Early life conditions, rapid demographic changes and older adult health in the developing world

Abstract

The demographic transition of the 1930s–1960s dramatically improved life expectancy in some developing countries. Cohorts born during this time are increasingly characterized by their survivorship of poor early life conditions, such as poor nutrition and infectious diseases. As a result, they are potentially more susceptible to the effects of these conditions at older ages. This study examines this conjecture by comparing obesity, diabetes, and hypertension in older adults born in the beginning portion of the 1930s–1960s across different mortality regimes using a subset of harmonized cross national data from seven low and middle income countries (RELATE, n=16,836). Using birthplace and height as indicators of early life conditions, results show (1) higher prevalence of obesity and diabetes and higher likelihood of obesity, diabetes and hypertension in middle income countries but, (2) no convincing evidence to indicate stronger effects of early life conditions on health in these countries. However, shorter adults living in urban areas were more likely to be obese indicating the overall importance of early life conditions and the potential negative impact of urban exposures during adulthood. Obesity results may foreshadow the health of future cohorts born in the later portion of the 1930s–1960s as they reach older ages (60+).

Introduction

There has been solid evidence of associations between early life conditions and older adult health (Almond and Curie 2010; Barker 1998; Crimmins and Finch 2006; Davey Smith, Hart, and Hole 1999; Elo and Preston 1992; Hayward and Gorman 2004; Wilkinson & Marmot 2003; van Ewijk, Painter and Roseboom 2013; Lillard et al. 2015). Although there have been a few studies that examine the effects of macro-level events in early life on adult health (van den Berg, Doblhammer & Christensen 2009) there have been almost no empirical studies regarding the degree to which important historical, demographic, and epidemiological changes in early life have influenced older adult health in the developing world. Reductions in infant and child mortality during the 1930s-1960s triggered by the medical and public health revolution led to dramatically increased life expectancy but without parallel improvements in standard of living (Preston 1976). These early life circumstances are predicted to affect elderly health in the developing world for at least the next 20–30 years (Palloni, Pinto-Aguirre, and Pelaez 2002) in diseases such as obesity, diabetes, and heart disease which are forecasted to increase in the developing world (WHO 2000; Kinsella & He 2009; Murray & Lopez 1996). Do the early life circumstances of the 1930s–1960s in any way help explain the increasing prevalence of these conditions in the developing world among older adults? The purpose of this study is to examine the evidence to answer this question by comparing the health of older adults in low and middle income countries born in the beginning decades of the period of the 1930s–1960s.

Public health interventions of the early to mid-20^th century

Public health interventions in the developing world in the early to mid-20^th century were focused primarily on reducing infectious disease by reducing infection through sanitary improvements but not directly improving diet and nutrition (Farley 2003; Clark 1930) because the synergy between nutrition and infection was not fully recognized until later in the century (Scrimshaw 1968). This period saw stagnant economic growth for many developing countries (Maddison 2006) and, even though reducing infection indirectly improves nutrition (Floud, Fogel, Harris, and Hong 2011), without economic improvements, many survivors continued to be exposed to poor nutrition and poor diet resulting, for the most part, in a stunted population.

The public health interventions of the early to mid-20^th century in developing countries initially benefited those living in urban areas because of limited coverage in rural areas. Although some smaller and poorer countries such as Costa Rica and Puerto Rico implemented country-wide interventions during the 1930s (Clark 1930; Rosero-Bixby 1990), public health interventions in other larger, present-day middle income countries, such as Mexico, were mostly concentrated in urban areas during this period (Rodríguez de Romo and Rodríguez de Pérez 1998). Present-day low income countries like China, Indonesia, Ghana, and India had less coverage of public health interventions even in urban areas during the 1930s–1940s (Banister 1987; Ramasubban 2008; Dyson 1997; Patterson 1981; Patterson 1979; Caldwell 1967; Nitisastro 1970).

Most of the population in the developing world during this period lived in rural areas. In contrast to the developed world (Preston and Haines 1991), environmental conditions in rural areas of the developing world during the first part of the 20^th century were precarious and, for the most part, worse than urban areas in exposure to poor diet and infectious diseases (Clark 1930; Rodríguez de Romo and Rodríguez de Pérez 1998)—circumstances still present in some rural areas (Sastry 1997).

As countries fully implemented public health interventions in both urban and rural areas, a larger population of infants and children survived. However, during the beginning period of the 1930s–1960s and prior to full country-wide implementation, increasing survivorship affected a smaller portion of the population which included mostly those living in urban areas in larger countries and, in only a few exceptional cases, included those living in both urban and rural areas in smaller countries.

Early life mechanisms

While there are several early life mechanisms that could explain older adult health (Barker 1998; Barker et al. 2010; Crimmins and Finch 2006; Davey Smith et al., 1999), the circumstances that produced unique cohorts increasingly characterized by their survival of poor early life nutrition and infectious diseases during the historical circumstances of the 1930s–1960s increase the chances of the manifestation of Barker-like effects at older ages. Low birth weight and stunted babies, a reflection of poor intrauterine and post-birth growth, have an increased risk of obesity, diabetes, and hypertension at older ages (Barker 1998). Further, the mismatch between being born in a resource scarce environment and then being exposed to a nutritionally richer environment later in life increases the risk of obesity, diabetes, and hypertension (Osmond and Barker 2000; Bateson et al. 2004) which, in turn, is associated with frailty and mortality (Kuh & Ben-Shlomo 2004).

Some developing countries experienced significant improvements in economic conditions and increased urbanization later in the 20^th century (Maddison 2006; Henderson 2002) which might have increased access to a more enriched nutritional environment during adulthood for the survivors of poor early life conditions from the 1930s–1960s, especially those living in urban areas. Diets increasingly higher in saturated fats pose an important source of risk to health (Popkin 2006) and the combination of poor early nutritional environment and exposures to Western-style diets could be a lethal combination. Not all economic growth has benefited the entire population in the developing world (López-Alonso 2007). Economic improvements largely benefiting urban dwellers may have increased exposure to this type of nutrition in addition to other types of exposures for urban dwellers.

The macro-level changes of the 1930s–1960s may elucidate a clearer manifestation of the Barker-type hypothesis because cohorts are less affected by mortality-driven selection than preceding cohorts due to public health interventions and medical technology but are increasingly characterized by survivorship of poor early nutritional and infectious disease environments. However, the manifestation of these macro-level changes (rapid changes in mortality due to public health interventions and medical innovation) played out differently in countries due to differences in the timing, pace, and reason for mortality decline between countries and mortality regimes. Thus, these differences may have produced cohorts with different mortality experiences which are now leading to different health patterns in later life (Palloni, Pinto-Aguirre, and Pelaez 2002).

At one extreme are developed, higher income countries (denoted as type A countries, Table 1) that experienced an earlier and more graded mortality decline at a higher standard of living. At the other extreme are low income countries (type E countries, Table 1) that experienced rapid mortality decline much later in the 20^th century primarily due to public health interventions and medical technology. In between are developing countries that experienced a mortality decline similar to the developed world (type B countries) and countries that are present-day middle income countries that experienced rapid improvement in life expectancy at some point during the 1930s–1960s (types C–D countries). Differences in mortality decline among cohorts born during the 1930s–1960s in different mortality regimes partially depend on the timing of exposure to the country-wide implementation of public health interventions. Thus, moving from country type A to country type E as country-wide public health interventions are fully implemented in both urban and rural areas implies moving towards countries with cohorts increasingly characterized by their survivorship of poor early life conditions and the effects of these conditions in older adulthood including higher fragility, stronger effects of childhood conditions on health, sharper SES inequalities, and increasing mortality risk.

Table 1.

General Framework for the Nature of Mortality Decline across Countries in the Early to Mid 20^th Century and Expected Health Patterns in Older Adults

Country Type A		Country Type E
Nature of Mortality Decline At Birth
Early graded	Mortality decline	Late, rapid
Higher fraction	Improvements in SES	Lower fraction
Smaller fraction	Public health interventions	Higher fraction
Expected Health Patterns In Older Adults
Less	Poor health	Greater
Weaker	Early life conditions	Stronger
Weaker	SES differentials	Stronger
Lower	Expected mortality risk	Higher

Source: Adapted from McEniry, 2014.

Notes: Country type A includes developed countries (e.g. England, Netherlands, US) and country type E includes low income countries (e.g. China, Ghana, India, Indonesia). In between these extremes are predominantly middle income countries (types B–D: Costa Rica, Mexico, South Africa) characterized by differences in the timing, pace, and reason for mortality decline during the 1930s–1960s.

Cohorts at the beginning of the 1930s–1960s period

While most of the mortality change occurred after the mid-1940s in the developing world during the period of the 1930s–1960s (Preston 1976), the cohorts of the 1940s–1960s have not all reached older ages (60+) and thus it is too early to fully examine those born in the mid-1940s through the 1960s. Nevertheless, recent evidence suggests that poor early life conditions are, in part, associated with high levels of disease and disability and reduction in life expectancy at older ages in the unique cohorts from the beginning part of the 1930s–1960s (Palloni and Souza 2013). Cohorts from settings such as country types C and D (now predominantly middle income countries) are now experiencing an increasing prevalence of chronic conditions in some regions of the world (Palloni et al. 2005) and evidence suggests the importance of early life nutritional environments and diabetes among those countries undergoing demographic changes at the beginning of the period (McEniry 2014). The reported prevalence of diabetes in country types C and D is higher than what appeared historically in developed countries (García-Palmieri et al. 1970; Gordon 1964; Hadden and Harris 1987; Harris et al. 1998; Wilkerson and Krall 1947).

Thus, examining the cohorts born in the late 1920s through mid-1940s experiencing an increasing prevalence of chronic conditions at older ages is a relevant starting point to examine the conjecture regarding the demographic changes of the 1930s-1960s. It may be possible to reach conclusions regarding the long-term consequences of demographic change in some now-middle income countries, especially among those most likely to have benefited from improving environmental conditions—those urban born older adults exposed to poor nutrition and infectious diseases in large, now middle income countries or those rural born older adults in those smaller countries with country-wide coverage of public health interventions in the beginning of the 1930s–1960s.

This Study

In this study we examine the degree to which the prevalence and likelihood of chronic conditions are related to the demographic changes of the 1930s–1960s. We focus on adult obesity, diabetes, and hypertension—conditions known to originate in early life (Barker 1998) and select two groups of older adults born in the beginning part of the 1930s–1960s period. The first group experienced increased survivorship of poor early life conditions (now predominantly middle income countries—Costa Rica, Mexico, South Africa, country types C and D) and the second group experienced very harsh environmental conditions in early life with little decline in mortality (China, Ghana, India, Indonesia, country type E, low income countries). With a recently compiled cross national data set of older adults, we use two measures of early life common to many surveys of older adults, rural/urban birthplace and height, to examine the merit of the conjecture. Using the general framework of Table 1, we expect to observe a (1) higher prevalence and likelihood of older adult obesity, diabetes, and hypertension, and (2) stronger effects of early life conditions (birthplace, low height) on these health conditions in older adults born in now-middle income countries as compared with those born in low income countries. The stronger effects of early life conditions should be more apparent for those most at risk--the most vulnerable population in now-middle income countries experiencing increasing survivorship in the 1930s–1940s. In terms of birthplace, this population includes those born in rural areas in the smaller type C countries where coverage of public health interventions extended into rural areas in the 1930s–1940s, and in urban areas for larger type C and D countries where coverage of public health interventions had not yet reached full implementation into rural areas. In terms of height, this population is comprised of shorter older adults.

Data and Methods

Data

The data to compare the health of older adults in low and middle income countries are drawn from a subset of the recently compiled Research on Early Life and Aging Trends and Effects (RELATE) database which contains harmonized cross sectional and panel data from major surveys of 147,278 older adults or households in 20 countries in Latin America, Asia, Africa, the US, England, and the Netherlands (RELATE 2013). These studies are based on probability sampling and are representative of the older adult population either nationally or in major country provinces. All studies had good interviewer training, good questionnaire design, concern for data quality, and high response rates. The RELATE data were harmonized where possible to make cross national comparisons across surveys possible (McEniry, Moen, and McDermott 2013).

The subset of RELATE data used for this study draws from seven surveys (n=16,836) of low and middle income countries including the Costa Rican Study of Longevity and Healthy Aging (CRELES), the Indonesian Family Life Survey (IFLS), and the WHO Study on Global Ageing and Adult Health Study (SAGE) from Mexico, China, Ghana, India, and South Africa. We include the SAGE survey because it has cross national data on low and middle income countries, uses similar sample designs and questionnaire construction, and is representative at a country level in some instances.

Measures

Early life conditions

Rural birthplace is used as an indicator of precarious environmental conditions in early life (exposure to poor nutrition, infectious diseases) and is defined according to questions asked of respondents regarding their birthplace and residence during childhood. Height is a marker of net nutritional status reflecting the impact of childhood nutrition and disease (Floud, Fogel, Harris, and Hong 2011) and we use the lowest quartile of adult height in the overall population to indicate stunting. For older adults born at the beginning of the 1930s–1960s period, the seven selected countries fell into two broad groups of mortality regimes at birth characterized by: (1) declining mortality and increased survivorship of poor early life conditions due to public health interventions (Costa Rica, Mexico, South Africa) (Rosero, 1990; Rodríguez de Romo and Rodríguez de Pérez 1998; Beinart & Dubow 1995) and (2) little mortality decline and continued high infant and child mortality (China, Ghana, India, Indonesia) (Banister 1987; Patterson 1981; Patterson 1979; Caldwell 1967; Nitisastro 1970; Ramasubban 2008; Dyson 1997).

Adult health

Obesity, diabetes and hypertension were of particular interest because they are known to originate in early life due to poor nutrition and/or the synergy with poor nutrition and infectious diseases (Barker 1998). Obesity was calculated using measured height and weight. A body mass index of ≥ 30 identified obese individuals (1=obese, 0=not obese). Elderly diabetes was defined by dichotomous variables (1=diabetic, 0=not diabetic) from self-reports which are based on questions asked of the respondent about whether a doctor has ever diagnosed them with diabetes. Although self-reports for diabetes have shown validity in some settings (Banks et al. 2006; Brenes 2008; Goldman et al. 2003), there is a strong likelihood that diabetes is severely under reported in some developing countries especially among those with limited access to quality health care. While biomarkers are preferable for the measurement of diabetes, most of the selected studies did not yet have publicly available biomarkers. Hypertension was defined as a dichotomous variables (1=hypertensive, 0=not hypertensive) using criteria described in the literature (Yan et al. 2012). Hypertensive respondents showed systolic ≥ 140 mmHg, diastolic ≥ 90 mmHg, or reported that they were taking medication to control hypertension.

Other variables

All statistical models controlled for age, gender, years of education, visits to a doctor, and current residence. Current residence was defined to be either rural (1) or urban (0). Visits to a doctor reflects, in part, preventive health care behavior and it was defined as a dichotomous variable to reflect at least one visit to a doctor within the last year.

Sample selection

We selected older adults who were born in the late 1920s to early 1940s for those studies which had measured blood pressure, obesity using measured height and weight, and self-reported diabetes in addition to childhood variables (rural birthplace, height). Imputation methods using Stata were used to address missing values (Raghunathan, Reiter, and Rubin 2003).

Analyses

Multivariate models using pooled country data for all seven countries and for only SAGE countries were estimated to examine the likelihood of older adult obesity, diabetes, or hypertension as a function of country and other predictor variables. Likelihood ratio tests were conducted comparing constrained and unconstrained models to determine the importance of including country dummy variables in models. Interaction terms between countries and birthplace and countries and low height were included in models to examine the likelihood of disease for those most at risk of manifesting the effects of poor early life conditions between low and middle income countries. Predicted probabilities for pooled models were calculated and compared between countries using average responses on model variables.

Before pooling the data, country-specific models were estimated and associations between predictor variables and adult health compared to ensure comparability regarding the direction of associations across the different countries. We also compared country-specific models and pooled results using weighted and non-weighted models for only the SAGE countries because SAGE countries had similar sample designs and questionnaire construction. There were few differences noted between models with and without sample weights. Similarly, there were few differences noted between models using all seven countries versus only SAGE. Thus, reported results are based on non-weighted models.

Results

Sample Characteristics

Demographic and health characteristics of older adults born in the late 1920s to early 1940s reveal a population of older adults who share some similarities but also differences in early life conditions and who differ in terms of adult health patterns, health systems use and changes in residence from rural to urban settings (Table 2). Across countries, there were lower levels of formal education achieved both for the respondent and parents, with particularly low parental education seen in Ghana and China. Average height for both males and females was similar across countries and suggests a stunted population. Born rural was fairly disparate across countries in that, with the exception of Costa Rica, a higher percentage of respondents were born in rural areas in the low income countries. Comparing rural birthplace with rural residence suggests country differences in migration from rural to urban areas particularly for Costa Rican adults where 72% were born rural, but only 38% currently reside in rural areas. A higher prevalence of obesity and diabetes was particularly notable in the middle income countries; the highest prevalence of obesity but lowest prevalence of diabetes among middle income countries occurred in South Africa. The prevalence of hypertension in the middle income countries was slightly higher than in the low income countries. The percentage of respondents who visited a medical doctor within the last year was highest in middle income countries like Costa Rica and was lowest in Indonesia.

Table 2.

Sample characteristics for cross national data on aging populations born during the late 1920s-early 1940s in selected countries

Variables/Countries	Middle Income Countries			Low Income Countries
	Costa Rica	South Africa	Mexico	China	Ghana	India	Indonesia
Age	68 (5.3)	69 (6.0)	72 (5.7)	70 (5.4)	70 (5.5)	69 (5.1)	62 (5.0)
Female (%)	52	62	50	53	49	48	55
Childhood
Born rural (%)	72	43	32	54	61	72	84
Father no education (%)	-	58	58	74	87	68	67
Years school	5.2 (4.1)	4.9 (4.7)	4.1 (4.2)	4.7 (4.7)	3.3 (5.0)	3.2 (4.4)	3.4 (3.9)
Height (avg)
Female	150.7	154.4	148.8	152.5	156.6	148.4	147.3
Male	164.4	159.1	163.7	163.0	164.8	162.5	158.6
Adulthood
Obese (%)	25	48	26	6	8	2	2
Diabetes (%)	22	11	20	9	4	7	4
Hypertension (%)	72	82	75	69	60	38	63
Visited doctor (%)	93	69	42	60	68	88	9
Rural residence (%)	38	39	27	45	60	70	62
Never smoked (%)	58	69	57	67	74	43	51
Smoked but not now (%)	32	10	23	9	15	6	6
Current smoker (%)	11	22	20	25	11	51	43

Source:RELATE 2013, weighted, low and middle income countries where both measured blood pressure and variables on childhood conditions were available; harmonized variables where applicable; older adults born during the late 1920s through early 1940s.

Notes: The table above is based on a total sample of 16,836 respondents. Sample sizes for individual countries were: Costa Rica (1,654), South Africa (1,475), Mexico-SAGE (1,169), China-SAGE (5,363), Ghana (1,946), India (2,723), and Indonesia (2,506). All numbers are either percentages (where indicated) or averages with standard deviations in parentheses. Age is at the time of the surveys. SAGE respondents are slightly older than other older adult respondents due to timing of the surveys. The averages appearing for diabetes are age-standardized and (Ahmad et al. 2001) are shown in Figure 3. Hypertension was defined to be systolic ≥ 140 mmHg or diastolic ≥ 90 mmHg or taking medication to control hypertension. Visited doctor indicates if the respondent visited a doctor or similar medical professional at least once during the year before the survey. Residence is the percent of respondents that lived in a rural area at the time of the survey.

Country effects and early life effects

For obesity, a likelihood ratio test comparing models with and without country dummy variables indicated significant country differences in predicting obesity (χ² (6) = 1894.02, p=0.000). The likelihood of being obese was much higher in the middle income countries such as Mexico (OR 5.29, 95% CI 4.23–6.62) and South Africa (OR 8.14, 95% CI 6.62–10.00) than in low income countries such as China (OR 0.66, 95% CI 0.54–0.82) and India (OR 0.39, 95% CI 0.29–0.53) relative to Ghana. On average across all SAGE countries, there was a higher likelihood of obesity for shorter individuals (OR 1.65, 95% 1.45–1.88) and a reduced likelihood for those born in rural areas (OR 0.66, 95% CI 0.57–0.76) (Table 3, Model 1). However, adding current residence (Table 3, Model 2) produced an attenuation of rural birthplace (OR 0.90, 95% CI 0.74–1.10) but showed a significantly reduced odds for those currently living in rural areas (OR 0.64, 95% 0.52–0.79). Similar results appeared for models using all seven countries (Appendix A). The odds of obesity increased slightly with increased levels of education (OR 1.03, 95% CI 1.02–1.05). There were no significant country interactions with height and an inconsistent pattern of country interactions for residence (results not shown). Predicted probabilities for obesity using only SAGE countries (Model 2) showed strong country differences between low and middle income countries. The predicted probability of being obese was 0.26 in Mexico and 0.37 in South Africa as compared with 0.04 in China, 0.07 in Ghana, and 0.03 in India (Figure 1). Across countries there was a higher predicted probability of being obese for shorter individuals living in urban areas than taller individuals in urban areas; however, there were higher probabilities of being obese and in the first quartile of height in Mexico and South Africa (Figure 2). Re-estimating models using continuous height produced significant results for height (OR 0.95, 95% CI 0.04–0.96—results not in table) and predicted probabilities across countries and height showed higher predicted probability of being obese for shorter individuals in Mexico and South Africa (Figure 3).

Table 3.

Models predicting obesity

	Model 1		Model 2
	OR	95% CI	OR	95% CI
Age	0.97	[0.96–0.99]	0.97	[0.96–0.98]
Female	1.96	[1.70–2.27]	1.94	[1.68–2.24]
Born rural	0.66	[0.57–0.76]	0.90	[0.74–1.10]
Low height	1.65	[1.45–1.88]	1.65	[1.45–1.89]
Years school	1.04	[1.02–1.05]	1.03	[1.02–1.05]
Health utilization	1.21	[1.04–1.40]	1.20	[1.03–1.39]
Residence			0.64	[0.52–0.79]
Smoking
Never (ref)	1.00		1.00
Past	0.77	[0.62–0.96]	0.76	[0.61–0.95]
Current	0.52	[0.43–0.62]	0.52	[0.43–0.62]
Country
Middle Income
Mexico	5.29	[4.23–6.62]	5.08	[4.06–6.37]
South Africa	8.14	[6.62–10.00]	7.93	[6.45–9.75]
Low Income
China	0.66	[0.54–0.82]	0.64	[0.52–0.79]
Ghana (ref)	1.00		1.00
India	0.39	[0.29–0.53]	0.40	[0.29–0.53]

Source:RELATE 2013, older adults born during the late 1920s through early 1940s; SAGE countries only, n=12,676.

Figure 1. Predicted probabilities for obesity by country.

Source: Predicted probabilities using Model 2, Table 3; holding model variables at means. Vertical lines show 95% confidence intervals.

Notes: There are noted differences in gender (female, male):

India (0.04, 0.02); Ghana (0.09, 0.05); China (0.06, 0.03); South Africa (0.44, 0.29), Mexico (0.33, 0.20). Similar results obtained for all seven countries (results not shown).

Figure 2. Predicted probabilities of being obese by country, height and urban residence.

Source: Predicted probabilities using Model 2, Table 3; holding model variables at means; Vertical lines show 95% confidence intervals.

Notes: Similar results obtained for all seven countries (results not shown)

Figure 3. Predicted probabilities for obesity by country and and height.

Source: Predicted probabilities using Model 2, Table 3, using continuous height (cm). Vertical lines show 95% confidence intervals.

Notes: Predicted probabilities showing gender by country produced no notable differences

For diabetes, a likelihood ratio test comparing models with and without country dummy variables indicated significant country differences in predicting diabetes (χ² (6) = 345.52, p=0.000) although the contrasts between countries were not as sharp as with obesity. The odds of self-reporting diabetes was highest in Mexico (OR 5.55, 95% CI 4.24–7.26), followed by India (OR 2.44, 95% CI 1.88–3.18), South Africa (OR 2.36, 95% CI 1.78–3.14), and China (OR 2.12, 95% CI 1.67–2.70) (Table 4, Model 1). On average across all low and middle income countries, being born rural decreased the likelihood of diabetes (OR 0.49, 95% CI 0.43–0.57) (Table 4, Model 1). Adding current residence made rural birthplace insignificant (OR 0.85, 95% CI 0.70–1.03) but urban residence significant (OR 0.45, 95% CI 0.36–0.55) (Table 4, Model 2). The odds of diabetes increased slightly with education (OR 1.05, 95% CI 1.03–1.06). Similar to obesity, there were no significant interactions between countries and height and an inconsistent pattern of significant interactions between countries and residence (results not shown). Models using all countries produced similar results (Appendix A).

Table 4.

Models predicting diabetes and hypertension

	Diabetes				Hypertension
	Model 1		Model 2		Model 3		Model 4
	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI
Age	1.01	[1.00–1.02]	1.01	[1.00–1.02]	1.03	[1.02–1.03]	1.02	[1.02–1.03]
Female	1.13	[0.98–1.30]	1.10	[0.95–1.27]	1.18	[1.08–1.29]	1.17	[1.07–1.28]
Born rural	0.49	[0.43–0.57]	0.85	[0.70–1.03]	0.88	[0.81–0.96]	0.96	[0.84–1.09]
Low height	0.87	[0.76–1.01]	0.89	[0.77–1.03]	0.93	[0.85–1.01]	0.93	[0.85–1.01]
Years school	1.05	[1.04–1.07]	1.05	[1.03–1.06]	1.01	[1.00–1.02]	1.01	[1.00–1.02]
Health utilization	1.64	[1.41–1.91]	1.63	[1.40–1.90]	1.25	[1.13–1.38]	1.25	[1.13–1.38]
Residence			0.45	[0.36–0.55]			0.89	[0.78–1.02]
Obesity	1.37	[1.14–1.64]	1.34	[1.12–1.61]	1.56	[1.35–1.81]	1.55	[1.34–1.80]
Smoking
Never (ref)	1.00		1.00		1.00		1.00
Past	0.95	[0.77–1.17]	0.94	[0.76–1.16]	1.06	[0.92–1.22]	1.06	[0.92–1.22]
Current	0.64	[0.54–0.76]	0.65	[0.55–0.77]	0.85	[0.77–0.94]	0.85	[0.77–0.94]
Country
Middle Income
Mexico	5.55	[4.24–7.26]	5.20	[3.97–6.81]	1.64	[1.39–1.94]	1.62	[1.37–1.92]
South Africa	2.36	[1.78–3.14]	2.26	[1.71–3.00]	2.54	[2.15–3.01]	2.52	[2.13–2.99]
Low Income
China	2.12	[1.67–2.70]	1.96	[1.54–2.49]	1.52	[1.36–1.70]	1.51	[1.35–1.68]
Ghana (ref)	1.00		1.00		1.00		1.00
India	2.44	[1.88–3.18]	2.50	[1.92–3.26]	0.49	[0.43–0.56]	0.49	[0.43–0.56]

Source:RELATE 2013, older adults born during the late 1920s through early 1940s; SAGE countries only, n=12,676.

For hypertension, a likelihood ratio test comparing models with and without country dummy variables indicated significant country differences in predicting hypertension (χ² (6) = 753.78, p=0.000). SAGE-only models showed high likelihood of hypertension in South Africa (OR 2.54, 95% CI 2.15–3.01), followed by Mexico (OR 1.64, 95% CI 1.39–1.94), and China (OR 1.52, 95% CI 1.36–1.70) (Table 4, Model 3). Initial effects of rural birthplace became insignificant when current residence was added although current residence is also not significant (OR 0.89 95% CI 0.78–1.02) (Table 4, Model 4). The odds of hypertension increased slightly with education (OR 1.01, 95% CI 1.00–1.02). Similar results were obtained for all seven countries (Appendix A).

Discussion

Demographic changes in the 1930s–1960s due to public health interventions and medical intervention in the absence of parallel improvements in standard of living produced cohorts that may be more susceptible at older ages to the long-term consequences of poor early life conditions. Expectations for a higher prevalence and likelihood of chronic conditions and stronger early life effects on adult health among older adults born at the beginning of this period and in now-middle income countries (country types C and D) were partially met for obesity and diabetes. We found a higher prevalence of obesity and diabetes in older adults born in middle income countries but no large differences in hypertension (measured), and a higher likelihood of being obese, diabetic and hypertensive in some selected middle income countries. There was no strong evidence of stronger effects of early life conditions in the middle income countries and urban residence was more predictive of disease than urban birthplace. However, overall, being short and living in urban areas increased the likelihood of being obese and shorter individuals from middle income countries showed a higher probability of being obese. While these broad patterns in the data suggest the overall importance of early life conditions in predicting obesity in both low and middle income countries, there is insufficient evidence using available data to make a strong claim that the demographic changes of the 1930s–1960s and increasing survivorship of poor early life conditions explain the notable higher prevalence of obesity and diabetes in middle income countries in the cohort of older adults born in the beginning period of the 1930s–1960s.

A few additional results merit further discussion. First, the results show important differences between countries in disease patterns, especially concerning obesity. These differences suggest different determinants of older adult health whether they be a result of macro-level events or population differences. The higher prevalence of obesity and diabetes is not surprising given that mortality due to diabetes has been increasing in some middle income countries (Palloni et al. 2005). The higher prevalence of self-reported diabetes is higher than the prevalence of diabetes among present day older adults and older adults several decades ago in the developed world (García-Palmieri et al. 1970; Gordon 1964; Hadden and Harris 1987; Harris et al. 1998; Wilkerson and Krall 1947). A high prevalence of conditions such as hypertension has been found in urban areas (Ibrahim & Damasceno 2012). Given that studies show increasing prevalence of these chronic conditions in even low income countries (Hossain, Kawar, and El Nahas 2007; WHO 2000; Kinsella & He 2009; Murray & Lopez 1996; Lloyd-Sherlock et al. 2014; Gao et al. 2013; Mendez-Chacon, Sanatamaria-Ulloa & Rosero-Bixby 2008), differences in the timing and development of disease as shown in this study remain an important consideration in understanding its determinants.

Second, the results indicate the overall importance of poor early life conditions for adult obesity. Height is a marker of net nutritional status, and problems in intrauterine growth, due to either poor nutrition or infectious diseases, lead to low birth weight babies and stunting (Barker 1998; Crimmins and Finch 2006) with its increased risk of obesity and other conditions later in life. All seven countries in this study had very low caloric intake during the 1930s–1940s (FAO 1946), reflecting poor nutritional and infectious disease environments for most of the population. These types of conditions in early life affect infant and child health (Adair et al. 2013; Yajnik 2013; Yan et al 2012) but they also increase the risk of poor adult health (Almond and Curie 2010; Crimmins and Finch 2006; Davey Smith, Hart, and Hole 1998; Elo et al. 2010; Elo and Preston 1992; Hayward and Gorman 2004).

Third, the results indicating the importance of urban residence are not surprising given that trends in the developing world show higher prevalence of chronic conditions in urban settings among older adults (e.g. Lloyd-Sherlock et al. 2014). However, that urban dwellers and shorter individuals are more at risk for obesity suggest the merit of early life theories which stress the possible mismatch between being born in impoverished nutritional conditions and being exposed to a more nutritional rich environment later in life (Barker 1998; Schmidhuber and Shetty 2005). All of the selected middle income countries were very poor countries in the 1930s–1940s. Developing countries such as Costa Rica, China, and Indonesia experienced rapid economic growth during the adulthood of the cohort examined. Economic growth may have benefited urban dwellers more, potentially improving diet but also potentially increasing exposure to a more Western-style diet high in saturated fats. All developing countries in this study have seen an increase in caloric intake (FAO 1946; FAO 2010) but exposure to a more Western-style diet in adulthood may be more likely in urban areas. Exposure to this kind of diet may partially explain results (Popkin, Horton and Kim 2001) and the combination of a critical early period with nutritional and lifestyle changes in later life might be a lethal combination for some older adults.

Fourth, the inconclusive result regarding the possible macro-level explanation for the noted differences in the cohorts of the 1930s–1940s—namely, the long-term consequences of demographic changes of the 1930s–1960s due primarily to countrywide public health interventions but without parallel improvement in diet and nutrition—does not negate its importance for potentially explaining differences in disease. However, it may be too early to fully explore the merit of this explanation. Although smaller countries were experiencing country-wide public health interventions with improvements in mortality in children and infants during the 1930s (Clark 1930; Rosero-Bixby 1990) and improvements were occurring in urban areas in the larger present-day middle income countries (Rodríguez de Romo and Rodríguez de Pérez 1998), the full impact of these demographic changes may not have occurred until after 1945 with the introduction of antibiotics and with complete country-wide implementation of public health interventions in the larger countries. Fully-implemented public health interventions would have resulted in increasing survivorship of the more vulnerable portion of the population, namely those living in rural areas, resulting in sharp contrasts in adult health between vulnerable members of the population and everyone else.

The results showing that increased years of education was associated with higher likelihood for obesity and diabetes suggests a possible reversal of expected patterns of health as described in other studies (Monteiro et al. 2004). However, the effects of low height were much stronger for obesity. Given that obesity is strong predictor of adult diabetes and hypertension, the results for obesity may be a precursor to what we will observe with diabetes and hypertension as countries undergo complete transitions. Thus, while the comparison between older adults born in the 1930s–1940s provides a hazy glimpse at the impact of macro-level events, the question remains as to whether the results foreshadow in any way the health of cohorts of older adults born later in the period—the 1940s–1960s in low income countries.

There are a number of other possibilities that explain the weaker results for diabetes and hypertension. Although self-reports for diabetes show validity in some settings (Banks et al. 2006; Brenes 2008; Goldman et al. 2003), underestimation of diabetes using self-reports is undoubtedly problematic in the developing world where access to good medical care and diagnostic tests is not wide spread. Those who are shorter are more likely to also be poorer with less access to these services. Additionally, there are undoubtedly large differences in health care infrastructure and urbanicity between low and middle income countries—middle income countries having a stronger health care infrastructure with a larger proportion of the population living in urban areas where older adults would be more likely to receive quality care and proper diagnosis of diabetes. In this study low income countries such as Ghana, India and Indonesia indeed showed a higher percentage of respondents living in rural areas. Even though we examined urban residents only, existing differences in health care infrastructure could have resulted in more severe underestimation in low income countries. If we also consider that these differences are also most likely to be correlated with mortality regimes and economic growth and development, we must conclude that this cross national analysis between low and middle income countries using self-reported diabetes must be subject to cautious interpretation.

The weaker results for hypertension are surprising in that several studies have shown strong associations between low birth weight, infant mortality rate, poor SES, and adult hypertension (Barker 1998, Johnson and Schoeni 2011; Lawlor and Davey Smith 2005; McGovern 2012). However, similar results obtained for height and diabetes suggests that the pathway from stunting to diabetes and hypertension may operate mainly through obesity (Popkin, Horton and Kim 2001). Thus, height may not be the best measure to sufficiently capture associations with adult diabetes and hypertension.

The study showed large differences among countries in terms of moving from rural to urban residences at some time during the life course. However, complete information about migration is limited in some surveys of older adults making it harder to determine the effects of this migration. It was also not possible to observe on an individual level factors which may have mitigated the effects of poor early life conditions such as economic growth or factors that may have either compounded their effects or have played a larger role in explaining adult health such as diet. Rapid economic growth may help explain adult health for those exposed to poor early life conditions (Steckel 2013) but speed of economic growth during adulthood might not be as important as who benefits from economic growth. Not all economic growth in the developing world has benefited the entire population (López-Alonso 2007). There is little information on individual diet during childhood or adulthood of older adult respondents in population-based studies and diets differ tremendously between Latin America and Asia making the interpretation of the results less clear in determining the relative importance of early life versus events such as the nutrition transition to a diet higher in saturated fat.

Additionally, the nature of the study prohibits the possibility of disentangling precise mechanisms in early life associated with older adult health because there is a complex synergy between nutrition and infection (Scrimshaw 1968). Birthplace is a broad measure that might also reflect epidemiological differences between rural and urban settings. Height could also be problematic given the imprecision of lowest quartile of height as a measure of stunting and lost height at older ages. Some of the SAGE countries are representative of a particular province or region but not the country as a whole. Mexico and South Africa may be special cases, although there is evidence of increasing prevalence of these chronic conditions, especially in the Latin American region (Palloni et al. 2002). The comparison by mortality regimes is based on analyses of historical mortality data from the early 20^th century (McEniry 2014) but these broadly defined mortality regimes may need refinement. Although we conducted analyses to show similar direction of associations between predictors and health outcomes by country, the estimated models make an important assumption that countries can be pooled together.

In spite of the study limitations, the topic of early life conditions and older adult health in the developing world remains important and individual-level survey data of older adults is one of the better sources of data that we have to examine the topic. Even though the examination of trends in health in the developing world is complex, cross national comparisons provide insight into health patterns and their determinants (National Research Council 2001). Rapid changes affecting developing countries and evidence from the middle income countries pose relevant questions about the long-term consequences of macro-level events on health at older ages. Further examination of those born in the late 1940s–1960s will be essential to more fully examine if these questions have merit.

Appendix A

Table A1.

Models using all seven countries predicting health outcomes

	Obesity		Diabetes		Hypertension
	OR	95% CI	OR	95% CI	OR	95% CI
Age	0.97	[0.96–0.98]	1.01	[1.00–1.02]	1.03	[1.02–1.04]
Female	1.97	[1.73–2.24]	1.07	[0.95–1.22]	1.21	[1.11–1.30]
Born rural	0.91	[0.79–1.06]	0.78	[0.68–0.90]	0.98	[0.89–1.08]
Low height	1.53	[1.36–1.73]	0.85	[0.75–0.97]	0.93	[0.87–1.00]
Years school	1.03	[1.02–1.04]	1.03	[1.02–1.05]	1.00	[1.00–1.01]
Health utilization	1.26	[1.09–1.45]	1.75	[1.51–2.03]	1.22	[1.11–1.34]
Residence	0.63	[0.54–0.73]	0.56	[0.48–0.65]	0.87	[0.79–0.96]
Obesity			1.59	[1.37–1.85]	1.61	[1.42–1.84]
Smoking
Never (ref)	1.00		1.00		1.00
Past	0.87	[0.74–1.04]	0.90	[0.77–1.07]	1.02	[0.91–1.15]
Current	0.47	[0.39–0.56]	0.63	[0.54–0.73]	0.84	[0.77–0.91]
Country
Middle Income
Costa Rica	3.50	[2.82–4.36]	5.04	[3.90–6.51]	1.69	[1.45–1.96]
Mexico	5.12	[4.09–6.42]	5.31	[4.06–6.94]	1.59	[1.35–1.88]
South Africa	8.12	[6.61–9.97]	2.22	[1.68–2.93]	2.53	[2.14–2.99]
Low Income
China	0.65	[0.53–0.80]	2.08	[1.64–2.64]	1.50	[1.35–1.68]
Ghana (ref)	1.00		1.00		1.00
India	0.40	[0.30–0.55]	2.47	[1.90–3.21]	0.50	[0.44–0.56]
Indonesia	0.27	[0.19–0.38]	1.55	[1.10–2.18]	1.77	[1.53–2.04]

Source:RELATE 2013; older adults born during the late 1920s through early 1940s; all seven selected countries, n=16,836.