Antecedents of teenage pregnancy from a 14-year follow-up study using data linkage

Australia has a relatively high teenage pregnancy rate among western countries at 19.8 births per 1,000 women aged 15-19 years [1, 2]. Of 30 OECD countries, Australia ranks 11th highest in teenage birth rate [3]. There are three main reasons why reducing the incidence of teenage pregnancy should be given a high priority. Firstly, teenage pregnancies are associated with increased medical risks during pregnancy and poorer birth outcomes for their babies [4–8]. Secondly, teenage mothers have poorer life-course outcomes in adulthood such as a higher risk of: dependence on welfare; being a sole parent; being a smoker; and having a lower socio-economic status [9–11]. Finally, the children of teenage mothers are more likely to be socio-economically disadvantaged, and have the potential to repeat the cycle of teenage pregnancies [12–14].

There is ongoing debate as to whether the adverse medical, social, educational and economic outcomes associated with teenage pregnancy are due to intrinsic risks of pregnancy and childbirth in the teenage years, or are associated with the social, economic and environmental circumstances of teenagers and their offspring [5]. Nonetheless it is widely accepted that even for teenagers who live in poverty and disadvantage, teenage pregnancy compounds the handicaps associated with social disadvantage, and initiatives aimed at reducing teenage pregnancy rates have the potential to expand the options and choices available for young people to overcome disadvantaged backgrounds [15–18]. With greater understanding of the antecedents of teenage pregnancy it may be possible to develop more effective interventions with the aim of reducing teenage pregnancy rates.

A range of antecedents of teenage pregnancy have been identified in the literature. Among the strongest and most persistent associations are measures of social disadvantage, inequality and social exclusion [10, 12, 19]. A pattern of intergenerational transfer has also been observed with daughters of teenage mothers being at high risk to become teenage mothers themselves [14], which has been associated with the transfer of disadvantage between generations [20], and with an association with childhood aggressive behaviours [21]. Aggressive and antisocial behaviours have also been independently associated with teenage pregnancy [19]. Dysfunctional family relationships, family breakdown and sole parent family structures are associated with higher rates of teenage pregnancy [19, 20].

In this study we used a large, well-defined cross-sectional population sample of children which was subsequently linked to health system data spanning 14 years. The Western Australian Child Health Survey (WACHS) featured a framework of variables based on a social-ecology model allowing for extensive family-based demographic, psychosocial, health, educational and direct measures of both parents and survey children [22, 23]. We hypothesised that there are factors in early life that are predictors of teenage pregnancy. As the girls in the WACHS were aged 4 to 16 years at the time of the survey, the factors measured in the survey were collected across a range of life stages. Using administrative record linkage to ascertain occurrence of births or abortions, we set out to: a) quantify the association of known risk factors for teenage pregnancy within our WA sample; b) investigate if other factors within the social-ecology framework were independently associated with teenage pregnancy outcomes; and c) investigate whether the observed associations changed with the age of the girl at the time when the risk factors were assessed.

Nearly 40% of girls who were identified as having both aggressive and delinquent behaviours became pregnant as a teenager, compared with fewer than 10% of girls who had neither of these behaviours. The combination of these behaviours had an increasing effect on both the proportion of girls that became teenage mothers, and the proportion of girls that had an abortion as a teenager.

The proportion of girls who became pregnant and gave birth before age 20 years decreased with increasing combined carer income, while the proportion of girls who had an abortion as a teenager did not vary by combined carer income. Some 12.3% of girls from low income families went on to become teenage mothers, while the proportion for girls from high income families was 2.7%.

Significant differences by family type were noted for those girls who became pregnant as a teenager, and those that became teenage mothers. Girls living in either step/blended or sole-parent families were more likely to become pregnant as a teenager than those living in original families. Some 16.8% of girls who were in a step/blended family at the time of the WACHS went on to become pregnant as a teenager, compared with 7.5% of girls from original families.

Girls were more likely to become pregnant as teenagers if their primary carer was a smoker at the time of the survey (14.6%) compared with those carers that were non-smokers (9.0%). Of those girls whose primary carer completed Year 9 or lower at high school, 22.3% became pregnant as a teenager compared with 8.4% of girls whose carer completed Year 10 or higher at high school. About one-fifth (20.6%) of girls whose mother had been a teenage mother herself went on to become teenage mothers, compared with 6.2% of girls whose mother had not been a teenage mother.

Level of family functioning, whether the teenager was breastfed, importance of religion and birth weight were not significantly associated with the proportion of girls who became teenage mothers, or the proportion that had an abortion as a teenager.

Teenage pregnancy was not independently and significantly associated with whether the girl's mother had been a teenage mother, level of family functioning, size of the family, whether the girl's primary carer had ever been treated for an emotional or mental health problem, SEIFA [29], type of school the girl had attended, how well the girl did at school (as reported by either the girl's primary carer or teacher), whether the girl needed professional help for emotional or behavioural problems (as reported by the primary carer), primary carer's marital status, primary carer's satisfaction level with the girl's progress in her education and learning skills, level of importance of religion to the girl's primary carer and whether the girl was breastfed.

To investigate whether age of the girl at the time of the survey affected the association between aggressive and delinquent behaviours and risk for teenage pregnancy, two age interaction models were fitted. Figure 1 shows the age interaction model which is unadjusted for other factors, while Figure 2 is adjusted for the other significant variables in the proportional hazards model (family type, combined carer income, highest school year completed by the girl's primary carer, whether the primary carer was a smoker and whether the girl lived in a metropolitan or non-metropolitan area at the time of the survey). Figure 1 shows that if a girl exhibited both aggressive and delinquent behaviours, she was between two to eight times as likely to become pregnant as a teenager compared with girls who showed neither of these problems. After adjusting for the other factors, Figure 2 showed that if a girl exhibited both aggressive and delinquent behaviours, she was between two to four times as likely to become pregnant as a teenager compared with girls who showed neither of these problems. The hazard curve was significantly elevated for girls who exhibited both aggressive and delinquent behaviours (χ² = 12.4; p < 0.001), but there was no significant difference between the hazard for girls who exhibited one of these problems compared with girls who showed neither (χ² = 1.53; p < 0.448).

Of the eight mental health problems measured by the Child Behavior Checklist [25], two were strongly associated with teenage pregnancy - delinquent behaviour (including breaking rules and norms set by parents and communities) and aggressive behaviour (including bullying, teasing, temper tantrums, arguing, fighting and threatening behaviours). While many of the factors assessed in the survey would be expected to be reasonably constant over time, such as carer education, one factor that may well vary depending on the age of the child at the time of the survey is aggressive and delinquent behaviours. One might expect that the relationship with pregnancy outcomes would be strongest for girls who were older at the time of the survey. For example, knowing that a 16 year-old is aggressive and delinquent might be expected to be a stronger predictor of teenage pregnancy than observing the same behavioural problems in a 4 year-old. Some young children with behavioural problems will grow out of these problems by their teenage years, while for other children the onset of these problems would be at older ages. To address this issue, we fitted an age interaction with aggressive and delinquent behaviours. The adjusted association with aggressive and delinquent behaviours was strongest among girls aged 11 years and over, however the presence of the combination of these two mental health problems was significantly associated with teenage pregnancy regardless of the age at which it was identified. While other studies have identified that delinquent teenagers often proceed to risk behaviours such as drug taking and early sexual initiation [33, 34], our study extends other findings that found an association between conduct problems in girls aged 8 years and later teenage pregnancy [35].

Previous literature has shown that children of teenage mothers are more likely to exhibit aggressive and delinquent behaviours. One study found that adolescent children of younger mothers (aged 18-years or younger at the time of their first antenatal visit) were more likely to be delinquent and aggressive than the children of older mothers [36]. Another study suggested that the young children of teenagers were more likely to have higher levels of aggression than the young children of older mothers [13]. Together the studies suggest that the daughters of teenage mothers are more likely to show aggressive and delinquent behaviours, while we have found that these behaviours increase the probability that they too will become teenage mothers. Other studies suggest that teenagers who exhibit aggressive and delinquent behaviours are more likely to engage in sexual risk behaviours, including those that result in teenage pregnancy [33, 37]. It is important to prevent and manage aggressive and delinquent behaviours early before they become entrenched, but additionally, our results suggest that addressing these behaviours as early in developmental pathway as possible may reduce the likelihood of teenage pregnancy.

Whether the girl's mother was herself a teenage mother was strongly associated with teenage pregnancy outcomes, but was not a significant independent predictor in the final model. This suggests that other factors included in the final model, such as family type, income and education level, and carer smoking mediate the relationship in teenage motherhood from one generation to the next. The circumstances of a teenage mother are often low education, poverty and difficult family circumstances. Once we accounted for these factors, the association is not strong enough to remain independently significant.

Primary carers who were smokers were more likely to have daughters who became pregnant as teenagers. Smoking by primary carers remained a significant predictor of teenage pregnancy even after adjusting for a range of other variables including socio-economic status as measured by SEIFA [29], highest school year completed by primary carer, family type, primary carer in receipt of government benefits, combined carer income and whether the teenager lived in a metropolitan or non-metropolitan area as a child. Despite the presence of a wide range of other demographic and psycho-social variables, it is possible that smoking status was masking another variable that was not collected in our original survey and was not available via the data linkage. One variable that we could not account for adequately was mental health of the carer, and there is an established association between mental health problems and smoking [38, 39]. This issue deserves further investigation. When carer smoking status was assessed in 1993, smoking rates were higher than they are now. Nevertheless the health consequences of smoking had been well publicised for many years at that stage, and it is quite likely that carers, principally mothers, who were smokers were systematically different from non-smokers in ways other than just the demographic and psycho-social variables measures in the survey.

Girls from step/blended families were significantly more likely than girls from original families to become pregnant as teenagers. The difference between sole-parent families and original families was significant in the bivariate analysis but not significant in the adjusted proportional hazards regression model. Our data do not allow us to speculate as to whether the higher teenage pregnancy rate was impacted by the original family breaking up, or the arrival of the new parent in the step/blended family, or a combination of these two issues. We did not know the girl's age at the time of the family break up, or her age at the time of the arrival of the step parent. Others have also found an effect of family type on rates of young motherhood [10, 24, 40, 41]. Girls are more likely to delay first sexual intercourse and use contraception at first intercourse if they had lived with both natural parents to sixteen years of age [42]. Sole parent and step/blended family status are associated with higher risks for aggression and delinquency in young people [43]. Girls from step/blended families had an elevated risk for teenage pregnancy after adjusting for education and income factors.

Combined carer income and the highest year of school completed by the primary carer had strong effects on predicting risk for teenage pregnancy, even after controlling for a host of other child specific and family variables. The daughters of families with a low combined carer income, and those of carers who completed Year 9 or lower at high school, were more likely to become pregnant as teenagers, findings which were consistent with other research [34, 36, 42, 44]. Individual or household based socioeconomic measures such as these were stronger predictors of teenage pregnancy than community based measures such as SEIFA [29]. In addition, one study found that girls from socioeconomically disadvantaged households tended to know less about sex and contraception, and to have negative or indecisive attitudes regarding condom use, increasing their risk of becoming pregnant as teenagers [45].

A major strength of this study is the increased analytic power which comes from linking data from our 1993 cross sectional survey to administrative data collected in the fourteen years that followed. An additional strength is the wide range of variables, based on a social-ecology model, available from the WACHS. With respect to the data on abortions of teenage pregnancies, there is evidence to suggest that self-reported data on abortions may lead to under-reporting [46–48]. Our data linkage design avoids the bias associated with self reports and provides an opportunity to obtain accurate and less biased data on abortions. Another strength of our study is the high linkage rate of the 2,304 WACHS children that were born in WA, 2,282 were linked to their birth records. Of the 432 WACHS children that were born outside WA, 355 linked to the morbidity, mental health or electoral roll records. The overall level of inaccuracies in the WADLS (including false positives and false negatives) has been estimated at 0.11% [49].

This study has several limitations. Data on pregnancies includes only abortions and live births (i.e. miscarriages have been excluded). We did not have complete ascertainment of abortions, as data from some private clinics was not available via the data linkage. We used the actual abortion proportion for WA to confirm that the numbers we had observed in our population were close to what was expected. The abortion proportion is derived by taking the number of abortions divided by the number of abortions and live births combined. The average abortion proportion in WA over 2002-2005 for women aged less than 20 years was 51.9% [50]. For the teenagers in our study, we had 72 abortions and 94 live births, an abortion proportion of 43.4%.

It is inevitable that some girls sampled in the WACHS had moved out of WA since 1993, and therefore had reduced opportunity to have a record on the administrative datasets that were used in the data linkage. However, analysis of the number of survey children now on the electoral roll in WA showed that approximately 86% had enrolled to vote in WA, suggesting that the capture rate was still high.

With the available sample size, the study may not have had sufficient power to detect associations with risk factors that occurred rarely or were associated with small effect sizes. It is thus possible that some of the risk factors we considered but did not find significantly associated with teenage pregnancy have associations that were too small for the power of our study to detect.

Finally, it may have been useful to analyse the CBCL scores from the teacher and parent reports separately, in particular, those children who were rated as aggressive and delinquent. One study suggested that parent and teacher reports of child behaviours are not always linked, meaning that often only one source reports a problem while the other does not [51]. Although there was not sufficient statistical power to separate the parent and teacher reports of child behaviours, both showed similar patterns.

It is known that girls who become pregnant as a teenager will face increased medical risks, and if the pregnancy is taken to term, there are also increased medical risks for their babies. We also know that teenage mothers, and their children, can suffer from poorer life outcomes. In addition, the children of teenage mothers are more likely to become teenage mothers, perpetuating the intergenerational cycle of teenage pregnancies. With these consequences in mind, we should aim to further reduce the incidence of teenage pregnancy, in particular those pregnancies that are unwanted and unintended.

At the outset, we hypothesised that there are factors in early life that are potential predictors of teenage pregnancy. Our findings have confirmed this, and knowledge of these predictive factors should be used to inform future preventive policy interventions. Our findings suggest that interventions to reduce teenage pregnancy rates could address other issues in addition to the traditional focus on sex education programmes. In particular, early intervention to treat behavioural problems may also be useful in reducing the incidence of unwanted teenage pregnancy.