MedModels GmbH is spun off as a sister company of Limebit GmbH and launches the AI-powered Patient Synthesizer on the market.
Limebit GmbH today announced the strategic spin-off of MedModels GmbH. The company will continue to develop its proprietary open source platform MedModels for processing and analyzing digital patient records and expand it to include enterprise solutions. The first core product is the "Patient Synthesizer".
This groundbreaking AI technology was developed to advance medical research in Europe by providing access to high-quality patient data without violating data protection regulations. The development of the MedModel project is funded by the German Federal Ministry for Economic Affairs and Climate Protection.
The Patient Synthesizer is a state-of-the-art AI system based on an autoregressive architecture similar to that used in large language models (LLMs). It learns the statistical distributions of real medical data sets and can then create new, completely synthetic patient records. These data reflect the statistical and structural characteristics of the original data without reproducing sensitive patient information. This ensures complete data protection and compliance. In pharmaceutical research, the Patient Synthesizer enables the secure generation of synthetic data sets, e.g., for real-world data-driven control arms in the contextualization of clinical trials, as demonstrated in an ongoing pilot project with a leading pharmaceutical company.
An independent performance comparison confirms the exceptionally high degree of consistency between the synthetic data and real data sets. A case study demonstrates that the Patient Synthesizer accurately replicates complex medical relationships, such as the simultaneous occurrence of events during a doctor's visit and the correlations between different characteristics, with a consistency of over 95% for multivariate correlations. Relevant longitudinal characteristics, such as distributions and frequencies of doctor visits, are also accurately preserved. The technology also shows strong resistance to re-identification attempts such as "membership inference attacks," underscoring its data protection properties.
The Patient Synthesizer enables data owners to securely unlock the value of their sensitive data sets, while providing health science companies with access to realistic data for research, development, AI training, and innovation. It offers a transformative solution that bridges the gap between strict data protection laws and the urgent need for data-driven insights. MedModels plans to expand the Patient Synthesizer in the coming months with modules for personalized medicine and AI training to better serve the needs of the global pharmaceutical industry.