Reproducing a Prospective Clinical Study as a Computational Retrospective Study in MIMIC-II

Area of Science:

• Clinical Informatics
• Computational Medicine
• Health Data Science

Background:

• Disseminated intravascular coagulation (DIC) is a complex clinical condition.
• Reproducibility in clinical research is crucial for validating findings.
• Electronic Health Records (EHRs) offer vast potential for retrospective studies.

Purpose of the Study:

• To computationally reproduce a prospective clinical study on DIC.
• To validate the use of EHR data for replicating clinical research findings.
• To assess the feasibility of using the MIMIC-II database for DIC research.

Main Methods:

• Developed SQL and Java electronic phenotypes to mirror prospective study's data sampling.
• Utilized R for statistical inference procedures identical to the original study.
• Employed the MIMIC-II database for a retrospective computational analysis.

Main Results:

• Identified 2,257 eligible patients within the MIMIC-II database.
• Achieved remarkable agreement between computational results and the prospective study's findings.
• Found that while some data elements were missing, statistically significant inferences were possible in most cases.

Conclusions:

• Computational reproduction of the prospective DIC study using EHR data was successful.
• EHR databases like MIMIC-II are valuable resources for validating and extending clinical research.
• This methodology supports the use of retrospective EHR studies to confirm findings from prospective trials.