THE FRAGILITY OF THE FUTURE AND THE TUG OF THE PAST: LONGEVITY IN LATIN AMERICA AND THE CARIBBEAN

Abstract

BACKGROUND

Cohorts that will attain age 60 after 2010 in the Latin American and Caribbean region (LAC) are beneficiaries of a massive mortality decline that began as early as 1930. The bulk of this decline is due to diffusion of low-cost medical technology that improved recovery rates from infectious diseases. This changes the composition of elderly cohorts in a distinct way: more among those who could experience the deleterious impact of adverse early conditions as adults survive to attain old ages.

OBJECTIVE

To compute bounds for the size of effects on old age mortality of changes in cohorts’ composition by exposure to early conditions. We calculate estimates for countries in the LAC region that span the entire range of post-1950 mortality decline.

METHODS

We use counterfactual population projections to estimate bounds of changes in the composition of cohorts by exposure to early conditions. These are combined with empirical effects of adverse early conditions on adult mortality to generate estimates of foregone gains in life expectancy at age 60.

RESULTS

under somewhat conservative assumptions life expectancy at age 60 will at best increase much more slowly than in the past and at worst reach a steady state or decline. Foregone gains may be as high as 20% of projected values over a period of 30 to 50 years, the time it takes for cohorts that reaped the benefits of the secular mortality decline to become extinct.

CONCLUSIONS

Changing composition of cohorts by early exposures constitutes a powerful force that could drag down or halt short-run progress of life expectancy at older ages.

1. Introduction

The rate of decline of the force of mortality at older ages after 1950 in Western Europe has been estimated to be close to 1% per year (Kannisto 1994; Kannisto et al. 1994). With initial levels of life expectancy at age 60 between 15 and 20 years this rate of decline yields gains in life expectancy at age 60 of about 0.10 years per year. This type of progress is not unique. Recent evidence from countries in the Latin American and Caribbean regions (LAC) (CELADE 2007; Chackiel 2004; Palloni and Pinto 2011) confirms increases in life expectancies at age 60 from about 18 years in 1950 to about 23 in 1995 in approximately linear fashion. The rate of progress is, here again, equivalent to yearly gains in life expectancy at age 60 of about 0.11 years per year, close to the rate of change in Western Europe. This empiric fits nicely with the idea that modern human mortality is on a distinct course (Oeppen and Vaupel 2002). But there are storm clouds ahead that deserve attention.

It has been pointed out that lifestyle changes embraced by newer cohorts of elderly people both in high and low income countries could oppose strong resistance to further improvements in longevity (Olshansky et al. 2005; Preston 2005). The smoking epidemic (Ezzati and Lopez 2003; Palloni, Novak and Pinto 2012) and increases in obesity and metabolic syndrome (Aschner 2002; Hossain, Kawar and El Nahas 2007; Kain, Vio and Albala 2003; Peña and Bacallao 2000; Barcelo et al. 2003) are powerful enough to threaten even the most optimistic of scenarios.

A second hurdle to continuous survival gains that applies to LAC countries is that members of birth cohorts that attain age 60 after the year 2000 are distinct and their past health and mortality experiences could derail future progress. The size of these cohorts is boosted by the fastest mortality decline ever recorded, a result of reduction in exposure, increased resistance and declines in fatality rates of common infectious diseases. Because the bulk of these changes were not achieved through improvements in standards of living, those who benefited from the changes are more likely to have experienced early deprivation that did old-age survivors from older birth cohorts. If theories linking early deprivation and poor early conditions to adult health and mortality prove to be correct, the past experience of these cohorts could slow down or halt further reductions of morbidity and mortality at older ages.

Some of the processes invoked by these two conjectures could well be related to each other because one of the mechanisms that explains rising prevalence of obesity and diabetes is indeed the interaction of changing life styles and early exposure to deleterious health conditions (Barker 1998; Gluckman and Hanson 2006; Langley-Evans 2004). Although future improvements in medical technology could suffice to blunt the impact of these two forces we have no way anticipating their timing or magnitude. Instead it is possible to estimate the impact that changes in the composition of cohorts by life style and behaviors (e.g. smoking, obesity) or by exposure to early conditions may have on adult health and mortality. It is known that in the absence of new technologies to treat heart disease and cancers, smoking will have powerful and enduring effects on life expectancy at older ages (Ezzati and Lopez 2003; Palloni, Novak and Pinto 2012). In this paper we estimate the additional burden associated with changing composition of cohorts by past exposures to poor early conditions and its impact on future life expectancy at adult ages. We do so by calculating bounds for life expectancy at age 60 for the period 2010–2050. These bounds are computed under different assumptions about (a) the magnitude of changes in composition of cohorts by early conditions and (b) the size of effects of early conditions on adult mortality. The calculations combine estimates of effects of early conditions on excess adult mortality risks from survey-based information with counterfactual population projections. We apply the procedure to selected countries in the LAC region that together span the entire range of mortality decline that took place after 1930: Argentina, Brazil, Costa Rica, Chile, Mexico, and Guatemala. The results suggest that changes in cohort compositions could have a durable influence on future older age mortality.

In Section 2 we provide background and identify factors involved in the relations of interest. In Section 3 we review the evidence for a linkage between early and late adult health, the “early-late health connection”, theorize about the operation of different mechanisms linking early and late health under a variety of mortality regimes and formulate a testable conjecture. In Section 4 we summarize results of a simple numerical exercise to illustrate the relevance of the main conjecture. In Section 5 we describe a method to assess the effects of past mortality and early-late health connection on life expectancy at age 60 and discuss empirical estimates. Section 6 concludes.

2. Background: Past and future survival gains

Despite recent bumps, the march towards longevity in most world regions appears to be uncontested (Oeppen and Vaupel 2002). The HIV/AIDS epidemic and the collapse of the Soviet Union are the only two blemishes that led to massive retrenchments of life expectancy in large populations. But elsewhere progress overwhelms occasional fluctuations and, in most cases, it is so imposing that we can easily ignore temporary and localized faltering in an otherwise impeccable record. But underneath the veneer of invincibility in this march towards longer life, hides a more complicated reality. The course of future mortality in most populations depends on two sets of factors. The first is related to prospective improvements in medical knowledge and technology that could reduce case fatality of chronic diseases, including heart disease, stroke, and some cancers. By and large the reductions in adult mortality experienced in the last two decades are largely attributable to innovations that increase early detection, improve prevention, and broaden access and adherence to improved treatments of some chronic diseases.¹

The second set of factors is related to changes in cohort composition of the elderly population. The distribution of deaths in modern mortality regimes is shifted toward older ages, with gradual increases in the modal age at death and plummeting variances (Cheung and Robine 2007; Kannisto 2001; Thatcher et al. 2010; Wilmoth and Horiuchi 1999; Horiuchi and Wilmoth 1998). Because of losses in the curvature of the survival function, future gains in life expectancy can only occur at older ages and will require larger proportionate decreases in age-specific mortality rates. And therein lies the rub: since the history of exposures and experiences of a cohort constitutes the raw material with which their mortality in later life is sculpted, it is at least possible that patterns of mortality rate accelerations, decelerations, or steadiness in the future will reflect the pulsation resulting from the inflow of cohorts with heterogeneous histories. This malleability of older age mortality patterns is a result both of exposures to past behaviors and constraints imposed by the nature of mortality regimes that different birth cohorts’ experience.

2.1 The influence of past behaviors

Different cohorts are exposed to risks influenced by the adoption of behaviors whose effects are visible only after considerable time lags. Patterns of smoking are the most obvious ones so much so that they explain between one and two thirds of the difference in life expectancy at age 50 between the US and other high-income countries (National Research Council 2010) and translate into four to five years of life expectancy losses in LAC countries (Palloni, Novak and Pinto 2012). These estimates are precise because of the tight connection between lung cancer, COPD and smoking. Diet and physical activity and the resulting trajectory of obesity are another obvious illustration. Indeed, it is believed, though not conclusively proven, that set-backs in the progress toward higher life expectancy in the US and other high income countries could soon emerge as a direct result of increasing obesity prevalence at all ages (Olshansky et al. 2005). But, unlike smoking, the impact of obesity is unclear because the relation between it, chronic ailments and excess mortality is not as tight as that between smoking, cancers, COPD and cardiovascular diseases.

2.2 The legacy of the past

A second mechanism that changes the composition of birth cohorts is less obvious and applicable mostly to countries whose secular mortality decline is recent and dominated by the revolution in medical technology that began after World War I. A large fraction of individuals born after 1930 in LAC and in the majority of low to medium income countries elsewhere, were able to survive beyond early childhood not because of improvements in standards of living and nutritional status but as a result of the spread of knowledge about infectious diseases, vector eradication campaigns and, most importantly, the diffusion of chemotherapy and medical technologies that decrease the lethality of infectious diseases. While exposure to diseases was also reduced through intelligent use of germ theory, large public health programs, and significant expansions of infrastructure (albeit with a great deal of regional variability), the main engine behind increases in life expectancy between 1930 and 1970 was increased resistance and recovery induced by the introduction of novel chemotherapy, including antibiotics and sulfa (Arriaga and Davis 1969; Stolnitz 1965; Palloni and Wyrick 1981; Palloni and Pinto 2011; Preston 1976, 1980). This suggests that members of cohorts born after 1930–1940, who will attain their sixtieth birthday after the year 2000, are more likely than members of previous birth cohorts to have experienced early conditions known to be associated with higher risks of adult chronic conditions, such as congestive pulmonary diseases, heart diseases, diabetes and other chronic conditions that dominate the landscape of mortality at adult ages (Barker 1998; Gluckman and Hanson 2006; Langley-Evans 2004). Apart from advances in medical technology, there are two factors that determine whether or not these early experiences express themselves as higher incidence of chronic diseases: (a) the strength of the linkage between poor early conditions and adult chronic illnesses; (b) the nature of the exposure to poor early conditions. Different mixtures of these factors generate variation in the impact that cohort compositional changes could have on ill-health and mortality of older cohorts.

3. Mortality regimes and early-late life connection

How strong is the early-late health connection? Do the effects of early conditions on adult health and mortality change across mortality regimes? If so, how?

3.1 The strength of early-late health connections

Accumulated evidence confirms that there are a number of mechanisms through which early childhood conditions affect the onset of adult chronic illnesses and, in particular, adult diabetes (type II), congestive pulmonary disease, and heart disease. Some of these mechanisms are highly specific, such as those associated with processes that start in utero, develop shortly before and/or around birth (“fetal origin hypothesis”) or during other critical periods (Barker 1998; Gluckman and Hanson 2006). Chronic illness induced after long lags by contraction of infectious diseases early in life is another example of a highly specific mechanism. Thus, for example, it is well-known that adult heart valve malfunction (aortic and mitral valve stenosis) develops late in life among individuals who contracted rheumatic fever (a streptococcus bacterial infection) during early childhood or adolescence. A somewhat different, less specific, set of pathways involves the delayed effects of inflammatory processes triggered by recurrent exposure to and contraction of infections and parasitic diseases during early ages (Crimmins and Finch 2006; Danesh et al., 2000; Fong 2000; Finch and Crimmins 2004).² And, finally, more diffuse and highly non-specific are pathways operating through socioeconomic conditions during early childhood, including stressful experiences, deteriorated environments, persistent poverty and deprivation (Ben-Shlomo and Smith 1991; Danese et al. 2007; Dowd 2007; Elo 1998; Elo and Preston 1992; Hertzman 1994; Kuh and Ben-Shlomo 2004; Lundberg 1991; Smith and Lynch 2004)³.

Empirically distinguishing between these various mechanisms is a thorny affair because, with some exceptions, they all lead to the same implication, namely, that the erosion of conditions that foster repeated exposure to and contraction of infections and parasitic diseases and/or maternal and child nutritional status could simultaneously reduce early childhood and subsequent adult and older age mortality of a birth cohort.

The existence of an early-adult health connection implies that if successive birth cohorts are exposed to changing conditions early in life, their older age morbidity and mortality are, to some degree at least, dependent on conditions set forth by their early experiences. If so, there is potential for within-cohort associations between mortality and morbidity at early and older ages. In such cases assessment of future changes in health status and mortality could be partially supported by examination of past levels and patterns of child mortality and health status. We argue below that the magnitude and direction of the association is strongly dependent on the regime of mortality decline: it is the nature of this regime that widens or narrows the opportunity for the expression of early-late health connections.

3.2 Regimes of mortality decline and the expression of early-adult health connections

Even if the empirical evidence for an early-late health connection were uncontroversial, it should not always generate lagged effects at older ages. Besides the obvious impact of advances in the medical diagnoses and treatment of chronic illness, there are other conditions that could blunt the adult health expression of early exposures. These relate to the nature of the mortality decline that enables a higher proportionate representation at older ages of individuals who, in the absence of mortality decline, would have never survived in the first place.

To facilitate discussion, Table 1 displays a cross-tabulation of types of mortality decline and early-late health connections. The horizontal axis identifies three mechanisms that could produce the early-late health connection. The first operates through impaired fetal, prenatal and postnatal growth and is associated with nutritional status (Barker 1998). The second depends on the onset and development of well-defined illnesses during early childhood that generate tissue damage or disable immune response but whose repercussions are seen only in adult life (Elo and Preston 1992). The third is unspecific and consists of the adult response to early exposure or contraction of an array of infectious illness that trigger sustained and enduring inflammatory responses (Crimmins and Finch 2006).⁴

Table 1.

Early-late health connection and types of mortality decline

Early-Late Health mechanisms and Mortality regimes
Type of early-late health connection(*)
	Nutritional Status	Experience with particular diseases (rheumatic heart fever/Helycobacterium pylori)	Experience with broad range of diseases (infections of digestive and respiratory tracts)
Mortality Regime
Standards of living	[1](++)	[2](?)	[3](+)
Public Health	[4](+)	[5](+)	[6](+)
Medical Innovations	[7](−−−)	[8](?)	[9](−−−)

(*) Numbers within squared brackets are cell numbers; symbols in rounded brackets represent the direction of change in adult mortality associated with the regime: + indicates that the mortality decline at older ages will be boosted; − indicates that the mortality decline will tend to be offset and, finally, ? indicates an indeterminate sign. A repeated sign (++ or −−−) signifies a strong relation

The vertical axis identifies mortality regimes according to the dominant force accounting for mortality decline: improvements in standards of living, public health or medical technology and innovation. No observed empirical case can implicate only one mechanism producing the early-late health connection any more than no observed mortality decline can be driven by economic advances to the exclusion of medical technology or public health. However, historical cases could be assigned to the single cell that best captures their main observed features and the typology could help us identify weak and strong forces that relate early exposures and older age health and mortality.

First, we situate ourselves in cell 1 representing the combination of changes in mortality driven by improvements in standards of living (primarily maternal and child nutritional status) with the prevalence of classic Baker effects. Birth cohorts born after the onset of the mortality decline should experience lower excess incidence of adult chronic conditions related to adverse fetal growth than cohorts born before the onset of the decline. Old age mortality at 60 or so years after the onset of mortality decline should decrease solely as a function of this cohort composition change. However, there is also a second order, weaker, effect inducing an increase of risks at older age since individuals who were better nourished as children are better equipped to survive the onslaught of early childhood infectious diseases to begin with. This will augment the pool of older people who were exposed to but survived early childhood problems and who will drag along the scars left by their past histories.

In cell 2 the regime of mortality decline is as before but the dominant early-late health connection operates through chronic illnesses associated with infectious diseases such as hepatitis B (liver cancer), helminthes and streptococcal infections. To the extent that better standards of living diminish exposure to infectious diseases (better housing, better clothing), older members of the cohorts exposed to these improvements will, on average, be less likely to manifest chronic conditions mediated by the second and third mechanism than cohorts born before the mortality decline. Since, in addition, improved nutritional status associated with better standards of living will defuse the classic Barker early-late health connection, the net impact at older ages will be to reduce old age risks of expressing the early-late health connection. As before, improved nutritional status is a double-edged sword and has a second order (weaker) influence as it tends to expand the representation at older ages of individuals who experienced early childhood infections that cold potentially damage adult health status.

Finally, cases in cell 3 correspond to a scenario where recurrent infections translate into persistent inflammation and higher risk of cardiovascular diseases, among others. A mortality decline driven solely by better standards of living will have a positive effect on older age mortality as reduced exposure usually follows improvements in standards of living. Here again, there is a second order negative effect as reduced exposure to illnesses leads to gains in nutritional status resulting from a lighter load of infections (Scrimshaw 1997; Scrimshaw and SanGiovanni 1997). This will expand the fraction of the older population who can express infectious conditions contracted early on.

Row 2 represents the experience of many high-income countries. In most of these the initial push toward massive mortality decline is rooted in a public health revolution that begins tentatively around 1850 and then becomes firmly anchored in germ theory after 1880. Public health interventions that reduce exposure to infectious and parasitic diseases will have the same effects irrespective of the dominant mechanism producing early-late health connection: they will diminish the inflammation load over a lifetime and reduce the risk of associated adult chronic conditions. Because there is a synergistic relation between nutritional status and infections (Scrimshaw 1997; Scrimshaw and San Giovanni 1997), public health interventions that reduce exposure can indirectly increase nutritional status and contract the pool of individuals who experience deleterious conditions in early childhood. This will attenuate the effects of the early-late health mechanism.

Row 3 refers to situations where mortality decline is driven by medical improvements that enhance resistance and recovery and improvements in survival are less due to decreased exposure than to decreased lethality of illnesses. Treatment with antibiotics, for example, is part of an arsenal of medical tools massively deployed after 1940’s that reduced the fatality rates of the most common infectious diseases such as typhoid and respiratory TB. But so is vaccination, the application of which reduces contraction rates even in the absence of changes in exposure. These two medical innovations, chemotherapy and vaccination, that propelled, albeit with different timing, the post 1930’s secular mortality decline, have opposite effects on the composition of cohorts by experience with early childhood diseases. Because of the synergisms between infectious diseases and nutrition, lower contraction rates (due to vaccination) and better resistance or recovery (due to medical treatment) could translate in better nutritional status even in the absence of increases in nutritional intake associated with raising standards of living. The relative composition of birth cohorts by exposure to early conditions will thus depend on the relative importance of these two factors. Since dissemination of new therapies preceded by almost two decades the widespread use of vaccination (Minna Stern and Markel 2005; Allen 2007) gains in survival during the initial phases of the mortality decline will increase the pool of individuals who at older ages are at risk of expressing the early-late health connection.

3.3 Latin America and the Caribbean

Historical conditions in LAC and in most developing countries fit in rows 2 and 3 of Table 1.

Although there is substantial variability in the time of its onset, many countries in the LAC region began an uninterrupted and sharp mortality decline around 1930 and, most definitely, after 1950. The forerunners in the decline (Argentina, Uruguay, Costa Rica and Cuba) are somewhat exceptional in that their trajectories resemble more the Western European style of mortality decline than their neighbors’ experience. For the first 20 to 30 years, the largest fraction of the decline is associated with decreases in mortality during early childhood mortality and the sharp rate of mortality decline coincides with a period during which medical treatments of infectious diseases (sulfa, antibiotics) make their debut in the area and begin to be widely used. Empirical investigations show that some of the improvements are associated with the deployment of health tactics (mostly vector eradication) that diminished exposure to some infectious and parasitic diseases (Stolnitz 1965; Arriaga and Davis 1969; Palloni and Wyrick 1981; Preston 1976).

With rare exceptions, raising standards of living and improved nutrition played only a minor role in the initial phases of the decline. Although these inferences are based on coarse estimates, they are robust enough to classify the LAC experience squarely in row 3 and, more weakly, in row 2. These estimates notwithstanding, the classification retains some ambiguity. First, while the deployment of new medical technology might indeed reduce lethality of infectious diseases (and therefore increase the pool of older people who can express the early-late connection), there will also be spill-over effects since the reduction of intensity and duration of illnesses indirectly improves nutritional status (thus blunting Barker’s mechanism). Second, vaccination campaigns, a rather late addition to the tool-kit of medical innovations deployed after 1950, act entirely through reduction of exposure and therefore diminish the load of infectious diseases and, again, enhance nutritional status. Both of these are second order effects and are likely dwarfed by the more direct effects of medical technology, namely, increased survival to older ages of higher risk individuals.

In summary, available evidence suggests that the experience of LAC belongs mostly in row 3 of Table 1. Considerations related to synergisms raise questions about the existence and importance of second order effects that depend on the importance of improvements in nutritional status as a secondary driver of the mortality decline. Choosing conservatively, we classify the LAC experience midway between rows 2 and row 3. If we adopt a broad interpretation of mechanisms for the early-late health connection we could exclude column 2 from the table. This leads to the following:

If second order effects associated with the synergism between infectious diseases and nutritional status are strong, then birth cohorts whose survival is enhanced by mortality decline will experience an added boost to their old age survival probabilities (cells 4 and 6). If, however, synergisms are weaker, these birth cohorts will experience excess old age mortality due to chronic illnesses that express exposure to early conditions, (cells 7 and 9).

Although this conjecture is not directly testable with existing data,⁵ Section 5 estimates bounds for the mortality effects of changes in the composition of older cohorts by early childhood experienced after 1950. We show that under realistic conditions the combination of early-late health connections and types of mortality decline in LAC lead to tangible impacts on mortality levels at older ages.⁶

4. Conventional and Barker-frailty

We argue that when a downward shift in mortality favors the expression of early-late health connections the composition of successive birth cohorts will change: younger cohorts who benefit more from the mortality decline will contain a larger fraction of individuals at higher risk of expressing early conditions and, consequently, to higher mortality. Older age mortality of cohorts born more recently will reflect not just period-effects but also the influence of cohorts’ compositional changes by early childhood exposures.

These compositional changes are similar but distinct from changes in composition by frailty invoked by a conventional frailty model (Vaupel et al. 1979; Vaupel and Yashin 1987). This model is based on the assumption that individuals are born with a random frailty value, γ, (γ>=1), such that higher values increase individual lifetime mortality relative to a standard in the same proportion at all ages. The distribution of γ is defined at birth and changes over the life of a cohort as members endowed with higher values of γ die out first. The results from this model are two. First, when the mortality regime and frailty distribution at birth are invariant the average mortality rates at older ages will increase more slowly than the individual mortality rates. Second, if the population experiences mortality declines and the frailty distribution at birth remains invariant, any improvement in mortality will increase the average level of frailty at older ages, i.e., the cohort will be, on average, more frail at all ages than it would have been in the absence of mortality changes or more frail than its predecessors. These effects will underplay the observed rates of mortality decline at all ages, but especially late in life.

The changes that interest us are a result of a different type of frailty, Barker-frailty⁷ for short: these are changes in the composition of cohorts by susceptibility to express exposure to early conditions that manifest themselves only at older ages and only as excess mortality due to a handful of chronic illnesses. For simplicity, assume that individuals are characterized at birth by a random Barker-frailty value, ε >=1, so that higher values signify increased susceptibility to express the early-late health connection. Elsewhere we formally demonstrate two properties of a mortality regime driven both by conventional and Barker-frailty. First, when there is no change in the mortality regimes nor in the conventional and Barker-frailty distributions, older age mortality rates will operate subject to laws of conventional frailty but mortality at older ages will be higher than expected under the conventional frailty model. This is because individuals who are scarred by early conditions, ε>1, and survive to older ages experience higher excess mortality at those ages than they did at younger ages. Unlike the conventional frailty factor, γ, Barker-frailty, ε, increases mortality rates relative to a standard set of rates and does so disproportionally at older ages. The net result is that the concavity of the mortality curve will be sharper than expected under a conventional frailty model but still duller than the concavity of the individual mortality rates.

Second, if mortality declines at all ages and frailty distributions are invariant, mortality rates at older ages could decline more slowly, not at all, or even increase as a result of changing composition of cohorts by ε, e.g. with a progressively larger influx of individuals with higher values of ε that boost mortality risks at older ages. If there were no links between early conditions and adult health and mortality, changes in cohort composition by ε would have no effects at all on older age mortality and only the impact of changes in conventional frailty composition, γ, will be observed, namely, a slower average mortality decline than the one actually affecting individual mortality. The tighter the link between exposure to early conditions and older age mortality is, the stronger the departure of older age mortality rates from the regime predicted by a conventional frailty model.

The operation of Barker-frailty requires two conditions. To simplify their description we assume that cohorts can be divided in two groups: one composed of individuals who would not have survived under the pre-transition mortality regime, A, and one with individuals who would have survived anyhow, B. The first condition is that mortality levels experienced before some target age, say 60, are higher in group A than in group B during the initial phases of mortality decline. The second condition is that differences in mortality at older ages (over 60) between A and B are at least in part the result of exposure to early conditions whose effects manifest mostly if not solely late in life. We refer to this mortality excess as Barker-effect.⁸ In most empirical cases changes in survivorship experienced during the early stages of the secular mortality decline are proportionately larger in subpopulations more likely to experience poor early conditions or higher values ε, so that their average value at any age before older ages will increase over time as a product of gains in survival.

4.2 A simple numerical example

To provide a sense of magnitude and some insight on the relations described above we briefly discuss the results of a highly stylized numerical example. The main features of the exercise are as follows: (a) individuals in the population are characterized by a latent variable ε with a Gamma distribution f(ε) with parameters 2 and 1 (mean and variance equal to 2). Whenever ε exceeds a threshold value (to be determined) the individual experiences poor early conditions and can potentially express excess mortality risks; (b) the mortality decline is defined so that the mortality rate at age y and t years after the onset of mortality decline is μ(y,t) =g(t)* μ_s(y,t), where μ_s(y,t) is a standard mortality rate and g(t) a (positive valued) function declining on t; individuals who belong to group A experience mortality rates equal μ_A(y,t) =g(t)* μ_s(y,t)*λ₁ at all ages below 60 and equal to μ_A(y,t) =g(t)* μ_s(y,t)*λ₂ at 60 and above with λ₂> λ₁. Members of group B experience μ(y,t) =g(t)* μ_s(y,t); (c) we set values of λ₂ and ₁ be as small as 1.5 and as large as 5; (d) we compute 100 cohort mortality “regimes” mimicking the mortality decline in the LAC region from 1941 to 2000, and then from 2000 to 2040 following the profile of UN mortality projections.

Figures 1a through 1c illustrate selected results of the calculations for a scenario where the fraction of individuals experiencing early conditions is about .20 (ε>=3), λ₁ =2 and λ₂ =4.⁹ Figure 1a shows the initial distribution of ε. Figure 1b displays survival probabilities at the onset and at the end of the mortality decline (life expectancies equal to about 45 and 80) and for members of group A in each case. Figure 1c displays the average values of λ₂ at age 60 throughout the decline. This figure shows the effects of mortality decline on compositional changes across cohorts over the period of the decline. In year 0 the values of λ₂ are lower than 1.75 (the average value at time 0 among those aged 60) irrespective of age. In years 50 and 75 (two highest lines) the mean values of λ₂ increase substantially reflecting extra survival among individuals born with poor prospects, e.g. higher values of λ₂.

Figure 1.

a: Distribution of values of Epsilon (early conditions)

b: S(y)- survival probabilities to age y at year 0 and 100

c: Average values of Lamda2 over time

d: Differences between average and baseline mortality

Finally, Figure 1d displays the average mortality rates and the standard mortality rates (group B) at age 65, 75 and 85 over the 100 years of observation. This is the ultimate outcome of changes in composition induced by mortality decline, viz., the average mortality (ages 65, 75 and 85) does not progress at the same rates as the baseline hazards rates do, and instead within some periods of time at least, does not progress at all or even increases.

Although this exercise is limited to only one out of many possible scenarios, the results are more general. Indeed, one can show that if, as assumed above, early conditions is treated as a binary variable, a mortality decline resembling the one experienced in LAC countries will produce results consistent with the conjecture formulated above as long as the fraction in group A exceeds .15 and the values of ₁ and λ₂ are larger than 1.5.¹⁰

5. Empirical bounds for the effects of the early-late health connection

Estimates from generalized stable populations indicate that between 30% and 40% of the rate of increase in the population aged 60 above between the years 2000 and 2020 in the LAC region is associated with the post-1940 mortality decline (Palloni et al. 2006). Thus the rate of increase in the number of elderly in the region is, in part at least, the product of augmented survival among individuals who were exposed to or who experienced infectious and parasitic illnesses but who survived them as a result of bolstered recovery rates. In the post-2000 period an increasing fraction of elderly will belong to birth cohorts whose members survived infectious and parasitic diseases that prior to the mortality decline would have killed them. To the extent that a nontrivial part of this mortality decline is dependent on the efficacy of chemotherapy, the fraction of adult individuals in a cohort who are likely to have experienced suboptimal nutrition and frequent episodes of infections and parasitic diseases during childhood will increase over time. Furthermore, if Barker-effects are strong, the prevalence of adult chronic illnesses and associated mortality risks will also increase.¹¹ Under these conditions life expectancy and healthy life expectancies at older ages could increase more slowly or cease to increase altogether even if ‘background’ mortality (i.e. mortality unrelated to early conditions) continues to decline. Below we calculate bounds for these effects and show that under plausible conditions they can significantly decelerate the rate of decline of older mortality.

5.1 Estimation of population exposed and excess mortality risk

We need to estimate two quantities. The first is the fraction of the population 60 and over that is more likely to be at risk of expressing early in life as adult chronic disease. This population survived health regimes that combined poor nutritional status and high exposure to infectious diseases with lower fatality rates. We call this the “exposed” population. The second is the Barker-effect or the excess mortality among those who are at risk or the exposed population. We use these two quantities to compute attributable mortality and the magnitude of years of life expectancy losses.

Population exposed

The subset of the population aged 60+ at time t created anew by the mortality decline that took place starting in 1930 is a function of improvements both in early childhood mortality (0–4) and at adult ages (5–59). The magnitude of the mortality changes experienced in most low-income countries, particularly in LAC, is quite large and an important fraction of it takes place between ages 0 and 5. Under a steady birth cohort size the improvement experienced between 1950 and 1990 will, on average, represent extra growth of the population at age 60 of about 53 percent.¹² This figure is an upper bound since successive cohorts did not experience all at once the mortality change that took thirty to forty years to unfold.

We can be more precise and partition the improvement into several components. Let B(t) stand for the size of a birth cohort, say in 1950; let S(1,O) and S(1,N) be the probabilities of surviving from birth to age 5 in the pre and post-transitional mortality regimes respectively, and S(2,O) and S(2,N) the conditional survivorship from age 5 up to age 60 exactly in the pre- and post-transitional mortality regimes, respectively. Finally, S(O)=S(1,O)*S(2,O) and S(N)=S(1,N)*S(2,N) are the unconditional probabilities of surviving from birth to age 60 in the pre and post-transitional mortality decline. The total (observed) population at age 60 at time t+60 is given by

$$P_{\mathrm{o}}\left(t+60,60\right)=B{\left(t\right)}^{\ast }S\left(N\right)=B{\left(t\right)}^{\ast }S{\left(1,N\right)}^{\ast }S\left(2,N\right)$$

The total (counterfactual) population that would have reached age 60 under the pre-transitional mortality regime is

$$P_{\mathrm{c}}\left(t+60,60\right)=B{\left(t\right)}^{\ast }S\left(O\right)=B{\left(t\right)}^{\ast }S{\left(1,O\right)}^{\ast }S\left(2,O\right)$$

The difference between P_o and P_c which we denote by Δ, is the sum of the weighted contribution of two components: one equivalent to the contribution of changes in survival between age 0 and 5 and the other the equivalent of the contribution of changes in the conditional survivorship from age 5 up to 60.¹³ The first component represents the growth of the population at age 60 that would have been observed if only early childhood mortality had changed and the survivors had been exposed to an adult pre-transitional mortality regime. The second component represents the growth of the population at age 60 that would have been observed if the mortality shift had only affected survivorship over age 5 but survival between birth and age 5 had followed the old mortality regime. The quantity Δ represents the population at age 60 added by the mortality decline in the interval (t, t+60). Alternatively, Δ is the population aged 60 in (t+60) that would not have been alive had it not been for the mortality improvements that took place between 1950 and 2010. The quantity Δ does not only include those who were saved because of changes in mortality between ages 0 to 5 but also those who survived because of mortality improvements at older ages. The quantity Δ can be calculated exactly by comparing the results of two population projections: one using the observed mortality schedules and one using the past (unchanged) mortality schedules.¹⁴

Δ is not a good measure of the size of the population exposed to the risk of expressing early-late health connections because it also includes a subpopulation that was saved due to mortality decline associated with improvements in standard of living and public health. This segment of the population is less susceptible to express Barker effects and we should be exclude it from the population exposed. To estimate these two subpopulations composing Δ we employ two procedures that can be used either jointly or separately. The simplest of these is to estimate the fraction of all mortality decline associated with improvements in income and living standards (as opposed to the diffusion of medical innovations and public health) and to partition Δ accordingly. The second is to use information on the relative contribution to the mortality decline of various causes of deaths. Since some causes of deaths are much more responsive to medical technologies than others (Palloni and Wyrick 1988; Preston 1976) one can use this knowledge to estimate bounds for the fraction of the total population ‘saved’ by the mortality decline that was more likely to have experienced poor early conditions. The first procedure (factor specific) requires knowledge of the fraction of the total mortality decline due to medical innovations and improvements in standard of living. The second procedure (cause-specific) is based on counterfactual projections by causes of death that result in estimates of the population ‘saved’ associated with each cause-of-death specific mortality decline. Finally, one could combine these two procedures and estimate cause-specific-factor-specific components. In what follows we only use the first procedure.

The above identification of subpopulations is not sufficient since we still need to assume that all individuals whose survival chances were ameliorated by medical technology are equally likely to express the effects of early conditions. This is too crude. By the same token, it would be too coarse to assume that all those whose survival chances were augmented by improvements in standards of living are not at risk of expressing effects of early conditions. In the absence of additional information we assume that in each case there is a lower and upper bound for the probability of expressing poor early conditions. As a lower bound we use the average proportion of low birthweight infants born during the period 1960–1980 (.10) The upper bound is the fraction of children who are stunted within the age interval 1 to 10 in a poor population (.18 in INCAP, Guatemala). Admittedly these are coarse bounds and introduce unavoidable uncertainty in our estimates.¹⁵ We now define the nature of the mortality excesses associated with the experience of early conditions.

Excess mortality risks or Barker-effect

The second parameter required to estimate the desired bounds is the magnitude of excess mortality risks associated with exposure to early conditions. In principle these excess risks are a function of the type of mechanism linking early conditions and adult morbid conditions: intrauterine deprivation could increase the risk of heart disease which, in turn, is translated into excess mortality due to ischaemic heart disease. But an alternative mechanism, such as malnutrition during the first five years of life may influence the likelihood of contracting diabetes 2 and produce excess mortality due to kidney failure or circulatory impairments. The mortality risks are different in each case and a precise estimation requires an assessment of the distribution of individuals according to their exposure to each of these distinct risks. Although this information is unavailable, we do have access to estimates of total excess mortality risks among those who experience some type of deprivation early in life. We assess excess mortality risks associated with early deprivation by estimating logit models (with suitable controls) for the probability of dying at ages above 60 using data from multiple wave panel surveys of elderly people in three LAC countries. These models yield remarkably consistent estimates of excess mortality, with relative risks contained within the range 1.5–2.5¹⁶

5.2 Computations

Following the rules established in 5.1 above, a birth cohort is first partitioned into 2 subpopulations: one corresponds to the population that would have been observed irrespective of mortality decline. This represents a fraction 1-Δ (baseline) of the birth cohort; the other subpopulation can only be observed as a result of mortality decline and represents a fraction Δ (counterfactual) of the birth cohort. Each of these two subpopulations is composed of a fraction whose survival was due to medical innovations and the complement, 1-λ, who survived due to improvements in living standards. Finally each of these four subpopulations is divided into a fraction that expresses early conditions, ϕ, and the complement, 1-ϕ, that does not. The total number of subpopulations thus created is 8 and for each cohort aged x at time t they will represent a fraction d(x,j,t) of the total birth cohort. Thus, d(x,1,j), the first subpopulation defined above, represents a fraction equal to (1-Δ)* λ* ϕ; d(x,2,t), the second subpopulation defined above, represents a fraction equal to (1-Δ)* λ* (1-ϕ), and so on (see Figure 2). Each of these subpopulations is associated with mortality risks that are solely dependent on whether or not the subpopulation expresses the effects of early conditions. If they do, they will experience a mortality excess equal to μ_o exp(β); if they do not they will experience background mortality, μ_o. The bottom of Figure 2 displays the 8 components of a birth cohort and identifies the mortality levels that each of them experiences. The terminal point of each branch in the figure is associated with the mortality risk characteristic of each subpopulation. The overall force of mortality at any age 60 and over and time t is calculated by averaging the mortality rate across all 8 subpopulations:

$$\mu \left(x,t\right)={\mu }_o\left(x,t\right){\sum }_{j=1,8}(d{\left(x,j,t\right)}^{\ast }\mathit{exp}({\beta }_j))$$

where μ_o (x,t) is a baseline mortality risk at age x and time t, d(x,j,t) is the distribution of the birth cohort aged x at time t by subpopulations, and exp(β_j) is the relative risk associated with subpopulation j.

Figure 2.

Decomposition of a birth cohort exposed to secular mortality decline driven by improvements in nutrition and medical technology

5.3 Simplifications

To simplify computations we introduce several assumptions. First, the only quantity that must per force change over time is the fraction ‘saved’ since this quantity depends on population projections. All other quantities, particularly the distribution of the population by early conditions, are assumed constant from 2000 onward. With one exception this assumption does not do violence to our argument and relaxing it will not impact the calculations in any significant way. The assumption will be unrealistic if the shift in mortality occurs simultaneously with, anticipates with a short lag, or even triggers a decline in fertility. This by itself is not sufficient to weaken estimates. Violation of the assumption will damage the estimates if the fertility decline so induced by mortality changes produces better birth outcomes and/or promotes higher quality investments in children. In this case one can no longer assume that the distribution by early conditions remains unchanged.

Second, the counterfactual projection sets all mortality rates between ages 0 and 60, not just infant and child mortality rates, to their 1950 levels. This means that we will include among those who could express early conditions at older ages a subgroup who survives the first five years of life but who would have died before age 60 under the old mortality regime. Because these individuals are, by definition, less likely to have been exposed to adverse early conditions we will exaggerate the effects of cohort changes. However, since mortality between ages 5 and 60 is low relative to child mortality, the bias will be small.¹⁷

Third, we will err on the conservative side and assume that the population size of the cohort component who would have survived in the absence of a mortality decline (component in the lower half of the graph), will not experience mortality excesses at all. This is a whopper simplification and can alter the bounds in significant ways but always in the direction of downplaying the effects of cohort compositional changes.

5.4 Data and baseline estimates

We proceed in two stages. First, we project forward the population aged 50 and over starting in 1950 and ending in 2050. We do this using UN-projected life tables for 2010–2050 (United Nations 2002). We then repeat the projection but keeping mortality at the same levels as in 1950. This leads to estimates of differences between the observed and counterfactual population for every year of the projection. Figure 3 displays the ratios of the observed to the counterfactual populations in the countries of interest for 2000–2050. The figure shows what we would have expected: countries with more recent mortality decline have much smaller counterfactual populations (Guatemala). Indeed, the rank order of the curves in Figure 3 corresponds exactly to the ordering from early to late mortality decline.

Figure 3.

Ratio of the total population aged 60+ from counterfactual projections (mortality at ages 0–59 kept at 1950 levels) to the coventional projected population

The second stage consists of calculating estimates of the excess mortality due to poor early conditions. The computations were carried out with data from three recent surveys of elderly people: MHAS (Mexican Health and Aging Survey), CRELES (Costa Rican Longevity and Health and Aging Study) and PREHCO (Puerto Rican Elderly Health Study). The procedure consists of estimating logistic models for mortality over age 60 over an inter-wave period that varied between 2 years (MHAS) and six years (PREHCO). The model includes an indicator of exposure to poor early conditions, controls for age, gender, education and SES. We then use the estimated effects of the variable for early conditions as a measure of the Barker-effect. In all cases the coefficients were properly signed and statistically significant. The relative risks were as small as 1.20 (CRELES) and as large as 1.9 (PREHCO).¹⁸ Needless to say these are not ideal estimates but coarse approximations because identification of early conditions by older respondents has not been validated and we know little about their correspondence with the early conditions that are of real significance (except perhaps for anthropometry that enables identification of early malnutrition). Even if our indicators were accurate, the estimates from the logistic models could be subject to errors due to misspecification of the model. In compensation though, it should be noted that we use a sample of old age survivors that underrepresents the population who experienced the worst conditions. As a consequence of this the estimated effects will be downwardly biased.

5.5. Results

We now calculate the mortality rates above age 60 and corresponding life expectancy at age 60 for several counterfactual scenarios. For each scenario we proceed as follows:¹⁹

1. we determine the counts of the ‘saved’ and ‘not saved’ populations by age. We call these S and NS;

2. we partition the saved population into a component that owes its survival to medical technology S*λ and another that owes its survival to socioeconomic conditions S*(1-λ);

3. each of the above populations is in turn divided into two subpopulations: one exposed to early conditions, S*λ *ϕ₁ and S*(1-λ) *ϕ₂ and another not exposed to early conditions, S * λ *(1-ϕ₁) and S*(1-λ) *(1-ϕ₂) where we allow the possibility that those saved by forces other than medical conditions have a lower probability of experiencing early conditions (ϕ₁>ϕ₂).²⁰

4. we calculate mortality rates for each of the subpopulations created above. We use as a baseline the UN mortality rates for 2000 and adopt the rates of decline assumed by the UN from 2000 though 2050. For each subpopulation exposed to early conditions we increase the baseline mortality by the relative risk or mortality excess estimated before.²¹

5. for each quinquennial period we calculate the average mortality rates across all subpopulations identified above and the associated life expectancy at age 60.

Figures 4a and 4b display the values of life expectancies at age 60 under three different scenarios for the two countries that produce extreme results, Argentina and Guatemala. At the beginning of the period under examination there can be only small differences between alternative scenarios since the fraction of the total projected population ‘saved’ must be very small. When the influx of cohorts born after 1950 into age groups over 60 begins to grow, the different scenarios start diverging and yield different life expectancies. A mild scenario (with low values of λ, ϕ, and relative mortality risks among those exposed to adverse conditions of about 1.5) applied to Argentina produces differences of the order of .1 to . 4 years of life expectancy at age 60. A harsher mortality regime among those exposed to deleterious conditions (relative risks of the order of 1.9) leads to larger differences, about .66 years of life expectancy. These values are small since Argentina’s mortality decline began very early, twenty to thirty years before and had somewhat different roots; as a consequence, the population saved after 1950 is much smaller and exposure to excess adult mortality is minimized. On the other hand in Guatemala, a country more typical among those that started the mortality decline after 1950, the potential losses in life expectancy at age 60 are larger. A benevolent set of parameters produces differences of about .80 years. But when those exposed to adverse early conditions are subjected to a harsher mortality regime the differences are as large as 1.5 years or close to 8 percent of the total life expectancy at age 60.

Figure 4.

a: Life expectancy at age 60 under different scenarios

b: Life expectancy at age 60 under alternative scenarios

Overall these are not massive differences but neither are they trivial. Consider the following: during the period 2010–2050 Guatemala is projected to add 2.3 years of life expectancy at age 60; thus, the potential losses calculated here are about half of the total projected gains. In Argentina the potential losses are, on average, one sixth of the expected gains. By the same token during the period 1980–2000 life expectancy at age 60 increased by about 5 years among the forerunners of the mortality decline (Palloni and Pinto 2011). Figures 4a and 4b reveal that as little as 8% and as much as 20% of this gain can be forfeited in a span of 50 years solely as a function of changes in cohort composition by exposure to early conditions.

Figures 4a and 4b correspond to an optimistic scenario, one where an entire section of the population, the “not saved” (NS), including individuals who would have survived in the absence of a mortality decline are willy-nilly spared exposure to early conditions and/or to excess mortality associated with them. This is a rather extreme assumption and surely leads to under-estimates of the effects we seek to identify. Thus the theoretical ‘losses’ displayed in Figures 4a and 4b must be considered as somewhat conservative, at least within the framework proposed here.

6. Conclusions

The estimates described before constitute bounds for the impact of changes in cohorts’ composition by early exposure on future life expectancy in the LAC region. These bounds and the implications for future life expectancy rest on two fundamental conditions. The first is that experiences with poor early conditions, from maternal health right before and during pregnancy, in utero growth and developmental impairments, malnutrition during the first five years of life, adversity and exposure to infectious diseases early in childhood, all leave an imprint that could be expressed as higher risks of adult chronic conditions and excess mortality risks. It must also be the case that even if the early-late health connections exist, they will not suppressed in the future by advances in medical knowledge or technology.

The second assumption is that the regime of mortality decline that a country experiences acts as a sorter, sifting through members of birth cohorts and enhancing survival to adulthood among those who, under conditions that prevailed in the past, would not have survived their first few years of life or, if they did, would not have attained their 60^th birthday. Furthermore, it must be the case that none of the mechanisms producing an early-late health connection are blunted by feedbacks embedded in the regime of mortality decline, for example, by strong synergisms between nutrition and susceptibility or resistance to infectious diseases.

If both conditions apply then the massive mortality shifts that took place after 1950 could have lasting consequences and may constrain future gains in life expectancy at older ages. We used rough procedures and computed coarse bounds to assess how strong the tug of the past could be. Our results are somewhat speculative but suggest that the impact of forces set in motion by mortality decline could be strong enough to require substantial background mortality improvements to offset the legacy of past mortality regimes.

Because our estimates of excess mortality associated with early conditions are conservative, the bounds we compute are likely to be too small. There is considerably more uncertainty surrounding the estimates of the size of the population that could manifest Barker frailty. In this case even our low bounds could exaggerate the final effects. These two sets of errors, one affecting estimates of excess mortality among an exposed population and the other affecting the estimates of the size of this population, are of opposite signs and will offset each other. The result could be that our overall bounds may be just about right. But this, of course, cannot be confirmed until we observe the actual mortality trajectory of the cohorts involved.

A final caveat: the effects we refer to in this paper are not the same as those that emerge as a result of cohorts’ past exposures and behaviors that may, but need not, be connected to early experiences, such as smoking and obesity. These trends are a separate threat of their own, and their impact will also take a long time to dissipate. Naturally, the joint occurrence of changing composition of cohorts by early exposures, on the one hand, and by past life style risks, on the other, constitutes a powerful combo that could drag down or arrest altogether short-run progress in life expectancy at older ages.