gms | German Medical Science

Background: Big Data methods in medicine are booming. We investigate which requirements have to be fulfilled for a successful application of such methods in clinical research, whether these are already realised and which future possibilities Big Data offers.

Results: Medical prediction models are much more complex than pattern recognition models and proof of clinical relevant performance has still not been given. Consideration of the intended patient population and randomized follow-up studies are key to promising translations to clinical surroundings. Without validation the results of Big Data studies should be interpreted with care. One example, where further studies are indicated, are polygenic risk scores (PRS). These summarise a patient's genetic risk to a specified common human disorder. Even though none of the scores was properly validated, neither prospectively nor by drawing upon existing long-term studies, and despite their obvious lack of predictive power, clinical use of the PRS for individual disease prediction has since been strongly advocated by some. Other important aspects for Big Data methods are causality versus correlation and the differentiation between available and suitable data. Not for all relevant research questions, large datasets are at hand, especially not for early stages of drug development. We give one example, where only 0.003 per cent of the Big Data datasets fitted the specific study aim. Nowadays and in the near future, large, high-quality datasets including thousands of patients are rare. These are, however, necessary to perform reliable and meaningful Big Data analyses in medical research.

Conclusion: Unsustainable promises and unfulfillable expectations should be avoided in the context of Big Data and be replaced by realistic views and evidence-based conclusions. The advent of Big Data in clinical research will increase the understanding that medicine is, and always has been, a data-based science. This will improve efforts to generate structured, standardized large datasets. These datasets will also include new data types like omics data, images or digital endpoints. This will lead to enhanced research quality and will consequently benefit the patients.

Hans-Ulrich Burger works at Hoffmann-La Roche AG, Basel.

Dietrich Knoerzer works at Roche Pharma AG, Grenzach.

The authors declare that an ethics committee vote is not required.