Objectives: We introduced statistical twin as aggregates of multiple virtual patients' data throughout the treatment at any chosen time point. The goal of this manuscript was to provide the proof of concept of statistical twin by evaluating the feasibility of detection of distinctive aggregates of patients throughout the perioperative trajectory (prerequisite for development of statistical twin).
Methods: We used a retrospective validated cohort of all comers with mitral valve disease treated (2014-2020) at a tertiary academic hospital. The end point was overall survival based on the decision of the heart team. We applied two-step cluster analysis to detect distinct aggregated virtual patients throughout the process of care.
Results: The cluster procedure resulted in 5 distant clusters with relatively equal numbers of patients. Effects of the treatment (surgery, transcatheter or optimal medical therapy) on survival were as follows: For optimal medical therapy, the expected survival ranged from 95% to 96% in 30 days to 58% to 75% in 10 years independent of baseline characteristics. However, for transcatheter interventions, the 5-year survival was 60-92% and was dependant on the initial characteristics of the virtual patient. Furthermore, survival following an uncomplicated operation of normal duration was higher through all observation periods. The aggregated virtual patients of cluster 5 would have a better survival rate at all times if the intervention were done by a dedicated surgeon.
Conclusions: It is possible to detect distinctive aggregates of virtual patients based on baseline characteristics and to capture the impact of perioperative events and external and other factors at multiple time points throughout the postoperative phase.
Keywords: Digital twin; Predictive models; Risk scoring; Statistical twin.
© The Author(s) 2024. Published by Oxford University Press on behalf of the European Association for Cardio-Thoracic Surgery.