How old are you, really? Communicating chronic risk through ‘effective age’ of your body and organs

Given a specified measure of ‘risk’, your effective age can be defined as the age of a typical ‘healthy’ person who matches your risk profile. So if your chronological age is 50, but your effective age is 60, this means that you are in the same risk category as a 60-year old who has ‘healthy’ risk factors, or at least the ones that are potentially modifiable.

The generic idea is shown in Fig. 1: a reference trajectory of a typical ‘healthy’ person is calculated, and then a subject’s actual risk level is mapped across to the find the age of a ‘healthy’ person with the matching level.

The same idea can be used for specific organs of the body. In particular, many cardio-vascular risk calculators will provide your ‘heart age’ or a ‘vascular age’ [10], and examples use a variety of risk measures:

The concept of lung-age was developed over 30 years ago [15]: using updated equations, calculators [8] derive the age for which your lung function results would be expected in a typical person of your age and height.

There are two crucial decisions in defining an effective age. The first is the specification of a ‘healthy’ person: for example, the Heart Age calculator on NHS Choices [7] uses a person of the same gender and ethnicity, and who is a non-smoker with no clinical conditions, a BMI of 26, systolic blood pressure 120, and LDL/HDL cholesterol ratio of 3.5. This choice of a ‘standard’ individual can have a major impact on the properties of the procedure. The ‘Ubble age’ calculator [6], based on nearly 500,000 UK Biobank participants [16], uses the population 5-year all-cause mortality as the reference level, but most of this risk comes from the few individuals who are already sick, and so greatly overestimates the risk of the vast majority of the population: it is one of those anomalous situations where nearly everyone is less than average. The consequence is that almost everyone’s Ubble age is considerably less than their chronological age – often around 10 years. In this case a better comparator may have been the 5-year all-cause mortality of a healthy population, or the median rather than mean risk.

The second decision is the choice of risk measure, and in particular whether it represents an instantaneous risk or ‘hazard’, or is defined as the chance of an adverse event of a fixed horizon, say 10 years. There is a strong relation to the work of Brenner and colleagues [17], who consider the difference between a person’s effective age (although they do not use this term) and their chronological age as advancing the time that they may suffer an adverse event, which they call an ‘advancement period’. If the risk is an ‘instantaneous’ measure, they term this a ‘rate advancement period’, and if it is a risk over a fixed horizon, say the chance of developing cancer in the next 10 years, they term this a ‘risk advancement period’.

For example, Liese and colleagues [18] used Rate Advancement Periods (RAPs) to communicate increased heart-attack risk from hypertension (RAP = 8 years) and smoking (RAP =11 years): the latter says that, compared with never/former smokers, smokers are expected to advance their risk of myocardial infarction approximately 11 years.

The difference between effective and chronological age could be called the ‘years lost/gained’ - for example if you are 50 years old, but your effective age is 60, you have ‘lost’ 10 years. It would be attractive if this difference did not depend on chronological age, so individuals could be told how specific behaviours added or subtracted so many years from their effective age: for example, that smoking 20 a day added 8 years to your effective age, however old you are at the moment.

Suppose we are concerned with a risk measure R, a chronological age t, and an exposure x, coded 1 if exposed, 0 otherwise. Then Brenner et al. [17] show that if there is an increasing function g such thatthen an individual with the risk factor (x = 1) and chronological age t ₀ will have the same risk as someone without the risk factor (x = 0) and age t ₁ = t ₀ + b/c, since

Thus under these conditions the ‘risk advancement period’ t ₁ − t ₀ = b/c does not depend on the chronological age t ₀.

We’ve also seen that, for example, different versions of ‘heart-age’ can depend on your risk of a heart attack this year, over the next 5 years, or the next 10 years. It would be desirable if this risk ‘horizon’ were irrelevant.

It turns out that there are reasonable assumptions under which both your years lost/gained do not depend on your current age, and your effective age does not depend on the risk horizon. In the Additional file 1 we prove that both these attractive properties hold given the following two conditions:

These conditions are equivalent to assuming, in Eq. (1), that h is the instantaneous risk of an event, and g is the logarithmic function, so that log(h) = a + bx + ct, or

Thus e ^b and e ^c are the hazard ratios associated with a unit increase in the risk factor, and a year of ageing, respectively.

Fortunately these two conditions are reasonably plausible in a wide range of situations. The first condition is the standard assumption in the Cox regression model used widely in epidemiological studies, while we now show that the second condition of an exponentially-increasing hazard can be assumed to hold over a wide age-range, at least for all-cause mortality.

Figure 2 shows the annual risk (or hazard) of death for an average person during each year of age in England and Wales between 2010 and 2012 [19] – this is traditionally known as the ‘force of mortality’.

We note the steep decline once one has survived early childhood, to a minimum of less than 1 in 10,000 at around 10 years old (in spite of the anxieties expressed about vulnerable children, nobody in the history of humanity has been as safe as a modern Western 10-year-old). Then, apart from a ‘risk-taking’ bulge between 15 and 25, particularly in young men, there is a reasonably straight line until around 95. This linear growth on a logarithmic scale corresponds to an exponential growth in annual hazard, meaning the risk of dying each year increases by a fixed proportion for each year we age, exactly the condition we seek.

We can check this observation by plotting the ratio of hazards for adjacent years (Fig. 3), which shows that the relative increase is fairly constant between ages 30 and 95 – this observation that dates back to Benjamin Gompertz in 1825 [20]. The increase in annual hazard associated with ageing 1 year is roughly around 1.1, which is e ^c in our previous notation - this means that the average chance of dying before a next birthday increases by around 10 % for each year of ageing, whether a man or a woman and regardless of age, if over 30. Equivalently, the average risk of dying before a next birthday doubles roughly every 7 years. Demographers have concluded that this ratio, 1.1, seems to be remarkably constant across populations and over time [21].

We now show that, given the two conditions of proportional and exponentially increasing hazards, we obtain an elegantly simple way of converting the estimated effects associated with different exposures to years of life lost or gained.

Standard output from epidemiological studies includes the estimated all-cause mortality hazard ratio, which is the increased annual risk of death for those with the risk factor compared to those without. The estimates are derived from Cox regression and apply over the range of ages being studied. The second column of Table 1 shows some examples from recent epidemiological studies.

The crucial insight is that these hazard ratios can be easily translated into changes in effective age. Suppose r is the hazard ratio associated with a risk factor, where r = e ^b and so b is the log(hazard ratio) obtained in a Cox regression analysis. The second condition says that each increased year of age is associated with a hazard ratio e ^c , and so the increased risk associated with the risk factor is equivalent to a change of t years in your effective age ifor equivalentlythe result derived in Eq. (1). Since c ≈ ln(1.1) ≈ 0.1, we have that t ≈ 10 ln(r). Applying this formula to the risk factors in Table 1, and rounding to the nearest integer, provides the values in the column headed ‘Change in effective age’. Translating each risk factor into a change in effective age provides a simple, but fairly rigorous, form of communication: for example, 2 h TV a night is associated with the same increased mortality risk as if you were 1 year older, for any age for which this hazard ratio holds.

There are three important caveats. Confronted with Table 1, the immediate temptation is to start adding the effects of the relevant exposures to get an overall number of years lost or gained. This assumes the values are all valid simultaneously, which would be true, for example, if all the hazard ratios had been calculated from a single Cox regression analysis with no interactions, However, combinations of estimates from different studies may not be too misleading provided the exposures are clearly distinct. This means there could be a reasonably sound basis for the sort of ‘real-age’ calculations shown on websites, although in practice the numbers used may often be more guesswork than based on the latest epidemiology.

Second, there must also be caution over any causal interpretation and assumption of reversibility – we should not casually say that changing behaviour will ‘take years’ off your age unless based on randomised evidence.

Finally, while we have argued that these two conditions hold for all-cause mortality, organ-specific effective age calculators are derived from other metrics such as the risk of cardiovascular events in current heart age calculators. These calculators assume a proportional hazard model and so the first condition holds, and the annual risk of cardiovascular events, while not specifically assumed to be exponential, will be strongly correlated with the risk of all-cause mortality. Thus the two conditions should roughly hold, and so the shift in effective age should be approximately independent of chronological age.

As a check, raised risk factors entered into QRISK2 that raised a chronological age of 60 to a ‘QRISK Healthy Heart Age’ of 64 were found to produce a similar difference of either 3 or 4 years when the chronological age was varied between 30 and 80.

Concepts such as effective age are clearly attractive and even gripping metaphors: a journalist recently listed the ‘age’ of each of their organs [22], and it is claimed that 27,000,000 people have been provided with their Real Age.

There is also some evidence that these ideas might help change behaviour and improve risk factors. A recent review [10] claims that vascular age is easily understood by patients and has a greater impact on care than presenting an estimated cardio-vascular disease risk score, citing a study [23] in which Heart Age “was more emotionally impactful in those participants at higher actual CVD risk levels”. A trial [24] randomised over 3000 subjects to either conventional medical advice, a risk score or Heart Age, and found that levels of metabolic parameters had improved significantly after 12 months of follow-up in both risk communication groups, but more in the group randomised to receive their Heart Age. Heart age had increased in the control group and decreased in the intervention groups.

For ‘lung age’, the Step2quit study [25] found double the quit-rates in smokers who were told their lung age (6 % vs 14 %), and this led to recommendations [26] to routinely use this concept. However a study [27] of 144 smokers cast doubt on the ability of lung-age to increase motivation, while it’s been suggested that the concept may even reduce motivation to quit for those with a ‘normal’ lung age [28].