A clinic measures blood pressure on one morning, a laboratory measures an inflammatory marker weeks later, and a researcher asks what years of exposure have done to the body. None of those observations is the same thing as a life history. The 2025 review Integrating allostasis and emerging technologies to study complex diseases asks how a theory of adaptation under demand can be studied across biological systems without reducing it to one number.
This article is a critical guide to Park and colleagues’ review in Communications Biology. Its focus is the paper’s proposals, evidence and limits, rather than a general introduction to chronic stress or allostatic load. The review ranges from neuropsychological conditions to immune disease and cancer, and it examines biomarker indices, multi-omics data and patient-derived cell models. Those methods answer different questions. Reading them as if they were interchangeable would exaggerate what the paper shows.
The central proposal
Allostasis describes regulation through change. A body does not hold every variable at one timeless value; it adjusts to demands, expected needs and environmental conditions. The associated concept of allostatic load concerns the cumulative burden of repeated or sustained regulatory activity. McEwen’s account of stress mediators helped explain why responses that protect in the short term can contribute to damage when repeatedly activated, insufficiently ended or badly timed.
Park and colleagues use this framework to ask about complex disease. Their review considers neuroendocrine, immune and metabolic processes together and argues that emerging technologies may make interactions across them easier to investigate. That is a research program. It does not mean a machine can already report a person’s total “stress burden,” or that a molecular signature has a single psychological cause.
The authors distinguish adaptation from overload. A rising hormone at the right time can be part of ordinary regulation. A persistent or poorly coordinated response may become consequential. The same biomarker can therefore have different meanings according to timing, disease, treatment and the system around it. A useful explanation must connect the measurement to a question and a time scale. A value without context is not a history of adaptation.
The review is broad and interpretive. It synthesizes work from several fields and proposes directions for future studies; it is not one longitudinal experiment spanning all the diseases it discusses. Its examples should be read according to the underlying study designs. A cell experiment can isolate one pathway. A patient cohort can describe associations over time. A clinical trial can test an intervention. None supplies all three forms of evidence alone.
From one stressor to a network of responses
The paper begins with familiar stress-response systems, including the hypothalamic-pituitary-adrenal axis and the sympathetic-adrenal-medullary system. These coordinate hormonal and autonomic changes. Their effects reach immune, cardiovascular and metabolic processes. That reach is why the allostasis framework attracts researchers studying illness beyond one organ.
But broad reach also creates an inference problem. Blood pressure may change because of activity, medication, sleep, pain, illness or the immediate measurement situation. An inflammatory marker may reflect infection, tissue injury or long-running social and environmental exposures. If a study finds both stress exposure and a marker, it must still distinguish the proposed pathway from alternatives. Temporal order, repeated measures and appropriate comparison groups matter.
The authors discuss conditions including addiction, depression, chronic infections, immune-mediated disease and cancer. They do not establish that “stress causes cancer” as a simple universal rule. Each condition has its own mechanisms and risk factors. A theory of whole-body regulation can help generate questions about those mechanisms, but it cannot replace disease-specific evidence. The paper’s value lies partly in inviting researchers to connect levels of explanation while keeping the level of each observation visible.
Consider a person living with persistent symptoms after infection. Immune activity, altered sleep, treatment, reduced income and uncertainty about the future may all influence physiological measurements. Labeling every change “allostatic overload” would hide the separate causal paths. A study needs to state which exposure was measured, which biological change followed and whether the change predicted a meaningful outcome. The framework is most useful when it disciplines those choices.
Why allostatic-load scores are difficult
Many studies form an allostatic-load index from several biomarkers. Researchers may select cardiovascular, metabolic, neuroendocrine and inflammatory measures; define a high-risk threshold for each; then add the number outside those thresholds. The resulting score can summarize a pattern across systems. It is also sensitive to which biomarkers were available and how “risk” was defined.
Carbone, Clift and Alexander’s 2022 scoping review examined 395 articles and found substantial variation in algorithms. More than half used high-risk quartiles from their own sample. A person can therefore cross a threshold because the comparison group changes, even when the underlying measure does not. Such an index may help compare groups within a study, yet it cannot automatically become a standard clinical reading comparable across places and years.
Beese, Postma and Graves reviewed reviews of allostatic-load measurement and identified variation in biomarkers as well as geographic gaps. A neighborhood exposure can matter, but administrative boundaries are imperfect proxies for lived environments. Researchers also have to consider who was sampled, what services and exposures were measured, and whether the same biomarker panel has similar meaning in different populations.
Park and colleagues explicitly call for more consistent biomarker panels. They note that narrow selections of immune measures miss aspects of immune allostasis. Standardization can improve comparison, but a universal panel is not automatically the answer to every question. The most informative measurements for an infection cohort may differ from those for a workplace study. Agreement on definitions must be accompanied by validation for an intended use.
Timing is equally important. Cortisol changes across the day. A single sample taken at an unknown hour cannot be interpreted as a complete daily rhythm. Repeated sampling may reveal dynamics that a cross-sectional snapshot misses, though it increases burden and can produce missing data. A good allostatic-load study must explain when measures were collected, which medications or illnesses could affect them and whether the outcome was measured later. Otherwise the index can appear more precise than its ingredients.
What multi-omics adds
The review describes genomics, transcriptomics, proteomics and metabolomics as ways to observe different molecular layers. Genetic variation, gene expression, proteins and metabolites are related but not identical. Integrating these data may reveal patterns that a small fixed biomarker panel misses. For example, a cohort might show linked immune and metabolic changes after a long illness.
The attraction is not simply that more variables make a better score. More data create more opportunities for spurious findings, batch effects and population bias. The person who gives a sample may differ from the population for whom an algorithm is later used. If the model is trained on one disease group, an impressive classification accuracy may fall when the setting, instruments or case definition changes. External validation and transparent handling of missing data are essential.
Park and colleagues discuss work using integrated molecular data in prolonged symptoms after COVID-19. The examples support investigation of persistent immune and metabolic patterns. They do not prove a single mechanism for every person with long COVID, nor do they turn a candidate signature into a clinical test. The relevant source studies have different samples, assays and endpoints; the review’s synthesis should not erase those differences.
Machine learning can find combinations of measurements that predict an outcome in a dataset. Prediction is not the same as explaining why the outcome occurred. A model may capture treatment history, age or collection site instead of a proposed stress mechanism. To claim mechanism, researchers need experiments, longitudinal evidence or other designs that can separate causal alternatives. The review’s promise of integrative analysis is strongest when paired with that methodological restraint.
What patient-derived cells can test
Induced pluripotent stem cells can be made from a person’s somatic cells and guided toward specialized cell types. Researchers can expose such cells to defined conditions and observe molecular or structural responses. Park and colleagues describe these models as a way to investigate adaptation under controlled conditions, including neural and organ-related questions. They also discuss organoids and multi-organ models designed to examine interactions between tissues.
A controlled model solves one problem by making an exposure and its timing more explicit. It introduces another: cells in a dish do not live within a full person with a vascular system, immune history, social environment and changing behavior. The review acknowledges that in vitro systems may lack cellular diversity and systemic interactions. A response to cortisol in cultured cells can illuminate a pathway without reproducing the lived causes or total consequences of chronic stress.
Patient-derived models also require careful comparison. Donor genetics, reprogramming, differentiation and laboratory batches can influence results. If cells from one group respond differently, the study must ask whether the difference follows the proposed exposure or another source of variation. Replication across donors and independent laboratories matters. A personalized cell line should not be mistaken for a complete prediction about that donor’s future health.
These limitations do not make the models unhelpful. They define their proper job: test a local mechanism under controlled conditions, then compare that mechanism with observations in people. The review’s most interesting research path is the connection among cellular experiments, repeated human measures and clinical outcomes, with each link tested rather than assumed.
How to read the paper’s disease examples
A reader can use four questions to evaluate any condition in the review. First, what stressor or exposure is specified? “Stress” could mean infection, inflammatory activity, economic strain, drug exposure or a task in a laboratory. Second, what time scale is studied: minutes, months or years? Third, what outcome is measured: a molecular signal, a symptom, a diagnosis or function in daily life? Fourth, which alternative explanations were addressed?
These questions prevent a common slide in language. A correlation between a biomarker and illness does not identify the marker as the source of illness. A cellular pathway activated by a stressor does not prove that the same pathway dominates every patient. An allostatic-load index associated with later health does not reveal which change would help a particular person. The paper can guide hypothesis formation while the strength of each claim comes from the underlying evidence.
For immune-mediated conditions, the authors ask whether sustained activation and feedback between systems can explain part of disease progression. For cancer, they describe possible interactions among chronic stress physiology, immune processes and tumor biology. Such connections require especially careful wording. Cancer has many causes and trajectories; the review’s systems framing should not be used to blame a patient for illness or imply that calmness is a treatment.
One practical gain of the allostasis perspective is that it draws attention to changing conditions, not just individual coping. Exposure to infection, insecure work, housing, pollution and limited care can shape physiology. These are candidate pathways to investigate, not details to be collapsed into a person’s “resilience score.” The Institute’s application is to ask which environmental and institutional conditions can be changed, then evaluate real health and participation outcomes.
What the evidence does not show
The review does not establish a standardized clinical test for allostatic load. The biomarker panels, cutoffs and algorithms are too varied for that claim without further validation. It does not show that an omics model can read an individual’s lifetime burden from one sample, or that a patient-derived organoid reproduces a whole body’s regulation.
It also does not prove that every complex disease is caused by chronic psychological stress. Biological adaptation is broad; exposures and mechanisms differ by condition. A systems framework should sharpen disease-specific research, not replace it with a single explanation. Claims about prevention or treatment require intervention evidence, including evidence about who benefits and who bears the burden.
Finally, the paper does not justify using biological data to judge a person’s character, employability or deservingness. Physiological measures can be useful in research and care when their meaning, uncertainty and purpose are clear. Without those conditions, a sophisticated dataset can become a misleading label. Privacy, consent and equitable interpretation are practical requirements of any application, though the review is primarily a research synthesis.
Return to the measurements
The clinic’s blood-pressure reading, the later inflammatory marker and the person’s account of the years between them remain three different pieces of evidence. Park and colleagues’ review proposes ways to study the connections. It asks researchers to combine systems and methods while acknowledging that the field has not settled how to measure allostatic load consistently.
The careful reader leaves with a stronger question than “How stressed is this person?”: which exposure, at what time, changed which process, with what consequence, and compared with which alternative? Emerging technologies can help answer parts of that question. The explanation still depends on study design, validation and the life the measurements are meant to illuminate.
Research references
- Park, I., Gwon, H., Jung, Y., et al. (2025). Integrating allostasis and emerging technologies to study complex diseases. Communications Biology, 8, 1526. View source
- McEwen, B. S. (1998). Protective and damaging effects of stress mediators. New England Journal of Medicine, 338(3), 171–179. View source
- Carbone, J. T., Clift, J., & Alexander, N. (2022). Measuring allostatic load: Approaches and limitations to algorithm creation. Journal of Psychosomatic Research, 163, 111050. View source
- Beese, S., Postma, J., & Graves, J. M. (2022). Allostatic load measurement: A systematic review of reviews, database inventory, and considerations for neighborhood research. International Journal of Environmental Research and Public Health, 19(24), 17006. View source
- McEwen, B. S., & Wingfield, J. C. (2003). The concept of allostasis in biology and biomedicine. Hormones and Behavior, 43(1), 2–15. View source