Context

The Caisse Primaire d’Assurance Maladie (CPAM) de Meurthe-et-Moselle is setting up a call center to encourage its beneficiaries to undergo free cancer screenings.

Need

  • Optimizing prevention campaigns: The Meurthe-et-Moselle CPAM wants to give priority to contacting beneficiaries eligible for free cancer screenings who are most likely to participate.
  • Data-driven prioritization: This prioritization of calls will be based on the use of data available to the CPAM concerning its beneficiaries.
  • Basic project requirements:
    • Absolute protection of personal data: The confidentiality of beneficiaries’ information must be guaranteed, excluding any access to Datalchemy.
    • Methodological robustness: The algorithmic methods used must be statistically robust, in line with current scientific research.
    • Project reversibility: Ensure that the project is passed on to internal teams so that they can take ownership of it.

Completed work

To meet this need, Datalchemy was commissioned to help the CPAM de Meurthe-et-Moselle. We were involved in :

  • Identification of relevant data: Relevant beneficiary data was identified through extensive discussions with CPAM’s business and IT teams, in order to understand and target the data essential to the project.
  • Processing real data using synthetic data. For the cleaning and preparation of relevant beneficiary data, an innovative approach was employed, involving the construction of a synthetic dataset, reproducing the characteristics of real data. This method made it possible to develop and validate a data processing code, directly applicable to CPAM confidential real data.
  • Training and prediction of an ML model for targeting: Beneficiaries most likely to undergo screening are identified using an ML model trained for this purpose on cleaned real data, enabling a list of people to be prioritized for contact by the CPAM call center.
  • Documentation and handover to CPAM teams to ensure project reversibility.

Results

  • The project demonstrated strong predictive performance, with model recall of 70%.
  • A list of beneficiaries to contact has been drawn up for the call center.
  • The project was carried out entirely without access to actual data, thus ensuring total confidentiality of personal CPAM data.
  • The project was successfully passed on to the CPAM teams.