## Main

Treatment advances leading to cure or longer survival for malignant diseases in childhood have resulted in over 360 000 childhood cancer survivors in the United States (Howlader et al, 2011). Morbidity associated with increased survival often involves many organ systems (Hudson et al, 2003) and has the potential to adversely impact psychological functioning (Zeltzer et al, 2008; Michel et al, 2010). The consequences of distress are considerable with increased rates of suicide ideation documented among survivors with depressive symptoms (Recklitis et al, 2006a, 2010). As fewer than 20% of long-term survivors receive follow-up care by an oncologist (Oeffinger et al, 2004), the responsibility of identifying psychological morbidities in this growing population often falls to primary-care providers.

### Statistical analysis

Descriptive statistics were calculated for all outcomes, predictors and covariates used in the analyses. We used a structural equation modelling (SEM) technique that allowed for the identification of discrete profiles of psychological distress (Mplus 6; Muthén and Muthén, 1998). Longitudinal latent profile analysis (LPA), a submodel of SEM, is a multiple-group structural equation model in which the group variable is unobserved. LPA uses measured variables (e.g., psychological distress) to identify different groups of survivors where it is hypothesised that group differences exist but the number and nature of the groups is not known. The grouping variable (e.g., class membership) is not measured or observed but is derived from the observed data. In LPA, multiple statistical indicators are used to select the best fitting model. Fit indices considered in the current analysis included (1) Bayesian Information Criterion, with the lowest value among competing models indicating best fit; (2) entropy, an estimate of how well the model classifies subjects, with values closer to 1.0 indicating better classification; (3) Vuong–Lo–Mendell–Rubin likelihood ratio test (VLMR) and sample size-adjusted VLMR with P-values <0.05 indicating a significant difference in model fit between two nested models that vary by one class; and (4) minimum class membership 5% to provide sufficient power for subsequent analyses. Models were fit with two through six profiles for each distress index to determine the optimal number of profiles needed to describe survivors from baseline through follow-up 3. Posterior probability testing was also conducted to determine how well each participant fit into their assigned class.

Once class membership was established via the above methodology, predictors of class membership were examined through logistic regression modelling with robust variance estimates to account for within subject correlation using SAS version 9.2 PROC Logistic (SAS Institute, Cary, NC, USA). Multivariable models, considering all possible combinations of predictors and covariates, with the smallest Akaike information criterion were selected as the final model for each distress outcome. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated for all predictors and covariates retained in the final model.

## Results

Characteristics of survivors who completed the BSI-18 at all three study time points were similar to those of survivors who completed the BSI-18 at only one or two time points (Table 1). A slightly larger proportion of females completed the BSI-18 at all three time points compared with males (P0.001). Survivors who completed the BSI-18 at all three time points were, on average, 27 years of age at baseline (17 years from diagnosis), 35 years at follow-up 2 (25 years from diagnosis) and 40 years at follow-up 3 (30 years from diagnosis).

The LPA modelling indicated that survivors fell into four meaningful classes of depression (posterior probability range: 0.97–1.0, mean: 0.98), anxiety (posterior probability range: 0.85–1.0, mean: 0.89) and somatisation (posterior probability range: 0.92–0.98, mean: 0.94) based on their symptom scores over time. Figure 2 depicts the four distinct longitudinal patterns of symptoms of depression (Figure 2A), anxiety (Figure 2B) and somatisation (Figure 2C) defined by our models: (class 1) survivors with few or no symptoms at all time points; (class 2) survivors with elevated symptoms at baseline that decreased over time; (class 3) survivors with few or no symptoms at baseline that increased over time; and (class 4) survivors with elevated symptoms that persisted over time. Supplementary online Table 1 provides the model fit indices for two to six class solutions for each distress outcome. Mean scores for each longitudinal class across all study time points are shown separately for depression, anxiety and somatisation (Table 2).

### Persistent distress symptoms

For each psychological outcome, a subgroup of survivors was classified as reporting persistently elevated or chronic symptoms of distress (Table 2: class 4: depressive symptoms: 8.9%; anxiety symptoms: 4.8%; somatic symptoms: 7.2% of survivors). Compared with the majority of survivors who reported few to no symptoms of distress over time (class 1), class 4 membership was predicted by the presence of a mild-to-moderate medical condition at baseline (depression: OR=1.6; 95% CI=1.2–2.2; anxiety: OR=1.6; 95% CI=1.1–2.5; somatisation: OR=1.8; 95% CI=1.2–2.9), perception of worsening physical health over time (depression: OR=2.9; 95% CI=2.0–4.1; anxiety: OR=3.4; 95% CI=2.3–5.4; somatisation: OR=4.4; 95% CI=2.8–6.8) and increased cancer-related pain (depression: OR=2.1; 95% CI=1.4–3.2; somatisation: OR=3.3; 95% CI=2.0–5.4). Change in marital status from previously being married to being single was associated with increased likelihood of persistent depressive symptoms (OR=2.3, 95% CI=1.1–4.6), whereas change in employment status to unemployed was associated with persistent somatic symptoms (OR=1.8, 95% CI=1.2–2.8; see Table 3). Radiation therapy, employment change and personal income change did not contribute to models predicting persistent symptoms of anxiety or depression.

### Increasing distress symptoms

Class 3 defined by our models comprises survivors who reported few or no symptoms at baseline with increasing symptoms of distress over time (depressive symptoms: 10.2%; anxiety symptoms: 11.8%; somatic symptoms: 13.0% of survivors), such that significant levels were observed within the group by the end of follow-up. Compared with survivors with few to no symptoms over time (class 1), class 3 membership across all symptoms was predicted by reported perception of worsening physical health over time (depression: OR=3.3; 95% CI=2.4–4.5; anxiety: OR=3.0; 95% CI=2.2–4.0; somatisation: OR=5.3; 95% CI=3.9–7.4). Class 3 membership for symptoms of somatisation was associated with perception of increased cancer-related pain over time (OR=2.4; 95% CI=1.6–3.6) and female sex (OR=1.6, 95% CI=1.3–2.0; see Table 3). Radiation therapy did not significantly contribute to multivariable models predicting increasing distress symptoms over time.

### Decreasing distress symptoms

Among the identified classes of symptoms, class 2 and class 4 are characterised by elevated distress symptoms at baseline, with a reduction in symptoms to subclinical levels observed over time for class 2 (depressive symptoms: 15.1%; anxiety symptoms: 15.2%; somatic symptoms: 11.6% of survivors) compared with persistent distress symptoms observed for class 4. A reduction in anxiety symptoms (i.e., class 2 membership) was associated with decreased likelihood of worsened cancer-related pain (OR=0.5; 95% CI=0.3–1.0), decreased likelihood of worsened physical health status (OR=0.4; 95% CI=0.2–0.6) and increased likelihood of change in marital status from being single to married (OR=1.6; 95% CI=1.0–2.5) relative to survivors in class 4. A reduction in depressive symptoms was associated with decreased likelihood of worsened physical health status (OR=0.4; 95% CI=0.2–0.6) and decreased likelihood of change in employment status to unemployed (OR=0.7; 95% CI=0.5–1.0). Survivors treated with non-cranial radiation were significantly less likely to report decreasing somatic symptoms (OR=0.6; 95% CI=0.4–0.9). Survivors with worsened physical health status were also less likely to have decreasing somatic symptoms over time (OR=0.3; 95% CI=0.2–0.5; see Table 4). Chronic medical conditions were unrelated to decreasing symptoms of psychological distress.

## Discussion

We report on longitudinal patterns of psychological distress in a large cohort of adult survivors of childhood cancer. Our findings are novel and compelling because not only do we demonstrate that subgroups of survivors are at-risk for persistent distress over the course of survivorship, but we also establish that in survivors psychological distress may emerge over several decades following their original cancer diagnosis. These findings have important implications for screening practices among health care providers caring for long-term survivors of childhood cancer.

Overall, the majority of survivors reported no or few symptoms of psychological distress at multiple time points over the 13-year follow-up period. This is consistent with previous cross-sectional reports indicating that elevated distress symptoms may affect only a subset of childhood cancer survivors (Zebrack et al, 2004, 2007; Zeltzer et al, 2008). That the majority of survivors reported no distress over time also supports emerging literature describing the perceived positive impact of cancer in many survivors, consistent with the phenomenon of post-traumatic growth (Zebrack et al, 2012). Moreover, we identified a subset of survivors who reported improvement in distress symptoms over time. Of interest, when compared with survivors who reported similar symptom levels at baseline but did not demonstrate a reduction in symptoms over time, survivors with decreasing symptoms were largely characterised by the absence of risk factors associated with distress (i.e., treatment factors and changes in sociodemographic variables), as opposed to the presence of unique protective factors.

We identified a subset of survivors who reported persistently elevated distress (depression: 8.9%, anxiety: 4.8%, somatisation: 7.2%) over the 13-year follow-up period. This finding highlights the potential pervasive course of these symptoms for some childhood cancer survivors and suggests that this group of survivors may benefit from early targeted intervention efforts to reduce distress or better manage physical health symptoms. We previously reported that 19% of survivors in CCSS initiated treatment with an antidepressant medication over a 10-year follow-up period (Brinkman et al, 2013); however, the effectiveness of such treatment in this population remains unclear. In fact, we found that initiation of psychoactive medication use in the current sample was associated with increased symptoms of distress over time. Nonpharmacologic approaches toward the management of distress have demonstrated effectiveness in the general population and provide an alternative treatment approach for survivors. Specifically, cognitive-behaviour therapy (CBT) is a well-established treatment for anxiety and depression (Butler et al, 2006), with long-term benefits comparable to or exceeding those observed with pharmacotherapy treatment alone (Shea et al, 1992; DeRubeis et al, 1999; Hart et al, 2012). Of note, we did not have data related to the use of nonpharmacologic treatment approaches in our cohort of survivors.

Importantly, we found a modest proportion of survivors characterised by few distress symptoms at cohort entry that increased steadily over time (depression: 10.2%, anxiety: 11.8%, somatisation: 13.0%). This pattern reflects a group of survivors previously undetected in cross-sectional follow-up studies of childhood cancer survivors. Moreover, this group represents survivors who would not be identified through a single screening of distress at the time of study cohort entry. Thus, these data highlight the need for regular, repeated screenings of psychological distress in adult survivors of childhood cancer. Such screenings may be particularly important for survivors who develop late medical morbidities that result in reduced health status.

Notably, we found that survivor perception of worsened physical health status and increased pain was significantly associated with reports of persistent and increasing symptoms of distress over time. Moreover, changes in socioeconomic factors such as marriage and employment were associated with persistent and increasing distress. The presence of psychological distress may be conceptualised within a biopsychosocial model of health, which holds that health and illness result from the complex interplay of biological, psychological, and social factors (Engel, 1981). Thereby, macrolevel processes such as social support (e.g., marriage, employment) and microlevel processes (e.g. chemical imbalances, organ dysfunction) interact to produce a state of health or illness.

Effective distress management requires timely and accurate identification of symptoms. As only a small proportion of adult survivors of childhood cancer are followed by an oncologist (Oeffinger et al, 2004), the burden of identifying distress symptoms in this survivor population often falls to general practitioners. Previous data suggest that most physicians working with oncology patients do not feel confident dealing with distress and few utilise standardised questionnaires to assess patient distress (Mitchell et al, 2008). Moreover, front-line clinicians demonstrate approximately 50% sensitivity and 80% specificity when assessing distress in cancer patients (Mitchell et al, 2011). These data, in the context of our findings, suggest that routine screening of psychological distress with a validated instrument (e.g., Brief Symptom Inventory) is warranted in adult survivors of childhood cancer. However, comprehensive care extends beyond screening and requires appropriate referrals for mental health evaluations and indicated treatment (e.g., CBT).

Although our study provides important insights toward understanding the trajectory of psychological distress in adult survivors of childhood cancer, we must acknowledge several limitations. The longitudinal analysis only included data from survivors who completed the BSI-18 at all three study time points. Although survivors who completed the BSI-18 at one or two time points did not differ from those who completed all three on key demographic and treatment factors at baseline, we cannot rule out the potential influence of selection bias. The BSI-18 provides a measure of distress symptoms over the previous 7 days. Although our study was longitudinal, the infrequent measurement of distress coupled with the short symptom reporting window, may result in an inaccurate estimation of survivors who experienced distress symptoms over the 13-year follow-up. Our study used, almost entirely, self-reported data to measure outcomes and predictor variables. We did not consider reciprocal relationships between predictors and outcomes. It is possible that distress levels influenced reporting of certain factors, such as perceived health status and pain. In addition, there are likely several unmeasured variables that may contribute to distress symptoms in this population of survivors. Future research is needed to examine comorbidities between types of distress and between distress and other factors such as fatigue, personality, social support and community integration. Finally, although we identified subgroups of survivors with divergent patterns of distress symptoms over time, age and time since diagnosis were fairly heterogeneous both within and across study time points. This limits our ability to generalise our findings to other groups of adult survivors of childhood cancer.

Our results show that well into adult survivorship, the majority of childhood cancer survivors do not present with elevated distress symptoms. However, we identified two subgroups of survivors who will likely benefit from intervention to mitigate or prevent the onset of symptoms: (1) survivors with persistently elevated distress and (2) survivors with emerging distress symptoms over time. Toward this goal, our data underscore the importance of routine screening of psychological morbidities in adult survivors of childhood cancer.