Work

Analyzing determinants of biological age through systematic review of prediction models

Year: 2026

Type: article

Abstract: Biological age can be estimated in many different ways, most often from biomarkers measured in routine blood tests. The aim of this systematic review was to identify data-driven models that estimate biological age from parameters obtainable in routine clinical care, to characterise the variables these models use, and to appraise how close such models currently are to clinical implementation. The review was reported in accordance with the PRISMA 2020 reporting guideline. Two authors independently searched PubMed, MEDLINE/CINAHL Ultimate, Web of Science and Scopus, and the SAGE Journals publisher platform, from database inception to 9 September 2026. The search was conducted in two waves: an original search executed on 19 September 2024 and covering the period up to that date, and an updated search executed on 9 September 2026, using a broadened string, covering the period since. Google Scholar was screened for grey literature. Studies published in English were eligible if they described a computational model estimating biological age in adults, listed the variables entering the model, and used at least one variable obtainable from a routine blood test; the last of these was an eligibility criterion and not a variable that was prioritised in the synthesis. Models based exclusively on DNA methylation or on other omics platforms, on imaging, or on perceived age were excluded. Risk of bias was assessed with PROBAST. Twenty-three primary studies were included in the quantitative synthesis: nine from the database searches, six from the grey literature search and eight from the updated search. Two further reports met the eligibility criteria but were review articles rather than primary studies; they were excluded from the synthesis and are discussed narratively. Across the 23 primary studies, 473 variable entries corresponding to 153 distinct variables were extracted and grouped into five categories: blood tests ( n = 352), other examinations and tests ( n = 67), body composition measurements ( n = 45), cognitive and physical measures ( n = 5) and other, non-routine molecular biomarkers ( n = 4). The most frequently used variables were creatinine (18/23), albumin (17/23) and total cholesterol (13/23) among blood tests, followed by alkaline phosphatase, glucose, HDL-cholesterol and white blood cell count (all 11/23), and C-reactive protein, mean cell volume and red cell distribution width (all 10/23). Systolic blood pressure (12/23) and diastolic blood pressure (6/23) were the most frequent among other examinations, and body mass index (8/23), height, weight and waist circumference (all 5/23) among body composition measurements. The sets of variables used to estimate biological age vary considerably, and the models published since 2024 are no exception, but a small core of inexpensive, routinely available laboratory parameters recurs across most models. This core could realistically be captured in electronic health records and used to compute biological age at the point of care. Before that is justified, the field needs standardised biomarker panels, external validation in independent populations, and evidence that acting on a biological age estimate changes clinical management and improves outcomes. At present such evidence is lacking, and biological age should be regarded as a risk-stratification research tool rather than an actionable clinical result. Not applicable.

Source: BMC Medical Informatics and Decision Making

DOI: 10.1186/s12911-026-03902-6