Patterns of Birth Cohort–Specific Smoking Histories, 1965–2009

Journal of Developmental Origins of Health and Disease, page 1 of 10.& Cambridge University Press and the International Society for Developmental Origins of Health and Disease 2010doi:10.1017/S2040174410000218

O R I G I N A L A RT I C LE

Birth cohort patterns suggest that infant survivalpredicts adult mortality rates

R. Meza, B. Pourbohloul and R. C. Brunham*

British Columbia Centre for Disease Control, University of British Columbia, Vancouver, Canada

Dramatic improvements in life expectancy during the 20th century are commonly attributed to improvements in either health care services orthe social and economic environment. We evaluated the hypothesis that improving infant survival produces improvements in adult (>40 years)mortality rates. We used generalizations of age-period-cohort models of mortality that explicitly account for the exponential increase of adultmortality rates with age (Gompertz model) to determine whether year of birth or year of death better correlate with observed patterns of adultmortality. We used data from Canada and nine other countries obtained from the Human Mortality Database. Five-year birth cohorts between1900 and 1944 showed consistent improvements in age-specific mortality rates. According to the akaike information criteria, Gompertz-Cohort models significantly better predicted the observed patterns of adult mortality than Gompertz-Period models, demonstrating that year ofbirth correlates better with adult mortality than year of death. Infant mortality strongly correlated with the initial set point of adult mortalityin a Gompertz-period-cohort. Selected countries exhibited elevated adult mortality rates for the 1920 and 1944 birth cohorts, suggesting thatthe period before the first year of life may be uniquely vulnerable to environmental influences. These findings suggest that public healthinvestments in the health of mothers and children can be a broad primary prevention strategy to prevent the chronic diseases of the adult years.

Received 2 November 2009; Revised 17 February 2010; Accepted 12 April 2010

Key words: adult mortality patterns, birth-cohort effects, Gompertz mortality, infant and fetal conditions

Introduction

Chronic diseases such as atherosclerosis, cancer, dementia anddiabetes among others are the principal causes of adult (>40years) mortality and are focus of much public health effort forprevention and control. Of complex etiology, these diseasesare age-related and display a complex web of causationlinking the environment to the genome with characteristiclong latency between cause and effect. Current control effortsinclude improving population level social determinants,influencing individual level behaviors and providing diseasespecific diagnosis and treatment.

Over 70 years ago, Kermack et al.1 first reported ‘somegeneral regularities’ in death rates in Britain and Sweden.In particular, they reported that declining age-specific adultmortality rates between 1845 and 1925 displayed patternsthat reflected declining mortality rates in successive birthcohorts. They hypothesized that adult mortality rates maybe determined by conditions experienced during the first15 years of life. Extending data to the United States, Jones2

noted that the birth cohort effects in adult mortality rateswere traceable to overall declines in death rates for most of themajor chronic diseases, such as atherosclerotic and hyper-tensive vascular disease, cancer and diabetes. More recently,Finch and Crimmins demonstrated a strong association

between early-age mortality and adult mortality in fourEuropean countries and suggested that the reduction in life-time exposure to infectious diseases and other triggers ofinflammation were key factors in the historical decline ofadult mortality observed in cohorts born before the 20thcentury.3,4 Thus, these and other investigators concluded thatconditions in early life appeared to broadly affect suscept-ibility to many categories of age-related chronic disease. Inpart, because the mechanisms underpinning the relationshipbetween early life events and chronic disease susceptibilityremain enigmatic, others have argued that improvementsin survival of age-related chronic disease are due to medicalimprovements in treatment and secondary prevention forchronic diseases.5

Strehler and Mildvan6 developed a general theory of mortalityand aging based on a mathematical model of adult mortalitythat Gompertz first reported. This model is based on theobservation that adult mortality rates increase exponentially withage. To explore whether early life events are statistically corre-lated to age-specific adult mortality rates, we used mortality datafrom Canada and nine other countries in a generalized Gom-pertzian model of mortality, which accounts explicitly for periodand cohort effects. This allowed statistical correlation of theobserved mortality patterns with either birth year or death year.We subsequently used regression analysis to determine whetherinfant mortality rates predicted adult mortality rates.

The findings provide strong statistical support for thecorrelation between early life and adult mortality rates suggesting

*Address for correspondence: R. C. Brunham, BC Centre for DiseaseControl, 655 West 12th Avenue, Vancouver BC V5Z 4R4, Canada.(Email [email protected])

that improvements in maternal and infant health is a majorbroad strategy for the primary prevention of the chronicdiseases of adulthood.7

Materials and methods

Data sources

Canada age-specific mortality data by gender were obtainedfrom the human mortality database for the years 1921–2004.8

Number of deaths by gender, age and calendar year and thecorresponding population bases were cross-tabulated by 5-yearcalendar periods (1921–1924, 1925–1929, 1930–1934, y,and 2000–2004) and 5-years age groups (ages 40–94 years).These data represent the mortality experience of individualsborn as early as 1835. Infant mortality rates were also obtainedfrom the same sources. To enable generalization among dif-ferent countries, similar data for Australia, Finland, France,Japan, Netherlands, Spain, Sweden, United Kingdom andthe United States were also obtained from the human mortalitydatabase. The data from Canada are used to illustrate theprincipal findings.

Mathematical models

We explore generalizations of age-period-cohort (APC) modelsthat have been used to examine the epidemiology of chronicdiseases in order to investigate age effects and period (temporal)trends in adult mortality. In these models, cohort effects repre-sent the changes in adult mortality attributed to factors thataffected, or benefited, most individuals born in a particular yearor years. On the other hand, period effects capture the changesin mortality due to factors that affected all adults living in aspecific calendar year or years. We used Gompertz mortalityfunctions to replace the non-specific effects of age in the tradi-tional APC models. Secular trends in period and cohort effectswere modeled in the usual fashion. Similar methods have beenused to analyze cancer incidence and mortality trends.9–11

We model adult age-specific mortality at age a occurring incalendar year j as

mijðaÞ ¼ bicjAeGa; ð1Þ

where AeGa is the Gompertz mortality function at age a greaterthan 40 years; cj, a coefficient that adjusts for calendar year j;and the coefficient bi adjusts for birth cohort i (i 5 j 2 a,stratified in 5-year groups; 1835–1839, 1840–1844, y,1945–1949 and .1950). To ensure identifiability and com-parability of the model parameters, the period and cohortcoefficients are normalized arbitrarily by setting b1860–1864 5 1and c1930–1934 5 1.

Statistical analysis

We stratify the mortality data in 11 age groups (40–44 years,45–49 years, y, 85–89 years, 90–94 years) and into 17 five-yearcalendar periods (1921–1924, 1925–1929, 1930–1934, y,

2000–2004). We fit models (equation 1) to the number ofobserved deaths stratified by age group, calendar period andbirth cohort to evaluate whether birth cohort (bi) or calendaryear of death (cj) correlates more strongly with the observedmortality rates. We obtain parameter estimates for eachmodel by maximizing the likelihood across all age-calendarstrata assuming that the number of deaths in each stratum isPoisson distributed with mean Naj 3 mij(a), where Naj is thepopulation at risk in age group a and calendar year j, andmij(a) is as given in equation (1).

We then use the akaike information criteria (AIC), a test ofgoodness of fit for multivariate models, to statistically dis-criminate between competing models.12 In particular, we usethe AIC to determine if age-period models (where cohorteffects are assumed equal to 1) or age-cohort models (whereperiod effects are assumed equal to 1) give better fit to thedata. We also use the AIC to determine if gender differencesin period effects, cohort effects or in the Gompertz mortalityparameters are statistically significant.

Results

Temporal trends in declining mortality rates

Figure 1 (top) shows the Canada age-specific mortality ratesnormalized relative to the birth-cohort of 1921. The diagonalprogression from high to low values of relative mortalityillustrates strong cohort effects, especially on adult mortality,in that diagonal lines represent the mortality experienced byspecific birth-cohorts as they aged. The distortion of thediagonal progression during 1920–1940 may be due to theimpact of the two world wars on Canadian adult mortality, inthat young adults experienced mortality rates higher thanexpected during this period. The middle age-distortion (ages30–50 years) between the 1960 and 2000 may be due to theeffects of the smoking epidemic on Canadian mortality(mainly among men), in that smoking attributed mortalityreached significant levels during this period.13

Figure 1 (bottom) shows age-specific mortality curves bybirth cohort and gender for Canada between 1900 and 1944shown on a natural (e) logarithmic scale. The rightward shiftin the adult mortality curves for different birth cohorts(diagonal lines) suggests that cohort effects appear to stronglyinfluence adult mortality rates.

Relative importance of period or birth cohort effectson adult mortality

To statistically explore this suggestion, we began our analysis byfitting Gompertz-period (GP) models (in which bi 5 1 inequation 1) and Gompertz-cohort (GC) models (in which cj 5 1in equation 1) to evaluate the relative impact of birth year oryear of death on the observed adult mortality rates. We usedgoodness of fit of the models as measured by the AIC todetermine, which effect is more important (the smaller the AIC,

2 R. Meza et al.

the better the fit of the model). Table 1 shows the AICs for theGC and GP model fits to Canadian men and women. The AICsof fits for other countries data are shown in the supplementary

information (SI). As judged by the AIC, the GC model isuniformly and significantly better than the GP model inexplaining the changing adult mortality rates. Results were

Fig. 1. Heat graph displaying Canada age-specific mortality relative to the 1921 birth-cohort (top). The figure visually shows the strikingbirth-cohort effect in the decline in mortality rates during the 20th century. Canada age-specific mortality rate by birth year (bottom).Male mortality for selected birth-cohorts (left panel). Female mortality for selected birth-cohorts (right panel).

Infant survival and adult mortality 3

analogous for all countries evaluated (SI, Table S1). Wealso fitted conventional age-period and age-cohort models andreached the same conclusions. These findings provide strongstatistical support for the hypothesis that birth cohort effects arethe dominant determinant of adult mortality rates.

We next compared the fit of the Gompertz models (whileadjusting for period and cohort effects) with other parametricmodels for adult mortality such as Weibull, power of age,logistic and Gompertz plus an extrinsic constant mortality.14

According to the AIC, the Gompertz model gives the best fitto the data in comparison with the other parametric modelsevaluated (see SI, Table S2).

Given that birth-cohort effects dominate as correlates ofadult mortality, we next adopted the following final estimationprocedure. First, we fitted Gompertz-cohort models simulta-neously calibrating the Gompertz mortality parameters (A andG) and the birth-cohort effects (bi). Afterwards, we kept theGompertz mortality parameters fixed and recalibrated simul-taneously the birth-cohort (bi) and period (cj) effects. We callthis the GC-P model. As shown in Table 1, the AICs for thefits of the GC-P model to Canadian mortality data are evenbetter. Similar results are obtained when using other countriesmortality data as shown in the SI (Table S1).

Finally, we tested if these models explain jointly theobserved female and male mortality patterns by country. Inparticular, we tested if gender differences were significant interms of period effects, cohort effects or in the Gompertzmortality parameters. We found that in all countries genderdifferences in all model parameters were statistically sig-nificant and therefore we kept separate models for womenand men (see SI, Table S3).

Differential gender and temporal birth cohort effects ondeclining mortality rates

Figure 2 (left) shows temporal trends in the magnitude ofthe estimated birth-cohort effects (from the GC-P models) onage-specific adult mortality rates for Canada. These can be

Table 1. Akaike information criteria* values for the GC and GC-Pmodels relative (difference) to the GP model**

GP GC GC-P

Canada male 0 224,698 229,992Canada female 0 214,890 223,332

GP, Gompertz period; GC, Gompertz cohort; GC-P, Gompertzcohort-period.

* 22 3 log(likelihood) 1 2 3 number of estimated parameters.** Relative values that weight the goodness of fit of the model

to empirical data. The lower the AIC, the better the model fit.

Fig. 2. Estimated birth-cohort effects for Canadian men (continuous line) and Canadian women (dashed line) (left). *May represent effectof smoking on male mortality. **May represent smoking effects on female mortality. Estimated period effects for Canadian men andCanadian women (right).

4 R. Meza et al.

interpreted as the relative adult mortality between individualsborn during different years in Canada. It is important tostress that birth cohort not only share one period of birth, butalso an entire life-sequence of environments located in time.Birth-cohort effects therefore represent the impact of not onlyfactors that occurred explicitly during the year of birth, butalso of any other factors or events that may have causedvariations in mortality across groups of individuals born inthe same years. Examples of the later are the levels of nutri-tion and hygiene experienced during childhood and the levelsof smoking, which varied significantly by birth cohort. Thefigure shows that adult mortality has decreased steadily bybirth-cohort since the 1850s. Interestingly, the decrease inmale mortality appears to have slowed down and even slightlyreversed around the 1900 birth-cohort. This may be due, atleast in part, to the effects of smoking on men’s health duringthe 20th century beginning with 1900 birth-cohort.15 Asimilar though less dramatic effect is seen among femalebirth-cohorts starting with the 1940s birth-cohorts, which isconsistent with the later onset of the tobacco epidemic among

women than men.15 Figure 2 (right) shows temporal trends inthe magnitude of the estimated period effects (from the GC-Pmodels) for Canada. These do not show a significant trend asthe birth-cohort effects, consistent with the conclusion thatbirth-cohort effects are the dominant determinant of adultmortality rates.

Infant mortality correlates with adult mortality

To investigate the relationship between infant experiences andadult mortality, we evaluated the correlation between a birth-cohort’s Gompertz adult mortality set point, A05biA (seeequation 1), and the corresponding infant mortality rate, m0.Figure 3b and c shows the values of A0 v. m0 for differentbirth cohorts in Canada and the corresponding R2 value oftheir linear regression. The large values of R2 for both menand women suggest a strong correlation between A0 and m0.On the basis of this regression, we can estimate the A0 setpoint for the 2000–2004 birth-cohort using its known infantmortality rate, m0. This leads to the prediction that the adult

Fig. 3. (a) Theoretical concept of the Gompertz mortality function. Based on Canadian data, the male adult mortality rates doubleevery 8.7 years (5log(2)/G). During the entire 20th century the set point of the Gompertz mortality (A0) has been decreasing steadily.The A0 value for the 2000–2004 birth-cohort is estimated from its known infant mortality rate, m0. (b) Theoretical (A0) set point ofadult mortality for specific male birth-cohorts (1920–1924, 1925–1929,y, 1950–1954) v. the corresponding infant mortality rates (m0).(c) Theoretical (A0) set point of adult mortality for specific female birth-cohorts (1920–1924, 1925–1929,y, 1950–1954) v. thecorresponding infant mortality rates (m0).

Infant survival and adult mortality 5

mortality rate experienced by a 45-year individual born in1900–1904 will be delayed to that of a 60-year-old individualin the 2000–2004 birth-cohort (see Fig. 3a).

The findings suggest that the health conditions experiencedduring in utero development and infancy by a birth cohortcan predict the starting level (A0) of the adult’s Gompertzmortality thereby determining the level of the correspondingadult mortality rates. Figure 3a shows an idealization of thishypothesis using Canadian data. Similar figures based onother countries data are shown in the SI (Figures S1–S5).

Improving infant mortality is inversely correlated withan increasing mortality doubling rate (G)

Figure 4 shows the Gompertz mortality parameter estimatesfor different countries. Values of G are inversely associated withvalues of A. Additionally, the parameter A is significantly higherfor men than for women in all countries, whereas the mortalitydoubling rate, G, is higher in women than in men. Thistranslates into age-specific mortality rates starting at age 40 yearsthat are lower for adult women than for men but which exhibitshorter mortality doubling times (,8 years for women v. ,9years for men). In other words, women’s mortality is system-atically lower than men’s until late life, when it catches upentirely. There is uniform consistency in the inverse relationshipbetween G and A for both men and women across multiplecountries suggesting this is a general effect.

We also explored trends in infant mortality rates throughtime. Figure 5 shows that the infant mortality rates in Canadahave decreased dramatically since the 1920s and that themale excess in infant mortality has substantially decreased inrecent years.

Fig. 4. Gompertz function parameter estimates by country and gender. Female estimates in gray and male estimates in black. The lowestvalue of G (0.070) corresponds to a mortality rate doubling time (MRDT) of 9.9 years. The highest value of G (0.092) corresponds to aMRDT of 7.5 years.

Fig. 5. Canada gender specific infant mortality rates from1921–2005.

6 R. Meza et al.

Evidence that the birth cohort effect acts duringfetal and infant development

Birth cohort effects on adult mortality may be due to envir-onmental influences acting during the developmental plasti-city of fetal and infant life or may be traceable to life longcharacteristics that associate with a specific birth cohort. Toexplore this further, we developed additional GC-P modelsby single year age groups and single calendar/birth years foreight selected countries. Figures 6 and 7 show declining adultmortality rates due to the birth cohort effect for males andfemales across the eight countries evaluated. To assess thesignificance of any departures from the observed trend for any

particular year, we also fitted auto-regressive integratedmoving average time series models to the birth cohort effectsper country and gender. Figures S6 and S7 in the SI show thestandardized residuals (difference between model predictionand observed value) from the fitted models. Although allcountries show overall declining rates, four countries (France,Netherlands, United Kingdom and Sweden) show a statisti-cally significant rise in adult mortality for the 1920 birthcohort. Additionally, Australia, Canada and the United Statesshowed marked rise in adult mortality for the 1900 cohortand the Netherlands and France for the 1944 birth cohort. Asthe elevated adult mortality rate was seen only among specificbirth cohorts and not in the several years prior (or after), these

Fig. 6. Estimated birth-cohort effects relative to 1867 by single years for Canada, UK, US and France showing men and womenseparately. For selected countries the 1900, 1920 and 1944 birth cohorts exhibit increased adult mortality rates.

Infant survival and adult mortality 7

data suggest that the first year of life is uniquely vulnerable toenvironmental influences that determine the onset level of theGompertz adult mortality function.

Discussion

This study used generalizations of APC models to analyze theage-specific mortality rates observed over the past century inCanada and nine other countries. The analyses provide strongstatistical support for the hypothesis that early life experiencesstrongly determine adult mortality patterns. Yashin et al.16

previously investigated the variation of the Gompertz mor-tality parameters in terms of period and birth-cohort using

mortality data from Sweden. However, to our knowledgeno statistical analysis of the exponential increase in adultmortality rates while simultaneously adjusting for both periodand cohort effects has been previously carried out before. Thisadjustment is essential as improvements in infant survivalrates have been occurring in parallel to improvements in thesurvival from adult chronic diseases, such as atherosclerosis,stroke and cancer.

The potential correlation between early life conditions andadult mortality rates and the significance of birth-cohort effectson mortality trends have been widely considered.1–4,17–21 Eversince the observations by Gompertz22 in 1825, it has been notedthat adult mortality exponentially increases with age doubling

Fig. 7. Estimated birth-cohort effects relative to 1867 by single years for Australia, Netherlands, Spain and Sweden showing men andwomen separately. For selected countries the 1900, 1920 and 1944 birth-cohorts exhibit increased adult mortality rates.

8 R. Meza et al.

every 8 to 9 years and used to develop theories of aging inhumans and other species.2,6,14,23–30 The dramatic decline in thelevels of infant and childhood mortality during the 18th and19th century was speculated to have contributed significantly tothe reduction of the levels of adult mortality for the cohorts bornin the 18th and 19th centuries.3,4,20 However, the relationshipbetween early and later-life conditions was expected to be lesssignificant for 20th century cohorts due to several factors, such asthe already low levels of childhood mortality at the beginning ofthe century, the rise of the tobacco epidemic, the widespreadadoption of routine childhood immunization and the remarkableadvancements in medical therapy during this century.3,21 Thus, itis of considerable interest that our analysis shows that the onsetlevel of the Gompertz mortality function has continued to shiftto lower values with later birth cohorts throughout the 20thcentury and remains of continuing demographic importance.

Birth-cohort effects can be due to a variety of populationtraits that act across the lifecycle, such as common behavioraland dietary characteristics and shared external, environmentalconditions among others. However, the regression of theGompertz mortality set point, A0, v. the infant mortalityrates, m0, suggests that a surprisingly large-fraction of theestimated cohort effects can be explained by early life con-ditions (for which m0 is a proxy variable). Additionally, theimpact of major period events such as the 1918 influenzapandemic or the second world war, which affected all age-groups across a population but were associated with elevatedadult mortality only in specific birth cohorts, localizes theperiod of maximum vulnerability before the end of the firstyear of life.31 Other authors reached similar conclusions usingdifferent approaches. In particular, Caselli et al.17 found that theadult mortality during the early-mid 20th century in Italydepended on the cumulative mortality experienced by birthcohorts up to age 15 years using APC models. Catalano et al.20

found a significant relationship between mortality before age5 years experienced in Sweden, Denmark and England andWales during the 18th, 19th and early 20th centuries, respec-tively, and the corresponding life expectancy using time seriesanalysis. Crimmins and Finch3 found a strong correlationbetween age 70–74 years mortality rates and the correspondingchildhood mortality in four northern European countries duringthe 18th and 19th centuries using regression analyses.

The inverse relation between the values of A and G indifferent countries was somewhat unexpected although waspreviously noted by Strehler et al.6 and others.29,30 Therelationship between A and G may suggest that there may be abiological limit to human life span in which optimal early-lifeconditions are somehow compensated by a shorter adultmortality doubling time. Alternatively, one could hypothesizethat the inverse relationship may be due to relaxation ofnatural selection during the early stages of life on the fitness ofadult individuals.30

How early life conditions influence adult susceptibility tochronic disease is uncertain. Multiple mechanisms have beensuggested. For example, it has been proposed that early life

inflammatory responses to disease are strongly correlated withcardiovascular diseases in the aged potentially due to pro-moting the early onset of atherogenesis.3 The high levels ofcell division associated with inflammatory responses in earlylife may also act to increase the risk of genetic alterations thatpredispose to cancer later in life.32 Chronic infection acquiredearly in life such as that due to cytomegalovirus has beenassociated with premature immunosenescence and age-relatedchronic disease.33 Of current interest are observations sum-marized by Gluckman et al.34 that adverse early life eventsepigenetically modify the genome and change phenotypicexpression of critical molecular pathways that regulate thestress and aging responses. McGowan et al.35 convincinglyshowed that early life events somatically modify the genome.In that report, it was shown that DNA methylation changesto the region encoding the promoter of the glucocorticoidreceptor gene in the brain of adults who committed suicidewas characteristic of individuals who suffered abuse as chil-dren. Thus, the data from this study may reflect a uniqueadaptive developmental plasticity of the fetal and infant epi-genome, which broadly determines the rate of aging and thussusceptibility to the chronic diseases of the adult years.Kenyon36 elucidated molecular pathways that regulate agingin multiple model organisms and proposed that the insulinresponse pathways regulate longevity in humans. It may bethat improved fetal and infant conditions set a molecularclock that determines aging through epigenome modification.Future research should focus on how specific experiencesduring the highly plastic early life period shape molecular andbiological responses to disease vulnerability later in life.

On a practical level these data show that long-term trendsin adult mortality rates may be largely determined by con-ditions laid down early in life. This suggests that continuedpublic health investment in the health of mothers and chil-dren is a broad primary prevention strategy that shouldreduce adult chronic disease susceptibility. Addressing dis-parities in infant mortality rates among different geographicjurisdictions and ethnic groups is a high priority likely toadvance population health and contribute to chronic diseaseprevention among at risk groups, although with a very longlag-time, approaching 80 years on average in the currentera. When maternal-child health programs are coupled withpolicies that address the social determinants of chronic dis-eases, with messages that inform personal choice and with thetargeted provision of evidence-based therapies to treat chronicdisease, large gains in mortality reduction could be feasible.Such coordinated public health strategies are applicable inboth developed and developing country settings.37

Acknowledgements

This work was supported in part by funding from the Pro-vincial Health Services Authority Centres for Populationand Public Health. R. M. acknowledges the support of theDivision of Mathematical Modeling at the UBC Centre for

Infant survival and adult mortality 9

Disease Control. B.P. acknowledges the support of theMichael Smith Foundation for Health Research (MSFHR –Senior Scholar Funds).

Statement of interest

None.

References

1. Kermack WO, McKendrick AG, McKinlay PL. Death-rates inGreat Britain and Sweden. Some general regularities and theirsignificance. Lancet. 1934; 31, 698–703.

2. Jones HB. A special consideration of the aging process, disease,and life expectancy. Adv Biol Med Phys. 1956; 4, 281–337.

3. Crimmins EM, Finch CE. Infection, inflammation, height, andlongevity. Proc Natl Acad Sci USA. 2006; 103, 498–503.

4. Finch CE, Crimmins EM. Inflammatory exposure andhistorical changes in human life-spans. Science. 2004; 305,1736–1739.

5. Cutler DM, Rosen AB, Vijan S. The value of medical spendingin the United States, 1960–2000. N Engl J Med. 2006; 355,920–927.

6. Strehler BL, Mildvan AS. General theory of mortality andaging. Science. 1960; 132, 14–21.

7. Butler RN, Miller RA, Perry D, et al. New model of healthpromotion and disease prevention for the 21st century. BMJ.2008, a337, a399.

8. Human Mortality Database, University of California, Berkeley(USA), and Max Planck Institute for Demographic Research(Germany). Retrieved 24 August 2009 from www.mortality.orgor www.humanmortality.de.

9. Luebeck EG, Moolgavkar SH. Multistage carcinogenesis andthe incidence of colorectal cancer. Proc Natl Acad Sci USA.2002; 99, 15095–15100.

10. Jeon J, Luebeck EG, Moolgavkar SH. Age effects and temporaltrends in adenocarcinoma of the esophagus and gastric cardia(United States). Cancer Causes Control. 2006; 17, 971–981.

11. Meza R, Jeon J, Moolgavkar SH, Luebeck EG. Age-specificincidence of cancer: phases, transitions, and biologicalimplications. Proc Natl Acad Sci USA. 2008; 105,16284–16289.

12. Akaike H. A new look at the statistical model identification.IEEE Transactions on Automatic Control. 1974; 19, 716.

13. Peto R, Lopez AD, Boreham J, Thun M, Heath Jr C. Mortalityfrom tobacco in developed countries: indirect estimation fromnational vital statistics. Lancet. 1992; 339, 1268–1278.

14. Ricklefs RE. Evolutionary theories of aging: confirmation of afundamental prediction, with implications for the genetic basisand evolution of life span. Am Nat. 1998; 152, 24–44.

15. Burns DM, Garfinkel L, Samet JM (eds). Changes in cigarette-related disease risks and their implications for prevention andcontrol. Smoking and Tobacco Control, Monograph 8, NIHPubl No 97-4213.

16. Yashin AI, Begun AS, Boiko SI, Ukraintseva SV, Oeppen J.New age patterns of survival improvement in Sweden: do theycharacterize changes in individual aging? Mech Ageing Dev.2002; 123, 637–647.

17. Caselli G, Capocaccia R. Age, period, cohort and earlymortality: an analysis of adult mortality in Italy. Popul Stud(Camb). 1989; 43, 133–153.

18. Elo IT, Preston SH. Effects of early-life conditions on adultmortality: a review. Popul Index. 1992; 58, 186–212.

19. Bengtsson T, Lindstrom M. Airborne infectious diseases duringinfancy and mortality in later life in southern Sweden,1766–1894. Int J Epidemiol. 2003; 32, 286–294.

20. Catalano R, Bruckner T. Child mortality and cohort lifespan: a testof diminished entelechy. Int J Epidemiol. 2006; 35, 1264–1269.

21. Yang Y. Trends in US adult chronic disease mortality,1960–1999: age, period, and cohort variations. Demography.2008; 45, 387–416.

22. Gompertz B. On the Nature of the Function Expressive of theLaw of Human Mortality, and on a New Mode of Determiningthe Value of Life Contingencies. Phil Trans R Soc Lond. 1825;115, 513–583.

23. Prentice RL, Shaarawi AE. A model for mortality rates and a testof fit for the Gompertz force of mortality. Appl Stat. 1973; 22,301–314.

24. Finch CE, Pike MC, Witten M. Slow mortality rateaccelerations during aging in some animals approximate that ofhumans. Science. 1990; 249, 902–905.

25. Olshansky SJ, Carnes BA. Ever since Gompertz. Demography.1997; 34, 1–15.

26. Vaupel JW, Carey JR, Christensen K, et al. Biodemographictrajectories of longevity. Science. 1998; 280, 855–860.

27. Kesteloot H, Huang X. On the relationship between human all-cause mortality and age. Eur J Epidemiol. 2003; 18, 503–511.

28. Bonneux L. Benjamin Gompertz revisited. Eur J Epidemiol.2003; 18, 471–472.

29. Hawkes K, Smith KR, Robson SL. Mortality and fertility ratesin humans and chimpanzees: how within-species variationcomplicates cross-species comparisons. Am J Hum Biol. 2009;21, 578–586.

30. Gurven M, Fenelon A. Has actuarial aging ‘slowed’ over thepast 250 years? A comparison of small-scale subsistencepopulations and European cohorts. Evolution. 2009; 63,1017–1035.

31. Mazumder B, Almond D, Park K, Crimmins EM, Finch CE.Lingering prenatal effects of the 1918 influenza pandemic oncardiovascular disease. J D O H D. 2009.

32. Medzhitov R. Origin and physiological roles of inflammation.Nature. 2008; 454, 428–435.

33. Sauce D, Larsen M, Fastenackels S, et al. Evidence of prematureimmune aging in patients thymectomized during earlychildhood. J Clin Invest. 2009; 119, 3070–3078.

34. Gluckman PD, Hanson MA, Cooper C, Thornburg KL. Effectof in utero and early-life conditions on adult health and disease.N Engl J Med. 2008; 359, 61–73.

35. McGowan PO, Sasaki A, D’Alessio AC, et al. Epigeneticregulation of the glucocorticoid receptor in human brainassociates with childhood abuse. Nat Neurosci. 2009; 12,342–348.

36. Kenyon C. The plasticity of aging: insights from long-livedmutants. Cell. 2005; 120, 449–460.

37. Daar AS, Singer PA, Persad DL, et al. Grand challenges inchronic non-communicable diseases. Nature. 2007; 450,494–496.

10 R. Meza et al.

Birth cohort patterns suggest that infantsurvival predicts adult mortality rates

Supplementary information

Rafael Meza, Babak Pourbohloul and Robert C. BrunhamUniversity of British Columbia Centre for Disease Control

1

0 20 40 60 80

5e−

055e

−04

5e−

035e

−02

Age

Age

−S

peci

fic M

orta

lity

Rat

e by

Birt

h Y

ear

(Log

sca

le)

●

●● ● ● ●

●

●

●

●

●

●

●

●

●

AoAoAoAo

Ao

G= 0.082 slope of lines

Ao ●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

Figure S1: Theoretical concept of the Gompertz mortality function. Basedon Canadian data, the women adult mortality rates double every 8.4 years(=log(2)/G). The departure from linearity for the earlier birth cohorts may beassociated with the high mortality due to childbearing for those cohorts.

BC Men BC Women

0 20 40 60 80

5e−

055e

−04

5e−

035e

−02

Age

Age

−S

peci

fic M

orta

lity

Rat

e by

Birt

h Y

ear

(Log

sca

le)

●●

●●

●

●

●

●

●

●

●

●

●

●

●

Ao

AoAo

AoAo

G= 0.076 slope of lines

Ao●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

0 20 40 60 80

5e−

055e

−04

5e−

035e

−02

AgeA

ge−

Spe

cific

Mor

talit

y R

ate

by B

irth

Yea

r (L

og s

cale

)

●

● ●● ● ●

●●

●

●

●

●

●

●

●

Ao

AoAoAo

Ao

G= 0.083 slope of lines

Ao●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

Australia Men Australia Women

0 20 40 60 80

5e−

055e

−04

5e−

035e

−02

Age

Age

−S

peci

fic M

orta

lity

Rat

e by

Birt

h Y

ear

(Log

sca

le)

●

● ● ●●

●

●

●

●

●

●

●

●

●

●

Ao

Ao

Ao

AoAo

G= 0.078 slope of lines

Ao

●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

0 20 40 60 80

5e−

055e

−04

5e−

035e

−02

Age

Age

−S

peci

fic M

orta

lity

Rat

e by

Birt

h Y

ear

(Log

sca

le)

●

● ● ● ●●

●

●

●

●

●

●

●

●

●

Ao

Ao

AoAo

Ao

G= 0.083 slope of lines

Ao●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

Figure S2: Theoretical concept of the Gompertz mortality function. Based onBritish Columbia and Australia data.

Netherlands Men Netherlands Women

0 20 40 60 80

5e−

055e

−04

5e−

035e

−02

Age

Age

−S

peci

fic M

orta

lity

Rat

e by

Birt

h Y

ear

(Log

sca

le)

●●

● ●

●●

●

●

●

●

●

●

●

●

●

AoAoAoAoAo

G= 0.089 slope of lines

Ao

●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

0 20 40 60 80

5e−

055e

−04

5e−

035e

−02

AgeA

ge−

Spe

cific

Mor

talit

y R

ate

by B

irth

Yea

r (L

og s

cale

)

● ●● ●

● ● ●

●

●

●

●

●

●

●

●

AoAoAoAo

Ao

G= 0.092 slope of lines

Ao

●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

Sweden Men Sweden Women

0 20 40 60 80

5e−

055e

−04

5e−

035e

−02

Age

Age

−S

peci

fic M

orta

lity

Rat

e by

Birt

h Y

ear

(Log

sca

le)

●●

●● ● ●

●

●

●

●

●

●

●

●

●

AoAoAoAoAo

G= 0.089 slope of lines

Ao

●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

0 20 40 60 80

5e−

055e

−04

5e−

035e

−02

Age

Age

−S

peci

fic M

orta

lity

Rat

e by

Birt

h Y

ear

(Log

sca

le)

●●

●●

● ●●

●

●

●

●

●

●

●

●

AoAoAoAo

Ao

G= 0.091 slope of lines

Ao●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

Figure S3: Theoretical concept of the Gompertz mortality function. Based onNetherlands and Sweden data.

UK Men UK Women

0 20 40 60 80

5e−

055e

−04

5e−

035e

−02

Age

Age

−S

peci

fic M

orta

lity

Rat

e by

Birt

h Y

ear

(Log

sca

le)

●● ● ●

● ●

●

●

●

●

●

●

●

●

●

Ao

Ao

AoAoAo

G= 0.081 slope of lines

Ao●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

0 20 40 60 80

5e−

055e

−04

5e−

035e

−02

AgeA

ge−

Spe

cific

Mor

talit

y R

ate

by B

irth

Yea

r (L

og s

cale

)

●● ● ●

● ●

●

●

●

●

●

●

●

●

●

AoAo

AoAoAo

G= 0.086 slope of lines

Ao ●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

France Men France Women

0 20 40 60 80

1e−

045e

−04

5e−

035e

−02

Age

Age

−S

peci

fic M

orta

lity

Rat

e by

Birt

h Y

ear

(Log

sca

le)

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

AoAo

AoAo

Ao

G= 0.072 slope of lines

Ao ●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

0 20 40 60 80

5e−

055e

−04

5e−

035e

−02

Age

Age

−S

peci

fic M

orta

lity

Rat

e by

Birt

h Y

ear

(Log

sca

le)

●

●●

●

●

●●

●

●

●

●

●

●

●

●

AoAoAoAo

Ao

G= 0.082 slope of lines

Ao●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

Figure S4: Theoretical concept of the Gompertz mortality function. Based on UKand France data.

Finland Men Finland Women

0 20 40 60 80

1e−

045e

−04

5e−

035e

−02

Age

Age

−S

peci

fic M

orta

lity

Rat

e by

Birt

h Y

ear

(Log

sca

le)

●

●

●●

●

● ●

●

●

●

●

●

●

●

●

Ao

Ao

AoAo

Ao

G= 0.07 slope of lines

Ao

●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

0 20 40 60 80

5e−

055e

−04

5e−

035e

−02

AgeA

ge−

Spe

cific

Mor

talit

y R

ate

by B

irth

Yea

r (L

og s

cale

)

●●

●● ●

● ●

●

●

●

●

●

●

●

●

AoAo

AoAo

Ao

G= 0.087 slope of lines

Ao●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

Spain Men Spain Women

0 20 40 60 80

5e−

055e

−04

5e−

035e

−02

Age

Age

−S

peci

fic M

orta

lity

Rat

e by

Birt

h Y

ear

(Log

sca

le)

●●

●

● ●● ●

●

●

●

●

●

●

●

●

Ao

Ao

AoAo

Ao

G= 0.075 slope of lines

Ao

●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

0 20 40 60 80

1e−

051e

−04

1e−

031e

−02

1e−

01

Age

Age

−S

peci

fic M

orta

lity

Rat

e by

Birt

h Y

ear

(Log

sca

le)

● ● ● ● ●● ●

●

●

●

●

●

●

●

●

Ao

Ao

AoAo

Ao

G= 0.083 slope of lines

Ao

●

Birth Year1880−18841900−19041920−19241940−19441960−19642000−2004

Figure S5: Theoretical concept of the Gompertz mortality function. Based onFinland and Spain data

Table S1: Akaike Information Criteria (AIC)* values for the Gompertz Cohort(GC) and Gompertz Cohort-Period (GC-P) models relative to the Gompertz Pe-riod (GP) model**.

GP GC GC-PBritish Columbia Men 0 -3108 -3428

Women 0 -614 -1660Australia Men 0 -15776 -28392

Women 0 -14264 -16296Netherlands Men 0 -10204 -16320

Women 0 -558 -19262Sweden Men 0 -3926 -7516

Women 0 -14870 -21602Finland Men 0 -10940 -13098

Women 0 -18004 -21942Spain Men 0 -19690 -106698

Women 0 -187074 -283934France Men 0 16260 -53174

Women 0 -113660 -191778UK Men 0 -80980 -104024

Women 0 -29804 -60998US*** Men 0 -123906 -166882

Women 0 94034 -68988Japan*** Men 0 -42506 -61270

Women 0 -175338 -188600

* -2*log(Likelihood)+2* no. of estimated parameters** Relative values that weight the goodness of fit of the model to empirical data.The lower the AIC, the better the model fit.*** US mortality data covers the period of 1933-2004. Japan mortality data coversthe period of 1947-2004. For other countries, the corresponding mortality datacovers the period 1921-2004.

Table S2: Akaike Information Criteria (AIC) values for alternative models of adultmortality relative to the Gompertz model*.

Alt-P Alt-C Alt-C-PPower of age Men 687484 712260 50196

Women 1911146 804060 296560Logistic Men 20310 58 60

Women 935814 1374 1284Weibull Men 3226198 712206 50488

Women 3431682 804046 75378

*Based on Canadian data. Relative values that weight the goodness of fit of themodel to empirical data. The higher the relative AIC, the worse the model fit incomparison with the Gompertz model.

Table S3: Akaike Information Criteria (AIC) values for common men/womenmodels of mortality relative to independent fits*.

GCCommon Gompertz parameters 1166084

Common birth cohort effects 60576Common model for women and men 504724

*Based on Canadian data.

1860 1880 1900 1920 1940 1960

−3

−1

12

34

Canada Men

Birth Year

Sta

ndar

dize

d re

sidu

als

1860 1880 1900 1920 1940 1960

−2

02

4

Canada Women

Birth Year

Sta

ndar

dize

d re

sidu

als

1860 1880 1900 1920 1940 1960

−4

−2

02

4

UK Men

Birth Year

Sta

ndar

dize

d re

sidu

als

1860 1880 1900 1920 1940 1960

−2

02

4

UK Women

Birth Year

Sta

ndar

dize

d re

sidu

als

1860 1880 1900 1920 1940 1960

−3

−1

01

23

US Men

Birth Year

Sta

ndar

dize

d re

sidu

als

1860 1880 1900 1920 1940 1960

−3

−1

12

34

US Women

Birth Year

Sta

ndar

dize

d re

sidu

als

1860 1880 1900 1920 1940 1960

−2

02

4France Men

Birth Year

Sta

ndar

dize

d re

sidu

als

1860 1880 1900 1920 1940 1960

−2

01

23

4

France Women

Birth Year

Sta

ndar

dize

d re

sidu

als

Figure S6: Standardized residuals from ARIMA(10,0,1) time series models of thebirth-cohort effects for Canada, UK, US and France. A residual larger than two inabsolute value indicates a significant departure of the corresponding birth-cohortmortality from the expected trend. For selected countries the 1900, 1920 and 1944birth cohorts exhibit significantly higher adult mortality rates.

1860 1880 1900 1920 1940 1960

−2

01

23

Australia Men

Birth Year

Sta

ndar

dize

d re

sidu

als

1860 1880 1900 1920 1940 1960

−2

01

23

4

Australia Women

Birth Year

Sta

ndar

dize

d re

sidu

als

1860 1880 1900 1920 1940 1960

−2

02

4

Netherlands Men

Birth Year

Sta

ndar

dize

d re

sidu

als

1860 1880 1900 1920 1940 1960

−3

−1

01

23

Netherlands Women

Birth Year

Sta

ndar

dize

d re

sidu

als

1860 1880 1900 1920 1940 1960

−2

02

4

Spain Men

Birth Year

Sta

ndar

dize

d re

sidu

als

1860 1880 1900 1920 1940 1960

−2

02

4

Spain Women

Birth Year

Sta

ndar

dize

d re

sidu

als

1860 1880 1900 1920 1940 1960

−3

−1

12

34

Sweden Men

Birth Year

Sta

ndar

dize

d re

sidu

als

1860 1880 1900 1920 1940 1960

−2

01

23

4

Sweden Women

Birth Year

Sta

ndar

dize

d re

sidu

als

Figure S7: Standardized residuals from ARIMA (10,0,1) time series models ofthe birth-cohort effects for Australia, Netherlands, Spain and Sweden. A residuallarger than two in absolute value indicates a significant departure of the corre-sponding birth-cohort mortality from the expected trend. For selected countriesthe 1900, 1920 and 1944 birth cohorts exhibit significantly higher adult mortalityrates.