Skip to content

Implementation of an artificial intelligence tool for the detection of drug safety problems

European Statement

Patient Safety and Quality Assurance

Author(s)

Noe Garin, Laia Lopez-Vinardell, Pau Riera, Adrian Plaza, Ivan Castellvi-Barranco, Jose Mateo-Arranz, M. Antonia Mangues

Why was it done?

APS is a rare disease with a high risk of thromboembolism. Recently, some data suggested an increased risk of thrombotic events with direct-acting anticoagulants (DOAC) compared with vitamin K antagonists in APS. Some agencies advise against the use of DOACs in these patients.

This methodology can be extrapolated to other risk situations, so this was a first step with AI to further detection of safety issues.

What was done?

We implemented an Artificial intelligence (AI) tool based on natural language processing (SAVANA®) to identify patients at risk of thromboembolism, defined as Antiphospholipid Syndrome (APS) diagnosis treated with direct-acting anticoagulants (DOAC). SAVANA® is an AI tool able to extract information contained in free-text from electronic clinical records.

A prior operation work was conducted, involving: direction, pharmacy, documentation, IT, SAVANA®, data protection. The work and previous meetings evaluated: feasibility, previous requirements, privacy issues, IT involvement and contract signings.

How was it done?

The implementation consisted of:
– Transference of medical record information to the SAVANA® cloud.
– Identification of the health problem (APS) and initial search.
– Search algorithm optimization in a multidisciplinary team.
– Evaluation of the search by SAVANA® by peer review in a sample of randomly selected cases (n=200).
– Precision and sensitivity analysis. Algorithm improvement.
– Obtaining the Gold Standard and validation.
– Definitive search for the detection of patients with APS in treatment with DOACs and performance of interventions.

What has been achieved?

The project implementation is at a very advanced stage. The algorithm has currently been evaluated and is being refined after precision and sensitivity analysis. Final validation and definitive identification of patients at risk is expected at the end of 2021. Patients detected during the implementation method have been evaluated with the haematology team.

What next?

This methodology can be implemented in any centre with computerized medical records. The use of AI is the only tool available for the identification of certain groups of patients when health problems are not coded. In other cases, its use regarding the extraction of lists allows a great capacity for analysis, absence of biases derived from human error, guarantee of reproducibility and complementary data obtention, mainly in samples of high size.

×

EAHP Forum

All the EAHP team is working on providing a Forum that can help connect all the members in Conversations and Groups to talk about important matters for the European Hospital Pharmacist.

The Forum will be accessible for all the EAHP members, you don’t have to create a new account to browse and participate.

Conversations and groups

The Conversations will be moderated by our team to provide documents and relevant topics for the community.

The Groups will connect all members that share a category. Members who work on the same assocation, on the same hospital, that have the same role, etc.

Stay tuned for the realase of the forum. Soon on EAHP.