Enhance pre-screening & cohorting stages

A tool enabling physicians to:
- secure and autonomous access to facility data
- facilitate and accelerate research projects by pre-screening patients
- reduce their dependence on technical teams

The Arkhn data architecture embodies the entire technical stack for deploying a facility's health data warehouse, from the methods for integrating data quality and documentation to the tools for accessing it.

Arkhn Explore is based on a minimal version of the Arkhn data architecture.

Arkhn adapts to the maturity level of your organization to create a comprehensive health data management framework:
- If the facility has its own health data warehouse, we can build on this to deploy Arkhn Explore.
- If the facility does not have its own health data warehouse: We deploy a minimal version of our data architecture to implement Arkhn Explore. It is also possible to deploy the full version of the Arkhn data architecture and access the Arkhn health data warehouse.

Our teams of data experts frame the project with the facility's staff and discuss it with the medical teams. The aim is to adapt to the level of maturity of each facility and to identify specific needs (use cases). We work with the professionals who master the data pathway in the centers (MID, ISN, Medical Team) to understand the patient and data pathways and identify their relevant sources.

We configure and integrate data into our tools. We verify data integration, carrying out a range of technical tests for data quality and the functional objectives defined in the use case (e.g. can Explore automatically find patients selected by hand in a preliminary study of the center?).

From a regulatory point of view, Explore's pre-screening view only contains aggregated data, and can therefore be accessed by all medical and research teams. Cohort extraction, on the other hand, is subject to approval by administrators (ISD, MID or other, depending on the hospital).

Pseudonymization of data is the mandatory step for research projects (with the exception of doctors who work with data from patients they see in their departments). Once the doctor has completed the pre-screening of the data, in other words, his feasibility study (which contains only aggregated data), he may then want to extract the identified patient data to carry out his study: this extraction includes a pseudonymization phase thanks to our NLP (Natural Language Processing) algorithms.Our pseudonymization algorithm uses state-of-the-art technologies (based on Transformer technology and BERT models), and is 99% accurate in identifying and masking a range of sensitive information (surname, first name, date of birth, PPI, address, email, telephone, etc.). It has been trained on a large database of reports manually annotated by medical experts.

A first category of NLP models are used to identify medical entities (procedures, diagnoses and prescriptions) within medical documents, and to standardize all available medical entities (structured or unstructured from documents) into standard ontologies (ICD10, etc.).
‍
A second category of models is dedicated to the pseudonymization of medical documents, notably for clinical research projects.
‍
NLP models systematically feed Explore's functionalities.
‍
Our algorithms use state-of-the-art technologies (based on Transformer technology and BERT models). They identify clinical entities (such as drugs) and standardize them into a reference terminology (such as CCAM). Our performance is 97.5% on the identification (recall) of treatments, examinations, diagnoses, symptoms and drugs.

- Number of patients meeting criteria, with reminder of selected criteria.
- Various views of aggregated statistical data on age, sex and living/deceased status (min, max, mean, standard deviation, histogram, etc.).