Smart Patient Label Challenge
Pfizer is seeking a ‘Smart Patient Label’ solution that can detect and link Summary of Product Characteristics (SmPC) to the appropriate sections of a corresponding version of a Patient Information Leaflet (PIL). When there is a change in SmPC, the solution should auto-detect the corresponding PIL changes. Ultimately, the capability to detect and link would be coupled with AI/ML techniques to auto-generate a PIL from a SmPC.
In Europe, the SmPC (the Summary of Product Characteristics) is the scientific label developed for all medicines. It describes the safety and efficacy of the product in scientific terms and is used by healthcare professionals who consult it prior to prescribing the product.
In the UK, the SmPC is available electronically.
The Patient Information Leaflet (PIL) is the patient-facing version of the SmPC. It’s written in patient-friendly terms and inserted in the commercial pack given to all patients. The license holder is responsible for keeping the SmPC and PIL up to date to reflect the latest benefit-risk information for the medicine. The license holder also ensures that the two documents are kept synchronized and consistent in terms of meaning.
In the initial phase of the challenge, collaborators are expected to develop an AI/ML-based solution to auto-detect and link the sections of the SmPC with appropriate sections of the PIL. The initial focus of the challenge is on SmPC and PIL pairs registered in the European markets. Training data will be sourced from Pfizer in the form of SmPC and PIL pairs. In addition, collaborators should reference public domain information (see Public Domain References section). Pfizer labeling subject matter experts (SMEs) will be made available to the selected collaborators during the execution stage of the challenge.
In the next phase of this challenge, when a change to the SmPC is made, the Collaborator will demonstrate that the solution is capable of detecting the context of the change along with all the relevant sections of the PIL that needs to change in order to link and recommend suggestions for the patient friendly language.
SMEs will review machine recommendations and provide ‘human-in-the-loop’ feedback that will be applied to the appropriate sentences in the PIL. Collaborators will have to demonstrate the solution’s capability to learn continuously from the reviewer’s feedback and progressively improve the solution’s accuracy to correctly identify, link, and recommend PIL language. The recommended test should be consistent with defined business rules and regulatory requirements. Ultimately the solution should scale to auto-generate de novo content, such as the creation of a PIL in its entirety, given an SmPC with appropriate contextual relevance.