Young adult cancer risk behaviours originate in adolescence: a longitudinal analysis using ALSPAC, a UK birth cohort study

Cancer is one of the leading causes of death in the UK [1] and has superseded cardiovascular disease as the leading cause of death in high income countries [2]. Tobacco smoking, [3, 4] alcohol consumption, [3, 5‐8] obesity, [3, 5, 9‐11] and unprotected sexual intercourse [3, 12] are known causes of cancer. There is also strong evidence that physical inactivity is associated with cancer incidence, [5, 13] with emerging findings indicating a causal association between physical inactivity and prostate, colorectal and breast cancers [14‐16]. These exposures are also known predictors of other adverse health outcomes and a significant burden to the NHS. In 2017–18, there were an estimated 489,300 hospital admissions and 77,800 deaths attributable to tobacco smoking, representing 16% of all deaths in the UK [17] and 10,660 hospital admissions with a primary diagnosis of obesity [18]. In 2018/19 there were 358,000 estimated admissions to hospital and 5698 alcohol-specific deaths, where the main reason for admission was attributable to alcohol [19]. While obesity is not a behaviour, for ease of description, we will refer to this group of exposures as behaviours henceforth.

Experimentation constitutes a normal part of growing up. However, these behaviours’ can become habitual and set the pattern for a life characterised by unhealthy practices [20]. While there have been significant declines in adolescent tobacco smoking and alcohol consumption in the UK, [21] the prevalence of other risk behaviours remains high. Twenty-eight per cent of children aged 2 to 15 years are overweight and among them, 17% of boys and 15% of girls are obese with prevalence increasing with age [22]. Physical inactivity and sedentary behaviour are common with only 18% of 5–16 years olds in England meeting current Chief Medical Officer guidelines of taking part in sport and physical activity for at least 60 min every day [18]. Young people also have the highest diagnosed rates of the most common sexually transmitted infections (STIs) of all age groups [23].

There is evidence that these behaviours co-occur in cross-sectional data, [24‐27] there is also evidence that single risk behaviours co-occur in longitudinal data, [28‐33] however less is known about how this multiplicity of risk behaviours pattern longitudinally across adolescence and into early adulthood. The aim of this study was to investigate the patterns of multiple cancer risk behaviours across adolescence (age 11–18 years) and their associations with subsequent cancer risk behaviours in early adulthood (age 24 years). Our primary hypothesis was that adolescents engaged in more cancer risk behaviours across adolescence would be at greater risk of early adult cancer risk behaviours at age 24 years. We have previously shown in cross-sectional analyses that multiple risk behaviours cluster by number of behaviours rather than producing distinct risk profiles [27]. Therefore, our secondary aim was to explore, in a longitudinal analysis, if multiple cancer risk behaviours cluster across adolescence to produce qualitatively distinct risk profiles, characterised by engagement in certain behaviours. To our knowledge this is the first study to develop longitudinal measures of multiple adolescent behaviours and explore associations with cancer risk outcomes in early adulthood.

Table 2 outlines the descriptive statistics for the complete case sample and the imputed data sample used in this research. For the complete case sample, high waist circumference with a prevalence of 411 (28.3%) was the most frequently occurring outcome at age 24 years. This was followed by harmful alcohol 273 (18.9%), obesity 183 (12.6%), daily smoking 121 (8.3%), high waist-to-hip ratio 114 (7.9%), and nicotine dependence 50 (3.5%).

Using repeated measures of cancer risk behaviours at four occasions across adolescence, we used two different methods to derive patterns of adolescent cancer risk. We found distinct groups of adolescents characterised by consistently high- and low-cancer risk behaviours during adolescence. We also found associations of large magnitude between adolescent and early adult cancer risk behaviours. We demonstrate for the first time that multiple cancer risk behaviours pattern longitudinally across adolescence and into young adulthood. Unlike previous research which focusses on either multiple risk behaviours at one time point (using cross-sectional data), or longer-term trajectories of single risk behaviours, our research combines the two. We have included a multiplicity of risk behaviours (five separate risk behaviours), using longitudinal data (at four different time points) covering a minimum 13-year period. Preventing multiple cancer risk behaviours during adolescence would likely reduce these behaviours in early adulthood and across the life-course, thereby reducing cancer incidence and mortality.

The strengths of this study were the long-term, longitudinal design. We were able to show associations between an exposure measure that covered the whole adolescent period (age 11–18 years) and outcomes 6 years later at age ~ 24 years. This has the advantage that most young people will have left education and entered the labour market and therefore our findings are indicative of a longer-term trajectory across different environments. We used both a cumulative score and latent classes to capture the exposure measure, which provided different insights, but were mutually reinforcing. We were also able to adjust for an extensive range of potential confounders, reducing the chance of residual confounding.

The weaknesses of the study include the potential for residual confounding and cohort attrition, as in all observational studies. However, for selection bias to pose a problem, our outcome measures would have to be conditionally related to whether a participant remains in the sample or missing not at random (MNAR) [45, 49]. We assume that our data is missing at random (MAR). Many of the measures were collected using self-report, which might mean participants favour responses they believe are more socially acceptable rather than choosing responses that reflect their true feelings. However, by the time ALSPAC participants had reached adolescence, following years of completing research questionnaires, it is likely that they will have developed trust in the maintenance of their anonymity, possibly reducing such bias. The risk behaviours were all reduced to binary variables in order to calculate the score total, which may lead to a loss of information. However, in line with our previous research and given the highly differential risks associated with these behaviours, at different levels and for different cancers, each risk has an equal weighting, as it would not be possible to accurately weight the risk behaviours differently. We were unable to include unhealthy diet as part of the adolescent exposure, as it was not collected in ALSPAC past the age of 13 years. We were also unable to include outcome measures relating to physical inactivity or sexual risk, owing to the extent of their missingness and the problem of perfect prediction as outlined in the methods. With reference to the latent class analysis, entropy was poor (6-class solution: 0.64 for complete case and 0.53 for the imputed sample and 4-class solution: 0.61 for complete case and 0.52 for the imputed sample). Therefore, while the classes may provide further insight into the patterns of adolescent behaviours, we cannot be confident that participants have been separated into the correct classes, which limits their utility as a targeting tool when considering public health interventions. They are also difficult to replicate because we would require the same number of similar exposures, at similar time points, which is not always available in other data sets. Finally, ALSPAC is not a nationally representative sample and therefore may limit the generalisability of our findings.

Given the co-occurrence of these behaviours and their associations with early adult risk behaviours, public health policies should adopt approaches that enable all health professionals who have contact with adolescents - including sexual health clinicians, general practitioners (GPs), public health workers and policy makers - to address a multitude of risks, at each contact. As a third of the cohort were identified to be persistently-moderate, or persistently-high engagers in risk throughout adolescence, it may be that early intervention is required to prevent engagement at an early stage in adolescence.

We found evidence that both the adolescent exposure measure and the outcomes at age 24 years (except harmful drinking) were socially patterned, i.e. the likelihood/odds of exposure to these risk behaviours increased for every incremental decrease in social position. However, interventions that focus on the most disadvantaged people, will not redress the social gradient and will only tackle a small part of the problem, while targeting subgroups excludes other deprived groups and risks stigmatising those targeted [50]. Further, because the social patterning for harmful drinking, ran in the opposite direction, adopting this strategy would be ineffective if we targeted interventions in this way. Universal interventions aimed at behaviour change at the individual level, have been shown to further exacerbate health inequalities, because more advantaged people and those who already have healthy behaviours tend to be quicker and more likely to take up these types of interventions [51]. Given this, and the social gradient of our exposure and outcome measures, interventions that are universal, but with a scale and intensity that is proportionate to the level of risk, [50] are preferable. Previous research relating to the social patterning of alcohol consumption has been mixed. For example, Melotti and colleagues found that adolescents that come from higher-income households in childhood were more likely to use alcohol. However, those with mothers with more educational qualifications were less likely to use alcohol [52]. Possible explanations that young adults with higher socioeconomic status in ALSPAC are more likely to be harmful drinking, include having increased disposable income available to them, which facilitates the purchase of alcohol, not available to less well-off individuals. An alternative explanation is provided by Luthar [53] who argues that young people from affluent families feel more pressure to achieve and are more isolated from their parents which contributes to their unhealthy behaviours. It is also possible that young people from higher socioeconomic backgrounds who are more likely to attend higher education, maintain drinking habits they acquired at university, into their early twenties.

A recent Cochrane systematic review showed that applying the World Health Organisation (WHO) health promoting schools framework improves some aspects of student health, including body mass index (BMI), physical activity, physical fitness and tobacco use, to an extent that is important at the population level. However, it found little evidence for improvement in zBMI (BMI, standardized for age and gender), and no evidence for alcohol use [54]. Another Cochrane Systematic Review looking at individual-, family-, and school-level interventions targeting multiple risk behaviours in young people, demonstrated that universal school-based interventions are most effective in preventing adolescent alcohol use, tobacco smoking, illicit drug use and antisocial behaviour, and increasing physical activity, but did not find strong evidence of benefit for family or individual-level interventions [55]. Therefore, efforts at preventing these behaviours should focus on developing appropriate proportionate universalist interventions in secondary schools.

We have shown that a multitude of behaviours known to cause cancer (tobacco smoking, alcohol consumption, obesity, unprotected sexual intercourse, and physical inactivity), originate in adolescence. Future research should focus on identifying the antecedents of adolescent cancer risk behaviours to identify opportunities for earlier intervention before the behaviours begin. Research is also required into whether the associations observed in ALSPAC are evident in similar cohort studies and outside of the UK.

We have shown that exposure to adolescent cancer risk behaviours greatly increased the odds of cancer risk behaviours in early adulthood. Interventions to reduce these behaviours should target multiple rather than single risk behaviours and should focus on adolescence. Given this, and the social gradient of our exposure and outcome measures, school-based interventions that are universal, but with a scale and intensity that is proportionate to the level of risk are preferable.