Biological Age Revisited

Dear Editor, In “Modeling the rate of senescence(definition): can estimated biological age predict mortality more accurately than chronological age?” published in the Journal of Gerontology Biological Sciences in June (1), Morgan E. Levine compared several techniques for estimating biological age (BA). In a large sample (n = 9,389) of people aged 30–75 at baseline, from the National Health and Nutrition Examination Survey, followed for 18 years, Levine calculated BA using either 7 or 10 of 21 available biomarkers. Four algorithms for calculating BA were presented; relationships with mortality were investigated using Receiver Operating Characteristic analysis. Of the four algorithms, three were based on multiple linear regression (2,3) and one on principal component analysis (4). The regression techniques employed two algorithms presented by Klemera and Doubal (2), one of which used Chronological Age (CA) to adjust the weights of the individual biomarkers in calculating BA. The results showed relatively good performance in term of the AUC of all models (superior to CA but with no significant difference between most algorithms). We also have suggested an algorithm for estimating BA, in a 2002 article (5) cited by the author who stated, however, that because BA had been derived in an older population (aged 65 and older), it “may not be useful in examining young- or middle-aged adults and, therefore, may not be ideal for use in prevention.” As it turns out, it is easy to test such an assertion, as our algorithm was perhaps simpler than those considered in the article (1). The 2002 algorithm is based on the deficit accumulation approach, suggested by our group (6) and elaborated by others (7–12), as a way to assess health status in individuals from adolescence onwards (13). BA is calculated based on the “calibration” of the relationship between CA and the index of deficit accumulation, which we call a frailty index (FI). The FI in turn is calculated as the ratio of the deficits present in an individual to the total number of deficits available in the study (14,15). The calibration equation (5) represents the average age-specific trajectory of the FI: The values of parameters a and b are usually estimated from the data using a regression technique. Note that the slope parameter b has been estimated in several databases to be about 0.03 (14). For example, the most recent estimates obtained from the Canadian National Population Health Survey (NPHS) for people aged 15–105 years gave the estimates of a = −4.16±0.08 and b = 0.035±0.001. The BA of an individual can be estimated (we purposely avoid the term “predicted”) from inversing equation (1), given the value of the FI of the individual: Conveniently for the individual case, BA can be compared with a person’s CA. After elementary transformation of equations (1) and (2) we can write: As is evident from equation (3), BA could be equal, greater, or lower than CA. BA therefore gives a ready metric for estimating by how many years an individual is younger or older than the average person of that CA in a given population. In that sense, people who are in a worse health can be considered as biologically older compared with those who are in a better health (biologically younger). Equation (3) contains only one parameter, the slope b. As noted, even though the estimate is, generally speaking, data dependent, it still shows a narrow range of values (14). In the case of those individuals no recorded deficits (FI = 0), their BA can be calculated by substituting FI = 0 by its next minimal value, which corresponds to one deficit present. The corresponding FI could be calculated accordingly: for example, where 50 deficits are being considered, FI = 0.02, with 100 deficits, FI = 0.01, etc. From equations (1) and (2), we can see that the theoretical limit to BA (when FI = 1) is −a/b = 126 years. Because maximally observed empirical limit of the FI = 0.7, the limit to BA (from equation (2)) is 116 years. This number is close to max life span observed in human. The FI is fundamental to this method of calculating BA; it is based on a simple count of deficits that are broadly defined, but biologically/clinically meaningful. The outcomes using predictive models based on this approach are highly generalizable; they typically show superior performance of the FI compared with other methods (5,9,13,16,17) including clinical settings (18–20). For now, how the methods for estimating BA presented by Levine will perform in other databases is not clear. What should be evident, however, is that their application in other data sets will require recalculation of the weights, which they employ; this level of precision is unlikely to be generalizable, even for biomarkers. Unless the model shows stable parameter estimates across databases, the theoretical understanding of BA proposed in the article will be restricted. The goal of measuring BA is to understand “the aging of the individual and not simply to describe mortality trajectories of the population”; this ultimately motivates aging research (21). To integrate multiple biological and clinical measures in estimating individual health status will require a systems biology approach and not just statistical algorithms. A.M. and K.R. have applied to various Canadian government schemes that support commercialization of university research related to the FI.