Health
Healthcare Solutions — Data, Knowledge and Intelligent Decision Support
Healthcare generates complex, heterogeneous and high-stakes data.
At amsafis, we integrate Machine Learning, Expert Systems and RAG-LLM technologies to transform clinical information into reliable predictions, interpretable reasoning and actionable decision support.
Our objective is simple:
to help clinicians and health organisations make better decisions with less uncertainty.
1. Epistasis and non-linear models in Health
(no GWAS, no additivism)
The biological and clinical world is not linear.
Diseases, adverse drug reactions, and complex phenotypes do not emerge from the isolated effect of individual variables, but from their simultaneous interactions.
We work explicitly from this premise.
We are epistasis researchers: we study gene–gene interactions and non-linear combinations as basic explanatory units.
For this reason, we do not perform GWAS or classical additive models: these approaches, although mainstream, structurally ignore interaction and reduce real complexity to marginal effects.
Health datasets contain non-linear relationships, conditional dependencies, and patterns that cannot be detected with linear statistics.
For this reason, we use approaches oriented toward:
- combined variables as entities in their own right
- explicit logical rules (AND / OR)
- contingency tables, decision trees, and non-linear regressions
- pattern mining with clinical meaning
- mechanism explanation, not only prediction
The objective is to explain diseases, explain adverse reactions, and make visible patterns that the dominant paradigm cannot see.
Example — Association rules in pharmacogenomics
Genetic analysis based on specific SNP combinations can reveal associations with adverse drug reactions that do not appear in univariate analyses or GWAS.
A concrete example of this approach applied to penicillin allergy can be consulted here.
2. Expert Systems for Transparent Clinical Reasoning
Expert Systems allow formalising medical knowledge into explicit rules, ensuring traceability, interpretability and control, which are essential in regulated environments.
They are ideal for:
- diagnostic support
- alert systems
- triage pathways
- rule-based quality and safety protocols
- hybrid reasoning combining rules + data-driven prediction
An example is the rare disease sandbox, which uses knowledge rules to identify possible diagnoses in interactive mode. Explore the interactive demonstration.
The same knowledge engine can be accessed both in a human-driven way and integrated into automated processes that allow the knowledge to be reused as an executable service within decision architectures or computational assistance.
3. RAG-LLM for Clinical Knowledge Retrieval
LLMs become clinically useful only when connected to verified sources, hospital documents or scientific guidelines.
Our RAG pipelines ensure:
- evidence-based answers
- citations and document traceability
- automatic summarisation of clinical reports
- safe interaction with structured and unstructured data
This enables clinicians to query complex documents naturally while maintaining a controlled and auditable system.
4. Integrated Decision Platforms
Real value emerges when the three components work together:
- ML detects risk or predicts outcomes
- Expert Systems encode medical knowledge and enforce clinical rules
- RAG-LLM retrieves scientific evidence and contextual information
Combined, they support:
- early warning and risk scoring
- automated triage
- patient-pathway optimisation
- resource planning
- quality and safety monitoring
A complete loop from data → knowledge → decision → action.
If your organisation needs support with predictive analytics, clinical reasoning, AI integration or decision-support automation, we can help you design a rigorous, safe and efficient solution.