Precision Medicine

Critical illness is rarely a single, uniform biological state. Patients who arrive in the ICU with the same clinical diagnosis can have markedly different underlying mechanisms, trajectories and responses to treatment. Yet critical care has often relied on protocols designed for populations rather than individuals. Precision medicine offers an opportunity to change this paradigm: not by abandoning evidence-based protocols, but by making treatment increasingly responsive to the biology and physiology of the patient.

Sepsis illustrates the challenge particularly well. Growing evidence shows that patients classified under the same syndrome can have distinct immune, metabolic, and inflammatory profiles. Advances in genomics, transcriptomics, proteomics and metabolomics are enabling researchers to identify molecular endotypes that may explain differences in disease progression and treatment response. The central challenge is no longer simply identifying subgroups, but determining which classifications can reliably identify patients who will benefit from a particular intervention (Gordon et al. 2024; Gay et al. 2026; Van Nynatten et al. 2026).

The ICU offers an environment designed for data-driven medicine. Continuous physiological monitoring generates streams of information alongside laboratory results, imaging, electronic health records and, increasingly, molecular measurements. Artificial intelligence (AI) and machine-learning methods can integrate these disparate sources to identify patterns that may be difficult to recognise using conventional approaches. Multimodal AI therefore has the potential to support earlier recognition of deterioration, phenotyping patients, and informing decisions about interventions (Yang et al. 2026; Singh et al. 2026).

Drug therapy presents another opportunity. Model-informed precision dosing can integrate patient characteristics and pharmacokinetic and pharmacodynamic information to help estimate an individual's drug exposure and guide dosing. In critically ill patients, whose physiology can change rapidly, such approaches may complement conventional dosing strategies.

However, the promise of precision medicine should be matched by scientific caution. Algorithms can inherit biases from their training data, molecular signatures may not translate into actionable treatment decisions, and apparently impressive predictive models still require rigorous prospective validation (Gordon et al. 2026; Yang et al. 2026). Precision medicine will succeed not when technology replaces clinical judgement, but when trustworthy biological and computational tools help clinicians answer a more fundamental question: what treatment is most appropriate for this patient, at this moment? The future of critical care may ultimately be less about choosing between protocols and more about intelligently adapting them to the individual.

As always, if you would like to get in touch, please email [email protected].

Jean-Louis Vincent