aion-clinical 1.0.6

Formal temporal and causal structure for consistent clinical data across systems.

AION — Structural Consistency for Clinical Data

Formal temporal and causal structure for consistent clinical data across systems.

---

Clinical data systems often produce inconsistent results.

Not because of missing interfaces —
but because time, semantics and causality are not modelled consistently.

AION makes clinical data consistent and verifiable across systems.

It provides a formal, executable structure to:

• represent clinical events over time
• define and validate relationships between them
• ensure consistent interpretation across systems
• enable reproducible analytics and reliable AI

Most systems exchange clinical data.

AION ensures that this data is interpreted consistently.

---

Quick Start

pip install aion-clinical

from aion.core import Interval, classify
import datetime

dt = lambda h: datetime.datetime(2024, 1, 1, h, 0)

anaesthesia = Interval(dt(7), dt(12))
operation   = Interval(dt(8), dt(11))

print(classify(anaesthesia, operation))
# -> AllenRelation.CONTAINS

---

Why AION?

Most systems can exchange clinical data.

The harder problem begins afterwards:

• data is interpreted differently across systems
• temporal relations remain implicit
• causal assumptions are hidden in project-specific logic
• analytics become difficult to compare and reproduce

AION addresses this structural layer directly.

Instead of modelling isolated data points only, AION models:

• intervals instead of points
• relations instead of implicit assumptions
• causal structures instead of static data
• verifiable consistency instead of hidden transformation logic

---

What is AION?

AION is a formal, executable model for clinical information systems.

It generalises earlier formal models (FM-1, FM-2 / CAIRN)
into a domain-independent structure for clinical data.

---

Core Capabilities

Temporal Model

• Complete Allen interval algebra (13 relations)
• Fuzzy intervals with probabilistic reasoning

Data Model

• Universal clinical event model (6-tuple)
• Type system with hierarchy and schema evolution

Query & Analytics

• Cohort algebra
• Temporal pattern languages

Causal Inference

• Causal graphs
• Do-operator
• Structure learning

AI Integration

• Formal AI component model
• Explainability (Shapley, counterfactuals)

Privacy

• Differential privacy (Laplace / Gaussian)
• Federated computation

Interoperability

• FHIR mapping via structural homomorphism

---

Example: Cohort Query

from aion.query.cohort import query_event_sequence

q = query_event_sequence("Diagnosis", "Procedure")
cohort = q.evaluate(ctx)

---

Example: Causal Analysis

from aion.causal.graph import CausalGraph
from aion.causal.do import DoOperator

g = CausalGraph()
g.add_edge("Diagnosis", "Procedure")

do = DoOperator(g, data)
result = do.intervene("Procedure", "Outcome")

---

Test Suite

pytest aion/tests/
# 326 tests, ~2s runtime

---

Architecture

aion/
├── core/          Allen algebra (13 relations), type system σ(τ),
│                  event 6-tuple, fuzzy intervals, process DAG
├── abstraction/   Episode formation B_{Φ,Δ}, clinical trajectories
├── query/         Cohort algebra, RTP patterns, TCFG, predicates
├── causal/        Causal graph, do-operator, backdoor adjustment,
│                  PC algorithm, bootstrap structure learning
├── privacy/       ε-DP Laplace/Gaussian, federated model, LDP
├── ai/            AI components A_l = (X,Y,f_θ,L,θ*), Π_l operator,
│                  feature extraction Ψ, risk estimation
├── explain/       Shapley attribution, counterfactual Δ*_cf,
│                  sufficient explanation S*_suf, Π_l^+ gate
├── schema/        Versioned type system, 7 schema change ops,
│                  migration function μ_{v→v'}
└── adapters/
    └── fhir/      Structural homomorphism (h_T, h_σ), Q_map index

---

Key Concepts

Event 6-Tuple (AION §5)

e = (p, a, τ, [t^B, t^E], α, ρ)

• p patient, a stay, τ type, [t^B,t^E] interval, α attributes, ρ references

---

Positioning

AION is not:

• a data format
• a data warehouse
• a FHIR replacement

AION is:

the structural consistency layer beneath clinical data systems

---

License & Citation

EUPL-1.2 — open source. Commercial licence: [e-mail]

@software{aion2026,
  title  = {AION: Algebraic Interval Ontology for Clinical Networks},
  author = {Matten, Friedhelm},
  year   = {2026},
  url    = {https://codeberg.org/fm2-project/aion},
  doi    = {10.5281/zenodo.19548857}
}

© Friedhelm Matten, ISCaD GmbH