It was 15 years ago that Professor Eric Topol, one of the most influential voices in the use of artificial intelligence (AI) in digital medicine, predicted the transformative potential that digital medical devices could have on precision healthcare.¹ This, of course, was provided that the medical profession and healthcare system would be able to step up to the challenge of embracing innovation and discovery. 

Ever since, the long-term ambition of precision medicine to replace a ‘one-size-fits-all’ approach with targeted and effective strategies for each patient has been supported by the unprecedented availability of new data, accounting for individual differences in genes, environment and lifestyle. However, because an abundance of data doesn’t equate to insight, it has become clear that outstanding challenges remain before the vision of precision medicine can be realised, particularly around multimodal data integration and biomarker analysis.

In endocrinology, numerous applications of AI have emerged, with an even larger number in development.^2–4 While AI is rapidly becoming a powerful tool in precision medicine, it alone cannot provide insight into endocrine function without theory. Sir Paul Nurse, who received the 2001 Nobel Prize in Physiology or Medicine, warned us of the need to develop theory as a framework for understanding. This, he emphasised, is crucial for the future of biology.⁵ But how can we unlock the potential of new data streams and uncover the regulation of endocrine systems in a way that leads to actionable clinical insights?

IT’S ABOUT TIME

 

Homeostasis is the process by which living beings maintain internal stability in the face of changes in their external environment. While internal stability often evokes the idea of fixed set points, the reality is that body functions are anything but static. This is particularly important for endocrine regulation, as hormonal imbalances might be better understood as departures from a healthy dynamic pattern, rather than fixed-time snapshot observations of hormonal excess or deficiency. Recognising the importance of a dynamic equilibrium to maintain health, particularly in relation to hormonal imbalances, is an essential step towards understanding endocrine dysfunction.

 

In the past decade, the ability to measure hormonal levels with unprecedented time resolution has introduced the challenge of defining, in quantitative terms, what constitutes a healthy dynamic pattern and what doesn’t. Novel wearable device technologies now allow the simultaneous collection of multimodal data at increasingly larger scales.⁶ This has facilitated the development of novel computational biomarkers that, when used in combination with machine learning algorithms, have the potential to support the diagnosis and management of endocrine disease.⁷

QUANTIFYING VARIABILITY

While computational biomarkers can offer a quantitative description of normative ranges of dynamic hormone profiles, they also highlight the problem of intra- and inter-individual variability. That is, the statistical spread of hormonal levels observed at any time varies widely in healthy individuals and depends on the time of sampling, on individual characteristics such as sex, age, body mass index, chronotype and glucotype, as well as on environmental factors. As you may already suspect, pathological states can further exacerbate variability. This raises the question of how we are supposed to develop precision medicine applications when even the ‘baseline’ variability in healthy sub-populations is rarely accounted for.

Quantifying variability is crucial to estimate the misalignment of hormonal rhythms in disease. However, this has been hindered by a lack of suitable measurements in real-life settings and by a lack of a theoretical framework to understand the dynamic effect of chronodisruptions. Recent technological advancements make it possible to address these challenges. There is now a novel ambulatory microdialysis device, worn on the waist, that enables high-frequency hormone measurements without the need for blood.⁶ There are also wearable devices that can track vital signals with unprecedented time resolution, including continuous glucose meters and finger rings that can record physical activity, heart rate variability, body temperature and sleep structure. Advancements in computational techniques, including AI and machine learning, now enable the analysis of multimodal data sets collected from wearable devices.⁸ For the first time, it may be possible to estimate the contribution of various sources of variability to dynamic hormonal profiles.

TOWARD DIGITAL ENDOCRINOLOGY

 

Historically, endocrine axes have been studied in isolation, but their inherent coupling with each other and with other physiological processes (e.g. inflammation) suggests that hormonal effects may be better understood as acting co-operatively. 

 

With multiple crosstalk points at different levels of organisation (e.g. cells, tissues, organs) and at timescales ranging from minutes to months, the networked organisation of endocrine systems enriches the diversity of dynamic responses, possibly contributing to its overall robustness to exogenous perturbations.⁹ It also makes them amenable to study via mathematical models that we can use to interrogate the processes leading to dynamic dysfunction and disease. 

Mathematical modelling is a powerful tool for understanding hormonal regulation because it requires us to describe it in non-ambiguous terms.¹⁰ By doing so, we not only test our understanding of endocrine regulation, but can also predict how targeting disrupted mechanisms can restore health. When combined with computational methods and carefully designed clinical studies, mathematical modelling can help us develop patient-specific models, or endocrine digital twins, with the potential to revolutionise precision medicine.

Mathematical models possess a remarkable ability to generalise and predict, but this doesn’t imply that they are the sole tool capable of transforming endocrinology into a mathematically oriented science akin to physics. It is not their objective either. Instead, mathematics is just another ‘microscope’ that allows us to see patterns that are not evident when other conventional tools are used. 

On the other hand, if the endocrinologist views the mathematician solely as a data analyst, or if the mathematician sees the endocrinologist as a mere data provider, a long-term, successful collaboration becomes impossible. To succeed in this endeavour, we must be willing to learn the basics of the other’s field of expertise, and to understand their philosophy, work style and problem-solving approach. The endocrinologist needs to develop an intuitive understanding of mathematical concepts, while the mathematician must acquire a comprehensive knowledge of the endocrine system that is being investigated and the available experimental techniques. A simple division of tasks is insufficient for a successful collaboration. 

The most innovative projects often begin with a novel and original question that captivates scientists across multiple disciplines. However, such questions can only arise when there exists a certain level of synergy. Fortunately for digital endocrinology, these exciting questions are abundant.

EDER ZAVALA
Assistant Professor in Mathematics, Centre for Systems Modelling and Quantitative Biomedicine, University of Birmingham

REFERENCES

1.    Topol EJ 2010 Science Translational Medicine https://doi.org/10.1126/scitranslmed.3000484.
2.    Assié G & Allassonnière S 2024 Journal of Clinical Endocrinology & Metabolism https://doi.org/10.1210/clinem/dgae154.
3.    Belkhouribchia J 2025 Frontiers in Endocrinology https://doi.org/10.3389/fendo.2025.1513929.
4.    Giorgini F et al. 2024 Journal of Endocrinological Investigation https://doi.org/10.1007/s40618-023-02235-9.
5.    Nurse P 2021 What is Life? Oxford: David Fickling Books.
6.    Upton TJ et al. 2023 Science Translational Medicine https://doi.org/10.1126/scitranslmed.adg8464.
7.    Grytaas MA et al. 2025 Research Square https://doi.org/10.21203/rs.3.rs-6095171/v1.
8.    Grant AD et al. 2022 Current Opinion in Endocrine & Metabolic Research https://doi.org/10.1016/j.coemr.2022.100380.
9.    Zavala E 2022 Journal of Neuroendocrinology https://doi.org/10.1111/jne.13144.
10.    Zavala E et al. 2019 Trends in Endocrinology & Metabolism https://doi.org/10.1016/j.tem.2019.01.008.
11.    Kim DW et al. 2020 Current Opinion in Systems Biology https://doi.org/10.1016/j.coisb.2020.07.007.