From Continuous Biological Sensing to Precision Pharmacotherapy: An Observability-Aware Dynamic Patient–Drug Modelling Framework

Abstract

Background. Precision medicine has historically relied on intermittent measurements, such as a
laboratory panel, a genetic test, or a clinic visit, to characterize a patient who is, in reality, a continuously
changing physiological system. Problem. Continuous biosensing, artificial intelligence (AI), digital
twins, multi-omics, and model-informed precision dosing (MIPD) are each advancing rapidly, and several
recent frameworks already combine them, including temporal causal inference on routine physiological
data, state-space digital twins for critical-care precision dosing, and proposed multi-omics/AI/digital-twin
architectures for predictive, preventive, personalized, and participatory (P4) medicine. This heterogeneity
of partially overlapping efforts makes it easy to overstate novelty and difficult to see what remains
structurally unresolved. Approach. This review maps the relevant literature against a common
architecture and argues that the unresolved problem is not integration but biological observability: the
gap between what is measured and what is physiologically true. Framework. We propose a conceptual,
mathematically explicit synthesis, the Observability-Aware Dynamic Patient–Drug Model (OAD-PDM),
which treats the patient as a partially observed dynamical system whose latent physiological state is
estimated from heterogeneous, asynchronous observations together with an explicit uncertainty and
observability annotation, and which couples that state bidirectionally to mechanistic
pharmacokinetic/pharmacodynamic (PK/PD) models rather than to an unconstrained predictive layer.
Evidence base. We illustrate the framework using three domains at different maturity levels: closed-loop
insulin delivery (clinically established), hybrid mechanistic-machine-learning vancomycin dosing
(actively converging, supported by recent comparative studies), and continuous gastrointestinal sensing
for oral drug absorption (experimental, demonstrated only in animal studies to date). We position the
framework against its closest existing counterparts rather than claiming to have invented latent-state
modelling, digital twins, or AI-assisted dosing. Limitations. The OAD-PDM framework is conceptual. It
has not been implemented or clinically validated, several of its proposed outputs lack established
estimation procedures, and its translational value depends on sensor reliability, external validation, and
formal uncertainty quantification that do not yet exist for most of the biological domains it spans.
Conclusion. We close with a staged, deliberately conservative research roadmap and an explicit statement
of limitations.