Population-level impact of adverse early life conditions on adult healthy life expectancy in low and middle-income countries

Abstract

Evidence from theories of Developmental Origins of Health and Disease (DOHaD) suggests that experiencing adverse early life conditions subsequently leads to detrimental adult health outcomes. The bulk of empirical DOHaD literature does not consider the nature and magnitude of the impact of adverse early life conditions at the population level. In particular, it ignores the distortion of age and cohort patterns of adult health and mortality and the increased load of chronic illness and disability that ensues. In this paper, we use a micro simulation model combined with empirical estimates of incidence and prevalence of obesity, type 2 diabetes (T2D) and associated disability in low- and middle-income countries, to assess the magnitude of adult delayed effects on healthy life expectancy and on compression (expansion) of morbidity at older ages. The main goal is to determine if, in what ways, and to what an extent can delayed effects due to early conditions influence a cohort’s profile of chronic illness and disability.

2. Introduction

Sustained increases in life expectancy during the twentieth century and rapid improvements in survival at older ages after World War II led to unprecedented increases in life expectancy at adult ages. This phenomenon sparked the emergence of a literature on the possible association between mortality and morbidity patterns. In the late 1970’s and early 1980’s researchers began questioning whether increases in survival were also accompanied by additional years of life in good health. Three competing theories offer explanations and predictions. First, the “failure of success” (Gruenberg 1977) approach posits that increases in survival would be accompanied by increased disease prevalence due to higher survival of individuals affected by poor health status. Second, the original compression of morbidity conjecture (Fries 1980) argues that the age of onset of chronic disease would be shifted to older ages at a faster pace than improvements in survival leading to shorter duration of morbidity (i.e. producing “compression of morbidity”). Finally, a third hypothesis, dynamic equilibrium (Manton 1982), suggests that the more likely scenario is a blend of scenarios predicted by the failure of success and compression of morbidity, that is, one in which the prevalence of chronic disease increases but does so hand in hand with a slower progression of the severity of illnesses.

The empirical evidence gathered in the last twenty or so years is mixed. Some studies point to the existence of compression (Cai and Lubitz 2007; Fries et al. 2011; Zhang et al. 2019) whereas others provide ammunition to demonstrate the opposite, namely, an expansion of morbidity (Seeman et al. 2010; Crimmins and Beltrán-Sánchez, 2011; Smith et al. 2013). Most of the original controversy and the bulk of empirical evidence is about conditions in high income countries, those where the onset of mortality decline dates back to the second half of the nineteenth century (Vallin 1991). Largely absent from the discussion is the trajectory of morbidity and disability in low- to middle-income countries, all of which experienced mortality declines whose timing and nature are radically different from those in high income countries.

Our objective in this paper is to theorize about what the evolution of morbidity could be in countries where the secular mortality decline is much more recent and took place in a highly contracted period of time. In particular, we ask whether conditions sustaining improvements in mortality are also propping amelioration of healthy life expectancy, shifting upward the age of onset of chronic illness and, more generally, driving a gradual compression of morbidity (and disability)?

The scarce empirical evidence available for low- to middle-income countries sides with the more pessimistic scenario and suggests that length of life with morbidity (and possibly disability) may be on the rise. Evidence from Mexico (Payne and Wong 2019), for example, suggests that recent cohorts of adults born in 1943–1952 lived significantly fewer years of healthy life at ages 60–69 (free of disability) than those born in 1932–1941. Additional evidence from the WHO-SAGE study in five low and middle-income countries in 2007–2010 (China, Ghana, India, Mexico, and South Africa) shows that people aged 50 or older are expected to live substantial fractions of their remaining life with disability in activities of daily living (Santosa et al. 2016). In India, for example, adults age 50 or older were expected to live about half of their remaining life with disability (53% and 34% for men and women, respectively), while in Mexico, Ghana, and South Africa the corresponding fraction is about 20%. Research from the Global Burden of Disease indicates that from 2005 to 2013, the number of disability adjusted life years (DALYs) increased for most specific non-communicable diseases, including cardiovascular diseases and neoplasms, particularly in less developed countries (Murray et al. 2015). In addition, between 1990 and 2013 and for both males and females in low-income countries, there was a faster increase in overall life expectancy than in healthy life expectancy, suggesting an expansion of morbidity (Murray et al. 2015). We argue below that a factor that could explain these empirical patterns is the emergence of adult delayed effects resulting from adverse early life conditions among members of the cohorts that attain older ages after the year 2000. We show that adult delayed effects are powerful forces that might generate expansion of morbidity among older adults even under conditions favoring continuous improvement in survival.

The goal of the paper is to articulate and provide evidence for a conjecture linking theories of adult delayed effects embodied in the Developmental Origins of Health and Disease (DOHaD) and theories about compression of morbidity. We do this by generalizing an existing formal demographic model (Palloni and Beltrán-Sánchez 2016, 2017) of relations between exposure to adverse early conditions, on one hand, and adult mortality on the other. The generalized model includes morbidity and disability in addition to mortality. In particular, we add consideration of two conditions that are increasingly common in low- to middle-income countries and are important targets of DOHaD theories (type-II diabetes (T2D) and obesity). The model extension enables us to assess the impacts on older age healthy life expectancy, on the distribution of ages of onset of illness and, finally, on compression/expansion of morbidity and disability. Inclusion of ill-health and disability contributes to the literature on compression of morbidity and mortality (Gruenberg 1977; Fries 1980; Manton 1982) and introduces an entirely novel element, namely, the role of birth cohorts’ early conditions on later life disease and disability burden. While predictions about within- cohort association between child conditions and old age mortality has been an object of study in the literature on cohort mortality (Finch and Crimmins 2004; Crimmins and Finch 2006; Beltrán-Sánchez et al. 2012), we are unaware of empirical research on within-cohort association between early childhood conditions and adult morbidity and disability.

The plan of the paper is as follows. Section 3 is a brief excursion through the central propositions of DOHaD theories. Section 4 identifies predictions about the impact of adverse childhood conditions on morbidity and disability. Section 5 describes a multistate model that includes two conditions and disability and introduces the skeleton of a microsimulation model. Section 6 summarizes results regarding years of life in health, illness and disability. Finally, section 7 summarizes, identifies limitations, and proposes a handful of generalizations.

3. DOHaD and DET theories and mediating mechanisms

3.1. A brief tour

In the last decade or so research guided by the Developmental Origins of Health and Disease (DOHaD) paradigm and related theories has grown rapidly (Gluckman and Hanson 2005; Bateson and Gluckman 2011; Rosenfeld 2016). While the original idea of delayed adult effects of early conditions has been around for quite some time in very disparate strands of literature (Derrick 1927; Kermack et al. 1934; Frost 1939; Preston and De Walle 1978; Kuh and Smith 1993), it was Barker’s work on “fetal programming” that provided the initial impetus for the creation of what is now a very active field of empirical research (Barker et al. 1989; Lucas 1991; Barker 1998; Hales and Barker 2001; Barker 2012).

3.1.1. Terminology

For the sake of terminological compactness, we adopt the following conventions. First, we use throughout the term “delayed effects” as shorthand for “adult delayed morbidity, disability and mortality effects” associated with one or multiple mechanisms that link early conditions and adult diseases and mortality, irrespective of the nature of initial conditions that triggered the adult response. Delayed effects also encompass risks of disability if and when the chronic illnesses involved do so. Second, we use the acronym DET (Delayed Effects Theories) as shorthand to refer to DOHaD as well as to related theories, regardless of differences between the mechanisms that each of them prioritizes. We are mindful of the heterogeneous nature of this body of theories and, when precision is needed, we will abandon the shorthand.

3.1.2. Mechanisms

A diverse set of mechanisms could transform episodic or recurrent early exposures prior to conception, in utero, perinatally, and during infancy and early childhood into delayed impacts on adult illness, disability, and mortality. The mechanisms are associated with organ-specific embryo and fetal cell growth and differentiation (Gluckman and Hanson 2005, 2006; Bateson and Gluckman 2011), epigenetic changes (Godfrey et al. 2007; Meaney 2010; Kuzawa and Eisenberg 2014; Gluckman et al. 2016), exposure to and contraction of early childhood diseases and sustained inflammation (Elo and Preston 1992; Fong 2000; Finch and Crimmins 2004; Crimmins and Finch 2006), and experiences with stressful conditions and environments (McEwen 1998; Meaney 2001, 2010). Furthermore, a large and influential body of empirical research documents the long-lasting impact of early nutritional status on adult health and mortality (Scrimshaw 1997; Scrimshaw and SanGiovanni 1997; Costa 2000; Fogel 2004) and long-term impacts of early disease exposure (scarring) on adult health (Bozzoli et al. 2009). Finally, there is widespread empirical evidence demonstrating that more diffuse exposures, such as poverty and severe deprivation in infancy and early childhood, can also have lasting impacts (Forsdahl 1978; Bengtsson and Lindstrom 2000; Forsdahl 2002; Van den Berg et al. 2006).

These mechanisms are the focus of various strands of theories with unique histories and distinct disciplinary foundations. Different as they may be, however, they all invoke perturbations of a phenotype during critical periods of growth and development that may then lead to disruptions in processes of embryonic organogenesis and morphogenesis, immune response, neurological development, metabolic regulation, and even the formation of adult preferences and behaviors. After variable but usually prolonged latency periods, these disruptions could manifest themselves as increased susceptibility to adult chronic illness. Chronic conditions in adulthood are the main precursors in the disablement process as they are strongly associated with physical functioning and cognitive performance (Verbrugge and Jette 1994). Thus, delayed effects have the potential to increase the prevalence of chronic disease and disability and to extend the length of time individuals spend with morbidity.

The mechanisms involved are complex and their biological underpinnings are poorly understood. As result, we are ill-equipped to trace precisely the pathways from initial exposure to final outcomes at the level of individuals, let alone discern different pathways for entire populations. Thus, it is out of necessity that our initial investigation of population level impacts of adult delayed effects associated with adverse early conditions ignores differences between mechanisms. Instead, we choose to focus only on their gross, overall impact on patterns of health status and mortality. Although this is a simplification to circumvent lack of knowledge about mechanisms involved—not unlike omitting consideration of mediating pathways in structural equation models— it has an important payoff for it enables us to identify the pathways that must be elucidated to understand long term impacts irrespective of the mechanism(s) conveying adult health manifestations. In particular, our aim is to assess how large do delayed effects should be to have more than trivial effects on aggregate population patterns of morbidity and mortality. The empirical evidence gathered within the DOHaD paradigm includes a bewildering variety of studies, from those resting on simple aggregate correlations (Barker and Osmond 1987) to those that follow birth cohorts exposed to adverse conditions in utero and during early life (Stanner et al. 1997; Hoddinott et al. 2008; Tofail et al. 2008; Stein et al. 2010; Lumey et al. 2011, 2012). The central objective of all these studies is to estimate the magnitude of effects of adverse early conditions (however measured) on multiple adult health outcomes, including mortality.

3.2. Missing pieces

Absent from DOHaD studies is consideration of the relation between delayed adult effects and adult patterns of morbidity, disability, and mortality at the population level. The bulk of research rests on evidence showing the size of effects for individual adult health status or mortality (e.g., Bozzoli et al. 2009). Missing from this body of work is an assessment of the population level impact of delayed effects. Because of this omission, there has been no investigation that we know of about the implications of different regimes of past mortality decline on the composition of the adult population by susceptibility to delayed effects. As an illustration, consider the following scenario shared by a majority of low- to middle-income countries. While their populations experienced a very rapid mortality decline, the most important forces supporting it were not, as was the case in Western Europe and North America, the product of improvements in nutrition and standards of living but rather medical innovations of various sorts (Preston 1976, 1980; Palloni and Wyrick 1981). As a result, the size of the older adult population swells with individuals who survive despite having experienced poor early conditions. In the absence of this type of mortality decline these individuals would not have attained older ages and could not possibly contribute to adult patterns of morbidity and mortality. Thus, the nature of improvements in survival experienced in these countries offers one among several pathways that leads to increases in the fraction of individuals who attain adult ages who are at higher risk of expressing delayed effects. A somewhat different pathway is a secular increase in births susceptible to manifest delayed effects, as may happen when a lower fraction of pregnancies are terminated prematurely, a result of improvements in maternal health and prenatal care.

Can the nature of post-1950 mortality decline in low- to middle-income countries influence the adult population composition by exposure to adverse early conditions in birth cohorts that will attain older ages in the next two to three decades? If so, how will aggregate, population level, adult morbidity, disability and mortality be affected? How large is the resulting excess of chronic disease and disability that these cohorts will face throughout their adult lives? Because these issues have large health policy implications, they should be at the forefront of population health research.

It is important to note that no variant of DET theories makes unambiguous predictions about adult disability. Although it is plausible that exposure to adverse early conditions may also worsen the risks of developing disabilities, we know of no formulation or empirical study alluding to it. As a consequence, the inclusion of disability as part of the set of outcomes associated with delayed effects must assume, as we do here, that its occurrence depends only on risks embedded in the presence of well-specified chronic illnesses.

We take a first step to answer these questions by using a microsimulation model to explore the implications of a relation between delayed effects and morbidity/disability. Our aim is not to do hypotheses testing. Instead, we wish to provide proof of concept and estimate the magnitude of effects implied by potential linkages between adverse early conditions, delayed effects and compression of adult morbidity and disability. If, under reasonable conditions, these effects turn out to be non-trivial, then future empirical research should explore in more detail the mechanisms involved and the population level implications.

4. Predicted impacts on morbidity and disability

In previous work we showed that delayed effects can have important impacts on adult mortality (Palloni and Beltrán-Sánchez 2016, 2017). In particular, in all populations exposed to conditions that make them susceptible to delayed effects, a mortality decline will produce potentially large, albeit lagged, increases of the fraction of the population surviving to adulthood who are scarred by adverse early experiences. It is known that when mortality declines, the rate of reduction of successive cohorts’ adult mortality rates could slow-down at all ages simply as a result of milder selection pressures exerted by a regime of improving mortality. This decelerating force is amplified when the population is also susceptible to adult delayed effects. In addition, delayed effects result in changes of the observed slope of mortality at older ages even though the background age pattern of mortality remains invariant. Finally, delayed effects will reduce the within-cohort association between adult and child mortality, even though it is fixed in the background mortality pattern.

All impacts of delayed effects on adult mortality must be mediated by excess chronic illness, excess disease-specific mortality, or both. The same applies to disability: to the extent that delayed effects are consequential for the incidence of chronic illnesses, they will also have an impact on disability. If birth cohorts are susceptible to express delayed effects, their levels of adult chronic illnesses and/or diseases-specific disability and mortality will be subject to two sources of upward pressure: the first is rooted in the increase of (standard) mean frailty of individuals who attain adult ages, a natural product of milder selection embedded in improving mortality regimes. The second originates in excess morbidity and mortality risks shared by a fraction of ‘new’ survivors who might have been exposed to adverse early life conditions and who are able to attain adult ages in the less punishing mortality regime. Thus, the increase in the fraction of survivors scarred by early experiences magnifies the decelerating force that naturally arises when only standard frailty prevails (Palloni and Beltrán-Sánchez 2017). Ultimately, some of these effects may not be observable at all if improvements in medical technology, e.g. screening and treatment of chronic illnesses or other innovations, partially or totally blunt the expression of adult delayed effects.

Under the foregoing conditions we should expect:

1. reductions of the rate of improvements of adult mortality;

2. increases (or smaller improvements) in the incidence of morbidity and associated dis-ability;

3. expansion (or reduced compression) of morbidity and disability and, as a consequence, decreases (or smaller increases) in healthy life expectancy.

The goal of the microsimulation implemented below is to evaluate these three predictions and to assess how large must delayed effects be to have significant impacts on age patterns of adult morbidity, disability and mortality in populations influenced by adverse early conditions.

5. A multistate model with chronic illness, disability and mortality

Consideration of delayed effects on morbidity and disability and their impact on healthy life expectancy and the extent of compression of morbidity and mortality requires the inclusion of at least one chronic illness and associated disability and mortality. We propose a multistate model that includes two conditions and associated disability and mortality. Because of their relevance for low- and middle-income populations, we will focus on obesity and type-II diabetes (T2D), the two most prominent conditions widely studied by multiple variants of DET.

Figure 1 is a graphic representation of a standard multistate hazard model where transitions from one state to another are defined by age (and possibly duration)-specific risks. In theory, one could write down expressions for the conditional survival probabilities associated with any state of origin (or destination) and determine the distribution of cohorts’ members by states at any age. From these quantities we could compute statistics summarizing population parameters such as the fraction of a cohort ever experiencing a visit to a particular state, the average age of first visit, or the duration of sojourn in some state. However, the formal expressions for these quantities are not tractable as they involve multiple integrals and must be evaluated using numerical procedures. Thus, the microsimulation model provides a shortcut to obtain the desired quantities.

Figure 1:

Graphic representation of the multistate model and transitions of interest

Note: T2D corresponds to type-II diabetes.

5.1. Empirics: estimates of the transition rates in the model

The parameters of the multistate hazard represented in Figure 1 could be estimated from a longitudinal study with full information on individuals’ early conditions and trajectories of adult health status. Since we cannot avail ourselves from such an ideal data set, we follow an alternative, albeit inferior, strategy. This consists of estimating the transition rates needed in the model in a ‘piece-meal’ fashion, namely, gathering information from different empirical studies that shed light on one or more transitions of interest and then combining these as if they described a single (artificial) population. Although this is not a ideal, it may suffice for our purposes. This is because we are not so much interested in precise estimates of transition rates as we are in assessing the magnitude of impacts of adult delayed effects. For this less ambitious purpose, it suffices that the transition rates we use for T2D and obesity incidence, for example, represent observed patterns in low- and middle-income countries, even though they may be just one out of multiple plausible ones. The range defined for the values of input parameters we used is empirically based on a detailed revision of nearly 70 papers that contain estimates of effects at the individual level on risks of obesity, T2D and disability among those who experience adverse conditions (see Appendix A). One study, Palloni and Souza (2013), provides estimates of the fraction of individuals who may have experienced adverse conditions in low- to middle-income countries. We use the middle of the range of their estimates as the upper bound in our simulation. Appendix B describes the nature of all inputs.

5.2. Simulation scenarios

The simulation design includes the following rules:

1. We simulate 101 cohorts with 10,000 people each. Each cohort is exposed to a mortality regime embedded in life tables from the Coale-Demeny models of mortality Coale and Demeny (1983). The oldest cohort (cohort 0) experiences survival probabilities from the West (female) model with a life expectancy at birth of 40 years. The youngest cohort (cohort 101) experiences much improved survival as it is exposed to a life table from the West (female) model with a life expectancy at birth of 80 years. Thus, the simulated cohorts undergo a mortality decline with gains in life expectancy at birth of about 0.4 years per year.

2. We consider 4 scenarios according to the fraction of a birth cohort susceptible to express delayed effects: 10%, 20%, 30%, and 40%.

3. We define four scenarios depending on the fraction of a birth cohort who eventually experiences a health transition (from healthy to obese, from healthy to T2D and for all higher order transitions)(Table 1): baseline, low, medium, and high incidence rates. Altogether there will be 4×4=16 scenarios to which a cohort can be exposed.

4. We generate 100 replicas of a cohort containing 10,000 individuals each for a total of 1 million simulated individual trajectories.

5. Each of the i = 1,...,10,000 people in the 100 replicas of cohorts contained in the 16 variants is subject to the multistate process according to the algorithm and microsimulation rules described in Appendix C.

Table 1:

Simulation scenarios

Transition state	Baseline	Low	Medium	High
Healthy to Obese	0.10	0.20	0.40	0.60
Healthy to T2D	0.05	0.10	0.20	0.30
Healthy to Disability	0.03	0.05	0.10	0.20
Obese to T2D	0.40	0.50	0.70	0.90
Obese to Disability	0.05	0.10	0.20	0.30
Obese & T2D to Disability	0.05	0.10	0.20	0.30
T2D to Disability	0.20	0.50	0.70	0.90

Note: T2D corresponds to type-II diabetes

To streamline the discussion of results we will focus on two cohorts only: the baseline cohort with the highest cohort mortality (cohort 0 with e₀ = 40) and the cohort born 100 years after the onset of the mortality decline that experiences the most favorable survival (cohort 101 with e₀ = 80). The contrast between these cohorts will illustrate the importance of a mortality decline. We expect the youngest cohort to experience the largest burden generated by the presence of delayed adult effects.

Throughout we assume that delayed effects are only expressed as excess risks of obesity and T2D. We ignore potential excess risks of disability in the absence of one of these other conditions. Nor do we assume that the conditional risks of second and third events (T2D, disability, or mortality) subsequent to the occurrence of the first event depend in anyway on delayed effect susceptibility status. Any excess risk for second or third order events is only associated with the state of occupancy at the time. Thus, for example, the risk of T2D among obese individuals at any age is the same for individuals who are susceptible to delayed effects and those who are not, but it is higher than the risk of T2D among non-obese individuals (regardless of delayed effect susceptibility status). This leads to lower bounds of impacts of interest for we assume that delayed effects do not increase the risks of second and third order events. The advantage of this design is that it minimizes the number of parameters of interest and provides a more robust simulation platform. The downside is that we stack the deck against identifying large effects.

5.3. Simulation scenarios and key indicators

We focus on three scenarios. The first (scenario N) includes cohorts free of delayed effects. In the second scenario (scenario Min) birth cohorts are exposed to minimum impacts: only 0.10 of the birth cohort is susceptible to delayed effects and only 0.10 of the population eventually experiences a transition from healthy to a condition (see Table 1). The third scenario (scenario Max) includes maximum delayed effects: 0.40 of the birth cohort is susceptible to delayed effects and 0.40 eventually experiences a health transition event. Throughout, we examine the influence of delayed effects on three sets of indicators.

First, we evaluate differences between scenarios in life expectancy at various ages. These differences will be a function of disparities in the incidence of obesity and T2D and, therefore, represent the mortality load associated with these two conditions. Thus, in scenarios with delayed effects (Min and Max) we should expect to see lower life expectancy at all adult ages when compared with scenario N. Furthermore, these differences should be largest in cohorts with the most favorable mortality experience.

Second, we calculate differences across scenarios in the average age of incidence of the first morbidity event and in age-specific prevalence of healthy individuals (e.g., absence of obesity, diabetes or disability). If delayed effects have an impact on the incidence (prevalence) of conditions, we expect earlier onset of morbidity and higher prevalence in scenarios Min and Max when compared with scenario N. Furthermore, the contrasts should be larger in cohorts with the highest life expectancy.

Third, we use an indicator of healthy life expectancy defined as the average years of remaining life in the absence of conditions. In addition, we compute three indicators of unhealthy life expectancy: (i) the fraction of residual lifetime after a given age lived with one or two conditions (obesity and T2D); (ii) the expected duration in each state with one or two conditions; and (iii) the fraction of residual lifetime lived under the worst combination of conditions (disability and obesity only, and disability, obesity and T2D). If delayed effects have an impact on healthy life expectancy, we expect to observe expansion of morbidity and disability (i.e., more years of unhealthy life expectancy) in Min and Max relative to scenario N. The magnitude of the expansion should be greatest in cohorts with the most favorable mortality pattern.

5.4. Sensitivity analysis

We also conducted sensitivity analysis using additional data on age-patterns of incidence of obesity and T2D from the Institute for Health Metrics and Evaluation website (Institute for Health Metrics and Evaluation, 2020) combined with varying levels in the parameters of interest (see Appendices D and E). We considered 36 scenarios resulting from the combination of two mortality regimes (low and high life expectancy), two levels of excess mortality, two levels for the fraction of the population that eventually experiences an event (high and low, see Table 1), and two fractions of births affected by DOHaD-type mechanisms (high (40%) and low (10%)).

6. Results

First, we review results for residual life expectancy at several adult ages in each scenario. Figure 2 displays differences in residual life expectancy at ages 20 and over between scenario N, Min and Max. The figure shows that differences between scenarios with and without delayed effects varies according to the cohort’s mortality levels. It is most irregular in the cohort with highest mortality (e₀ = 40) and most regular in the cohort with lowest mortality (e₀ = 80). This should be expected for we use identical incidence rates of illnesses (T2D) and conditions (obesity and disability) in all cohorts, irrespective of their mortality level (see Table 1). As a consequence, cohorts with the lowest levels of life expectancy will experience higher excess mortality associated with illnesses and conditions that express delayed effects than cohorts with milder mortality levels. Furthermore, the excess mortality should lead to larger reductions of years of life lived at ages old enough to be affected by these illness and conditions (after age 30 or so) and young enough to be associated with larger residual life expectancy. This explains the sharper peak at age 35 or so in the highest mortality cohorts and the smoother curve for cohorts with less severe mortality levels.

Figure 2:

Differences in conditional life expectancies at various ages between no delayed effects and two delayed effects scenarios (Min and Max) for three cohorts with low (e0 = 40), medium (e0 = 60), and high (e0 = 80) life expectancy (LE), respectively.

Note: Min and Max correspond to 10% and 40% of the cohort is susceptible to delayed effects with low and high transition rates of events, respectively (see Table 1 for transition rates).

Figure 2 shows positive values indicating higher residual life expectancy across all ages in scenario N. For example, in a cohort with low life expectancy (e₀ = 40), a scenario with delayed effects leads to residual life expectancy at ages 35 and 60 that are respectively 3.8 and 1.4 years lower than in a scenario with no delayed effects. The differences are smallest in the cohort with the highest life expectancy. This is because the more favorable survival probabilities at all ages partially offset losses due to increases in the number of individuals expressing delayed effects at adult ages. Results from sensitivity analysis show similar patterns (Appendix E). Differences in residual life expectancy are larger for the Max scenario, and smaller in magnitude when life expectancy is higher. That is, the “mortality tax” of delayed effects decreases when the room for more taxes (more mortality) decreases.

We note that the differences shown in the graph are lower bounds. First, by design, these differences reflect excess mortality associated with only two out of potentially more conditions. Second, and also by design, the simulation assumes that delayed effects only augment risks of first order events and leave unchanged the risks of second or third other events. The contrasts in residual life expectancy we observe across scenarios is a function only of the frequency of each of the lineages in the state space, e.g. each of the allowed trajectories and the excess mortality embedded in them (see Figure 1). The simulation design does not allow mortality excess among those who can express delayed effects, regardless of condition.

Second, Table 2 displays the mean age of incidence of diabetes, obesity and disability in each simulation scenarios. The figures show that in both, the low and high life expectancy cohorts, the mean age of incidence of T2D is younger in Min and Max relative to scenario N (no delayed effects). Thus, for example, in the low life expectancy cohort, the average age of incidence is about 50 years in scenario N and about 35 in scenarios Min and Max. In the high life expectancy cohort, the mean age of incidence for all conditions is also lower and the differences relative to scenario N are somewhat lower than those in the low life expectancy cohort. Equivalently, these figures indicate that delayed effects could potentially increase the duration of life spent with T2D by about 10 to 16 years due to earlier incidence. The figures for obesity and disability display a similar pattern: there is a decrease in the mean age of incidence in scenarios Min and Max, potentially higher duration of life lived with the conditions and, finally, slightly smaller differences between Min and Max and N in cohorts with the highest life expectancy.

Table 2:

Mean age at first event by simulation scenarios for two birth cohorts with low and high life expectancy (LE)

Scenario	Low LE (e0= 40)	Difference1	High LE (e0= 80)	Difference1
Diabetes
Absence of DOHaD	49.72		49.72
Max	34.28	–15.44***	39.74	–9.99***
Min	33.96	–15.76***	39.25	–10.48***
Obesity
Absence of DOHaD	36.46		36.46
Max	28.19	–8.28***	30.65	–5.81***
Min	27.22	–9.25***	28.21	–8.25***
Disability
Absence of DOHaD	43.74		43.74
Max	30.41	–13.33***	32.09	–11.65***
Min	29.48	–14.26***	31.35	–12.39***

^***p<0.000

¹Differences taken with respect to absence of delayed effects.

Third, we estimate age specific prevalence and total years of life lived in the healthy state and compute ratios relative to the scenario with no delayed effects (Figure 3). Figure 3a displays ratios of numbers of healthy survivors in the Min and Max scenarios to those in scenario N for the two cohorts. The ratios decline steeply at older ages in both the Min and Max scenarios and dip as low as 0.2 at age 50 in the latter. There are no major differences in these ratios incohorts with highest and lowest life expectancy because the effects of mortality on the ratios affect both numerators and denominators and cancel each other out.

Figure 3:

Ratios of the prevalence in the healthy state (absence of conditions) and total years of life lived in the healthy state relative to a scenario with no delayed effects

(a) Ratios of survivors in the healthy state relative to no delayed effects

(b) Ratios of total healthy years lived above age x relative to no delayed effects

Note: LE stands for life expectancy; “No delayed effects” corresponds to background mortality with no delayed effects, Min and Max correspond to 10% and 40% of the cohort is susceptible to delayed effects with low and high transition rates of events, respectively (see Table 1 for transition rates).

What are the impacts of delayed effects on the expected years of healthy life? Figure 3b shows ratios of healthy years of life (absence of conditions) in the Min and Max scenarios relative to those in scenario N. As expected, the disparities are more pronounced in the lowest mortality cohort where individuals aged 60+ in a Max scenario could live less than 15% of their remaining life in the healthy state relative to those in scenario N. In a regime of delayed effects, improvements in years of healthy life will be slower in all cohorts with more favorable mortality patterns. These contrasts should be tempered, however, since the “lineage” that remains healthy at older ages is a minority that represents less than 20 percent in all scenarios.

Figure 4 display indicators for expected years in unhealthy state. To avoid cluttering, we only show figures for obesity and T2D. The results for disability are similar although they could slightly overstate its impact since we do not allow individuals to recover from disability once they acquired it. Figure 4a displays the fraction of a lifetime after a given age that will be lived with obesity, including the most extreme combination of conditions (obesity/T2D and disability). The differences across scenarios are large, particularly between scenarios N and Max. At their highest, during middle adulthood, populations affected by delayed effects attain values hovering around 0.40 to 0.45, e.g., they could spend close to half of residual lifetime with obesity. A similar pattern applies to T2D (Figure 4b) albeit with slightly smaller magnitudes. Thus, at their peak, during middle-adulthood, the values are slightly over 0.30 in the low mortality cohort. Finally, Figure 4c displays expected duration in the state of obesity (including all combinations involving T2D and disability). Here again, the differences between scenarios are substantial. For example, in the lowest life expectancy cohort, individuals in the Max scenario are expected to live close to 12 years after age 20 in the obese state. In contrast, this figure is less than 3 years in scenario N. These differences are slightly more pronounced in the cohort with lowest mortality as they experience a larger influx of susceptible individuals expected to live with obesity. Sensitivity analyses show similar age pattern (see Appendix E, Figure E.2) although with slightly lower magnitudes. For example, the fraction of residual life time with obesity is about 0.25 in the Max scenario (high excess mortality, high incidence and high fraction of births affected by DOHaD) (Appendix E). These results show that delayed effects could lead to significant expansion of morbidity both directly by increasing risks of T2D and indirectly, by increasing the risks of obesity and associated chronic illnesses.

Figure 4:

Overall impact of delayed effects on the expected years of life lived with obesity and type II diabetes

(a) Fraction of years of life with obesity

(b) Fraction of years of life with type II diabetes

(c) Expected years of life in all states with obesity

Note: LE stands for life expectancy; “No delayed effects” corresponds to background mortality with no delayed effects, Min and Max correspond to 10% and 40% of the cohort is susceptible to delayed effects with low and high transition rates of events, respectively (see Table 1 for transition rates).

Finally, we assess the impact of delayed effects on the subpopulation with comorbidities (Figure 5). Figures 5a, 5b and 5c display the fraction of years of life lived in combinations of the least favorable conditions allowed in the model. Figure 5a is for years of life lived with disability and obesity but not T2D, Figure 5b is for years of life lived as obese and T2D but no disability and, finally, Figure 5c is for years of life in the worst of conditions, namely, with obesity, T2D and disability. In the highest and lowest life expectancy cohorts, at least 15% of the remaining years of life at age 65 are lived with obesity/disability. In contrast, when there are no delayed effects there is a negligible fraction of life lived with these comorbidities. The differences between scenarios in Figures 5b and 5c are not large but they all suggest an expansion of time spent with comorbidities. Although they implicate a smaller fraction of individuals, these subpopulations arguably represent those that experience the largest health burden and require some of the most expensive interventions (medical assistance, medical treatment, rehabilitation).

Figure 5:

Overall impact of delayed effects on the expected years of life lived with comorbidities

(a) Fraction of years of life lived as obese/disabled but no type II diabetes

(b) Fraction of years of life lived in obesity and type II diabetes but no disability

(c) Fraction of years of life lived in obesity/type II diabetes/disability

Note: LE stands for life expectancy; “No delayed effects” corresponds to background mortality with no delayed effects, Min and Max correspond to 10% and 40% of the cohort is susceptible to delayed effects with low and high transition rates of events, respectively (see Table 1 for transition rates).

7. Discussion

7.1. Summary

We assessed the size of impacts of delayed effects implied by DET theories on aggregate population patterns of morbidity, disability, and mortality. The question we addressed is about the differences we might observe between populations with and without a contingent of cohorts primed to manifest adult delayed effects. Elsewhere we showed that when considering only gross effects on mortality, the impacts can be substantial and influence the rates of adult mortality decline as well as age patterns of mortality (Palloni and Beltrán-Sánchez 2016, 2017). This paper extends the simple two-state model and considers impacts on two conditions, obesity and T2D, as well as on mortality and disability via excess incidence of obesity and T2D. We show that in populations with conditions likely to be experienced in low- and middle-income countries, the incidence and prevalence of the two most common conditions (obesity and T2D) will increase and will be associated with decreases of healthy life expectancy. There are sizable differences in the fraction of the population in a healthy state and expected years of healthy life. We find a steep decline by age in the fraction of healthiest survivors (those with no conditions): a scenario with a large population susceptible to delayed effects leads to only one-fifth of the population aged 50 or older being in a healthy state. Moreover, delayed effects can reduce the average remaining years of healthy life and the fraction of residual life lived in a healthy state. Thus, people aged 60+ in a high delayed scenario are expected to live less than 15% of their remaining life in a healthy state.

These results are consistent with recent empirical findings. Thus, using data from Mexico, Payne and Wong (2019) found that although there were no differences in the overall mortality levels at ages 60–69 between two cohorts (born in 1943–1952 and in 1932–1941), those born in 1943–1952 lived significantly fewer years of healthy life at ages 60–69 (free of disability) than the older cohort. Furthermore, research based on the WHO-SAGE study in China, Ghana, India, Mexico, and South Africa show that older adults in these countries are expected to live at least 20% of their remaining life with disability. Similar patterns were obtained in an examination of recent trajectories of disability-adjusted life years (DALYs), particularly in less developed countries (Murray et al. 2015). These results are in line with the simulation and suggest that, as indicated at the outset, the story of compression in low- and middle-income countries like Mexico, China, Ghana, India, and South Africa might be different than in high-income countries. Mexicans born in 1943–1952 are primed for delayed effects since their survival to older ages is due to large improvements in infant and child survival associated with dissemination of medical technologies, not improvements in standards of living (Palloni and Wyrick 1981). This is precisely the population most susceptible to express delayed effects as expansion of morbidity.

Our results also indicate that delayed effects could extend the length of time individuals live with obesity and T2D, two of the chronic conditions that have grown fastest in low- and middle-income countries. For example, there is an earlier age at onset of obesity and T2D in the presence of delayed effects (before age 45) indicating that they could lengthen the time lived with the condition. In a regime of high prevalence of delayed effects scenario, about half of the residual lifetime at ages 40 to 60 is spent with obesity and about one-third of their residual life could be spent with T2D. By the same token, in a scenario with high prevalence of delayed effects and low mortality, the fraction of years of life lived simultaneously with obesity, T2D and disability at age 65 could be as much as 12% higher than in a regime with no delayed effects. Furthermore, these sub-population lineages that are not in health arguably contain some of the most burdened pockets of unhealthy individuals. Computation of costs associated with assistance, medical interventions, medications, and rehabilitation will be disproportionately influenced by the relative size of these “minorities”. For example, some research suggests that most of the health care expenditures in the US are concentrated amomg the very sick in the elderly as 5 percent of this population accounts for the majority of health expenditures (Berk and Monheit 2001). It could be argued that the cost burden is significant only at older ages, where the differences between populations with and without delayed effects are smaller. However, a key feature of a regime with delayed effects is that critical ages beyond which delayed effects are felt may be subject to substantial stochastic variability. Lower means and higher variance of these ages will aggravate the health consequences. This is the case of obesity: the mean age of incidence has declined considerably, and its variance has increased (Hughes et al. 2011; Cunningham et al. 2014). It would not be surprising that similar changes could be affecting the mean and variance of T2D’s age of onset, a condition closely associated with obesity.

On average, the estimated magnitude of effects is neither small nor huge but, in all likelihood, these are lower bounds. In. Appendix E we show results of sensitivity analyses that show similar results a when using alternative inputs. As we learn more about the pathways through which adverse early conditions induce delayed effects, we will be in a better position to represent them and estimate their impacts.

Conservatively, we could state that low- to middle-income countries primed for the expression of delayed effects may experience significant expansion of morbidity and disability while simultaneously undergoing a contraction of the room for further gains in survival at older ages.

7.2. Limitations

A paper such as this contains a hefty number of shortcomings. We will single out the six most relevant ones in descending order of importance. First and foremost, we ignore mediating mechanisms. This is not trivial as the many variants of DET theories propose surprisingly different mechanisms, each associated with heterogeneous outcomes. Furthermore, and as suggested by life course epidemiology (Kuh et al. 2003), there could potentially important indirect impacts of delayed effects on adult health outcomes. For example, a chain risk model suggests a sequence of linked exposures that could increase cumulative disease risks as an outcome of detrimental exposures (Kuh et al. 2003).

Second, the model rests on the rather blunt assumption that birth cohorts are composed of two discrete subgroups possessing (or not) a propensity to express delayed effects. This is an oversimplification as it is more realistic to represent susceptibility to delayed effects as a phenotypic trait varying continuously. Introducing this modification is not technically taxing but does require untestable assumptions about the form of the underlying trait distribution.

Third, any exercise with the microsimulation model for projection and policy design demands much more that the synthetic estimates of inputs we used here. And these, in turn, demand hard-to-access empirical data that are robust enough to represent experiences of true cohorts.

Fourth, the multistate model is highly simplified and ignores return transitions. This is not trivial as it precludes occurrence of events that may reverse and attenuate effects. For example, an individual may become obese at some point in her life but, through diet and exercise, she may move back to a state with normal weight. Similarly, although T2D may not be cured, its effects could be reduced via proper treatment.

Fifth, the role of variability of critical ages beyond which delayed effects manifest themselves is a theme quite debated by DET theories and with potential relevance for the size of impacts. Unfortunately, these ages depend on both the type of adverse effects to which individuals are exposed and the adult chronic condition most likely to express them, two understudied areas.

Finally, the micro simulation model includes variability that only represents Monte Carlo noise, not true variability associated with underlying inputs, e.g., the parameters estimates used throughout to assign values ot the transition rates.

7.3. Extensions: epigenetic, social and behavioral processes

Although the paper does not deal directly with mediating pathways, we hope to have shown that the micro simulation model offers ample room for specifying multiple mediating pathways conjectured by DET theories. There are two mediating paths of relevance. The first involves physiological as well as molecular processes that take place early in life and are the result of deleterious exposures of various types. To represent these processes, it is necessary to integrate parameter estimates from clinical studies relating changes of epigenetic signatures (via excess methylation of CpG islands, depressed histone acetylation, other changes to the nucleosome, and small interfering microRNA) that could be induced by early nutritional deficiencies, stress, and other adversities. These epigenetic changes might increase the propensity to develop child and adult obesity and other metabolic disorders.

Second, we should study in depth how the inclusion of events experienced across the life course modifies (reduce, amplify) the impact of early conditions. This is of strategic importance for the design of interventions as it is known that some impacts of adverse early conditions are indeed reversible by a variety of exposures and individual behavior.

The inclusion of these two dimensions requires simultaneous consideration of molecular, socioeconomic, environmental, and behavioral components. And, undoubtedly, it also calls for a different class of models, including hybrid models of complex systems.