Survival advantage in selected populations Vanessa G. di Lego - - PDF document

Survival advantage in selected populations

Vanessa G. di Lego (CEDEPLAR-UFMG) Cassio M. Turra (CEDEPLAR-UFMG)

Introduction

In the demographic study of mortality, there has been growing attention to population subgroups that are more likely to first benefit from mortality progress or benefit more intensively than others. Some authors define these subgroups as “vanguard” populations and the reasons for the increased interest in these groups are threefold. First, mortality trajectories

• f vanguard populations can be instrumental to disentangling the pathways to longer lives

(Evgueni et al. 2014). In addition, the survival advantage of these selected groups can help reveal the distribution of mortality gains within and between countries at different stages

• f the health transition. This seems to be particularly important in a context of increasing

mortality differentials not only across countries, but also at the sub-population level that has characterized the second half of the 20th century (Mackenbach 2003; E. M. Andreev et al. 2011; Caselli, Meslé, and Vallin 2002; McMichael et al. 2004; Moser, Shkolnikov, and Leon 2005). Third, there is growing availability of high-quality mortality data for vanguard groups, including in middle-income economies, which has offered the opportunity of novel survival 1

analyses in different socioeconomic contexts (Lego, Turra, and Cesar 2017; Luy, Flandorfer, and Di Giulio 2015; Luy 2003). A recent work (Evgueni et al. 2014) explored the mortality paths between vanguard and non-vanguard population subgroups using individual census-linked mortality data in three nordic countries: Finland, Denmark and Sweden. The authors defined as vanguard groups the married and highly educated. The study compares trends in life expectancy and mortality by cause of death between this vanguard group and the rest of the population from the 1970s to the 1990s. The results show no sign of convergence between the higher and lower mortality groups, indicating that non-vanguard have their own pathways to transition to lower mortality schedules, which in turn are related to specific determinants of mortality changes. Along that line, some studies have examined the survival advantages of subgroups that represent small but selected fractions of the populations, including certain learned societies (E. M. Andreev et al. 2011; Winkler-Dworak and Kaden 2014), master athletes (Lemez and Baker 2015; Teramoto and Bungum 2010), religious groups (Enstrom and Breslow 2008; McCullough et

• al. 2000; Luy, Flandorfer, and Di Giulio 2015; Luy 2003), and the military (D. L. Costa 2012;
• D. L. Costa 2003; D. L. Costa and Kahn 2010). Usually, data on military mortality has been

employed by those who want to learn more on survival under extreme conditions, such as war and famine, as well as on the mechanisms responsible for mortality differentials over the life

• cycle. However, in countries that have not been involved in internal conflicts or foreign wars,

the military do live in favorable conditions, with officers being positively selected with respect to health (Lego, Turra, and Cesar 2017). Also, when we consider that in developing countries there is often little or inaccurate information on SES gradients in mortality that allow for 2

a thorough trajectory analysis (C. M. Turra, Renteria, and Guimaraes 2016; Silva, Freire, and Pereira 2016), military data may help reveal the upper limits of mortality gradients and the gap between the most and less advantaged groups of the population. In addition, the tradition of the military in collecting vital data can be valuable in low and middle-income countries where it remains unclear the magnitude and determinants of mortality differentials. Taken all those aspects into consideration, this work addresses the following question: if such differences in mortality trajectories between vanguard and non-vanguard population subgroups are even found in countries with developed and strong welfare systems (Evgueni et al. 2014),how are those trajectories operating in developing country contexts? A second related question is: are the characteristics of the vanguard population subgroup in a developing conext the same as that in a developed one? In order to explore this question we use a novel and longitudinal military dataset for Brazilian Air Force personnel, considering them as a vanguard population subgroup in Brazil. In addition, we use other available data on subgroups that experience lower mortality, such as the recently developed life tables for insured Brazilian lives (year 2015) (De Oliveira et al. 2016), US military life table (year 2015), the Retirement Plans Experience Committee of the Society of Actuaries (SOA-USA) life

• tables. We also use UN Brazilian life tables (years 1950-2000) and official mortality data for

Brazil (IBGE, year 2015), in order to improve our comparisons. Lastly, we derive life tables from recent mortality estimates for Chilean males with more than 13 years of schooling (years 2001-2003) (Sandoval and Turra 2015), as an indicator of vanguard mortality experience in a Latin American country with available data on deaths by education. 3

Background

Although it is reasonable to assume that members of the military are healthier than the average citizen, it is still somewhat surprising that they experience lower mortality than most groups in society, even in the context of extreme settings such as famine (D. L. Costa 2012; S Horiuchi 1983), and war (Buzzell and Preston 2007). Buzzell and Preston (2007) estimated the death rates of United States troops in Iraq since the beginning of the U.S

• invasion. According to the authors, death rates for black males aged 20-34 in 2002 living

in Philadelphia were 9% higher than for troops in combat in Iraq. A much earlier work from MacIntyre et al. (1978), which followed the U.S Navy’s cohort of 800 survivors from World War II and the Korean conflict, showed that pilots experienced not only lower all-cause mortality, but also lower mortality from cardiovascular, cancer, and accidental causes when compared to the U.S civilian population. The main explanation, according to the author, was the generally good socioeconomic background of members of the military. He also speculated about the positive genetic influence of long-lived parents, above average intelligence, and the health and fitness orientation of the military aviators (MacIntyre et al. 1978). Some scholars have argued that the survival advantage of the military accrues from a selection bias at enlistment or recruitment, sometimes called the “healthy soldier effect” (McLaughlin, Nielsen, and Waller 2008; Shah 2009), in which the selection of the healthier and fitter results in lower mortality rates. Variations in risk of death among the military officers, particularly during times of war, reflect several underlying factors including military branch and service component, as well as rank in service. Buzzell and Preston (2007) showed that these characteristics affect exposure to combat and therefore, the variability of mortality risk 4

across subgroups. Other research has used military data to approach epidemiologically the pathways through which more risky and stressful situations affect survival and causes of death within a life cycle perspective. Models of mortality selection posit that insults at younger ages can yield individuals that are more robust at older ages (S Horiuchi and Wilmoth 1998; Manton and Stallard 1981; Preston et al. 1996). However, there is also evidence of positive associations between hazardous life events and morbidity and mortality later in life (Finch 2004; S Horiuchi 1983; Kannisto, Christensen, and Vaupel 1997; Preston et al. 1996). Recent research evidenced that veterans from the American Civil War who experienced greater stress in battle had higher mortality rates at older ages and were at greater risk of developing Post-Traumatic Stress Disorder (PTSD) (D. L. Costa and Kahn 2010). Other work on PTSD show that exposure to military trauma can impact physical health in later years among veterans both in the U.S and Europe (E. C. Clipp and Elder 1996; G. H. Elder et al. 2009; Chatterjee et al. 2009). Some scholars have also taken advantage of military data to explore the relationship between nutritional status and exposure to infections earlier in life with subsequent mortality levels (Fogel and Costa 1997). D. L. Costa (2003) found that stomach ailments while in the army significantly predicted mortality from all causes, and from some specific chronic diseases. In addition, respiratory infection significantly predicted mortality from respiratory illnesses and acute illnesses, when other wartime disease covariates were excluded (D. L. Costa 2003). In another set of studies, Fogel and Costa (1997) used data on Civil War soldiers to show that poor body builds increase vulnerability to both contagious and chronic diseases, and can be powerfully predictive of morbidity and mortality at later ages (Fogel and Costa 1997; 5

• D. L. Costa 2012).

Another important line of research places military survival advantage within the occupational

• literature. This perspective focuses on military as an occupation among others, and thus

also include analysis from military at peace times. In this case, the framework is that of the healthy worker effect, i.e., fact that workers have lower death rates than the general population, since good health is related to the ability of securing and maintaining employment, while the general population includes sick and disabled people who may be at greater risk of dying (Choi and Noseworthy 1992; Goldsmith 1975; Shah 2009). The military occupation is a specific case of this phenomenon, where individuals experience an even higher degree of selection, coined in the literature as the “healthy soldier effect” (Guest and Venn 1992; Kang and Bullman 1996; McLaughlin, Nielsen, and Waller 2008). Kang and Bullman (1996) showed that strict physical criteria for recruitment, a requirement to maintain a certain standard of physical well-being, and better access to medical care during military service explain the lower mortality among military personnel relative to the general population. McLaughlin, Nielsen, and Waller (2008) summarized the evidence comparing mortality rates of military personnel with the general population and quantified the magnitude of the healthy soldier effect focusing

• n nineteen primary studies that investigated the association between service in the military

and mortality rates. They showed that most studies presented standardized mortality ratios (SMR) less than 1, indicating that military populations had lower all-cause mortality than the general population. A study that focused only on US retired civil airline pilots during years 1980-1989 also showed that their life expectancy was 5 years higher than the average white male, suggesting that not only military pilots, but also civil aviators, present a survival 6

advantage relative to their national counterparts (Besco et al. 1995). However, one study showed that military cooks and pay clerks presented higher all-cause mortality than the general population, indicating that the healthy soldier effect is not applicable to all military rank or career types (Coggon and Wield 1993). In addition, research on the submariners in the Royal Navy showed that external deaths accounted for a SMR of 115 relative to the general British population (Inskip, Snee, and Styles 1997), also supporting that deaths due to external causes are not trivial in the military population. In previous research (Lego, Turra, and Cesar 2017), we estimated mortality levels of a cohort

• f 808 officers from the Brazilian Air Force (BAF) born on average in 1935, and compared

them with males in Brazil and in low mortality countries. The results showed that BAF life expectancy was higher than the average Brazilian male and comparable to Sweden, France and Japan in 2000. We also found that younger pilots were in a higher risk of dying on duty when compared to other officers, but experienced lower mortality rates from other causes at advanced ages. These first findings suggest how this specific population subgroup is very longevous and can be considered as a vanguard population subgroup for analysis. However, the small number of cases of this previous work was an important limitation that led to some statistical uncertainty when comparing BAF estimates with those from other populations. In this present work we update the previous estimates to include 39,799 individuals, with a median follow-up time of 38.39 years and maximum follow-up 72.91 years. We expanded the timeframe to consider military personnel who enrolled in the institution from 1943 to 2000, and were followed until the end of period study (December 3rd, 2015) or death (total

• f 4,454 deaths throughout the period). The analysis of this Brazilian subgroup, approached

7

within a vanguard mortality perspective, may contribute to improve our understanding

• f the phenomenon, especially in a developing country scenario where lack of reliable and

longitudinal data leaves demographers and other researchers with greater uncertainty as regards mortality schedules.

Data

We use different sets of data for performing the estimates. The data from the Brazilian Air Force military was retrieved from their personnel information system, collecting data from individuals who entered the institution from 1943 to 2000. It is a retrospective longitudinal database where we can track each individual from birth to death, and capture their transitions into the institution and later on the pension fund system. We were able to initially get raw data on N=97,625 individuals and 10,638 deaths. However, we had to exclude those career types that do not necessarily enter the pension fund system and so do not have a proper follow-up, which include soldiers and temporary officers. In addition, we excluded incomplete cases where precise dates for transition (i.e., dates of birth, death and retirement) were not available or inaccurate. This left us with a total of N=39,799 individuals and 4,454 deaths. In order to descriptively show the survival follow-up of all officers, we present Lexis diagrams split into two 30-year distribution groups, which can be considered as (left-hand side of pannel) pre- and (right-hand side of pannel) post mortality transition periods in Brazil, considering that the transition shows its first signs of mortality decline in the 1940s, but unfolds mostly between 1970-1991(Sastry 2004). Figure 1 shows the time span encompassing 8

Figure 1: Cohort Follow-up of Brazilian Air Force officers, 1940-1970 and 1970-2000.

• ur cohort of military members, allowing us to perform a unique mortality analysis that

accounts for pre- and post-transition mortality experiences. This dataset is also composed of individuals from different career types in the military. As a matter of detailing the data characteristics to the reader, we refer to general scheme on Figure 2. As the pyramid shows, the non-commissioned ranks are the basis of the military hierarchical structure and have a larger contribution in terms of absolute size, particularly sub-officers. On the top of the pyramid we have the commissioned officers, a group composed by Academy officers (pilots, infantry and intendants), the most selected members of the military career, and the non-Academy officers (professionals who finished their undergraduate

• r technical course outside the military institution, such as engineers, doctors, dentists,

pharmacists, religious ministers, and other technical fields). The difference between those two career paths is that commissioned officers undergo a stricter process of recruitment 9

and immersion in military life, possibly reaching the highest ranks in the Air Force; while non-commissioned ranks are trained to take part in the military service (which includes basic service such as cooking and cleaning), music, maintenance, infra-structure, communication (air traffic control), health areas, and are not able to reach the highest ranks. Within commissioned officers one can also distinguish between Academy and non-Academy

• career. To become an Academy officer one must enroll in the institution through military

schools and academies (BAF 2014). The course lasts from 4 to 7 years in military schools, living weekly in those institutions and only going home during the weekends. It is a complete immersion in military life and culture, where participants receive full nutritional and physical support, wages, and daily military training. The eligible age range for entering military school is from 14 to 19 years old (high school course), and for the military academy is from 17 to 23 years old (undergraduate course)(Brasil 2011). On the other hand, Non-Academy

• fficers undergo competitive exams to enroll in the institution. Despite less strict, they also

go through military training (depends on the field of specialization, but in general lasts 12-24 months) and can reach higher ranks in the Air Force, as well as receive full wage upon retirement. In addition, they have different career paths and rank limits. For example, doctors and engineers can become generals, but the highest possible rank for dentists and pharmacists is colonel, while for specialists with technical skills is captain. None of them can reach the highest possible rank in the Air Force (Lieutenant General, reserved only for career pilots) and their age limit for enrolment is 36 [@]. In addition, candidates who apply for the Academy undergo the strictest stages of selection in order to enroll as pilots, intendants (administrative officers), or infantry (operational and intelligence officers). The 10

stages comprise: 1. Written exams (Math, Physics, Portuguese, English and Essay); 2. Psychological inspection; 3. Health inspection; 4. Physical Conditioning; and, exceptionally for pilots 5. Tests for measuring motor coordination specifically targeted at the ability to fly. With respect to the psychological criteria, they are mainly related to adaptation to extreme conditions without losing focus or logical reasoning, as well as organization, adoption of norms and cooperation skills. Regarding the health inspection, it includes complete blood tests, images of all bone structure of the body, teeth and gum health, anthropometric measures, circulatory exams, neurological exams, and hearing and sight tests (Brazil 1980). Candidates cannot have diabetes or other 189 diverse health related issues to enter the military career. Also, it is not surprising that the criteria for selecting pilots are stricter than those for officers who want to follow the administrative path. For example, youngsters who want to become aviator officers must not weight under 58.65 kg or over 93.53 kg, and must be 164 to 187 cm

• tall. In addition, they must be under 23 years of age. To enter for non-commissioned ranks the

main eligibility criteria is to have at least a high school degree. They are then trained to take part in the military service (which includes basic service such as cooking and cleaning), music, maintenance, infra-structure, communication (air traffic control), and health areas. Most non-commissioned ranks (all present in this dataset), just as the commissioned ones, have the right to retire and leave beneficiaries, and so also take part in the pension fund system. They cannot move to the higher ranks in the military, entering as Sergeants, Stewards or Sub-officer and remaining on those ranks throughout their lives. Lastly, we remind the reader that rank is the hierarchical system of the military. For sake of analysis we have grouped that variable into the following broader categories: General, Senior, Intermediary and Subaltern officer, representing the ranks for commissioned officers; and Sub-officer, Sargeant and Corporal 11

Figure 2: Data characteristics, according to BAF hierarchical structure. representing the career ranks for the non-commissioned. We, however, do not distinguish between commissioned and non-commissioned ranks in these first mortality estimates. We also use three different sets of data that come from life tables employed and developed by actuaries: 1. the Male RP-200 Rate (2000 Mortality Table - Male Aggregate - Combined Healthy Participant, Retirement Plans Experience Committee of the Society of Actuaries/SOA- USA), which is used by the Brazilian Department of Defense for military pension; 2. the recently updated Brazilian private insurance life table BR-EMS 2015 (De Oliveira et al. 2016), to be adopted by private Brazilian insurance companies; and 3. life tables used by the US Department of Defense (DoD) Office of the Actuary for their military pension scheme. This life table is published on an yearly basis in their Statistical Report on the Military Retirement 12

System (we used here as a reference the fiscal year 2015 report: MRS/July 16, 2016). In addition, to improve the quality of our comparitive analysis, we use Human Mortality Database (HMD) estimates for years 1940-2000 for Sweden, Japan and USA, coupled with Brazilian official mortality statistics (IBGE 2015), and UN Brazilian life table estimates reconstructed by 5-year cohort groups (1950-2000). Lastly, in order to provide a survival experience standpoint from a neighbooring Latin American candidate for vanguard subgroup, we built life tables based on the recent adult mortality rate estimates for the highly educated Chilean males (2001-2003)(Sandoval and Turra 2015).

Problems and Limitations

The first step in organizing and cleaning the data we adopted was to deal with inaccuracies. Data need to be consistent in order to be fit for statistical analysis. Consistency can be understood to include in-record consistency (no contradictory information is stored in a single record), cross-record consistency (statistical summaries of different variables do not conflict with each other), and cross-dataset consistency (dataset is consistent with other datasets pertaining to the same subject matter ) (Jonge and Loo 2013). In this database, the main problem is either missing, improper, or unreasonable dates. A previous investigation performed by the Ministry of Defense for improving data quality for all Armed Forces had already indicated the existence of some of these issues, and our estimates were compatible with their figures for the Air Force (DoD 2016). In the following section, we show additional verification tests that we performed in order to assess for data quality. However, considering 13

the problems we have in regular databases, and especially in longitudinal ones (especially missing or loss to follow-up), the following tests and investigation evidences the high quality

• f this database.

We used the R package editrules to perform constraints, Boolean tests and other verification

• procedures. Since it is not uncommon for data inconsistency to be dependent upon multiple

variables, the sources of error become highly obscure, and that is why we chose to use this package to accurately track down any problems and keep them organized in an edit file that can be run by other users. The criteria imposed were based on the Brazilian Air Force Law restrictions billed by the Civil House in Brazil (Brasil 2011). Since we are dealing with a database that encompasses different types of military career, we know from the Brazilian Law 12.464/Art. 20 that individuals must not be younger than 14 years of age or older than 45 years of age to be eligible for the different types of military career we are dealing with. So, we constrained age at entrance to 14 < x < 45. There are 1,213 individuals who violated this constraint (1.2% of the total sample). Out of these cases, 92% or 1.1% of the total sample refers to ages at entrance x ≤ 14 (including some cases with negative ages) and the rest (8%

• r 0.1% of the total sample) are cases with entrance at ages 45 years and older. For additional

verification, we also assumed a 110-year lifespan limit, in order to restrict survivorship to that

• age. 10 cases presented unreasonable deaths before entrance (i.e., negative ages at death),

and an additional 354 cases presented ages at death that were beyond the lifespan limit. The latter were mainly due to problems in the missing birthdates that show up in the system, leading to unreasonably high ages at death. We dropped those cases in order to keep only complete and reasonable cases for full follow-up. 14

Analytical Approach

Mortality rate estimates

First, we employ a smoothing spline Poisson regression model as a function of age for the analysis of mortality data and computing smoothed mortality rates with confidence intervals. This model is a robust non-parametric approach that uses penalized likelihood estimation method and provides approximate Bayesian posterior confidence intervals (Gu 2014; Gu and Wahba 1993). It is also often regarded as a penalized generalized linear model (PGLM) in which the log likelihood is modified by a difference penalty on the regression coefficients (I. D. Currie, Durban, and Eilers 2004; M. Durban, Currie, and Eilers 2002). Demographers have discussed in depth the advantages of this approach to estimate mortality rates, specifically when one has a smaller number of deaths (but not many null number of deaths) and a pattern that varies from one year to another. Mainly, the method limits artificial fluctuations in estimated values(Shiro Horiuchi et al. 2014).For more details on the penalized likelihood estimation procedure please also refer to Paul H. C. Eilers and Marx (1996) and Gu and Kim (2002). For Poisson model with penalized likelihood estimation, let Dx be non-negative numbers that represent deaths observed at age x and Ex be non-negative numbers that represent the exposure of persons under observation at age x. In this case,Dx is a Poisson distributed random variable with parameter Exλ(x): Pr(Dx = k) := 1/k!(λ(x)Ex)k exp(−Exλ(x)) (1) 15

Where in Equation 1 we assume that the force of mortality is constant within integral ages. The model terms are sums of unpenalized and penalized terms and to every penalized term there is a smoothing parameter. In addition to the smoothing splines approach, we also estimate the one-dimensional regression P-splines, which are B-splines with penalties using the R package “Mortality Smooth” (Camarda 2012). The latter has the advantage of combining the flexibility of non-parametric methods with the stability and simplicity of parametric smoothers (Paul H. C. Eilers and Marx 1996, Paul H C Eilers and Marx (2010)). In order to estimate this set of mortality rates, we first compute mortality rates for the whole group from 1943-2000 and compare with other HMD countries using the same approach. In addition, we compute mortality rates for officers who go to academy and receive a deeper military training and also for two equally sized cohort groups for initially exploring any difference between entrance cohorts. We use the R package “gss” for performing all estimates (Gu 2014).

Life tables

Second, we use the mortality rate estimates from the previous section as imputs for computing the BAF male Life tables by 5-year period (from 1940 to 1995) using the method described in S. Preston, Heuveline, and Guillot (2001). Using the same method, we use as imputs the UN mortality rate estimates from 1950-1995 available by 5-year period and reconstruct the Brazilian cohort, and also compute the life tables for the highly educated Chilean males from 2001-2003. The resultant life expectancies starting at exact age 40 from year 1950 to 1995 for Brazil are used for comparing mortality trajectories. The life expectancies from the other population subgroups (Brazilian insured lives, US military retirement, Brazilian military 16

retirement) are simply retrieved from their respective life tables and their mortality levels are compared and discussed.

Results

Figure 3 presents the mortality rates estimated using smoothing spline Poisson regression for different populations between 1940-2000. We summed cohort deaths and exposure by single-year ages (15-100) throughout the years for a couple of HMD countries, in order to capture all the mortality heterogeneity that is present throughout those years and also present in our cohorts of officers. With this, we are able to provide an overarching comparison of the mortality experience of BAF officers with those countries throughout those years. Figure 3 indicates that BAF mortality for those cohorts entering the institution between 1943-2000 is lower than all the countries considered for the same periods. This suggest a high degree

• f survival advantage among BAF officers compared to the male average of other countries.

When we analyze more closely BAF officers’ mortality rates according to academy criteria (underwent academy training or not), as shown in Figure 4, we can see that the mortality rates below age 30 are significantly higher for academy officers than for those who did not undergo this heavier military training, even when considering different smoothing methods. This set of mortality rates has broader confidence intervals after the age of 80, since there is a lower number of deaths and exposure at those ages, leading to higher uncertainty. For the purposes of this work, however, we will use the mortality experience of all career types considered together, and the mortality rates estimated from the p-spline method are used as 17

Figure 3: Mortality rates (log-scale) using smoothing P-spline Poisson regression for BAF, Japan, and Sweden between 1940-2000. 18

Figure 4: Mortality rates (log-scale) according to different smoothing methods, no academy and only academy officers, ages 15-90, 1943-2000 Cohorts. imputs for the life table estimation. These estimates, together with previous results, indicate that Brazilian Air Force officers are a suitable candidate for a vanguard population subgroup within the Brazilian population, for whom we have accurate data. As Figure 5 indicates, our results for the Brazilian case are consistent with that of Evgueni et al. (2014): BAF population, considered here as our vanguard subgroup in Brazil, has a mortality trajectory paralell to their national counterparts. That is, between 1950-1990, period where most of the mortality transition unfolds in Brazil, there are no signs of convergence between BAF personnel and Brazilian civilians. On the other hand, Figure 5 shows the mortality experience of this cohort set of BAF personnel relative to other population subgroups that are probably longevous: the insured lives of the Brazilian private market, US military personnel, highly educated Chilean males 19

Figure 5: Life expectancy of vanguard and non-vanguard population, Brazil, 1950-1990. 20

Figure 6: Adult Life expectancy at exact ages, selected male population subgroups, 1950- 2015.\label{ex.comp_all and the actuary period life table used by Brazilian military personnel to estimate pension benefits for the members of the Air Force. Not surpringly, the US armed forces experience the highest adult life expectancy at all ages, starting from age 40, relative to all other subgroups. More striking, however, is that they are immediately followed by the Brazilian insured lives’ experience. The life table used by BAF actuaries present a survival curve that is right below that of the insured lives. Highly educated Chilean males experience higher life expectancy than the Brazilian general population in 2015 after age 55, and also than the BAF cohorts that we analyze more closely and than the Brazilian cohort of 1950. Nonetheless, there are important time frame differences to account for. Still, even considering that there has been an increase in life expectancy on all 21

subgroups, highly educated Chilean males in 2001-2003 experience a higher life expectancy at all ages after age 55 than the Brazilian population in 2015. This figure also indicates that BAF military personnel are just one (and possibly not the best) of the possible candidates for vanguard population subgroup. However, one must account for the fact that we are comparing here the experience of historical cohorts with accurate mortality data with other period and recent mortality experience.

Discussion

In this study, we update the mortality estimates from a previous work on Brazilian Air Force military data and its implications for demographic and health research. The relevance of other work that focused on military data, as a means of analyzing the extremely negative effects of warfare related issues on health and mortality, such as starvation, stunting and Post Traumatic Stress Disorder is unquestionable. However, we argued that the military, and especially career

• fficers in a conflict-free context, are also paramount to understanding positive effects on health

and mortality. Even more importantly, in a developing country scenario where lack of reliable and longitudinal data leaves demographers and other researchers with great uncertainty regarding mortality schedules, the focus on subgroups that offer high-quality information places us closer to the whereabouts of the “true” mortality distribution. In addition, we claimed that studying more closely vanguard groups of the population might contribute to a better understanding of the upper limits of life expectancy and the mechanisms underlying survival advantages. With gained access to the pension fund system of the Brazilian Air Force, 22

we were able to examine the mortality pattern and level for this highly selected subgroup of individuals relative to their national counterparts and other countries or other population

• subgroups. The individuals are not only physically, psychologically and cognitively selected to

become officers, but they are also exposed to a healthy and controlled environment since their admission in the institution. All the aforementioned factors, coupled with the fact that the military institution has a tradition of collecting good quality vital data, and that researchers suffer from a lack of accurate information on deaths by SES status in Brazil, enable us to evaluate if mortality trajectories among those who are potentially more longevous pave the way towards lower mortality in a developing country scenario. Our results stress BAF’s survival advantages relative to their national counterparts’ experience and also to other low mortality countries when making cohort-cohort comparisons. However, when compared to Brazilian insured lives mortality estimates for 2015, this advantage loses relative importance. Nonethless, one must remember that members that first enrolled in this BAF’s cohort were born on average in 1935, many years before Brazil’s demographic

• transition. The lack of further cohort data prevents us from performing more appropriate

comparisons. This research also points to the importance of focusing on subgroups of the population as a strategy to study mortality. The remarkable mortality decline during the second half

• f the twentieth century was also followed by an unprecedented growth in mortality and

health inequalities, not only between countries, but also between subgroups in a specific population(Mackenbach 2003; V. M. Shkolnikov et al. 2012). That scenario has led to a greater interest on studying vanguard groups, who benefit from mortality progress first or 23

more intensively than the rest of population. Previous work has shown that there were no signs of convergence in life expectancy between vanguard and non-vanguard groups, even in egalitarian Nordic countries. On the contrary, divergence occurred in all age ranges(Evgueni et al. 2014). Therefore, examining the longevity of vanguard groups such as military officers provide one possible indicator for the current upper bounds for male survival in Brazil, a country characterized by extreme socioeconomic, racial and regional inequality. Finally, mortality data, particularly at older ages, are defective in Brazil. Although our sample is limited to a particular subgroup, our data are accurate, longitudinal and contain complete information on ages at birth and death in a substantially better way than in

• ther traditional mortality datasets in Brazil. Therefore, we argue that analyzing selected

subgroups or vanguard populations in such a context is key to provide more accurate clues about mortality and health outcomes for the entire population.

Acknowledgements

I thank the CNPq Brazilian institution for funding my research and providing me with the

• pportunity to develop my work. I also thank all members of Cedeplar, my home institution,

for all the intelectual and friendly support. 24

References

Andreev, Evgeny M., Dmitri Jdanov, Vladimir M. Shkolnikov, and David A. Leon. 2011. “Long-term trends in the longevity of scientific elites: Evidence from the British and the Russian academies of science.” Popul. Stud. (NY). 65 (3): 319–34. doi:10.1080/00324728.2011.603428.

• BAF. 2014. “ICA.” http://www.fab.mil.br/icas/ICA{\_}160-6{\_}MAR{\_}2014.pdf.

Besco, Robert O, Satya P Sangal, Thomas E Nesthus, and Stephen J H Veronneau. 1995. “A Longevity and Survival Analysis for a Cohort of Retired Airline Pilots.”

• Brasil. 2011. “Dispõe sobre o ensino na Aeronáutica e revoga o Decreto-Lei no 8.437, de 24 de

dezembro de 1945, e as Leis nos 1.601, de 12 de maio de 1952, e 7.549, de 11 de dezembro de 1986.” Brasília, D.F: Civil House. http://www.planalto.gov.br/ccivil{\_}03/{\_}Ato2011- 2014/2011/Lei/L12464.htm.

• Brazil. 1980. “Lei n 6.880, Estatuto dos Militares de 9.12.1980, Federative Republic of Brazil.”

Brasília, D.F: Executive. Buzzell, Emily, and Samuel H. Preston. 2007. “Mortality of American Troops in the Iraq War.” Popul. Dev. Rev. 33 (3): 555–66. doi:10.1111/j.1728-4457.2007.00185.x. Camarda, Carlo G. 2012. “MortalitySmooth : An R Package for Smoothing Poisson Counts with P-Splines.” J. Stat. Softw. 50 (1): 1–24. doi:10.18637/jss.v050.i01. Caselli, Graziella, France Meslé, and Jacques Vallin. 2002. “Epidemiologic transition theory 25

exceptions.” Genus 58 (1): 9–51. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.118.4299{\&}rep=rep1{\&}t Chatterjee, Sharmila, Avron Spiro, Lynda King, Daniel King, and Eve Davison. 2009. “Research on aging military veterans: Lifespan implications of military service.” PTSD Res.

• Q. 20 (3): 1–8.

Choi, Bernard C. K., and A Lynn Noseworthy. 1992. “Classification, Direction, and Prevention of Bias in Epidemiologic Research.” J. Occup. Environ. Med. 34 (3): 265–71. doi:10.1097/00043764-199203000-00010. Clipp, Elizabeth Colerick, and Glen H.-J. Elder. 1996. “The aging veteran of World War II: Psychiatric and life course insights.” In Aging Posttraumatic Stress Disord., 19–51. Coggon, D., and G. Wield. 1993. “Mortality of army cooks.” Scand. J. Work. Environ. Heal. 19 (2): 85–88. doi:10.5271/sjweh.1493. Costa, Dora L. 2003. “Understanding mid-life and older age mortality declines: evidence from Union Army veterans.” J. Econom. 112 (1): 175–92. doi:10.1016/S0304-4076(02)00159-8. ———. 2012. “Scarring and Mortality Selection Among Civil War POWs: A Long-Term Mortality, Morbidity, and Socioeconomic Follow-Up.” Demography 49 (4): 1185–1206. doi:10.1007/s13524-012-0125-9. Costa, Dora L., and Matthew E. Kahn. 2010. “Health, Wartime Stress, and Unit Cohesion: Evidence From Union Army Veterans.” Demography 47 (1): 45–66. doi:10.1353/dem.0.0095. Currie, Iain D, Maria Durban, and Paul HC Eilers. 2004. “Smoothing and forecasting 26

mortality rates.” Stat. Model. an Int. J. 4 (4): 279–98. doi:10.1191/1471082X04st080oa. De Oliveira, Mário Moreira Carvalho, Milton Ramos Ramirez, Ricardo Milton Frischtak, Rafael Brandão de Rezende Borges, Bruno Costa, and Ricardo Cunha Pedroso. 2016. “Mortality tables for the Brazilian insured population.” Rev. Bras. Estud. Popul. 33 (3): 653. doi:10.20947/S0102-30982016c0010.

• DoD. 2016. “Avaliação Atuarial do Sistema de Pensões Militares das Forças Armadas.”

Brasília, D.F: Ministry of Defense/SECRETARIA DE ORGANIZAÇÃO INSTITUCIONAL – SEORI. Durban, M, I Currie, and P Eilers. 2002. “Using P-splines to smooth two-dimensional Poisson data.” Proc. 17th Int. Work. Stat. Model. Chania, Crete, 207–14. http://www.ma.hw.ac.uk/{~}iain/research/crete.pdf{\%}5Cnpapers2://publication/uuid/74FA21E2- A627-4A85-BCBE-840A743BFC76. Eilers, Paul H C, and Brian D. Marx. 2010. “Splines, knots, and penalties.” Wiley Interdiscip.

• Rev. Comput. Stat. 2 (6): 637–53. doi:10.1002/wics.125.

Eilers, Paul H. C., and Brian D. Marx. 1996. “Flexible smoothing with B -splines and penalties.” Stat. Sci. 11 (2): 89–121. doi:10.1214/ss/1038425655. Elder, G. H., E. C. Clipp, J. S. Brown, L. R. Martin, and H. S. Friedman. 2009. “The Lifelong Mortality Risks of World War II Experiences.” Res. Aging 31 (4): 391–412. doi:10.1177/0164027509333447. Enstrom, James E., and Lester Breslow. 2008. “Lifestyle and reduced mortality 27

among active California Mormons, 1980–2004.” Prev. Med. (Baltim). 46 (2): 133–36. doi:10.1016/j.ypmed.2007.07.030. Evgueni, M A, A J Dmitri, V Denny, M France, and V Jacques. 2014. “Do Vanguard Populations Pave the Way towards Higher Life Expectancy for Other Population Groups ?”

• Popul. - E. doi:10.3917/pope.1404.0531.

Finch, Caleb E. 2004. “Inflammatory Exposure and Historical Changes in Human Life-Spans.” Science (80-. ). 305 (5691): 1736–9. doi:10.1126/science.1092556. Fogel, Robert W., and Dora L. Costa. 1997. “A Theory of Technophysio Evolution, With Some Implications for Forecasting Population, Health Care Costs, and Pension Costs.” Demography 34 (1): 49. doi:10.2307/2061659. Goldsmith, J R. 1975. “Letter: What do we expect from an occupational cohort?” J. Occup.

• Med. 17 (2): 126–31. http://www.ncbi.nlm.nih.gov/pubmed/1142022.

Gu, Chong. 2014. “Smoothing Spline ANOVA Models: R Package gss.” J. Stat. Softw. 58 (5). doi:10.1007/978-0-387-98135-2. Gu, Chong, and Young-Ju Kim. 2002. “Penalized likelihood regression: General formulation and efficient approximation.” Can. J. Stat. 30 (4): 619–28. doi:10.2307/3316100. Gu, Chong, and Grace Wahba. 1993. “Smoothing Spline ANOVA with Component- Wise Bayesian ‘Confidence Intervals’.”

• J. Comput.

Graph. Stat. 2 (1): 97–117. 28

doi:10.1080/10618600.1993.10474601. Guest, C S, and A J Venn. 1992. “Mortality of former prisoners of war and other Australian veterans.” Med. J. Aust. 157 (2): 132–5. http://www.ncbi.nlm.nih.gov/pubmed/1630377. Horiuchi, S. 1983. “The long-term impact of war on mortality: old-age mortality of the First World War survivors in the Federal Republic of Germany.” Popul. Bull. UN. 15 (15): 80–92. http://www.ncbi.nlm.nih.gov/pubmed/12265835. Horiuchi, S, and J R Wilmoth. 1998. “Deceleration in the age pattern of mortality at older ages.” Demography 35 (4): 391–412. doi:10.2307/3004009. Horiuchi, Shiro, Nadine Ouellette, Siu Lan Karen Cheung, and Jean-Marie Robine. 2014. “Modal age at death: lifespan indicator in the era of longevity extension.” Vienna Yearb.

• Popul. Res. Volume 11: 37–69. doi:10.1553/populationyearbook2013s37.

Inskip, H., M. Snee, and L Styles. 1997. “The mortality of Royal Naval submariners 1960-89.”

• Occup. Environ. Med. 54 (3): 209–15. doi:10.1136/oem.54.3.209.

Jonge, Edwin de, and Mark van der Loo. 2013. “An introduction to data cleaning with R.”

• Stat. Netherlands, 53. doi:60083 201313- X-10-13.

Kang, Han K., and Tim A. Bullman. 1996. “Mortality among U.S. Veterans of the Persian Gulf War.” N. Engl. J. Med. 335 (20): 1498–1504. doi:10.1056/NEJM199611143352006. Kannisto, Vaino, Kaare Christensen, and James W. Vaupel. 1997. “No increased mortality in later life for cohorts born during famine.” Am. J. Epidemiol. 145 (11): 987–94. http: 29

//www.ncbi.nlm.nih.gov/pubmed/9169907. Lego, Vanessa di, Cássio M. Turra, and Cibele Cesar. 2017. “Mortality selection among adults in Brazil: The survival advantage of Air Force officers.” Demogr. Res. 37 (October). Universidade Federal de Minas Gerais, Centro de Desenvolvimento e Planejamento Regional: 1339–50. doi:10.4054/DemRes.2017.37.41. Lemez, Srdjan, and Joseph Baker. 2015. “Do Elite Athletes Live Longer? A Systematic Review of Mortality and Longevity in Elite Athletes.” Sport. Med. - Open 1 (1): 16. doi:10.1186/s40798-015-0024-x. Luy, Marc. 2003. “Causes of Male Excess Mortality: Insights from Cloistered Populations.”

• Evol. Heal. Dis. 29 (December): 647–76. doi:10.2307/1519702.

Luy, Marc, Priska Flandorfer, and Paola Di Giulio. 2015. “Ageing in an aged society: experiences and attitudes of Catholic order members towards population ageing and older people.” Ageing Soc. 35 (01): 1–36. doi:10.1017/S0144686X13000421. MacIntyre, N. R., R. E. Mitchell, A. Oberman, W. R. Harlan, A. Graybiel, and E. Johnson.

• 1978. “Longevity in military pilots: 37-year followup of the Navy’s ’1000 aviators’.” Aviat.
• Sp. Environ. Med. 49 (9): 1120–2.

Mackenbach, J. P. 2003. “Widening socioeconomic inequalities in mortality in six Western European countries.” Int. J. Epidemiol. 32 (5): 830–37. doi:10.1093/ije/dyg209. Manton, Kenneth G., and Eric Stallard. 1981. “Heterogeneity and its Effect on Mortality Measurement.” In Methodol. Collect. Anal. Mortal. Data, edited by Jacques Vallin, J. H. 30

Pollard, and L. Heligman, 265–99. Liege, Belgium: International Union for the Scientific Study of Population. McCullough, M E, W T Hoyt, D B Larson, H G Koenig, and C Thoresen. 2000. “Religious involvement and mortality: a meta-analytic review.” Health Psychol. 19 (3): 211–22. http://www.ncbi.nlm.nih.gov/pubmed/10868765. McLaughlin, Ruth, Lisa Nielsen, and Michael Waller. 2008. “An Evaluation of the Effect of Military Service on Mortality: Quantifying the Healthy Soldier Effect.” Ann. Epidemiol. 18 (12): 928–36. doi:10.1016/j.annepidem.2008.09.002. McMichael, Anthony J, Martin McKee, Vladimir Shkolnikov, and Tapani Valkonen. 2004. “Mortality trends and setbacks: global convergence or divergence?” Lancet 363 (9415): 1155–9. doi:10.1016/S0140-6736(04)15902-3. Moser, Kath, Vladimir Shkolnikov, and David A Leon. 2005. “World mortality 1950-2000: divergence replaces convergence from the late 1980s.” Bull. World Health Organ. 83 (3): 202–9. doi:/S0042-96862005000300013. Preston, Samuel H., Irma T. Elo, Ira Rosenwaike, and Mark Hill. 1996. “African-American Mortality at Older Ages: Results of a Matching Study.” Demography 33 (2): 193ro. doi:10.2307/2061872. Preston, Samuel, Patrick Heuveline, and Michel Guillot. 2001. Demography: Measuring and Modeling Population Processes (Paperback). 1st ed. Oxford: Wiley-Blackwell. Sandoval, Moises H., and Cássio M. Turra. 2015. “El gradiente educativo en la mortalidad 31

adulta en Chile.” Rev. Latinoam. Población 9 (17): 7–35. Sastry, Narayan. 2004. “Trends in Socioeconomic Inequalities in Mortality in Developing Countries: The Case of Child Survival in Sao Paulo, Brazil.” Demography 41 (3): 443–64. doi:10.1353/dem.2004.0027. Shah, Divyang. 2009. “Healthy worker effect phenomenon.” Indian J. Occup. Environ. Med. 13 (2): 77. doi:10.4103/0019-5278.55123. Shkolnikov, Vladimir M, Evgueni M Andreev, Dmitri A Jdanov, Domantas Jasilionis, Øystein Kravdal, Denny Vågerö, and Tapani Valkonen. 2012. “Increasing absolute mortality dispari- ties by education in Finland, Norway and Sweden, 1971–2000.” J. Epidemiol. Community Health 66 (4): 372–78. doi:10.1136/jech.2009.104786. Silva, Lariça Emiliano da, Flávio Henrique Miranda de Araújo Freire, and Rafael Henrique Moraes Pereira. 2016. “Diferenciais de mortalidade por escolaridade da população adulta brasileira, em 2010.” Cad. Saude Publica 32 (4). doi:10.1590/0102-311X00019815. Teramoto, Masaru, and Timothy J. Bungum. 2010. “Mortality and longevity of elite athletes.”

• J. Sci. Med. Sport 13 (4): 410–16. doi:10.1016/j.jsams.2009.04.010.

Turra, C. M., E. Renteria, and R. Guimaraes. 2016. “The Effect of Changes in Educa- tional Composition on Adult Female Mortality in Brazil.” Res. Aging 38 (3): 283–98. doi:10.1177/0164027515620245. Winkler-Dworak, Maria, and Heiner Kaden. 2014. “The longevity of academicians: evidence from the Saxonian Academy of Sciences and Humanities in Leipzig.” Vienna Yearb. Popul. 32

• Res. Volume 11: 185–204. doi:10.1553/populationyearbook2013s185.

33