Calendar age is used almost universally in many models of risk factor research, largely because it is almost always known with greater precision than many other risk factors, such as environmental or lifestyle factors (overweight, physical inactivity, etc.). Information on calendar age is most important for understanding risk differences between people whose ages differ by decades, whereas for people of the same age, information on calendar age is obviously completely useless. In recent years, scientists have been trying to identify common features of chronic diseases in order to understand their shared mechanisms at the cellular and subcellular levels. As noted by American researchers James Kirkland and Tamara Chkonia of the Mayo Clinic (Rochester, Mich. Rochester, Michigan), it is becoming clear that diseases such as dementia, atherosclerosis, diabetes, cancer, arthritis, and many others have much in common in their pathogenetic mechanisms with processes associated with aging, including mild inflammation, cellular senescence, accumulation of damaged macromolecules, and stem and progenitor cell dysfunction (see Kirkland, Tchkonia 2015). Changes in body function with age occur at different rates at different levels. At the level of the individual, the trajectories of aging processes are highly divergent: individual trajectories diverge, and these divergences increase with age. Some people age faster than others due to multiple factors: heredity, environmental factors, socioeconomic circumstances. This diversity in aging rates (changes in numerous characteristics) leads some researchers to the idea that people of the same chronological age differ in their ‘biological age.’ Thus, biological age can also be seen as a metaphor for the heterogeneity of aging rates. For biological age to be more than just a metaphor, it must be measured. As early as 1955, Sir Peter Medawar noted that the lack of agreed definitions and measurements indicated the immaturity of gerontology as a science. Later, Tom Kirkwood (Kirkwood, 1988) pointed out that despite many successes, the fundamental problem of measurement had not been solved. Even now, several decades later, it cannot be said that the measurement problem has been completely solved, but significant advances in biology, in the development of modern technologies for the analysis of biomolecules, as well as mathematical modelling and machine learning methods for data analysis, make it possible to approach the problem of measuring biological processes and raise the question of the possibility of measuring biological age.

Methods for estimating biological age in an individual are based on the comparison of biomarker values in a given individual with some ‘normative’ values typical for a given population. If the value of a given biomarker in a particular individual differs from the average (for people of the same age) value of the biomarker in a given population, then we can say that the biological age of this person differs from his or her calendar age. For example, if in a 70-year-old individual the value of this indicator corresponds to the value typical for 50-year-olds, we can say that this individual is 20 years ‘younger’ than his or her chronological age. And vice versa, if the value of this indicator corresponds to the typical value of 90-year-olds, we can say that this individual is 20 years ‘older’ than his or her chronological age. The normative curve of dependence of the mean value of a biomarker plays the role of a coordinate system (reference system) for determining the measure of biological age. For most biomarkers, the dependence of the mean biomarker value on age is linear or can be linearised after a simple transformation (e.g. using a logarithmic scale). The problem is that for a number of biomarkers the difference between the values of some indicators in an individual and the ‘normative’ values for a given age may be positive, while for other biomarkers this difference may be negative. Of course, it all depends on how statistically significant this difference is, which can be assessed using statistical analysis methods. Obviously, different biomarkers, generally speaking, can give different estimates of biological age. An individual may be ‘older’ on one indicator but ‘younger’ on another. The problem is that different biomarkers have different age dynamics and it is necessary to decide how they should be ‘weighted’ in order to obtain an index of biological age. A number of ways of such weighting have been proposed, among which the most common are: multivariate linear regression and the principal component method. In recent years, due to the availability of large amounts of genetic, epigenetic, and other data, as well as the development of machine learning techniques, appropriate models have found application, which we will discuss in the following sections of this chapter. Although the necessity of using a battery of biomarkers to calculate biological age is undisputed, the idea of how many such biomarkers should be in this battery has been changing over the past few decades. In the early days, it was thought that a small number of biomarkers was sufficient (e.g., Hochschild used 12 biomarkers). In recent years, due to the development of genetic and epigenetic studies, the number of factors included in regression models of biological age (and its modifications such as DNA age) exceeds hundreds (Horvath 2013, Hannum et al. 2014, Enroth et al. 2015). This has been made possible by the development of powerful applications of multivariate statistical analysis such as elastic net, lasso-estimator, based on machine learning techniques (Friedman et al. 2010), as discussed in the following sections. How can biological age determined from biomarkers be verified? One approach to verifying biological age is to use normative survival curves as a frame of reference to construct a biological age metric. Biological age (if correctly determined) should not only predict mortality more accurately than calendar age, but ideally it should do so in accordance with Gompertz’s law when calendar age is replaced by biological age. In other words, a person whose biological age is 70, no matter what their calendar age is, should have the same mortality risk as someone whose calendar age is 70. To test this hypothesis, it is necessary to have data on a group (sub-population) of individuals with a similar biological age and to monitor the survival of this sub-population. This kind of data is difficult to obtain. Available databases usually contain information (e.g. physiological characteristics) of several thousand individuals. For such data, calculating risks consistent with Gompertz’s law is in many cases impossible. Instead, it is possible to compare the accuracy of mortality prediction by knowing both the calendar age and the calculated biological age. If the latter is more accurate in determining mortality risk than the former, this would indicate in favour of using biological age. Moreover, if calendar age does not affect the accuracy of the prediction against the background of biological age, this may be in favour of replacing calendar age with biological age. A problem is the extent to which the results obtained on a very limited sample can be generalised to different populations. Significant and coordinated efforts of different scientific teams are needed to solve this problem.

A team of researchers from Duke University in North Carolina (USA) applied the Clemer and Doubal algorithm to calculate biological age in a group of relatively young individuals (n=954) from the Dunedin Study birth cohort database. Their biological age was determined at the time they were 38 years old using 18 biological markers of ageing (see Belsky et al. 2015). Although these individuals were not found to have chronic diseases, their biological age was distributed between 28 and 61 years, indicating that some members of this cohort were biologically younger or older than others. Since biological age should reflect the changes occurring in a person, it can be assumed (as Belsky and colleagues did) that those with an estimated biological age greater than their calendar age of 38 are aging faster than those with a calculated biological age less than 38. The researchers also had access to longitudinal data on those 18 biomarkers of aging used in calculating biological age. The availability of longitudinal data corresponding to 26, 32, and 38 years of age allowed them to calculate in each individual a measure called the ‘rate of aging’ by the authors. For this purpose, the 18 biomarkers were standardised: reduced to the same value using the Z-score known in statistics (the difference between the biomarker value and the mean divided by the squared deviation) and averaged over all 18 biomarkers. From these data, the rate of aging was calculated for each individual by summing the slopes of the regression model (mixed effects of growth curves). Belsky and colleagues showed 1) individuals who had a higher than average biological age had a higher rate of aging, 2) healthy subjects with an accelerated rate of aging show impaired physical function compared to those with a lower rate of aging, 3) increased risk of stroke and dementia, and 4) were more likely to complain of poor health. This work presents a compelling argument in favour of using biological age to estimate rates of ageing. Indeed, data on people of the same calendar age show how heterogeneous they are physiologically, and their rate (pace) of aging is determined by the value of biological age. In other words, those who are biologically older age faster, which brings to mind Gompertz’s law: the acceleration of mortality with age. Work on measuring biological age will be important for assessing the health of individuals and for investigating ways to influence to improve health and slow down the processes associated with aging. This work shows how information about ageing processes can be obtained in young people and used to combat ageing. The Clemer-Doubal algorithm discussed above was used to estimate biological age. The fact that calendar age is considered by Klemera and Doubal as a biomarker of aging (i.e., calendar age is included in the definition of biological age) is, in our opinion, a serious obstacle to using this method to calculate biological age, unless one understands this kind of biological age as simply an empirical measure of population heterogeneity. The term ‘biological age’ has an aura of attractiveness, but how accurately and completely does it reflect the real biological processes underlying aging? Does this debate reflect some immaturity in the science of biological age? Despite these criticisms and questions (sometimes rhetorical), we must recognise that the results reported by Belsky and co-workers are very useful from a practical point of view. To what extent would the application of other algorithms for calculating biological age change the results obtained by Belsky and co-workers? In our opinion, if other measures of population heterogeneity (including other biological age algorithms) were used instead of this algorithm, the results would be similar. We are aware that this statement requires evidence, which will undoubtedly be obtained and published in the near future. Source: A. Fomenko, A. Baranova, A. Mitnitsky, S. Zhikrivetskaya, A. Moskalev. ‘Biomarkers of human aging’ Photo: www.candidaplan.com