Lecture
Prognostic Patterns - Can we predict Disease from Data?
- at -
- ICM Saal 4a
- Type: Lecture
Lecture description
The conventional evaluation of laboratory results is typically based on a one-by-one comparison of each result and its corresponding reference interval. However, the multidimensional interplay among numerous laboratory parameters is expected to contain far richer diagnostic information. In this context, we are conducting the PREDICD study, which is based on anonymized data collected retrospectively from about one million hospital patients who underwent at least one laboratory test between 2014 and 2024. In total over 200 million laboratory test results and more than eight thousand main and secondary ICD-10 diagnoses were included. The present study explores and compares machine learning (ML) approaches for the identification of latent diagnostic patterns by integrating laboratory values with demographic information and ICD-10 diagnoses. The objective of this study is to develop an ML-supported model capable of generating probabilistic diagnostic suggestions directly from a patient's laboratory profile, including rare and under-researched conditions. Novel associations between laboratory results and the manifestation of disease identified by this approach will be systematically compared with existing diagnostic signatures and validated according to clinical relevance. Ultimately, the insights gained by the PRED-ICD study may form the basis for an expert decision-support system, thereby further refining data-driven diagnostic approaches and individualized therapy.
