2006 Chicago Actuarial Association Workshop Predictors of Exceptional Human Longevity Dr. Leonid A. Gavrilov, Ph.D. Dr. Natalia S. Gavrilova, Ph.D. Center on Aging NORC and The University of Chicago Chicago, Illinois, USA

2006 Chicago Actuarial Association Workshop Centenarians represent the fastest growing age group in the industrialized countries Yet, factors predicting exceptional longevity and its time trends remain to be fully understood In this study we explored the new opportunities provided by the ongoing revolution in information technology, computer science and Internet expansion Jeanne Calment ( )

2006 Chicago Actuarial Association Workshop Revolution in Information Technology What does it mean for longevity studies? Over 75 millions of computerized genealogical records are available online now!

2006 Chicago Actuarial Association Workshop Computerized genealogies is a promising source of information about potential predictors of exceptional longevity: life-course events, early-life conditions and family history of longevity

2006 Chicago Actuarial Association Workshop Computerized Genealogies as a Resource for Longevity Studies Pros: provide important information about family and life-course events, which otherwise is difficult to collect (including information about lifespan of parents and other relatives) Cons: Uncertain data quality Uncertain validity and generalizability

2006 Chicago Actuarial Association Workshop For longevity studies the genealogies with detailed birth dates and death dates for long- lived individuals (centenarians) and their relatives are of particular interest In this study 1,001 genealogy records for centenarians born in were collected and used for further age validation

2006 Chicago Actuarial Association Workshop

Internet Resources Used in Centenarian Age Verification Social Security Administration Death Master File is publicly available at the Rootsweb website: Head of household indexes and census page images for 1900, 1920 and 1910 federal censuses are provided by Genealogy.com Individual indexes of enumerated persons by 1900, 1920 and 1930 federal censuses and census page images are provided by Ancestry.com

2006 Chicago Actuarial Association Workshop Steps of Centenarian Age Verification 1. Internal consistency checks of dates 2. Verification of death dates – linkage to the Social Security Administration Death Master File (DMF) 3. Verification of birth dates – linkage to early Federal censuses (1900, 1910, 1920, 1930)

2006 Chicago Actuarial Association Workshop A typical image of ‘centenarian’ family in 1900 census

2006 Chicago Actuarial Association Workshop Results of Centenarian Age Verification 1001 records consistency checks 990 records used for further verification 990 records were linked to the SSA Death Master File Linkage success rate 77% (80% for centenarians born after 1890) In 3% of cases centenarian status was not confirmed 548 records found in DMF for persons born in were then linked to early US censuses Linkage success rate 80% when using Genealogy.com and 91% after supplementation with Ancestry.com. In 8% of cases a 1-year disagreement between genealogy and census record was observed

2006 Chicago Actuarial Association Workshop Conclusions of the Age Verification Study Death dates of centenarians recorded in genealogies always require verification because of strong outliers (1.3%, misprints) Birth dates of centenarians recorded in genealogies are sufficiently accurate - 92% are correct; for the remaining 8% only one-year disagreements Quality of genealogical data is good enough if these data are pre-selected for high data quality

2006 Chicago Actuarial Association Workshop Birth Order and Chances to Become a Centenarian Cases centenarians born in the United States between 1890 and 1899 Controls – their siblings born in the same time window (1,119 controls) Model: log(longevity odds ratio) = ax + bx 2 + cz + d where x – birth order; z – family size; a,b,c,d – parameters of polynomial regression model

2006 Chicago Actuarial Association Workshop Birth Order and Survival to 100 Source: Gavrilova, N.S., Gavrilov, L.A. Search for Predictors of Exceptional Human Longevity. In: “Living to 100 and Beyond” Monograph. The Society of Actuaries, Schaumburg, Illinois, USA, 2005, pp

2006 Chicago Actuarial Association Workshop New Developments Can birth order effect be confirmed by more rigorous approach – a strictly within-family analysis? Method of conditional logistic regression allows us to compare centenarians with their siblings within the same family. This eliminates confounding caused by between- family variation.

2006 Chicago Actuarial Association Workshop First-born siblings are more likely to become centenarians (odds = 1.8) Conditional (fixed-effects) logistic regression Number of obs = 950 LR chi2(2) = Prob > chi2 = Log likelihood = Pseudo R2 = Variable Odds Ratio P>|z| [95% Conf. Interval] First-born status Male sex

2006 Chicago Actuarial Association Workshop Can the birth-order effect be a result of selective child mortality, thus not applicable to adults? Approach: To compare centenarians with those siblings only who survived to adulthood (age 20)

2006 Chicago Actuarial Association Workshop First-born adult siblings (20+years) are more likely to become centenarians (odds = 1.95) Conditional (fixed-effects) logistic regression Number of obs = 797 LR chi2(2) = Prob > chi2 = Log likelihood = Pseudo R2 = Variable | Odds Ratio P>|z| [95% Conf. Interval] First-born status | Male sex |

2006 Chicago Actuarial Association Workshop Even at age 75 it still helps to be a first-born child (odds = 1.7) Conditional (fixed-effects) logistic regression Number of obs = 557 LR chi2(2) = Prob > chi2 = Log likelihood = Pseudo R2 = Variable Odds Ratio P>|z| [95% Conf. Interval] First-born status Male sex

2006 Chicago Actuarial Association Workshop Are young fathers responsible for birth order effect? Conditional (fixed-effects) logistic regression Number of obs = 950 LR chi2(2) = Prob > chi2 = Log likelihood = Pseudo R2 = Variable Odds Ratio P>|z| [95% Conf. Interval] Born to young father Male sex

2006 Chicago Actuarial Association Workshop Birth order is more important than paternal age for chances to become a centenarian Conditional (fixed-effects) logistic regression Number of obs = 950 LR chi2(3) = Prob > chi2 = Log likelihood = Pseudo R2 = Variable Odds Ratio P>|z| [95% Conf. Interval] First-born status Born to young father Male sex

2006 Chicago Actuarial Association Workshop Are young mothers responsible for the birth order effect? Conditional (fixed-effects) logistic regression Number of obs = 950 LR chi2(2) = Prob > chi2 = Log likelihood = Pseudo R2 = Variable Odds Ratio P>|z| [95% Conf. Interval] Born to young mother Male sex

2006 Chicago Actuarial Association Workshop Birth order effect explained: Being born to young mother! Conditional (fixed-effects) logistic regression Number of obs = 950 LR chi2(3) = Prob > chi2 = Log likelihood = Pseudo R2 = Variable Odds Ratio P>|z| [95% Conf. Interval] First-born status Born to young mother Male sex

2006 Chicago Actuarial Association Workshop Even at age 75 it still helps to be born to young mother (age <25) (odds = 1.9) Conditional (fixed-effects) logistic regression Number of obs = 557 LR chi2(2) = Prob > chi2 = Log likelihood = Pseudo R2 = Variable Odds Ratio P>|z| [95% Conf. Interval] Born to young mother Male sex

2006 Chicago Actuarial Association Workshop Back to a broader comparison of ‘centenarian’ and ‘non-centenarian’ families

2006 Chicago Actuarial Association Workshop Case-Control Study of Early-Life Conditions and Exceptional Longevity Cases households where centenarians (born in ) were raised (from centenarian records linked to 1900 census) Controls – 1% random sample of households with children below age 10 enumerated by 1900 census (from Integrated Public Use Microdata Sample, IPUMS: )

2006 Chicago Actuarial Association Workshop Childhood Residence and Survival to Age 100 Odds for household to be in a ‘centenarian’ group A – New England and Middle Atlantic (reference group) B – Mountain West and Pacific West C – Southeast and Southwest D – North Central

2006 Chicago Actuarial Association Workshop Household Property Status During Childhood and Survival to Age 100 Odds for household to be in a ‘centenarian’ group A – Rented House B – Owned House C – Rented Farm D – Owned farm (reference group)

2006 Chicago Actuarial Association Workshop Paternal Immigration Status and Survival to Age 100 Odds for household to be in a ‘centenarian’ group A – Father immigrated B – Father native-born (reference group)

2006 Chicago Actuarial Association Workshop No Association was Found (so far) Between Chances to Become a Centenarian and Paternal literacy Child mortality of siblings

2006 Chicago Actuarial Association Workshop Month of Birth Predicts the US Life Expectancy at Age 80 Computed using the Social Security Administration data Source: Gavrilova, N.S., Gavrilov, L.A. Search for Predictors of Exceptional Human Longevity. In: “Living to 100 and Beyond” Monograph. The Society of Actuaries, Schaumburg, Illinois, USA, 2005, pp

2006 Chicago Actuarial Association Workshop Seasonality (month-of-birth effects) for US life expectancy

2006 Chicago Actuarial Association Workshop Within-Family Study of Month-of-Birth Effects on Exceptional Longevity Cases - Centenarians born in Controls – Their own siblings Method: Conditional logistic regression Advantage: Allows researchers to eliminate confounding effects of between- family variation

2006 Chicago Actuarial Association Workshop Month of Birth and the Likelihood to Become a Centenarian Method: Conditional logistic regression for odds to become a centenarian, using siblings as within- family control. 921 observations

2006 Chicago Actuarial Association Workshop Month of Birth and the Likelihood to Become a Centenarian for Adult Siblings (20+ years) Method: Conditional logistic regression for odds to become a centenarian, using siblings as within- family control. 787 observations

2006 Chicago Actuarial Association Workshop Conclusions The shortest conclusion was suggested in the title of the New York Times article about this study

2006 Chicago Actuarial Association Workshop

Conclusions The accuracy of 'longevity risk' estimates can be greatly improved by using such 'trivial' information about person's childhood as: Mother's age at person's birth Month of birth Place of birth and some other characteristics of parental family Most important, these findings indicate that a subsequent large-scale research project on early-life determinants of human longevity is likely to produce more new results, very important for actuarial science and practice

2006 Chicago Actuarial Association Workshop Acknowledgments This study was made possible thanks to: generous support from the Society of Actuaries and the National Institute on Aging

2006 Chicago Actuarial Association Workshop For More Information and Updates Please Visit Our Scientific and Educational Website on Human Longevity: And Please Post Your Comments at our Scientific Discussion Blog: