Big data hurdles in precision medicine and precision public health

Abstract

Background: Nowadays, trendy research in biomedical sciences juxtaposes the term 'precision' to medicine and public health with companion words like big data, data science, and deep learning. Technological advancements permit the collection and merging of large heterogeneous datasets from different sources, from genome sequences to social media posts or from electronic health records to wearables. Additionally, complex algorithms supported by high-performance computing allow one to transform these large datasets into knowledge. Despite such progress, many barriers still exist against achieving precision medicine and precision public health interventions for the benefit of the individual and the population.

Main body: The present work focuses on analyzing both the technical and societal hurdles related to the development of prediction models of health risks, diagnoses and outcomes from integrated biomedical databases. Methodological challenges that need to be addressed include improving semantics of study designs: medical record data are inherently biased, and even the most advanced deep learning's denoising autoencoders cannot overcome the bias if not handled a priori by design. Societal challenges to face include evaluation of ethically actionable risk factors at the individual and population level; for instance, usage of gender, race, or ethnicity as risk modifiers, not as biological variables, could be replaced by modifiable environmental proxies such as lifestyle and dietary habits, household income, or access to educational resources.

Conclusions: Data science for precision medicine and public health warrants an informatics-oriented formalization of the study design and interoperability throughout all levels of the knowledge inference process, from the research semantics, to model development, and ultimately to implementation.

Fig. 1

Revisitation of precision medicine. a Precision medicine moves from a genetics-centered personalization of treatment on to a dynamic risk assessment and optimization of current and future health status through immutable (e.g. genetics) and actionable factors (e.g. behavior). b Disease phenotypes are reclassified on the basis of new system-level evidence, identifying pathophysiological endotypes associated with common, known phenotypes. The logos are trademarks of their respective companies and institutions, and their uses do not represent affiliation or endorsement

Fig. 2

The health avatar: a virtual representation of a person with all their associated health information, and intelligent ways to manage and predict their future health status. The health avatar is centered on the personal health records and integrated with healthcare, commercial governance, and research entities

Fig. 3

Precision public health. Community, societal and ecological factors must be accounted on top of the individual-based, fine-grained approach for precision medicine. The map is an edited version of a Wikimedia Commons image (https://commons.wikimedia.org/wiki/File:United_States_Administrative_Divisions_Blank.png, licensed under the Creative Commons Attribution-Share Alike 3.0 Unported)

Fig. 4

Semantic integration on data, study design and inference. The logos are trademarks of their respective companies and institutions, and their uses do not represent affiliation or endorsement. TensorFlow, the TensorFlow logo and any related marks are trademarks of Google Inc. The logos are used for informative purposes only, and the list included here is not exhaustive

Fig. 5

The social-ecological model with associated information domains and data sources for a multi-domain study design

Fig. 6

Machine learning models: white- and black-boxes. Increasing model complexity can lead to better approximation of functions and enhance prediction performance, but can lead to a decrease in interpretability of the model

Fig. 7

Domain-guided and complexity-guided model selection. a Hypothetical data set with two domains and their merged domain, on which models of increasing complexity (linear regression, decision tree, and deep learner) are fit and compared. For example, using the receiver-operating characteristic, (b) the predictive performance of a prime model can be assessed using a single domain or merged domains, or (c) different models can be compared within the same domain space