Interaction between fertility and longevity in a long-living population : Villagrande (1830 – 2012)

Michel POULAIN1,2, Anne HERM1,2, Dany CHAMBRE and Gianni PES3

1 Estonian Institute for Population Studies, Tallinn University, Estonia 2 IACCHOS, Université catholique de Louvain, Louvain-la-Neuve, Belgium. 3 Department of Biomedical Sciences – Università degli Studi di Sassari -

Please do not quote this contribution and use its results before publication

1. Question addressed

The interactions between fertility and duration of life are multiple and complex and have been discussed long but findings are not fully consistent. Based on historical genealogical data over 200 years Freeman (1935) found no evidence of association between number of children and the length of post-reproduction life of mother while there was slight association between age at marriage and duration of life. By reviewing the results of a large number of studies on post-reproductive longevity Helle et al. (2005) showed that over more than 30 investigated studies, the most reported positive relationship exists between late reproduction and longevity of mothers. At opposite, the age at first childbirth has in most studies no significant relation to survival of mothers above age 50 as well as the total number of children. Nevertheless, in a historical French-Canadian cohort of the 17th-18th century experiencing natural fertility, Müller et al (2002) found that higher number of children was linked with increased post-reproduction survival while Westendorp et al. (1998) have shown the opposite with British aristocratic families. Dobblhammer et al. (2000) demonstrated that mothers with larger number of children have lower relative mortality risks in contemporary populations in England and Austria.

Individual data are largely preferable for analyzing the association between reproductive behaviour and longevity but very often the small size of the populations involved introduces limitations for the statistical signification of the findings. Individual data used often represent historical pre-transition populations (Helle et al, 2005, Smith et al. 2009, Muller, 2002). A recent study by Dribe (2012) on Swedish historical population (1790-1880) demonstrated a positive effect of number of children ever born on mortality of mothers, but not of fathers. He showed also that the mortality risk is higher for women who had their last child relatively earlier. Some studies used representative longitudinal data or combined aggregated data with individual data on vital events for more recent (post- transition) populations (Hank et al. 2010, Tabatabaie et al., 2011, Astolfi et al 2007). Compared to studies on pre-transition populations their findings showed that links between late maternity and the chances for survival in older age of mothers seem to be less obvious. Few researches consider the relation between the reproductive history and exceptional longevity. According Tabatabaie et al., (2011) centenarians in Ashkenazi Jewish community have had both their first and last childbirths at later age and had fewer children than individuals who did not survive up to such an old age. Caselli et al. (2012) found by analysing Sardinian centenarians that women who reach 100 years seems to have on average fewer children, and apparently at an older age, particularly for their last children

Some historical data involve the follow-up of population during a large period that covers the pre-transitional, transitional and post-transitional periods. Such data allows analyzing the interaction between fertility and mortality during the fertility transition and more specifically to investigate the relation between the decline of infant mortality and the decrease of fertility (Van Poppel 2012). Among all possible interactions between fertility and longevity we choose the following question that is of particular interest for the Sardinian population:

Question addressed. How the reproductive history of mothers is associated with their survival above 50 years of age? More specifically we would like to test at individual level if a later age at last child is associated with better survival for mothers, an association found at aggregated level of municipalities in by Astolphi et al. (2007, 2009).

The aim of this contribution is to address this question in a population characterised by exceptional longevity.

2. The context and the choice of the population investigated

The data collected on the population and families of the Sardinian village of is among the most suitable for study of possible interaction between fertility and longevity for several reasons. 1. Sardinian late fertility transition and high level of marital fertility has been reported by numerous researchers (Livi Bacci 1977, Bernardi 2007). Still in 1950’s the province of in the mountainous part of Sardinia showed the highest values for the Coale Index on marital fertility Ig among all Italian provinces during the fertility transition (figure 1).

2. Sardinia as whole but more specifically some in-land regions of the island, mostly in the Nuoro province, were still recently characterised also by late maternity (Astolfi et al. 2007, 2009).

3. Sardinia and more specifically the Nuoro province are also experiencing an extreme situation as far as the proportion of oldest olds and longevity are concerned. At the 1981 census, the Nuoro province presented the highest proportion of oldest olds (P85+/P60+), (Cantalini et al. 1990) but also the higher male life expectancy (Salaris 2008). An area named Longevity and composed by 14 municipalities located in the Ogliastra and the Nuoro provinces has been characterised by an exceptionally high male longevity (Poulain et al. 2004).

4. The idea of a possible relation between late maternity and higher longevity has been launched in Sardinia by Astolfi et al. (2007). Based on aggregated data on Sardinian municipalities, they demonstrated that areas with persistent late maternity are also characterized by longer life span, suggesting that the extension of both reproductive and post-reproductive period might be influenced by common determinants. Considering ‘late maternity’ through the proportion of babies born to mothers aged 35 years and older, they identified different spatial pattern in Sardinia where late maternity accompanies relatively low neonatal mortality. Also Tentoni et al (2011) suggested that the late maternity could be associated with a longer life span and correlated with the genetic structure of the local population. The similarity of the spatial patterns on maternal reproductive traits and longevity suggests this association at aggregated level. Sardinia was said to be a candidate for studies on the association of the life longevity with reproductive longevity (Astolfi et al. 2009) and the Longevity Blue Zone in the Nuoro province represent the most interesting place for a more in-depth analysis of the relation between late fertility and longevity.

5. More recently Caselli et al (2012) have investigated the relation between reproductive behaviour and extreme longevity based on data on Sardinian centenarians and various control groups. The preliminary results demonstrate that women who live longer seems to have on average fewer children, and apparently at an older age, particularly for their last children. More concretely they found that having a child at 40 years or older has a protective effect on mother’s survival, while the total number of children as well as the age at first childbirth seem to have no significant effect.

6. Pozzi (2008) analysed data at individual level using civil records of birth and death for the population of Alghero, a city on the sea-side, for the period 1866-1920, and two villages located in the Longevity Blue Zone, and , from 1866 till 1935. The lowest infant mortality in Sardinia already identified by Coletti (1908) and Gatti (2002) was confirmed and the situation was more pronounced in Urzulei compared to Alghero (Table 1). At the opposite, the child mortality from 2 to 5 is found higher in Sardinia (Gatti 2002, Salaris 2008). Pozzi expressed that unfortunately the period covering years following WWI are not explored at micro level.

Finally, for the setting for our investigation, the village of Villagrande Strisailli was chosen as the village of the Longevity Blue Zone showing the highest extreme longevity index, e.g. the larger proportion of newborns reaching age 100 in the whole Sardinia (Poulain et al. 2004).

3. Setting, data and methods

The village of Villagrande is located at 700 meters above sea level in the , but the altitude of its territory ranges from sea level to Punta La Marmora at 1,834 meters. On 1 January 2010, 3,441 inhabitants lived in Villagrande (ISTAT) with agro- pastoral activities and traditional life style still prevalent. Despite the fact that until the 1960s this region was among the poorest within the island, recent economic developments brought the population of this area close to the average welfare standard of the Italian population.

Demographic genealogical data on the whole population of the village has been collected. The start and end points of the database as marked by earliest birth included and last deaths ranges from year 1826 up to 2012. The families have been reconstructed for couples that registered their civil marriage in Villagrande starting 1851 up to 1955. Data are based on parish records up to 1865 and on civil records starting from 1866. The civil records were satisfactory crosschecked by information from parish records for the first decade of civil registration to ensure he exhaustive coverage of newborns.

A total of 1067 family nucleus composed by a couple having at least one child were reconstructed including 5788 children. 1029 different women are involved in the 1067 families, meaning that 38 women married twice and had children from more than one husband. Among these women, we identified 702 women that are mothers in completed families in a sense that both spouses marry only once and both survived up to the end of the fertility period of the mother. For 87 women alive on 1st January 2012, the data will be right-truncated.

In addition to simple cross-tabulations, we use the Cox proportional hazard method to assess the relative mortality risks of mothers in different generations associated to their age at first and last childbirth, and the Kaplan-Meier regression methods to compare the survival of different groups mothers.

As mentioned by Van Poppel et al. (2012) for a population in the course of demographic transition such relationships can be hard to follow and diverse results might appear due to the changes in both the fertility and mortality patterns. Particularly with a declining average age at last childbirth in parallel to the decrease in mortality, the relationship between late maternity and longevity could be biased.

4. Fertility and longevity evolution in Villagrande

Fertility transition

Fertility levels and trends could be estimated by using Coale’s index with data on the number of children born and mothers at risk extracted from the family database.

The computed Ig are compared with similar data for the province of Nuoro, Sardinia and Italy (figure 1). These trends confirm that the population of Villagrande is experiencing higher fertility rates and a very late fertility transition as the TFR only decreases after 1950.

*Data from Livi Bacci (1977) p. 66 ** Data from Livi Bacci (1977) p. 141

The Villagrande family database allows computing marital fertility rates by age group of mothers up to 1950 and figure 2 shows that the fertility level of women above 35 years start to decrease only in the 1930’s.

Figure 2. Marital fertility rates in Villagrande by five years age groups from 1870-1960 (mobile average on 11 years).

Figures 3, 4 and 5 display the variation of age at first and last child and total number of children born to mothers by ten years birth cohorts of mothers. The age of first child does not show any specific trend while the age at last child decreases significantly for the generations of mothers born after 1890, corresponding of births occurring after 1930. The total number of children per mother increased for mothers born in the second half of the XIXth century and decreased for generations born after 1900 for which children were born after 1930.

The family life have started relatively late in Villagrande. According the Villagrande database, the first child of mothers was born in average at age 26 years for cohorts born in mid XIXth century and two years later for generations born more near to the end of XIXth and at the beginning of XXth century. I we consider that these marriages were contracted, one or two years earlier. Compared an average age at marriage of 22 years, in 1866-1885 in Alghero, marriages in Villagrande started considerably later.

From another hand, we observe remarkable delay of the end of the reproduction period among Villagrande married women. The mean age of women at the birth of the last child in completed families was above age 40 for all generations of mothers born until 1910. Similar data for Alghero in 1866-1885 show more than 2 years earlier end of fertility for women. Also data from AKEA study on Sardinian centenarians born between 1900-1905 show lower average age of mothers at birth of first and at last child compared to the same birth cohorts in Villagrande. According to Caselli et al. (2012) the mean age of mother at first birth among Sardinian female centenarians was about 27 years and at last child 39 years while in Villagrande for cohorts born in 1900-1910 the latter was 40.5 years.

Longevity improvement

There is a strong improvement of longevity in the middle of the 1870’s up to the middle of the 1890’s. Thereafter the trend is more stable with two thirds of those reaching 50 who survive above 80.

Figure 6. Evolution of the expected duration of life for those surviving at 50 (e50) and Survival rate from 50 to 80 years (in percent).

Infant and child mortality

The probability to die before 5 years varies largely around 25 et 30% up to 1930. From then to 1965, the fall is clear to reach about 5%. These changes present large differences between infant and child mortality as shown on figure 7.

Figure 7 shows that infant mortality was never high (more than 15%) in Villagrande and this cannot be explained by non-registration of children who died in their first days of life. The distribution of the length of intergenesic intervals does not indicate the existence of missing children while children recorded exactly with those found in the church baptism register for the years 1866-1875. The infant mortality is declining starting from 1930 and the decrease is from 13% à 5% in 30 years up to 1960. The child mortality was largely higher between 1870 and 1880 (about 20%) but it decreases to stay between 10% and 15% up to 1920. Then the decrease was permanent up to 1% from 1955 (figure 8). Table 1 hereunder compares the levels of infant and child mortality in Villagrande and the neighbouring village of Urzulei in comparison with Alghero. The similarity between figures in Villagrande and Urzulei reinforces the conviction that levels of infant and child mortality in Ogliastra were significantly lower.

Table 1. Infant and child mortality rates (figures for Alghero are from L. Pozzi and for Urzulei from P. Melis)

Alghero Alghero Urzulei Urzulei Villagrande Villagrande Villagrande 1866- 1890- 1866- 1901 - 1866 - 1900 1901-1935 1936-1975 1890 1920 1900 1935 N= N=1411 N=2211 N=2022 Neonatal 61 44 29.8 35.4 34.7 51.6 46.5 Postneonatal 146 126 90.5 103.8 71.4 74.4 26.5 IRM 198 165 117.6 135.5 105.6 122.1 71.7 1 to 2 144 109 86.7 77.2 88.7 63.4 22.9 3 to 5 130 90 115.9 69,7 91.3 62.7 18.5 1 to 4 257 191 192.6 141.5 171.9 122.1 41.0 0 to 4 404 324 287.6 257.8 259.4 229.3 109.8 Q0/q1-4 0.56 0.57 0.61 0.96 0.61 1.00 1.75

5. Interaction between fertility and mortality of mothers investigated at individual level

Question addressed. How the reproductive history of mothers is associated with their survival above 50 years of age?. In this investigation we will characterise the reproductive history by the age at first child, age at last child and the total number of children. Methodologically spoken, applying the KM and Cox PH methods to the total number of complete families is misleading as it introduces a strong bias linked to the fact that for the younger cohort experiencing the fertility transition, the age at last child decreases while their longevity was increased compared to earlier cohorts. Accordingly the results emerging when considering the whole period with all families grouped show a strong negative impact of the age at last child on survival above 50. Nevertheless by analysing separately the behaviour of mothers under natural fertility regime and mothers engaged in the fertility transition, we found that characteristics of the reproductive history of women may have an impact on their survival above 50.

We analyse these relationships by using the Cox proportional hazard method to estimate the relative mortality risk of mothers above 50 in relation with their fertility characteristics (age at first child, age at last child and total number of children born).

Table 2 displays the RMR –estimated by running Cox PH model separately with age at first child, age at last child and total number of children as continuous covariates and the survival above 50 as outcome. For each model we compare two groups of generations of mothers, the first group covering the natural fertility period and the second the transitional period. Significant results are only found for the first group of mothers following a natural fertility regime: (i) the later the age of the first childbirth, the longer the survival of the mother above 50, (ii) the later the age at last childbirth, the longer the survival of the mother above 50, (iii) the number of children has no significant impact on the survival of mother above 50. Table 3 presents similar estimations when the Cox PH model is runned with the three covariates together: the three corresponding RMR are below 1 confirming the results from the previous models but no one reach a significant level.

Table 2. Relative mortality rates for mothers above 50 by controlling in separate models age at first and last childbirth and total number of children in natural fertility period and transitional period.

AGE AT FIRST N Of which are Exp(ß) p-value CHILD truncated RMR Born up to 1910 461 0 0.977 0.004*** Born after 1910 241 87 1.015 0.359 AGE AT LAST N Of which are Exp(ß) p-value CHILD truncated Born up to 1890 283 0 0.964 0.043** Born after 1890 419 87 0.997 0.842 NUMBER OF N Of which are Exp(ß) p-value CHILDREN truncated Born up to 1900 360 0 1.021 0.313 Born after 1900 342 87 0.996 0.886

Table 3 Relative mortality rates for mother during the natural fertility period controlled for age at first child and last child and total number of children.

Mothers born up to 1890 N = Exp(ß) p-value 95% confidence interval 283 RMR Age at first child (cont.) 0.988 0.589 0.944 1.033 Age at last child (cont.) 0.971 0.275 0.922 1.023 Total number of children 0.987 0.814 0.885 1.101 (cont.)

6. Discussion

We may conclude on this base that later ages at first and last childbirth favour longer survival after 50 but this is only valid for mothers involved in the natural fertility regime. This result is compatible with the findings of Caselli et al. (2012) based on Sardinian centenarians. At the opposite the impact of the total number of children seems to be not statistically significant The discussion (to be added) will revisit the different explanations proposed in the scientific literature to support our findings.