Statistical distance as a measure of physiological dysregulation is largely robust to variation in its biomarker composition

• Published in: PloS one Vol. 10; no. 4; p. e0122541
• Main Authors: Cohen, Alan A, Li, Qing, Milot, Emmanuel, Leroux, Maxime, Faucher, Samuel, Morissette-Thomas, Vincent, Legault, Véronique, Fried, Linda P, Ferrucci, Luigi
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 13-04-2015; Public Library of Science (PLoS)
• Subjects: Aberration; Adult; Age; Age Factors; Aged; Aged, 80 and over; Aging; Biomarkers; Chronic Disease; Chronic illnesses; Complexity; Computational Biology - methods; Computer Graphics; Cross-Sectional Studies; Data Interpretation, Statistical; Female; Frailty; Geriatrics; Gerontology; Health care; Humans; Longitudinal Studies; Male; Mathematical analysis; Medicine; Middle Aged; Mortality; Older people; Physiology; Population; Population (statistical); Population studies; Predictive Value of Tests; Prospective Studies; Sensitivity; Sensitivity analysis; Software; Statistics; Values; Womens health; Young Adult
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0122541

Physiological dysregulation may underlie aging and many chronic diseases, but is challenging to quantify because of the complexity of the underlying systems. Recently, we described a measure of physiological dysregulation, DM, that uses statistical distance to assess the degree to which an individual's biomarker profile is normal versus aberrant. However, the sensitivity of DM to details of the calculation method has not yet been systematically assessed. In particular, the number and choice of biomarkers and the definition of the reference population (RP, the population used to define a "normal" profile) may be important. Here, we address this question by validating the method on 44 common clinical biomarkers from three longitudinal cohort studies and one cross-sectional survey. DMs calculated on different biomarker subsets show that while the signal of physiological dysregulation increases with the number of biomarkers included, the value of additional markers diminishes as more are added and inclusion of 10-15 is generally sufficient. As long as enough markers are included, individual markers have little effect on the final metric, and even DMs calculated from mutually exclusive groups of markers correlate with each other at r~0.4-0.5. We also used data subsets to generate thousands of combinations of study populations and RPs to address sensitivity to differences in age range, sex, race, data set, sample size, and their interactions. Results were largely consistent (but not identical) regardless of the choice of RP; however, the signal was generally clearer with a younger and healthier RP, and RPs too different from the study population performed poorly. Accordingly, biomarker and RP choice are not particularly important in most cases, but caution should be used across very different populations or for fine-scale analyses. Biologically, the lack of sensitivity to marker choice and better performance of younger, healthier RPs confirm an interpretation of DM physiological dysregulation and as an emergent property of a complex system.