At the height of the Covid pandemic, we had around a hundred patients in intensive care at our hospital in Lyon, whereas normally this number does not exceed 10 people hospitalized in this department.
In acute dialysis, treatment parameters are calculated from the patient's biometric data and the machine's technology, with each device having its own rates and approach. The practitioner calculates them manually over 30 to 60 minutes, with a risk of error at the very moment when there is the least room for it. An initial version of the application existed, without an algorithm or automation: the parameters still had to be calculated manually.
Prove within ten days that a single formula could reliably and instantaneously produce these parameters on the target devices, for a presentation to investors in Paris.
Analysis of clinical cases, design of the formula and algorithms, development of the application and AI assistant. Service provided under contract through Isalyx.
L'enjeu clinique
Figures provided by Dr Viet Nguyen in La Dépêche (November 2023). Acute dialysis artificially performs the blood purification functions normally carried out by the kidneys.
Dix jours, 170 heures
Mathematical formula derived from the analysis of clinical cases, parameterized by the patient's sex and biometric data.
Fresenius Multifiltrate Pro (CVVHD Citrate, CVVHD) and Baxter Prismax and Prismaflex (citrate CVVH, CVVHDF, CVVH).
Dedicated monitoring algorithms for each device, powered by the same calculation engine.
Obtain a prescription or monitoring plan by conversing with the assistant instead of filling out the form.
The decision is automated: manual calculation is removed from the practitioner's workflow.
“Dializy Plus” subscription journey and credit system integrated into the demonstration.
- The calculation remains deterministic: the assistant does not calculate; it calls the engine via function calling and presents the result.
- Assistant grounded in a medical corpus through RAG: outside this corpus, it has no basis for answering, keeping it focused on the dialysis domain.
- LLM hosted in Europe for the governance and sovereignty of health data.
- Patient data anonymization
- No patient data stored
- LLM hosted on a European server · Governance and sovereignty of health data.
Équipe
- Yohan DebusscherAlgorithm design, development & AI assistant
- Dr Viet NguyenMedical expertise & clinical validation