Interindividual differences in the dynamics of symptom development over time are best captured by person-centered approaches, which reflect the heterogeneity of symptom development across time. In a systematic review focusing on trajectory studies of depressive symptoms in late adolescence and emerging adulthood (15–24 years), the vast majority of studies found three to four trajectories, with a lower number of trajectories being more likely among younger adolescents (Schubert et al., 2017). In most studies, 40–55% of the assessed subjects fell into the trajectories with stable low or no depressive symptoms (stable low/absence of symptoms). A second reoccurring trajectory was found for a group of subjects (5–15%) with constantly high depressive symptoms (persistent depression/stable high). Furthermore, there was at least one group with less severe and less stable symptoms with varying increasing or decreasing slopes over time. In these intermediate trajectories external factors might play a more important role compared to the trajectories with low and high persistent symptoms (Schubert et al., 2017). Only a few studies investigated trajectories of depressive symptoms from adolescence to emerging adulthood in European samples, where mental disorder treatment is often covered by mandatory public health insurance. Evidence for depressive symptom trajectories has been reported for samples in the United Kingdom (Gaysina et al., 2011; St Clair et al., 2012), the Netherlands (Diamantopoulou et al., 2011; Lubke et al., 2016; Maciejewski et al., 2019), and Sweden (Lallukka et al., 2019).

It is known that specific demographic factors are associated with higher depressive burden. The identification of risk and protective factors might be useful for understanding the etiological factors in different symptom trajectories. Female adolescents, for example, have a higher incidence of major depression, a more chronic trajectory, an earlier age of onset, and a higher increase in depressive symptoms compared to male adolescents (Crockett et al., 2020; Edwards et al., 2014; Hargrove et al., 2020; Mezulis et al., 2014). In a meta-analysis of depressive symptom trajectories in children and adolescents, high or increasing depressive symptom trajectories were also predominantly predicted by female gender (Shore et al., 2018). There is evidence for a decrease of gender differences in emerging adulthood and a stabilization during adulthood (Salk et al., 2017).

Other predictors of increased depressive symptoms in adolescence and young adulthood include lower parental income and poverty, often preceded by lower parental education (Korhonen et al., 2017; Najman et al., 2010; Quesnel-Vallée & Taylor, 2012; Wickrama et al., 2009). However, evidence suggests stronger associations between socioeconomic status and externalizing problems compared to internalizing problems. When family income increased, externalizing symptoms of children (e.g., aggressive behavior) seemed to decrease, but internalizing symptoms persisted (Costello et al., 2003). Furthermore, cross-sectional and longitudinal findings suggest that female adolescents may be more vulnerable to economic deprivation than their male peers regarding their depressiveness (Walper, 2009). Higher education, being associated with higher family income, may also be a protective factor (Mendolicchio & Rhein, 2011).

Family type has also been shown to matter, with results indicating higher depressive symptoms among young people from single-parent households (Björkenstam et al., 2017; Laukkanen et al., 2016; Sieh et al., 2013), whereas living in a two-parent household was identified as a protective factor for developing a depressive disorder (Costello et al., 2008). However, stepfamilies can bear increased risks of children’s and adolescents’ adjustment problems (Jeynes, 2006). Increased stress in the context of parental separation and/or stepfamily formation and compromised family relationships may contribute to a higher risk of depression (Jensen & Lippold, 2018; Shafer et al., 2017).

Finally, there is evidence that adolescents with an immigration background and those who belong to an ethnic minority experience higher levels of depressive symptoms and an earlier age of onset compared to people with a nonimmigrant background, not only in U.S. samples (Brown et al., 2007; Costello et al., 2008), but also in European samples (Belhadj Kouider et al., 2014). An immigration background and membership in an ethnic minority is considered a risk factor for experiencing acculturation distress, facing discrimination, racism, and struggles in the process of identity development, which are, in turn, associated with an increased risk of the development of depressive symptoms (Desalu et al., 2019; Meca et al., 2019). We sought to replicate these findings in the present study.

This is the first study investigating trajectories of depressive symptoms from adolescence to emerging adulthood in a German community sample. The first aim of this study was to investigate whether subgroups of adolescents transitioning into young adulthood differ in their course of depressive symptoms, using longitudinal data drawn from seven annual waves of the German Family Panel (pairfam). Taking prior research (Schubert et al., 2017) into consideration, we expected to identify four to six trajectories, including trajectories with stable low and stable high symptoms of depression and at least two trajectories with intermediate symptoms with variations in slope (intermediate increasing and intermediate decreasing). A second aim was to analyze whether and to what extent sociodemographic variables that have been identified as risk factors for depressive symptoms (gender, family type, education, immigration, and economic deprivation) predict trajectory membership. The results of earlier studies led us to expect that female gender, lower level of education, economic deprivation, single-parent and stepfamily status, and immigration background would be associated more with trajectories of higher depressive symptoms with reference to trajectories with no, low, or decreasing depressive symptoms.

Our analyses are based on Waves 2–8 of the pairfam study. A detailed description of the study has been provided by Huinink et al. (2011). The pairfam study started in 2008/2009 with a nationally representative sample of three birth cohorts (1971–1973, 1981–1983, and 1991–1993). A total of more than 12,000 computer-assisted personal interviews for all cohorts were conducted in pairfam’s Wave 1 (2008/2009), followed by annual reinterviews. For the present study, only the cohort born in 1991–1993 was considered, as adolescents and their subsequent course of development were our focus. The first wave was not included because depressive symptoms were assessed only from the second wave onward. All participants of the respective birth cohort who completed Waves 2 (15-19 years) until Wave 8 (21-25 years) were considered with a 30% tolerance of missing data (Collins et al., 2001). The sociodemographic predictors were selected according to their last assessment date.

Participants were 3,682 adolescents and young adults who participated in the pairfam study (49.3% female). Their age ranged from 15 to 19 years in Wave 2, that is, T1 in our analyses (M = 17.03 years, SD = 0.88) with 96.8% of the sample being in late adolescence (16-18 years). At T1, 95.7% of the participants reported living with their parents, and 4.2% had already moved out (three missing values). Most participants were German natives (76.3%), with the remaining having some form of immigrant background (21.9%) or missing data (1.8%). At T1, 7.9% of the participants had left school without a degree or were currently enrolled in a lower-level secondary school (Hauptschule), 34.7% were enrolled in a middle-level secondary school (Realschule), 42.4% were enrolled in a higher level secondary school (Gymnasium), and 15.0% were enrolled in some other type of school. Thirty percent of the parents (Wave 2) had completed a lower education level, 40% a middle education level, and 30% a higher education level. At T1, 65.2% of the youth were living in nuclear families, 11.6% were living in stepfamilies, and 21.3% were living with a single parent (2% missing data). The mean value for economic deprivation was M = 2.3 (SD = 1.0; Range = 1-5) with a median of 2.

Analyses were conducted in Mplus Version 7.0 (Muthén & Muthén, 2017) using full information maximum likelihood. We chose the modeling techniques of latent class growth analysis (LCGA) and growth mixture modeling (GMM), to capture heterogeneity of depressive symptom trajectories. The main difference between LCGA and GMM lies in the modeling of the variance for intercept and slope. In LCGA, a special form of GMM, the variance for intercept and slope within the classes is fixed at 0, which allows for fewer parameters to be estimated (Berlin et al., 2014). We estimated a series of models (see Supplemental Table 1) with up to six trajectories for each modeling technique (LCGA vs. GMM). For the selection of the most suitable model, we used multiple fit indices as well as prior empirical evidence (Schubert et al., 2017). The relevant fit indices included information criteria such as the Akaike information criterion (AIC), where better fitting models have a lower AIC. Further, we implemented likelihood ratio tests where we assessed the relative fit of the model by comparing it to structurally similar models with one less class. Entropy was used to assess distinctiveness of classes, where higher entropy signals higher distinctiveness and more accurate assignment of individuals to classes (Infurna & Grimm, 2018). Further analyses were performed with R 2022.07.1. Once latent classes were identified, multinomial logistic regression was used to determine whether and to what extend sociodemographic variables predict a trajectory membership (dependent variable). One category of the dependent variable is chosen as the reference category, and other categories are compared to the reference. To correct for multiple comparisons, p-values were adjusted using the Benjamini-Hochberg procedure with a False Discovery Rate (FDR) set at 10%. This threshold was chosen in line with common practice in exploratory and multivariate research, where a more liberal FDR is acceptable to increase sensitivity and avoid Type II errors (Benjamini & Hochberg, 1995; Storey & Tibshirani, 2003). The correction was applied across all analyses, including those reported in the Supplementary Materials.

The model comparisons are shown in Table 1. The entropy values for the LCGA solutions were generally higher than those for the GMM solutions, which speaks for a better assignment of the individuals to the trajectories within the LCGA solutions.

On the other hand, the lower (= better) information criteria (AIC, Bayesian information criterion [BIC], and sample-size-adjusted BIC) within each class were obtained with GMM, which indicates a better model quality. The likelihood ratio tests show that the LCGA approach offers a five-class solution as optimal, whereas the GMM approach produces a three-class solution as most suitable. When analyzing these two models in more detail, only the LCGA solution identified the characteristic trajectories of very high and very low burden. In summary, the entropy and the consistency with previous empirical data support the five-class solution of the LCGA, so the classification of this model was used for further analyses.

Figure 1 shows the final class solution with the associated gradients and the respective sizes of the classes. Class 1 comprises young people with stable low depressive symptoms (34%); Class 5 depicts the small group with stable high depressive symptoms (4%).

In addition, three classes have an intermediate starting point at T1, showing either a slight decrease (Class 2: 8%), a slow increase that levels off at the starting point of those with a strong increase (Class 3: 4%), or a strong increase across time that rises up to the level of the stable high group (Class 4: 9%), Class 2 has the highest starting point among all three intermediate groups and drops to a final level close to those with a slow increase.

Since economic deprivation is associated with family type and immigration status, we first calculated a model without economic deprivation as a predictor variable (see Supplemental Table 2). The group with stable low symptoms was used as a reference group. Female gender was a significant predictor for all four classes associated with depressive symptoms. Living in a stepfamily and having an immigration background were predictors for membership in the groups with strongly increasing symptoms and chronic trajectory. Living in a single-parent family was also a predictor for a chronic trajectory. Multicollinearity was low with a variance inflation factor (VIF) ranging from 1.64 to 4.27, with a model fit of AIC = 4,969.26.

Second, we included economic deprivation in the model with a better model fit of AIC = 3,764.56 and revealed high multicollinearity with a VIF of 9.90.

Table 2 reports the descriptive statistics of sociodemographic variables for each trajectory. The results of the four multinomial logistic regressions for all four models with odds ratios are reported in Table 3, 4, and 5.

In Model 1, female gender and higher economic deprivation were significant predictors for all four classes associated with higher depressive symptoms with reference to the class with stable low depressive symptoms. Education, family type, and immigration background were not associated with memberships in trajectory classes.

In Model 2, female gender and economic deprivation were associated with a lower likelihood of a slow increase in depressive symptoms (Class 3) with reference to the class with decreasing symptoms (Class 2). Economic deprivation was a predictor for membership in the class with stable high symptoms (Class 5) with reference to the class with decreasing symptoms (Class 2).

In Model 3, female gender and economic deprivation were predictors for being in a trajectory with intermediate strong increasing symptoms (Class 4) with reference to the intermediate slow increasing trajectory (Class 3). In Model 4, odds ratios were non-significant for the trajectory with stable high symptoms (Class 5) with reference to the trajectory with strong increasing symptoms (Class 4).

The person-centered approach based on longitudinal data of a community sample, which covered the time span from late adolescence to emerging adulthood, is one particular strength of our study. Although this approach represents a reduction in complexity, as measurement points of depressive symptoms were limited to a 1-year time span, there is evidence that depressive symptoms in adolescents remain rather stable over a 1-year time span (McLaughlin & King, 2015). LCGA is a powerful tool for identifying latent subpopulations (depressive symptom trajectories) within a sample according to patterns of responses to observed variables (Muthén & Muthén, 2017) but class assignment is based on probabilities and may neglect infrequent patterns in the course of depressive symptom. Trajectories might also vary according to methodological differences in study design. It can further be discussed whether the low-stable and intermediate-slow increase groups are in fact meaningfully different, as the differences between these two trajectory groups for gender and economic disadvantage are much smaller than for the comparisons between the low-stable group and the other groups. Overall, however, our findings are well in line with previous research and provide new evidence pertaining to the phase from adolescence to emerging adulthood, which represents an important time window of development. Another limitation is the use of the measurement instrument to assess depressive symptoms. A measure that enables greater comparability with other studies (e.g. CES-D) would have been of advantage. We assessed gender, family type, parental education, economic deprivation, and immigration background as predictors for depressive symptom trajectories, but there may be other predictors that were not considered in the current study but could be important in understanding the development of depressive symptoms. Examples include family support, school climate, peer belonging (Gregory et al., 2020), and—not least—treatment use. The analyzed predictors are also not independent of each other. The risk of poverty is higher for single parents and income depends on the level of education (OECD, 2024). The fact that we did not control for possible changes in the predictors (e.g., economic deprivation, family status) across the different waves is also a limitation of the study. Sociodemographic variables do not cover the complexity of risk factors as risks can arise from the interplay of multiple factors within the guise of the analyzed variables. The FDR rate of 10%, which was used for alpha error inflation instead of the more conservative rate of 5%, can also be discussed critically. Although links between trajectories of depressive symptoms and adolescents’ help-seeking behavior would be of great interest, such issues require a different strategy of data analysis and will be addressed in future analyses. The data analysis was not preregistered, which should be considered for future analysis.

Understanding trajectories of depressive symptoms as well as risk and protective factors in the development of depressive disorders is crucial for identifying youths at risk for depression and increasing the probability of receiving adequate treatment. Longitudinal data on depressive symptoms in late adolescence transitioning into adulthood have demonstrated heterogeneous trajectories, enabling us to identify female gender and economic deprivation as risk factors for group membership in trajectories associated with higher and more stable levels of depressive symptoms. On an intermediate subclinical level of depressive symptoms, female gender was also associated with a higher probability of membership in a trajectory with decreasing symptoms over time than in one with increasing symptoms, suggesting more protective factors among females for coping with subclinical depressive symptoms. Family status and immigration status were not significant risk factors per se but were instead mediated by economic deprivation. It is of great importance to take measures to reduce female gender discrimination and economic deprivation as risk factors to mitigate the risks of developing depressive disorders. Such measures (e.g., political and societal initiatives) should target the underlying etiology of both constructs, taking into consideration the complexity and interaction of biopsychosocial variables. These results also emphasize the need for early detection and treatment of depressive symptoms, especially in the trajectories of stable high and strongly increasing symptoms.