Digital twins (DT) are defined as (physical and/or virtual) machines or computer-based models that are simulating, emulating, mirroring, or “twinning” the life of a physical entity, which may be an object, a process, a human, or a human-related feature. With the aggregation of multi-modal data such as clinical and demographic data from electronic health records (EHR), physiologic data from remote patient monitoring, and healthcare utilization data from administrative databases, DTs can advance health care efficiency to identify the right patient at the right time for the right treatment.

We have developed a health DT via deep phenotyping — the integration and automated processing of clinical data and demographics from EHR and physiologic data derived from remote patient monitoring (RPM) — and demonstrate its use for predicting future states.

We successfully developed and conducted initial validation of a fit-for-purpose digital twin closing the loop between virtual and real-world data. The approach included a combination of algorithms with complementary strengths. We apply an adaptive ensemble of approaches to improve models of future time point prediction including counterfactuals, and developed uncertainty estimations calculating both aleatoric and epistemic uncertainty for the digital twin.