provenance of clinical information. A common API alone is not sufficient if connected systems interpret the same data differently.

Positioning: The Missing Layer

Existing classes of solutions solve important but different problems. They may overlap with some functions described here, but they are not primarily designed to provide a persistent, cross-institutional memory of documented clinical experience and observed outcomes as a day-to-day working layer for physicians:

Class of solutions What they give the physician What they typically do not provide as their primary function
Clinical decision support & predictive analytics (Webbiomed, MosMedAI and similar) Risk scores, preliminary diagnoses, EMR analysis Persistent shared memory of documented real cases with outcomes across institutions
Evidence engines (UpToDate, DynaMed, ClinicalKey, OpenEvidence) Access to guidelines and scientific literature Systematic reuse of the healthcare system’s own accumulated clinical experience
Medical AI search tools Fast retrieval of normative and clinical sources Systematic accumulation and reuse of real-world institutional outcomes
EHR-native solutions Support inside one MIS or hospital network A cross-system clinical memory layer spanning participating institutions
Federated learning / research platforms Ability to train models without moving raw data A mass-scale, day-to-day working tool for practising physicians that exposes comparable historical cases and their documented outcomes

Our niche is different.

This is not “another medical AI”. It is infrastructure for the accumulation and reuse of clinical experience.

The market today mostly answers:

“What does medicine recommend?”

The proposed system adds a second, equally necessary question:

“What actually happened in real similar documented cases — and what was the result?”

The combination of evidence-based knowledge and real-world clinical memory can become a foundation for the next generation of physician decision support. The two sources of knowledge should complement each other rather than be treated as interchangeable: evidence describes what is supported by research and established medical knowledge, while clinical memory describes what has been observed in the participating healthcare system.

Value for the Healthcare System

For the physician

  • Faster orientation in complex cases.
  • Access to the experience of leading centres without leaving the working MIS.
  • Transparent recommendations: the doctor sees the supporting cases, confidence and limitations.

For the system

  • Clinical experience becomes a systemic asset instead of personal knowledge that walks out the door.
  • Quality of care can become less dependent on the particular institution a patient happens to reach.
  • A single interface creates a practical standardisation point for different MIS.

Core principle

Final clinical responsibility always remains with the physician.

The system must show not only a conclusion, but the evidence it rests on and the boundaries of its applicability.

Pilot Proposal: Start Where Experience Matters Most

We propose to begin with neurology — a domain in which differential diagnosis is complex, accumulated expert experience is especially valuable, and access to specialised centres is uneven.

Pilot contour

  • 1–2 large neurological centres → sources of expert clinical experience
  • Several primary-care and inpatient institutions → users of the system
  • 1–2 MIS vendors → validation of the standard connection model

What the pilot must verify

  • Value — Does search time for analogous cases and tactical decision-making decrease?
  • Quality — Are real cases, decisions, relevant context and outcomes structured completely and correctly enough?
  • Trust — Do physicians find the recommendations useful, understandable and verifiable?
  • Integration — Can an external MIS connect to a single interface without a custom bilateral integration while preserving semantic consistency?
  • Safety — Does the de-identification, access-control, traceability and recommendation-limitation model work in a closed or federated environment?
  • Evidence quality — Can the system distinguish documented observations from reconstructed reasoning and observed associations from conclusions that would require causal evidence?
  • Privacy — Can sufficient clinical context be preserved for useful case comparison without creating unacceptable re-identification risks?

Roadmap

  1. Discuss the concept with the relevant departments of the Ministry of Health of Russia and interested National Medical Research Centres.
  2. Form a working group (clinicians, digitalisation specialists, legal experts, information-security specialists, data architects and MIS developers).
  3. Develop a minimal clinical-case standard, semantic model and the technical assignment for the pilot.
  4. Define governance, access-control, privacy, provenance and clinical-safety requirements before connecting real clinical data.
  5. Launch a limited pilot and evaluate it against pre-agreed metrics.
  6. If the hypothesis is confirmed — define the path to scaling, federated connection of regions, and further standardisation of clinical-experience exchange.

Closing Thought

EGISZ already helps the healthcare system remember which documents were created and which information was recorded.

The next step is to help the system systematically transform documented clinical practice into reusable experience: what was observed, which decisions were made, what treatment followed, and what outcomes were subsequently observed.

Digital Memory of Medical Institutions is a proposal to build exactly that layer — an infrastructure of reusable clinical experience that can raise the quality of care, reduce dependence on individual specialists, and make the collective knowledge of Russian medicine more accessible to every authorised physician who needs it.

This proposal is intentionally conceptual. It is not a claim that the complete technical, clinical, legal or organisational solution has already been designed. It is the beginning of a much larger journey towards implementing an innovation of this scale. Many challenges will only become visible through practical work, and they will have to be discovered, tested and overcome step by step.

The objective of the proposal is therefore not to replace clinical judgement or established medical science, but to create the infrastructure through which the healthcare system can learn more systematically from its own documented experience while preserving evidence, uncertainty, provenance and human responsibility.

Related Publications