REVERSE model: a novel approach in medical research.

Abstract

Background: Randomized controlled trials are considered the gold standard but they are limited by high costs and external validity. The REVERSE model is introduced to address these challenges.

Methods: The REVERSE model encompasses two sequential phases. First, in the data mining phase, compatible datasets are identified and merged by using matching or stricter inclusion/exclusion criteria, thereby reducing selection bias. Second, a randomization phase addresses the inherent biases of the selected datasets. For a dichotomous scenario, the data are organized into four sub-cohorts according to the concordance with the original and new assignments: two concordant and two non-concordant. New decision factors are tested in concordant groups. Patients in non-concordant cohorts were excluded. ROMICAT-II was used to reproduce the findings from both the ROMICAT-II and ROMICAT-I trials, with results reported as the median of 10,000 applications.

Findings: The REVERSE model successfully replicated the results of ROMICAT-II and ROMICAT-I using only ROMICAT-II data. For ROMICAT-II, the median (interquartile range) of all median differences between length of hospitalization stay with cardiac computed tomography angiography (CCTA) and standard diagnostic strategy after 10,000 applications matched the trial's findings 100% of the time (18.06 h [17.76-18.32] vs. 18.1 h; p < 0.05). For ROMICAT-I, median of all REVERSE plaque prevalence (PP) at CCTA matched the observed PP at CCTA from ROMICAT-I (49.63% [48.2-51.2] vs. 49.7%). The REVERSE PP fell within 49.63% ± 5% in 9733 (97.33%) applications.

Conclusion: The REVERSE model allows repurposing existing datasets to explore novel research questions while mitigating inherent biases through stringent inclusion criteria matching and randomization.