There are two major goals in this exploratory study: (1) To find out whether early-life conditions may have significant effects on adult lifespan. We also tried to determine whether our data set on European aristocratic families could be useful to explore the role of early-life conditions, and to be used in future more detailed studies. (2) To determine whether early-life conditions may have significant effect on sex disparities in adult health and longevity. These sex disparities are well documented (Van Poppel, 2000), but they still have to be explained and fully understood. For example, the following research question could be posed: are the long-lasting effects of early-life conditions identical for both sexes, or, on the contrary, they are sex-specific? This question stimulated us to conduct this study on the sex specificity of the effects of early-life conditions on adult lifespan.

The study of sex differences has also important methodological implications for research work on early-life effects. This is because in many cases the available data sets are limited in their size and/or the studied outcomes are rare events, thus creating a temptation to pool the data for both sexes together in order to increase the statistical power (Blackwell et al., 2001). It is important, therefore, to find out whether gender differences in response to early-life conditions are indeed similar (so that the data could be pooled together with simple adjustment for sex just by one indicator variable), or they are fundamentally and qualitatively different (so that each sex should be studied separately).

In this study we addressed these scientific problems (fundamental and methodological) by studying the effects of early-life conditions on adult lifespan of men and women separately, using the methodology of historical prospective study of extinct birth cohorts. We found significant sex differences in adult lifespan responses to early-life conditions that justify the need for further full-scale research project on related topic.

The database on European royal and noble families (a family-linked database) was developed as a result of 6 years of our continued efforts because of extensive data cross-checking and data quality control. The earlier intermediate versions of this database were already used in our previous studies (Gavrilov, Gavrilova, 1997a; 1997b; 1999a; 2000a; 2001b; Gavrilov et al., 1997; Gavrilova, Gavrilov, 2001; Gavrilova et al., 1998). To develop this database we have chosen one of the best professional sources of genealogical data available—the famous German edition of the "Genealogisches Handbuch des Adels" (Genealogical Yearbook of Nobility). This edition is known world wide as the "Gotha Almanac" — "Old Gotha" published in Gotha in 1763-1944, and "New Gotha" published in Marburg since 1951 (see Gavrilova, Gavrilov, 1999, for more details). Data from the Gotha Almanac were often used in early biodemographic studies of fertility (see Hollingsworth, 1969, 199-224, for references) and proved to be useful now in the studies of human longevity (Gavrilov, Gavrilova, 1997a; 1997b; 1999a; 2000a; 2001b; Gavrilov et al., 1997; Gavrilova, Gavrilov, 2001; Gavrilova et al., 1998).

Each volume of the New Gotha Almanac contains about 2,000 genealogical records dating back to the xiv^th-xvi^th centuries (to the founder of a particular noble genus). More than 100 volumes of this edition are already published, so more than 200,000 genealogical records with well-documented genealogical data are available from this data source. The high quality of information published in this edition is ensured by the fact that the primary information is drawn from the German Noble Archive (Deutsches Adelsarchiv). The Director of the German Noble Archive (Archivdirektor) is also the Editor of the New Gotha Almanach. Our own experience based on cross-checking the data, has demonstrated that the number of mistakes (mostly misprints) is very low in the "New Gotha Almanac" (less than 1 per 1000 records), so this source of data is very accurate compared to other published genealogies.

The information on noble families in the New Gotha Almanac is recorded in a regular manner. The description of each particular noble genus starts with information on 2-3 generations of founders of male sex only. Then three to four the most recent generations are described in more detail, including information on individuals (e.g., first and last names; event data: birth, death, marriage dates and places; descriptive data: noble degrees, occupation if available, information on death circumstances if available), information on parents (e.g., first and last names; event data: birth and death dates and places), information on spouses (e.g., first and last names; birth and death dates and places; first and last names of parents) and information on children (detailed as for each individual).

The process of data computerization is not yet completed—instead this is an ongoing project because of tremendous amounts of published data available for further computerizing. The present study represents, therefore, our intermediate findings.

These data were used as a supplement to the main data source since their quality was not as high compared to the Gotha Almanac. Although data on European royalty were recorded in computerized data sources ("Royal92", British Peerage CD, see above) with sufficient completeness, data on lower rank nobility (landed gentry) were less complete and accurate. The same was true for the data on Russian nobility. All supplementary data were matched with the Gotha Almanac data, in order to cross-check the overlapping pieces of information. This cross-checking procedure allowed us to increase the completeness of the database by complementation of information taken from different sources.

Each record in the database represents an individual’s event data (birth and death dates and places) and individual’s descriptive information, that is, identification number, sex, first and last names, nobility rank, occupation, birth order, cause of death (violent/nonviolent), ethnicity, marital status, data source code number, data source year of publication. Individual information is supplemented by data for parents (identification numbers, first and last names, birth, death and marriage dates, cause of death) and spouses. Thus, the database that is used in this project is organized in the form of triplets (referred to as the "ego" and two parents). This structure of records is widely used in human genetics and is adequate for studies of parent-child relationships. Similar database structure was used in the recent study of kinship networks (Post et al., 1997).

The genealogical data sources were checked for the following: (1) completeness, in reporting birth and death dates, which is crucial for calculating individual life span, the variable of particular interest in our study; (2) accuracy, whether the percentage of mistakes and inconsistencies between reported dates (such as, for example, birth by the dead mother) is low enough to be acceptable; (3) representativeness, whether the characteristics of investigated data sets (distribution by age, sex, marital status, age at death, etc.) is close to demographic characteristics of populations in similar geographic areas, historical periods and social groups. In our study we referred to the well-known publication by Thomas Hollingsworth (1962) on British peerage as a standard for European aristocracy, to check for data representativeness.

The completeness in birth and death dates reporting in the New Gotha Almanac was very high: dates of all vital events were reported for nearly 95% of all persons. Such high completeness is not common for many other genealogical data sources. For example, for British Peerage data published by Burke almanac in most cases there are no birth dates for women, which makes calculation of their life span impossible. In fact, this problem (with British aristocratic women) was first noticed a century ago by Karl Pearson (Beeton and Pearson, 1899, 1901). He used the British Peerage data to study the longevity inheritance and had to exclude women from his consideration for the following reason: "The limitation to the male line was enforced upon us partly by the practice of tracing pedigrees only through the male line, partly by the habitual reticence as to the age of women, even at death, observed by the compilers of peerages and family histories." (Beeton and Pearson, 1901, 50-51)

The accuracy of data published in the New Gotha Almanac is also very high: the frequency of inconsistent records is less than 1 per 1000 records while for many other genealogical data sources it falls within 1 per 300-400 records.

As for representativeness, the comparison of our data with Hollingsworth’s analysis of British peerage (Hollingsworth, 1962) revealed good agreement between his findings and our data on mortality patterns, including male/female gap in life expectancy (up to 7-10 years of female advantage in lifespan).

The genealogies for the members of European aristocratic families presented in the "Gotha Almanac" are of descending type, tracing almost all the descendants of relatively few founders. This is an important advantage of this data source over other genealogies that are often of ascending type (pedigrees). It is known in historical demography that the ascending genealogies are biased, over-representing more fertile and longer-lived persons who succeed to become ancestors, and for this reason such genealogies should be treated with particular caution (JettE9 and Charbonneau, 1984; Fogel, 1993).

Another important advantage of this data set is that the data are not spoiled by selective emigration (common problem for local registers), because every person is traced until his/her death in this data set.

Thus, the genealogical data published in the Gotha Almanac are characterized by high quality and accuracy. We have, however, encountered two problems regarding the data completeness, which are discussed below, along with proposed solutions.

Tab. 1
Characteristics of the data set

Birth cohort(year of birth)	Mean Age at Death [*]± Standard Error (years)
	Daughters(sample size)	Sons(sample size)
1800-1809	65.8 ± 0.7(539)	63.8 ± 0.6(543)
1810-1819	66.4 ± 0.7(563)	63.5 ± 0.6(629)
1820-1829	66.5 ± 0.6(684)	64.0 ± 0.6(668)
1830-1839	67.8 ± 0.6(684)	63.6 ± 0.5(587)
1840-1849	69.5 ± 0.6(834)	63.7 ± 0.5(890)
1850-1859	71.3 ± 0.5(1,015)	64.1 ± 0.5(1,081)
1860-1869	73.9 ± 0.4(1,414)	66.3 ± 0.4(1,472)
1870-1880	76.0 ± 0.3(1,968)	65.8 ± 0.4(1,840)

* Mean age at death is calculated for those persons who survived by age 30. This variable refers to “adult lifespan” in this study. The study data set consists 8,052 of males and 7,979 females.

Data presented in Table 1 demonstrate no significant sex bias in our data set.

The underreporting of children who died in infancy may be also a serious problem, especially for studies that include fertility analysis. Fortunately, in the Gotha Almanac the families that belong to the higher nobility rank (kings, princes, earls) are described with remarkable completeness. In particular, all ever born children are recorded, including those who died the same day when they were born. Another indicator of data completeness is the normal sex ratio at birth (101 to 108) observed among these families (based on analysis of our sample). In our database, over 90 aristocratic genuses belonged to the upper nobility were recorded completely, although data for lower rank nobility were not yet completed. Underreporting of children is not a problem for this particular study that is focused on adult life span for those who survived by age 30 years.

In this study we applied a multivariate regression analysis with nominal variables, which is a very flexible tool to control for effects of both quantitative and qualitative (categorized) variables. This method also allows researchers to accommodate for complex non-linear and non-monotonic effects of predictor variables. The beauty of this method is that it does not require any assumptions about the analytical function describing the effects of predictor variables. Instead, the model allows researchers to calculate directly a conditional mean lifespan in a group of individuals with a particular set of predictor variables values. The regression coefficients obtained in this model (named as differential intercept coefficients) have a clear interpretation of additional years of life gained (or lost) due to change in a particular predictor variable.

We applied the methodology of prospective historical study to the data for extinct birth cohorts, free of censored observations. We focused our study on old cohorts born in 1800-1854, for which secular changes in life expectancy were minimal. We also tested a long list of explanatory and potentially confounding variables (described below) to avoid possible artifacts.

Life span of adult (30 +) progeny (sons and daughters separately) was considered as a dependent outcome variable in multivariate regression with dummy (0-1) variables using the SAS statistical package (procedure REG).

In this study we used the following multivariate model:

The independent predictor variables included 14 types of binary variables:

(1) calendar year of birth (to control for historical increase in life expectancy as well as for complex secular fluctuations in lifespan). The whole birth year period of 1800-1854 was split into 5 year intervals (11 intervals) presented by 11 binary (0-1) variables with reference level set at 1850-1854 birth years. For data set of more recent birth cohorts (born in 1855-1880) five 5-year intervals were used with reference level at 1875-1880 birth cohort.

(2) maternal lifespan (to study maternal lifespan effects through combined genetic effects and shared environment). The maternal lifespan data were grouped into 5-year intervals (14 intervals) with the exception of the first (15-29 years) and the last (90 + years) longer intervals with small number of observations. The data were coded with 13 dummy variables with reference level set at 55-59 years for maternal lifespan.

(3) paternal lifespan (to study paternal lifespan effects through combined genetic effects and shared environment). The data were grouped and coded in a way similar to maternal lifespan (see above).

(4) maternal age when a person (proband) was born. This variable is used to control for possible confounding effects of maternal age on offspring lifespan. The data for mother’s age were grouped in 5-year intervals (6 intervals to cover the age range of 15-60 years) with the exception of the last longer interval of 40 + years with small number of observations. Maternal age of 25-29 years is selected as a reference category.

(5) father’s age when a person was born. This explanatory variable is used to study paternal age effects on offspring lifespan. The data were grouped and coded in 5-year intervals (9 intervals to cover the age range of 15-80 years) with the exception of the first (15-24 years) and the last (60-79 years) longer intervals with small number of observations.

(6) birth order. This variable is represented by 8 binary variables with the first birth order initially taken as the reference level.

(7) nationality. The nationality of individual is represented by a set of 6 categories—Germans, British, Italians, Poles, Russians and ‘others’. Germans (the largest group in our sample) is selected as a reference group.

(8) cause of death (‘extrinsic’ versus ‘natural’). The death is coded as extrinsic or premature in the following cases: (1) violent cause of death (war losses, accidents, etc.), (2) death in prison and other unfavorable conditions (concentration camp, etc.), (3) death from acute infections (cholera, etc.) and (4) maternal death (for women only). Deaths from all other causes combined were considered as a reference outcome. The proportion of reported ‘extrinsic’ deaths in our data set was about 5% for males and about 1% for females.

(9) loss of the father in the formative years of life (before age 10). This is a binary variable coded as 1 when father was lost before the age 10 and coded as zero otherwise.

(10) loss of the mother before age 10. This binary variable is coded as 1 in those cases when mother was lost before the age 10 and coded as zero otherwise.

(11) month of birth. This variable was included into analysis, because previous studies indicated that month of birth may be an important predictor of adult lifespan (Gavrilov, Gavrilova, 1999a; Doblhammer, Vaupel, 2001), particularly for daughters (Gavrilov, Gavrilova, 1999a). This variable was represented as a set of 11 dummy variables with those born in February considered as a reference group. The main focus of this particular study is on sex-differences in the month-of-birth effects that were not well studied before.

(12) death of sibling before age 18. This variable was included into analysis, because previous studies demonstrated that death of siblings during childhood may be an important predictor of late-life mortality (Alter et al., 2001). This binary variable is coded as 1 in those cases when at least one sibling died before age 18 and coded as zero otherwise.

(13) nobility rank of the father. Families are categorized in groups of ruling royal families, non-ruling princes, dukes and counts (coded as 1) and other nobility (coded as 0).

(14) family size. Families are categorized in groups according to sib ship sizes: 1-2, 3-4, 5-6, 7-8, 9-10, and 11 +. Later some categories were collapsed together during an iterative fitting procedure.

First, the numbers of males and females are rather similar in all studied birth cohorts (no apparent sex bias). The sex ratio in the entire data set is 1.01 (8,052 males/7,979 females). In the data set of cohorts born in 1800-1854 the sex ratio is 3,994/3,879 = 1,03. This is close to the normal sex ratio, in contrast to the sex ratio of 1.42, observed in the British peerage database with many missing records for women (Gavrilov, Gavrilova, 1999b). Thus, it seems to be possible to study sex differences in response to early-life conditions without particular concerns about selective sex bias in this data set.

Second, the values for mean lifespan are rather high—more than 63 years for males and 65 years for females. It indicates that lifespan in this socially elite population is to some extent comparable with modern lifespan values observed now in some countries of the world.

Third, there is a significant increase in lifespan over studied historical period, in particular for females (10 years gain). Therefore, the data should be adjusted for secular trends in lifespan (which has been done in this study) and/or analyzed for more narrow time periods (which was also done in this study). Finally, the temporal changes in lifespan are clearly not linear (no significant improvement in lifespan during the first 30 years of the xix^th century birth cohorts), and sometimes even not monotonic which justifies the method of analysis used in this study (multivariate regression with nominal variables and treating the year of birth as categorized predictor variable).

Tab. 2
Effects of month-of-birth on female lifespan. Birth cohorts: 1800-1854

Month-of-birth	Net effect [*] on adult lifespan, years (point estimate)	Standard Error	P value
February	0.00	Reference level
March	0.88	1.29	0.4956
April	1.51	1.31	0.2494
May	2.21	1.31	0.0914
June	1.99	1.29	0.1212
July	3.08	1.29	0.0173
August	1.24	1.27	0.3297
September	1.29	1.33	0.3301
October	2.26	1.32	0.0861
November	2.07	1.29	0.1092
December	2.86	1.30	0.0276
January	1.01	1.30	0.4348
February	0.00	Reference level

* Net effect corresponds to additional years of life gained (or lost) compared to the reference category. The data are point estimates of differential intercept coefficient adjusted for other predictor variables using multivariate regression with nominal variables.

Effects of month-of-birth on female lifespan. Birth cohorts: 1855-1880

Month-of-birth	Net effect [*] on adult lifespan, years (point estimate)	Standard Error	P value
February	0.00	Reference level
March	0.51	1.14	0.6577
April	2.39	1.13	0.0348
May	2.51	1.10	0.0222
June	1.17	1.11	0.2913
July	0.94	1.11	0.3913
August	1.39	1.12	0.2142
September	1.12	1.10	0.3120
October	1.20	1.10	0.2765
November	2.00	1.18	0.0909
December	2.32	1.14	0.0408
January	0.57	1.14	0.6198
February	0.00	Reference level

* Net effect corresponds to additional years of life gained (or lost) compared to the reference category. The data are point estimates of differential intercept coefficient adjusted for other predictor variables using multivariate regression with nominal variables.

It is interesting to note that the months of February and August are “bad” months to be born, and this finding corresponds well with some previous published data. For example, a similar bimodal month-of-birth distribution was found for birth frequencies of cystic fibrosis disease with peak births in February and August (Brackenridge, 1980). Further studies are required to find out whether this is just a coincidence of findings or a general seasonal pattern.

Seasonal effects on human survival in historical population were demonstrated for infant mortality in Sardinia (Breschi, Livi Bacci, 1986). However, the fact that such an early circumstance of human life as the month of birth may have a significant effect 30 years later on the chances of human survival at adult ages is particularly intriguing. It indicates that there may be critical periods early in human life particularly sensitive to seasonal variation in living conditions in the past (e.g., vitamin supply, seasonal exposure to infectious diseases, etc.) with long-lasting effects in later life.

It is known that the deficiency of vitamins B₁₂, folic acid, B₆, niacin, C, or E, appears to mimic radiation in damaging DNA by causing single- and double-strand breaks, oxidative lesions, or both, and may contribute to premature aging (Ames, 1998). The seasonal lack of these vitamins in late winter/early spring, in coincidence with one of the two critical periods in fetus or child development (the third critical month of pregnancy and the first months after birth), may explain a dramatic life span shortening among those born in August and February. This finding is also consistent with the reliability theory of aging, which emphasizes the importance of the initial level of damage that determines the future length of human life (Gavrilov, Gavrilova, 1991; 2001b).

These general explanations, however, do not match well with data for males presented in Table 4. In contrast to females, the male lifespan does not depend on month of birth, at least in this particular data set. Males born in 1855-1880 also do not demonstrate any relation between their lifespan and month of birth (Table 5). This observation is the first example in our study when sex differences in response to early-life conditions are observed.

Effects of month-of-birth on male lifespan. Birth cohorts: 1800-1854

Month-of-birth	Net effect [*] on adult lifespan, years (point estimate)	Standard Error	P value
February	0.00	Reference level
March	0.21	1.19	0.8576
April	-0.23	1.20	0.8473
May	1.21	1.15	0.2945
June	0.90	1.18	0.4437
July	-0.84	1.16	0.4716
August	-0.53	1.16	0.6459
September	-0.12	1.16	0.9168
October	0.21	1.21	0.8647
November	-0.41	1.20	0.7297
December	1.50	1.20	0.2106
January	-0.31	1.17	0.7936
February	0.00	Reference level

* Net effect corresponds to additional years of life gained (or lost) compared to the reference category. The data are point estimates of differential intercept coefficient adjusted for other predictor variables using multivariate regression with nominal variables.

Effects of month-of-birth on male lifespan. Birth cohorts: 1855-1880

Month-of-birth	Net effect [*] on adult lifespan, years (point estimate)	Standard Error	P value
February	0.00	Reference level
March	0.71	1.12	0.5244
April	-0.94	1.13	0.4059
May	1.32	1.13	0.2439
June	0.83	1.12	0.4610
July	-0.33	1.09	0.7608
August	-0.37	1.12	0.7414
September	0.44	1.09	0.6888
October	-0.09	1.12	0.9330
November	-0.16	1.13	0.8851
December	-0.90	1.16	0.4391
January	1.29	1.14	0.2557
February	0.00	Reference level

* Net effect corresponds to additional years of life gained (or lost) compared to the reference category. The data are point estimates of differential intercept coefficient adjusted for other predictor variables using multivariate regression with nominal variables.

The sex specificity of the month-of-birth effects on adult lifespan is a puzzling observation, but it is also a reassuring one from the methodological point of view. Indeed, the data for men and women are taken from the same sources and are represented by the same set of parental variables (because they are brothers and sisters to each other). Therefore, any possible flaws in data collection and analysis (such as omission of important predictor variable, for example) should produce very similar biases both in males and females data. Instead we observe a clear-cut sex-specific effect, which is reassuring from the methodological perspective.

While discussing the greater response of female lifespan to the season of birth, it is interesting to see whether other traits such as female childlessness are also affected by the month of birth. Indeed, studies on Dutch women found that the birth distribution of childless women, as compared with fecunds, was best represented with bimodal curve with zeniths in January and July (Smits et al., 1997). It is interesting to note that the two peaks for childlessness (January and July) seems to correspond well with the two observed minimums for female adult lifespan observed in our study (February and August—just only one month shift compared to childlessness findings).

Our finding that the month of February is “bad” month to be born for female corresponds well with schizophrenia studies. The risk of schizophrenia is higher for persons, whose birth date is close to February, and this seasonal effect is more marked among females (Dassa et al., 1995). It was also found that pre-natal exposure to influenza epidemic is associated with later development of schizophrenia in females but not in males (Takei et al., 1993; 1994).

While discussing studies of month-of-birth effects, it is important to be aware of methodological problems and pitfalls. In some cases a simplistic approach is applied to study the effects of month of birth on human lifespan: mean ages at death are calculated for people born in different months using cross-sectional data, i.e., death certificates collected during a relatively short period of time (Doblhammer, Vaupel, 2001). This methodology can produce both false positive and false negative findings. For example, if the seasonality of births and infant mortality were more expressed in the past, then the month-of-birth distribution of people would differ in different age groups of the population, thus producing a spurious month-of-birth effect on lifespan (if erroneously estimated through mean age at death). This mistake happens because the mean age at death depends on the age distribution of living people, which may differ depending on month-of-birth. Thus, even if the month of birth does not affect adult lifespan, nevertheless a false positive finding may occur, simply because the effects of population age structure are not taken into account. On the other hand, month-of-birth effects could be overlooked by this cross-sectional method if the seasonal effects on age-specific mortality rates are proportional. This false negative finding happens because proportional changes in death rates produce a proportional changes in the numbers of deaths in all age groups, and such proportional changes in numbers have no effect on the mean age at death. Thus, a false negative finding may occur, because cross-sectional analysis of death records is blind to proportional changes in age-specific death rates. In our study we avoided this simplistic cross-sectional analysis of death records as a flawed methodology. Instead we applied a cohort approach by following people born in the same calendar years until the last person died (method of extinct generations).

Finally, we would like to comment on the importance to control for socioeconomic status while studying the effects of month of birth. This is very important issue because there are significant differences in birth seasonality between different social classes (Smithers, Cooper, 1984; Bobak, Gjonca, 2001). Therefore, studies of aggregated data for whole countries (Doblhammer, Vaupel, 2001) may simply reflect the well-known differences in procreation habits of different socioeconomic classes. In our study we control for socioeconomic status by stratification (only aristocratic families are included into analysis).

Effect of paternal age at person’s birth on female lifespan. Birth cohorts: 1800-1854

Paternal Age	Net effect [*] on adult lifespan, years (point estimate)	Standard Error	P value
15-24	-1.27	1.38	0.3588
25-29	-0.98	0.86	0.2566
30-34	-0.18	0.76	0.8105
35-39	0.0	Reference level
40-44	-0.31	0.85	0.7184
45-49	-0.10	1.08	0.9285
50-54	-4.08	1.44	0.0046

* Net effect corresponds to additional years of life gained (or lost) compared to the reference category. The data are point estimates of differential intercept coefficient adjusted for other predictor variables using multivariate regression with nominal variables.

Shorter lifespan of daughters conceived to older fathers could be explained by age-related accumulation of mutations in DNA of paternal germ cells (Crow, 1997; Gavrilov, Gavrilova, 2000a; 2001a). Advanced paternal age at person’s conception is an important risk factor for such disease of adult age as schizophrenia (Malaspina, 2001; Malaspina et al., 2001), and such disease of old age as sporadic (non-familial) Alzheimer disease (Bertram et al., 1998).

The practical importance of these findings is obvious: the age constrains for the donors of sperm cells in the case of IVF (in vitro fertilization) perhaps need to be revised to exclude not only the old donors, but also those donors who are too young. Of course, more detailed studies are required, before such an important practical recommendation could be made.

Again, all these interesting ideas and suggestions seems to fail when data on males are analyzed (Table 7).

Male lifespan as a function of paternal age at reproduction. Birth cohorts: 1800-1854

Paternal Age	Net effect [*] on adult lifespan, years (point estimate)	Standard Error	P value
15-24	-2.06	1.28	0.1106
25-29	-0.62	0.81	0.4436
30-34	-0.58	0.69	0.4056
35-39	0.0	Reference level
40-44	-1.22	0.79	0.1207
45-49	0.27	0.95	0.7789
50-54	-0.07	1.31	0.9600

* Net effect corresponds to additional years of life gained (or lost) compared to the reference category. The data are point estimates of differential intercept coefficient adjusted for other predictor variables using multivariate regression with nominal variables

In contrast to females, the male lifespan does not depend on paternal age at person’s birth, at least in this particular data set. This observation is the second example in our study when sex differences in response to early-life conditions are observed.

Since only daughters inherit paternal X-chromosome, a hypothesis was suggested that mutation accumulation in X-chromosome of old sperm cells may be responsible for specific lifespan shortening of daughters conceived to older fathers (Gavrilov, Gavrilova, 1997a; 1997b; 2000a; 2001b).

Effect of sibling death (before age 18) on adult lifespan

Birth cohort	Net effect [*] on adult lifespan, years (point estimate)	Standard Error	P value
Females:
1800-1854	-0.13	0.61	0.8371
1855-1880	-1.87	0.61	0.0022
Males:
1800-1854	-0.21	0.55	0.7090
1855-1880	-1.69	0.62	0.0062

* Net effect corresponds to the years of life lost in a group with deceased sibling compared to the reference category when no siblings died before age 18. The data are point estimates of differential intercept coefficient adjusted for other predictor variables using multivariate regression with nominal variables.

The second research direction is related to the finding made by Frans van Poppel and his colleagues that women’s fecundability is associated with month of birth (Smits et al., 1997). We plan to replicate this finding and to include fecundability variable in the future data analyses as the outcome variable, as well as the predictor/confounding variable for adult lifespan.

The third possible research direction is related to findings of George Alter and his colleagues on the importance of interfamily differences in adult mortality (Alter et al., 2001). It would be interesting to take into consideration the “family effects” using the random effects model and to see how it affects our preliminary findings made in this study.

Finally, we acknowledge that the findings presented in this study should be interpreted with caution and need to be replicated on other data sets and with variety of analytical methods. However, the results of this study clearly indicate the need for separate analysis of data for males and females when late-life consequences of early-life conditions and events are explored. There is a definite need for subsequent full-scale studies of the effects of early-life conditions on sex-specific health outcomes in later life, and our pilot study presented here justifies the need of further work in this direction.