Introduction
Depression is a common mental illness believed to affect more than 300 million people worldwide (WHO
2017). Depression is comorbid with other psychiatric conditions including anxiety, bipolar disorder and schizophrenia (Malhi and Mann
2018; Thapar et al.
2012), and is often associated with increased substance misuse, impaired educational attainment and increased risk of suicide (Fergusson et al.
2007; Marmorstein
2009).
One avenue for research has focused on childhood and adolescent depressive symptoms as a potentially modifiable risk factor for more severe depression during adulthood (Hill et al.
2014; Musliner et al.
2016) and previous research has shown that elevated levels of depressive symptoms within a population (and even those that never rise above the threshold for depression) are associated with a greater risk for depression in later life (Ellis et al.
2017; Yaroslavsky et al.
2013). Evidence concludes that the period between late childhood and adolescence may be important for subsequent mental health and social functioning, thus identifying and treating elevated depressive symptoms during this time could limit or prevent depression in later life through support and services (Fergusson et al.
2005; Thapar et al.
2012).
Research has yet to fully understand the nature of trajectories of depressive symptoms across adolescence and into young adulthood and identifying the key characteristics and time course of depressive symptoms across this period will ultimately aid in improving treatments and interventions. The current study attempts to build upon previous research by estimating trajectories of depressive symptoms between late childhood and young adulthood, and identifying critical points along those trajectories that could shed light on the nature of how these trajectories of depressive symptoms develop and change over time.
Previous studies have examined trajectories of depressive symptoms between childhood and young adulthood in an attempt to explore the nature and risk factors underlying greater depressive symptoms, for reviews see Musliner et al. (
2016), Shore et al. (
2018) and Schubert et al. (
2017). Indeed, research suggests that trajectories of depressive symptoms tend to increase from childhood through adolescence, before decreasing in young adulthood (Ge et al.
2006; Natsuaki et al.
2009). Evidence also suggests that these trajectories may peak in mid-to-late adolescence, towards the ages of 15–17 years old (Costello et al.
2008; Ferro et al.
2015b).
Other research has suggested that depressive symptoms may increase post adolescence. For instance, Ge et al. (
2001) demonstrated that trajectories of depressive symptoms were highest around the ages of 17–18 years old, but were unable to explore depressive symptoms further than this age due to a cease in data collection. Other research has suggested that trajectories of depressive symptoms are highest in young adulthood (towards age of 20) but then decline until older age (Sutin et al.
2013). In this instance, Sutin and colleagues did not have data preceding young adulthood so they were unable to explore if trajectories were higher around adolescence. It may be that trajectories of depressive symptoms could peak outside adolescence, yet do not have preceding and succeeding data around this period to substantiate these claims (Brendgen et al.
2005; St Clair et al.
2012). Likewise, many studies also use small sample sizes, which can make it difficult to infer about population level changes of depressive symptoms.
Observing the longitudinal nature of trajectories of depressive symptoms enables researchers to identify critical points at which to intervene to limit or prevent more severe depression. The notion of identifying critical points is not new (Bartley et al.
1997), but this application to trajectories of depressive symptoms has yet to be fully explored. Using depressive symptoms data from childhood to young adulthood (12 to 25 years), Ferro et al. (
2015b) identified the age of maximum depressive symptoms to occur between the ages of 15 to 17 years old, whilst Rawana and Morgan (
2014) found that depressive symptoms were highest at approximately 17 years old. Similar results have also been observed by Natsuaki et al. (
2009) who plotted trajectories that indicated the age of maximum depressive symptoms was 16 years old. Other research has also found evidence of sex differences in regards to an age of maximum depressive symptoms with females reaching this maximum earlier than males (Adkins et al.
2009; Edwards et al.
2014).
While identifying the age of maximum depressive symptoms is important for characterising the nature of depressive symptoms, calculation of the age of peak velocity may also be important (i.e., the age at which depressive symptoms are increasing most rapidly). Research has suggested that identifying depressive symptoms early or before depression already manifested may be a key step towards preventing greater depressive symptoms or severe depression from occurring (Cuijpers and Smit
2004; Kessler et al.
2001). By identifying the age of peak velocity of depressive symptoms, it may be possible to highlight a critical point where interventions and treatments could be implemented to reduce or limit greater depressive symptoms from escalating. However, no studies have calculated the age of peak velocity of depressive symptoms and studies that have previously identified the age of maximum depressive symptoms have done so using heuristics, graphs or figures rather than empirically calculating these ages.
Trajectories of depressive symptoms are not homogeneous within the population (Musliner et al.
2016), and a number of risk factors can influence trajectories of depressive symptoms (Weeks et al.
2014; Yaroslavsky et al.
2013). Consistent evidence has shown that females tend to have higher trajectories of depressive symptoms compared to males (Dekker et al.
2007; Yaroslavsky et al.
2013). Although in several studies where heterogeneous trajectories are identified, there is some evidence that females are associated with more intermediary trajectories such as “increasing”, “late onset” or “early high” (Costello et al.
2008; Olino et al.
2010). These studies suggest there may be mechanisms that underpin membership into varying trajectories, but the extent to which this can be explained has yet to be fully understood. Females also appear to have a higher peak along these trajectories of depressive symptoms (Adkins et al.
2009; Ge et al.
2001; Natsuaki et al.
2009), and may reach this peak earlier than males (Edwards et al.
2014), yet it is also not clear why this is the case.
Discussion
The longitudinal nature of trajectories of depressive symptoms between adolescence and young adulthood is not fully established, and research has yet to identify critical periods of trajectories of depressive symptoms that could potentially be used to target a stage of development where depressive symptoms might be increasing at the fastest rate. A greater understanding of the nature of depressive symptoms would aid researchers and clinicians in developing and improving treatments and interventions. The purpose of this study was to explore trajectories of depressive symptoms from childhood to young adulthood between males and females and to identify and compare critical points along these trajectories for both populations.
This study’s results showed that females and males have different population-averaged trajectories of depressive symptoms, with varying ages of peak velocity of depressive symptoms. Using multilevel growth-curve modelling on 8 waves of data between ages 11 to 22, results suggest that females were on average associated with higher trajectories of depressive symptoms compared to males, with the exception of between 10 and 11 years old where on average males had higher depressive symptoms. Both male and female population-averaged trajectories increased during adolescence before declining in young adulthood, yet females on average had a higher rate of deceleration (depressive symptoms slowing down) from age 20 and both sexes had trajectories that first plateaued and then started to decline in young adulthood. Evidence suggested that that on average, females had an earlier age of peak velocity of depressive symptoms (i.e., age at which depressive symptoms is increasing most rapidly), but little evidence to indicate that females had an earlier age of maximum depressive symptoms (i.e., age at which depressive symptoms is highest on the trajectory). Finally, depressive symptoms scores at the age of peak velocity and age of maximum depressive symptoms were both higher for females on average compared to males.
These findings support earlier hypothesis and previous research that trajectories of depressive symptoms increase from late childhood, through adolescence and begin to decrease in young adulthood (Adkins et al.
2009; Ferro et al.
2015b; Natsuaki et al.
2009). One possible explanation for increased depressive symptoms during adolescence is that young people face a number of social, psychological and biological changes during this stage of development (Adkins et al.
2009; Thapar et al.
2012). These changes include transitioning between schools, making new friends, taking exams and experiencing puberty. As many studies highlight that trajectories of depressive symptoms increase during this period, efforts should be made to monitor individuals who show heighted depressive symptoms as they may be individuals at a greatest risk of depression or higher levels of depressive symptoms.
This study was also able to expand on previous work by estimating points on population-averaged trajectories marking the average age of peak velocity, of maximum depressive symptoms and the depressive symptoms scores at both of these ages. These results show the age of peak velocity of depressive symptoms was almost 3 years earlier for females. The results suggest that depressive symptoms are increasing most rapidly for females at approximately 13.7 years old and for males at 16.4 years old, and that treatment and interventions could be implemented at different ages for males and females. These findings have implications for clinical services, schools and parents, who should be made aware that females are more likely to be younger when depressive symptoms are increasing most rapidly, with males following at a later stage. Likewise, as females appear to have higher trajectories for a longer of period of time, more awareness could be targeted towards services, schools and parents over this period of heightened and increasing depressive symptoms. Identifying features such as the age of peak velocity may help determine at what age depressive symptoms are getting worse most rapidly, but also highlight a key characteristic of trajectories of depressive symptoms that is potentially modifiable. Future research should primarily examine if the age of peak velocity identified here is universal to other cultures and countries and should also look to examine other predictors of these ages to see if they are potentially modifiable. Given that individuals with higher starting points (intercepts), had steeper trajectories (slopes), it is important to identify depressive symptoms early and when they are increasing as the current results suggest that those who start with higher depressive symptoms are at a greater risk of continuing to have higher trajectories of depressive symptoms.
This study’s findings suggested that the age of maximum depressive symptoms was approximately a year apart between males (20.7 years old) and females (19.7 years old), although evidence for a difference was weak. Several studies have suggested that this age should occur around mid-to-late adolescence, approximately between 15–17 years old (Ferro et al.
2015b; Natsuaki et al.
2009; Rawana and Morgan
2014). However, other evidence has indicated that this age of maximum depressive symptoms occurs later in development (Ge et al.
2001; Sutin et al.
2013), which coincides with the current results. There are several potential explanations as to why the current results observed a much later age of maximum depressive symptoms compared to some previous research. In several studies, the number of depressive symptoms measurements from follow-up are low and are therefore unable to pick up more nuanced changes in depressive symptoms over time. Depressive symptoms are dynamic and can change rapidly over a short period, thus researchers need frequent measurements to track subtle changes (Wesselhoeft et al.
2013). In the present study, the measurements were assessed on average never more than 3.5 years apart, which is more regular than many of the previous studies. It could be that frequently assessing depressive symptoms allowed detected changes and characteristics of the trajectories that were not observed in previous studies.
Another explanation is that variation in the age of maximum depressive symptoms is the result of contextual differences between the cohorts used. For example, similar ages of maximum depressive symptoms were observed using data from the National Longitudinal Study of Adolescents Health (Costello et al.
2008; Natsuaki et al.
2009), and the Canadian-based National Longitudinal Survey of Children and Youth (Ferro et al.
2015b; Rawana and Morgan
2014). However, other research using different North American cohorts have observed later ages of maximum depressive symptoms (Ge et al.
2001; Sutin et al.
2013). ALSPAC is one of the few longitudinal studies in the UK that currently has repeat depressive symptoms data across this period of development so it hard to conclude whether similar effects would be observed in other UK studies at this point. Other studies that examine depressive symptoms in longitudinal settings around the world use alternative methods to derive latent classes of trajectories of depressive symptoms (Costello et al.
2008; Yaroslavsky et al.
2013), and as such it is harder to compare critical points from these studies as there are often 4 or 5 trajectories from each study (each with their own age of maximum depressive symptoms). However, an important point to consider is that this study is the first to empirically estimate this age and present it with measures of uncertainty rather than simply describing it from figures. It is unlikely that this will explain the observed differences with other studies, but in the interest of interpretability and to help characterise the nature of trajectories of depressive symptoms, future research should estimate and report this age as this will help clarify whether there cross cultural differences for critical ages in the development of depression. Similarly, this is the first study to calculate and then compare critical points from two population trajectories (males vs. females). Interestingly, the current results supported previous research that suggested no sex differences between the ages of maximum depressive symptoms (Ge et al.
2001; Natsuaki et al.
2009), but contradicted other research suggesting that females have an earlier age of maximum depressive symptoms compared to males (Edwards et al.
2014). However, the authors here did not use the same number of measurements, used a quadratic polynomial model and only had data that went up to approximately 18 years old. Future research should examine depressive symptoms regularly around these ages to see if the critical periods identified are similar across contexts. Such research will inform services, schools and parents about the nature of depression and how to prevent and treat it around these ages.
Still, it is not clear why females on average have higher trajectories of depressive symptoms and why females and males differ in their ages of peak velocity of depressive symptoms. One explanation is that women tend to experience puberty earlier than men and evidence has suggested that early pubertal timing may be a mechanism responsible for depression and higher depressive symptoms (Joinson et al.
2012; Thapar et al.
2012). Research has also shown that an earlier age of menarche is positively associated with higher depressive symptoms (Joinson et al.
2013; Joinson et al.
2011) and a causal mechanism for greater depressive symptoms (Sequeira et al.
2017). Transitioning through puberty is associated with other psychological and social changes, and individuals who transition early may not have developed the cognitive and emotional skills to combat these changes, and therefore experience lasting effects of depressive symptoms. Likewise, early pubertal changes could result in increased responsiveness to stressors in females, resulting in higher depressive symptoms (Thapar et al.
2012). These findings suggest that individuals with higher starting points, had higher trajectories. Therefore an earlier age of higher depressive symptoms may set an individual up for a higher trajectory which takes longer to recover from. This could explain why females have higher trajectories compared to males, although more research using the timing and changes in pubertal status for both males and females would be needed to substantiate this claim.
Additionally, Angold et al. (
1998) found that depression was higher for females between the ages of 9 and 16, and this seemed to coincide with both pubertal status and timing of puberty. Of note, the number of girls with depression in their study was highest at around age 14, which coincides with the age of peak velocity of depressive symptoms in the present study. This suggests there may be some common mechanism at this age that is underpinning depressive symptoms, and how this manifests. The age of peak velocity from the current study tends to match the ages at which both males and females experience puberty and so one possible reason why the current study observed varying ages of peak velocity of depressive symptoms between males and females may be through the role of puberty.
Despite a number of strengths, this study has several limitations that should be highlighted. One limitation that arises with the data used here, and more generally with longitudinal data, is attrition and the role this plays in biasing results towards individuals who respond. The sample size in this study decreased from 7,335 at the first wave of data collection to 3,840 by the eighth occasion opening the possibility to potential attrition bias. Analysis on individuals with depressive symptoms data at both the first and last occasion, compared to those with depressive symptoms data at the first occasion but not the last, revealed differences in the overall symptoms scores and underlying demographics. However, this is consistent with previous research (Edwards et al.
2014), and suggests that studies with attrition could be underestimating the effect of sex differences in trajectories of depressive symptoms as females were more likely to have not responded at the last occasion. Previous studies have also imputed missing data utilising a missing at random approach (MAR) but found that bias due to systematic missingness in ALSPAC is not substantial (Bould et al.
2014; Pearson et al.
2017). The multilevel growth-curves models used in the present study also use FIML to account for missing data and sensitivity analyses revealed that the main effects of this study (i.e., the trajectories and the critical points) were not substantively affected by the inclusion of covariates associated with missing data and attrition bias. Nevertheless, future studies should highlight missing data patterns and attempt to account for data that could potentially be missing not at random.
Another limitation in this study stems from the choice of model used and the assumptions made with multilevel growth-curve models. Modelling trajectories of depressive symptoms appropriately is challenging. A cubic polynomial model was chosen given it is a more parsimonious approach in comparison to splines and fractional polynomials and that it has been used in previous studies that show nonlinearity (Ferro et al.
2015a; Rawana and Morgan
2014). However, alternative approaches such as a restricted cubic spline model may fit the data better at the expense of parsimony. Similarly, a quartic polynomial model may be a better model if the depressive symptoms data continues to rise. Cubic terms (and other polynomials) may also perform poorly at the start and end of the trajectories, as well as potentially producing artificial turns in the data that do not exist (Tilling et al.
2014). Checks were made to ensure that no artificial turns occurred in the data by comparing against other models and the underlying data and descriptive statistics for plausibility. Additionally, the age of peak velocity of depressive symptoms and age of maximum depressive symptoms were calculated well within the range of the trajectories so any bias from potentially mismodelling the start and end of the trajectories is minimised.
A similar limitation in this study is in regard to highlighting variability with the multilevel growth-curve model. A problem with population growth curves like the ones used in the present study is that it is harder to convey how much variability exists across the population, compared to other methods such as latent class growth analysis or growth mixture modelling, which typically stratify population trajectories into multiple subpopulation trajectories. Future research could derive critical periods from latent class analysis to examine if certain groups of trajectories have earlier ages of peak velocity or later times of maximum depressive symptoms. Such research could further highlight variability in critical points across multiple trajectories.
Acknowledgements
We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists and nurses. This publication is the work of the authors and A.S.F.K and G.L. will serve as guarantors for the contents of this paper. The UK Medical Research Council and Wellcome (Grant Ref: 102215/2/13/2) and the University of Bristol provide core support for ALSPAC. A comprehensive list of grant funding is available on the ALSPAC website. This research was specifically funded by Wellcome (08426812/Z/07/Z), Wellcome and the MRC (076467/Z/05/Z; 092731; 092731; 092731), the MRC (MR/M006727/1), NIH (PD301198-SC101645). A.S.F.K is funded by an ESRC Advanced Quantitative Methods Studentship. NJT is a Wellcome Trust Investigator (202802/Z/16/Z), is the PI of the Avon Longitudinal Study of Parents and Children (MRC & WT 102215/2/13/2), is supported by the University of Bristol NIHR Biomedical Research Centre (BRC-1215-20011), the MRC Integrative Epidemiology Unit (MC_UU_12013/3) and works within the CRUK Integrative Cancer Epidemiology Programme (C18281/A19169). ES works in the Medical Research Council Integrative Epidemiology Unit at the University of Bristol which is supported by the Medical Research Council and the University of Bristol (MC_UU_00011/1).
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.