Birth order and alcohol-related mortality by ethnic origin and national context: Within-family comparisons for Finland and Sweden

Background: Previous studies have found that birth order is an important predictor of later life health, including hospitalisation for alcohol use. We examine the relationship between birth order and alcohol-related mortality in two national contexts, within native families who differ on ethnic origin. Methods: We study the association between birth order and alcohol-related mortality after age 17, using Finnish register data for cohorts born 1953-1999 and Swedish register data for cohorts born 1940-1999. We apply Cox proportional hazard models and use sibling fixed effects that eliminate confounding by factors shared by siblings. We separate full-sibling groups by ethnic origin, which for Finland means mother’s and father’s Finnish or Swedish ethnolinguistic affiliation. For Sweden, we distinguish native-born according to whether one or both parents were born in Sweden or Finland. Results: We find a positive correlation between birth order and alcohol-related mortality, but only for ethnic Finns in Finland and primarily men. Within these sibling groups, second-borns have an alcohol-related mortality risk that is 9% higher than that of first-borns, third-borns 19% higher, fourth-borns 23% higher, and fifthor higher-borns 48% higher. No such birth order associations can be found for any of the other ethnic groups analysed in Finland or Sweden. Conclusions. Our findings suggest that cultural-related behaviours typical for ethnic groups, and the national context in which they are studied, are relevant for whether any association between birth order and alcohol-related mortality can be observed. Differences in the social interplay within the family may be an important factor.


Introduction
Studies on the interrelation between birth order and health have found that later-born siblings generally perform worse than earlier-born siblings with regard to various health outcomes, such as depression, mental distress, psychiatric deviation, anxiety, self-esteem, physical fitness, and mortality (Modin, 2002;Mittendorfer-Rutz et al., 2004;Riordan et al., 2011;Barclay and Myrskylä, 2014). Similar to studies on intelligence, cognitive and non-cognitive skills, and educational attainment (Blake, 1989;Black et al., 2005;Bjerkedal et al., 2007;Kristensen and Bjerkedal, 2007), a number of potential explanations to the birth order pattern have been suggested. Researchers have argued that siblings are part of a dynamically changing environment that may become less cognitively stimulating when the family grows in size, parental resources may decrease, the likelihood of communicable diseases may increase, and there may be biological depletion of mothers with additional births (Zajonc, 1976;Blake, 1981;Strachan, 1989;Hertwig et al., 2002;Riordan et al., 2006;Batty et al., 2007). Another set of explanations relate to the social environment within the family. Offspring may occupy different niches in order to avoid inter-sibling competition, and there could be within-family bullying at the expense of later-born siblings (Sulloway, 1996;Zweigenhaft and Von Ammon, 2000;Tucker et al., 2013).
Recent studies have found that higher birth order relates to higher mortality from external causes, primarily suicides, but also accidents and events of an undetermined event (Bjørngaard et al., 2013;Rostila et al., 2014;Barclay and Kolk, 2015;Saarela et al., 2016). A fundamental contribution of these analyses is the use of the sibling fixed effects approach. It means that within-family variation only, and not between-family variation, is analysed. This minimises residual confounding from unmeasured time-invariant factors that are shared amongst siblings, such as common genetic factors, parenting style, socioeconomic environment at childhood, parental health behaviours, and parental patterns of drug abuse. Using sibling fixed effects, this paper is the first to study if alcohol-related mortality is associated with birth order.
Alcohol-related deaths have been estimated to account for three million deaths, or more than five per cent of all deaths, worldwide every year. The harmful use of alcohol is a causal factor in more than 200 disease and injury conditions, and about five per cent of the global burden of disease and injury is attributable to alcohol (WHO, 2018). For people aged 15-49 years, alcohol use is the leading cause of death (GBD, 2018). Sibling influence can be considered highly important for the uptake and continuation of alcohol consumption. Younger siblings may be introduced by older siblings to developmentally inappropriate alcohol behaviours at a younger age than otherwise would have been the case, with direct and indirect influences on health and mortality (Elliott, 1992;Harakeha et al., 2007). Access to alcohol may therefore be facilitated, and the exposure is likely to be higher for siblings with higher birth order than for those with lower. In correspondence, a study on birth order and hospitalization for alcohol use based on Swedish population register data has found that later-born siblings are hospitalized at a notably higher rate than first-borns, and that there is a monotonic increase in the hazard of hospitalization with increasing birth order (Barclay et al., 2016). Whether this pattern carries over to alcohol-related mortality, which is a much more severe indicator of excessive alcohol use, is not known, and is in the focus of this paper.
We improve on previous research also by conducting similar analyses across two national contexts, Finland and Sweden, and for native groups that differ by ethnic origin. Ethnic origin relates to cultural norms and behaviours. They affect alcohol use and vary by context, meaning that the birth order pattern for alcohol-related mortality may differ across ethnic groups and countries (Ahern et al., 2008). They are likely more emphasised in contexts where 4 alcohol intoxication, drunkenness, hangovers, and alcohol-induced pass-outs are more of a norm, and particularly so if values related to the family as an institution are deviant.
The multigenerational population registers of Finland and Sweden provide novel opportunities to study how alcohol-related mortality relates to birth order, and to do so for people who differ by ethnic background. Both countries have substantial populations of ethnic Swedish and ethnic Finnish origin, who differ in alcohol-related mortality. Swedish speakers in Finland account for five per cent of the total population. Finns constitute the third largest foreign-born group in Sweden, or seven per cent of all foreign-born individuals, where a large share have raised families in Sweden. Until 2017, they constituted the largest group. Both these minority groups have managed to keep their cultural roots and identities in spite of a substantial degree of intermarriage. Swedish speakers in Finland have a very long history, while many ethnic Finns in Sweden are more recent migrants, primarily arriving during the country's economic expansion in the 1960s and 1970s (Leiniö, 1984;McRae, 1997). In each national context, people in the minority group have formed a permanent and stable community.
In ages 18-50 years, the ethnolinguistic group of Finnish speakers in Finland have approximately three times higher rates of alcohol-related mortality than the native group of Swedish speakers in the country (Blomgren et al., 2004;Saarela and Finnäs, 2016). The Finnish speakers report also more frequent drunkenness, suffer more frequent hangovers and have alcohol-induced pass-outs significantly more often than Swedish speakers (Simpura, 1990;Paljärvi et al., 2009). Similar differences exist between ethnic Swedes and ethnic Finns in Sweden (Ågren and Romelsjö, 1992;Hjern et al., 2004;Westman et al., 2015). Socioeconomic, demographic and area-level variables explain only a small part of the differentials in alcohol-related mortality between ethnic Finns and ethnic Swedes. One may therefore assume that they relate to group-specific cultural norms that affect alcohol use, and to variation in social networks and family bonds that protect from unhealthy drinking behaviours (Saarela et al., 2016;Saarela and Rostila, 2019;Saarela and Kolk, 2020). Empirical support for such claims can be attained from analyses that examine within-family variation in alcoholrelated mortality, using data that include information about parental ethnicity.
In an international perspective, both Sweden and Finland have strict alcohol policies, with only slight differentials. Both societies are described as 'dry' drinking cultures, though with more sporadic and heavy drinking oriented towards intoxication (Bruun and Rosenqvist, 1985;Karlsson et al., 2012). Alcohol and booze have been perceived as particularly rooted in the Finnish self-perception, although in an international perspective, Sweden and Finland are quite similar in this respect (Peltonen, 2000). In the working-age population, the alcoholrelated mortality rate is more than twice higher in Finland as compared to Sweden, and roughly four times higher in men than in women in each country (Saarela and Kolk, 2020). Many of the underlying factors are nevertheless similar in both countries, and particularly the educational gradient in alcohol-related mortality (Martikainen et al., 2013;2014;Mackenbach et al., 2015;Nordahl et al., 2014;Östergren et al., 2018).
The extent to which cultural norms affect alcohol use and, thus, alcohol-related mortality, may vary by context. In immigrant parents, cultural-related drinking behaviours and family values may assimilate toward the norms of a new context and cease to affect intrafamily behaviours of the offspring (Kulis et al., 2012). If not, and the social environment within the family is obsolete, within-family behaviours may be abiding and become even more pronounced within another environment (Caetano and Clark, 1999).
Thus, there have been many studies on alcohol-related mortality in both Finland and Sweden, but none have examined birth order effects within families. We use population register data from two generations of the population in Finland and in Sweden to examine sibling groups who are the offspring of majority-culture parents and sibling groups with minority-culture or mixed cultural background. Based on this setting, the study seeks to answer whether alcoholrelated mortality relates to a person's birth order, and whether any such interrelation is dependent on ethnic origin and national context.

Data
We use Finnish register data that cover the period 1971-2017 and Swedish register data that cover the period 1971-2016. Persons in these multigenerational population registers can be linked to the mother and the father, which means that we can construct sibling groups.
All study persons in Finland, and their mother and father, were registered as a Finnish speaker or as a Swedish speaker. Foreign-born immigrants and their children were excluded. Immigration and intermarriage across other ethnic lines in Finland have until recently been rare. The data on Sweden are restricted to index persons born in Sweden, whose mother and father were born in Sweden or Finland. Due to substantial differences in group size, we study people with Finnish ethnic background (Finnish) separately from those with Swedish or mixed ethnic background (Swedish or Mixed) in Finland. The first group consists of Finnishregistered persons with a Finnish-registered mother and a Finnish-registered father. The second group consists of all other combinations of Swedish-or Finnish-registered index persons with Swedish-or Finnish-registered parents. In Sweden, we separate persons with both parents born in Sweden (Swedish) from those with one or both parents born in Finland (Finnish or Mixed). We lack sufficient statistical power to split the data by different combinations of parental ethnicity and sex.
With the data from Finland, we study full-sibling groups in which all siblings were born 1953-1999, to ensure that both parents can be identified (Karhunen and Uusitalo, 2017), and that all persons were at least 17 years old when they entered the study window. A similar setup is used for Sweden, i.e., full-sibling groups in which all siblings were born 1953-1999, but we can extend also with cohorts born 1940-1952. Birth order is based on the birth registers of each country. Here it refers to birth order within the full-sibling group, as we perform within-family analyses by ethnic origin.
For the entire period 1971-2017 in Finland and 1971-2016 in Sweden we can separate alcohol as the main cause of death. For the period 1987-2016 in Sweden, it is possible to conduct also analyses with alcohol as the main or contributing cause of death. We consequently have one data setup for Finland, i.e., for cohorts born 1953-1999, period 1971-2017, and follow-up from age 17, and three data setups for Sweden; (a) cohorts born 1953-1999, period 1971-2016, and follow-up from age 17, (b) cohorts born 1940-1999, period 1971-2016, and followup from age 17-30, and (c) cohorts born 1940-1999, period 1987-2016, and follow-up from age 17-30. For the sake of comparison and completeness, we perform parallel analyses for mortality from any other main (or contributing) cause.
Descriptive statistics of the study populations are found in Table 1. There are in total 8,066 alcohol-related deaths in the data from Finland, and 2,571, 7,799, and 17,340 in the three different setups of the Swedish data. The alcohol-related mortality rate is higher in the Finnish group than in the Swedish or Mixed group in Finland (deaths per 1,000 person years are 0.144 vs. 0.069), and lower in the Swedish group as compared with the Finnish or Mixed group in Sweden (0.031 vs. 0.057, 0.063 vs. 0.067, and 0.190 vs. 0.201).

Statistical analyses
To analyse the association between birth order and alcohol-related mortality we apply Cox proportional hazards models with sibling fixed effects. The failure event is death from an alcohol-related cause. We run parallel models in which death from any other cause is the failure event. The baseline hazard for mortality is time since age 17. Right-censoring occurs at first emigration or at the end of the observation period. The sibling fixed effects approach is based upon a within-family comparison. By comparing siblings in the same family we consequently adjust for all time invariant observed and unobserved factors that are shared by the siblings within a family. The approach can be considered superior to the regular Cox proportional hazard model because of its ability to minimise confounding from unobserved or unmeasured variables. The analyses are based upon stratified Cox regressions where siblings share the same baseline hazard. The sibling group variable is the shared mother plus father id (i.e., full-siblings). A requirement for the analyses is that sibling groups must have variance in the outcome, meaning that there must be at least two siblings in the group, and at least one must have died from an alcohol-related cause, or from any other cause in the parallel models. The analyses were performed using SPSS 26 and Stata 14.
Separate analyses are conducted for sibling groups with Finnish and Swedish or Mixed origin in Finland, and for sibling groups with Swedish and Finnish or Mixed origin in Sweden, to study if birth order associations differ by ethnic origin and national context. We adjust for each sibling's (index person's) birth year, sex, the mother's age when giving birth, and educational level. All variables are categorised. Birth year is used to adjust for cohort trends in alcohol use. Sex accounts for differences in alcohol use between men and women. Mother's age at child birth captures potential changes in parental consumption of alcohol with age. Father's age at child birth was excluded due to low explanatory power. Education, measured as the highest level attained, is included to proxy health-related behaviour. There is no need to adjust for factors like parental education or parental socio-economic status because the sibling comparison removes the confounding effect of factors shared by the siblings.
We also conduct analyses where education is removed from all models, to evaluate how important this variable is for the overall pattern observed, and models in which we interact birth order with sex, to see if any birth order association is similar for men and women. Table 2 summarises the results from the fully adjusted family fixed effects models that examine the relationship between birth order and alcohol-related mortality as the main cause, on the one hand, and birth order and mortality from any other cause, on the other hand. The first four columns with estimates refer to Finland and separate Finnish sibling groups from those with Swedish or Mixed background, while the four latter columns with estimates are for Sweden and separate Swedish sibling groups from those with Finnish or Mixed background. The cohorts analysed are the same in both countries, or those born 1953-1999, and the followup is from age 17. For ethnic Finns in Finland, there is a clear birth order pattern for alcoholrelated mortality. Within these sibling groups, second-borns have an alcohol-related mortality risk that is 9% higher than that of first-borns (95% CI: 1.01-1.18), third-borns 19% higher (95% CI: 1.05-1.35), fourth-borns 23% higher (95% CI: 1.02-1.48), and fifth-or higher-borns 48% higher (95% CI: 1.16-1.89). No such birth order associations can be found for any of the other ethnic groups studied in Finland or Sweden. For ethnic Swedes in Sweden, the alcoholrelated mortality risk rather decreases with birth order. For mortality from any other main cause than alcohol, there is no evident birth order pattern in any ethnic category. For ethnic Swedes in Sweden, mortality from any other cause increases slightly with birth order. This is consistent with previous results from Sweden (Barclay and Kolk, 2015), which have found a positive correlation between birth order and all-cause mortality.

Results
( Table 2 about here) Excluding educational level from these models does not change the conclusions to any considerable extent (Table S1 in the Supplementary materials). In the context of ethnic Finns in Finland, the estimates for second-borns, third-borns, fourth-borns, and fifth-or higherborns are only slightly higher than in the fully adjusted models, or 1.12 (95% CI: 1.04-1.21), 1.22 (95% CI: 1.08-1.39), 1.27 (95% CI: 1.06-1.52), and 1.51 (95% CI: 1.19-1.91). Thus, differences in educational level between siblings contribute only modestly to the association between birth order and alcohol-related mortality. Similar conclusions apply to those with Swedish background in Sweden; excluding education gives a slightly less attenuated pattern for how birth order relates positively to mortality from any other cause than alcohol. The negative association between birth order and alcohol-related mortality for these sibling groups, on the other hand, is slightly less emphasised when education is excluded. Table 3 summarises the results based on the two additional setups of the Swedish data, which extend to cohorts born 1940-1999. The first four columns with estimates refer to main cause mortality in 1971-2016, while the latter four columns with estimates refer to mortality from main or contributing cause. The number of alcohol-related deaths is much higher with these setups as compared with the first one in Table 2, and particularly so for ethnic Swedes, but not any single estimate for alcohol-mortality is statistically significant at the 5% level. Thus, in Sweden, there are no significant associations between birth order and alcohol-related mortality in any of the ethnic groups studied. Estimates for associations between birth order and mortality from any other cause than alcohol have no sizeable effects or they are statistically not significant. In corresponding models where education is excluded, there are not either any sizeable or significant birth order associations (Table S2 in the Supplementary materials). The only exception is alcohol-mortality as the main or contributing cause, for which there is a slight level difference between first-borns, second-or third-borns, and fourthor higher-borns. Thus, in this context, sibling differences in educational attainment contribute modestly to the sibling differences in mortality.
( Models that interact birth order with sex show that the associations between birth order and alcohol-related mortality for ethnic Finns in Finland apply primarily to men (Table S3 and  Table S4 in the Supplementary materials). Second-born men have 4% higher (95% CI: 0.94-1.14) alcohol-related mortality than first-born men, third-born men 16% higher (95% CI: 1.01-1.34), fourth-born men 26% (95% CI: 1.03-1.54) higher, and fifth-or higher-born men 8 46% higher (95% CI: 1.13-1.90). For women, the birth order pattern is not monotonous and less emphasised. For Sweden, there is a negative association between birth order and alcoholrelated mortality for both men and women with Swedish background born 1953-1999. A similar conclusion applies to the positive association between birth order and mortality from any other cause in the same birth cohorts. For the cohorts born 1940-1999 in Sweden, there is no association between birth order and alcohol-mortality or mortality from any other cause in these models. There is a slight positive correlation between birth order and mortality from alcohol as the main or contributing cause in men with Swedish background, but none of these estimates are statistically significant.

Discussion
Using high quality population register data from Finland and Sweden and a sibling comparison, this study has found that later-born siblings are more likely to die from an alcohol-related cause than their first-born sibling. However, the correlation can be observed only for ethnic Finns in Finland and primarily men, and not for any of the other ethnic groups analysed in Finland or Sweden.
A number of theories have been proposed for why higher birth order should be associated with negative outcomes in adulthood, such as elevated mortality and behaviours with negative health consequences like excessive alcohol consumption. Researchers have suggested that later-born children receive less parental resources as they have to compete with a larger number of siblings, such as in the resource dilution model (Blake, 1981). The confluence hypothesis similarly suggest that later-born children will grow up in a cognitively less stimulating environment (Zajonc, 1976). Another set of explanations have focused on personality differences, suggesting that later-born are more likely to engage in risky and rebellious behaviour, where drinking may be a typical example (Sulloway, 1996). Older siblings may also introduce drinking and smoking to younger siblings, meaning that laterborn siblings take up such behaviours at earlier ages and will be more heavily exposed during their course of life (Blane and Berry, 1973;Elliott, 1992).
Alcohol use has been proposed as one mechanism behind the relationship between mortality and birth order (Barclay et al., 2016). We find only partial support for this explanation, in the form of ethnic Finnish men in Finland, while associations are ambiguous or close to zero in the other groups studied. One explanation may be that the groups analysed are subject to cultural contexts that differ in terms of family support, bonding, and parental monitoring, which are known to be associated with lower levels of alcohol use, and in familism and the nuclear family, which generally serve as protective factors (White et al., 2006;Ramirez et al., 2012;Ewing et al., 2015). The interplay between siblings in social behaviour may be different in Finland, and for ethnic Finnish men in particular, than in Sweden (Saarela et al., 2016;Saarela and Kolk, 2020). In Sweden and among ethnic Swedes in Finland, other social contexts such as peers, schools and neighbourhoods, may potentially be more important and have protective effects (Saarela and Rostila, 2019). Ethnic Swedish parents may also, hypothetically, more commonly intervene to reduce sibling rivalry, bullying, and competition. It needs to be stressed, however, that the use of population register data implies that we cannot measure cultural norms or values in an explicit manner, nor drinking, alcohol-related behaviours or family relations directly. Morbidity differences, for example, may explain some of our findings.
To our knowledge, previous research that has used data from different countries to compare birth order associations is very scarce. A study on birth order and earnings in Sweden and Finland (Björklund et al., 2004) finds an overall negative correlation, but lack statistical power to detect any variation across countries. This line of research may nevertheless be helpful in order to understand our findings.
Another explanation for the lack of a universal effect of birth order on alcohol-related mortality is that we have been concerned with (late) adulthood and medical conditions that are fully attributable to alcohol. Barclay et al. (2016) find more attenuated and non-significant birth order effects for alcohol hospitalisation at ages 20 and above, as compared with ages below 20. In this perspective, the pattern observed for alcohol hospitalisation at adulthood carries over to alcohol-related mortality. Thus, birth order effects on alcohol behaviour may no longer be of major importance at advanced ages, as the influence of the family of origin tends to diminish over time (Blane and Berry, 1973;Cruz et al., 2012;Walsh et al., 2014).
This study has concerned ethnic groups which are firmly integrated and rooted, and have good access to social support and government services. This limits the generalisability to contexts in which ethnic minorities exist due to recent migration, where they may be less integrated and more affected by the migration history. On the other hand, our setting allows us to assess a more direct association with ethnic origin and cultural behaviour, not affected by socioeconomic and other disadvantages, which often is the case when ethnic groups are being compared.
To conclude, we demonstrate that, when using total population data of all alcohol deaths observed over 45 years and for complete birth cohorts of siblings, birth order effects on alcohol-related mortality are not universal, though more substantial effects exist in some contexts, such as for ethnic male Finns in Finland. We apply sibling comparison models, which allows us to be certain that the birth order effects observed are related to dynamics within the family of origin, and not to variation between the type of families in which the children of various birth orders have grown up.   224,614 4,055,924 75,278,194 3,782,099 118,637,392 4,310,452 87,326,716 3,640,284 The data consist of full-sibling groups in which all siblings are born 1953-1999, or 1940-1999, and for whom both parents can be identified. Each index person is observed from age 17, or from age 18-30 if born before 1953, until death, first emigration, or end of the observation period. For Finland, the index persons and their parents are either Finnish-registered or Swedish-registered. For Sweden, index persons are born in Sweden, while their parents are born in Sweden or Finland. Number of deaths, sibling groups, siblings, and person years refer to to the complete cohorts as decribed above. Numbers used in the analyses of within-family variation are provided at the bottom of the results tables. The variable distributions are per person years and refer to the complete cohorts as described above. Alcohol-related mortality refers to the ICD-8 codes 291, 303, 571, 5728X, E849, E851, E860, E980, N979, and N980 for deaths in 1971-1986, to the ICD-9 codes 291, 303, 3050, 3317, 34570, 3457A, 3457X, 3575, 3594, 4255, 535, 571, 5771, 8609, 980, E849, and E851 for deaths in 1987-1995, and to the ICD-10 codes E244, F10, G312, G405, G621, G721, I426, K292, K70, K860, O354, P43, X45, T51, Y90, Y91, Z502, Z714, and Z721 for deaths in 1996-2017. Sweden, born 1940-1999, Finland, born 1953-1999, Sweden, born 1953-1999, Sweden, born 1940-1999, period 1987-2016 Table A1. Estimates for within-family birth-order effects on alcohol-related mortality and mortality from any other main cause, by ethnic background in Finland and Sweden, period 1971-2016, cohorts born 1953-1999