Purpose:
BECAUSE-ICU will build the first iteration of a data warehouse with large-scale, real-world data from Danish intensive care units, in an OMOP common data model. Then, we will use the data warehouse to replicate results of previous clinical trials, and predict the results of ongoing or imminent trials.

BECAUSE-ICU will continue current work on a proof-of-concept extract-transform-load pipeline that transforms complex source data from machines and manual registrations, stored in tens of thousands of files, into an OMOP common data model. The data warehouse will live on a secure high-performance computing infrastructure and follow data engineering best practices, including version control and logged access control. Code and software will be containerised and shared under lenient licenses.

Real-world causal evidence will be generated with a variety of conventional and machine learning methods for causal inference, using both frameworks built specifically for OMOP’ed data and general-purpose frameworks. We will build undirected graphs of head-to-head comparisons of interventions and exploit the network metanalytic framework to synthesise one effect-size estimate for each exposure-outcome pair and scrutinise the results by comparing direct with indirect evidence.

BECAUSE-ICU is a true (clinical) data science project combining methods from DataOps, machine learning, data visualisation and clinical epidemiology.

BECAUSE-ICU builds the foundation for a lasting data infrastructure, built to scale and give faster answers to more questions using better data.

BECAUSE-ICU will alleviate challenges of current clinical trials in critical care by identifying performant enrichment and stratification schemes.

BECAUSE-ICU code and software will be open source so others can reuse and repurpose our work for their needs without locking themselves into ecosystems of specific vendors.

BECAUSE-ICU will (also) enable efficient build-up studies to identify and prioritise promising candidates for Intensive Care Platform Trial domains (see below) through fast cohort characterisation and causal inference estimation. This is important as only 10-15% of current critical interventions rest on high-level evidence.