Trajectories of adjustment disorder symptoms in post-treatment breast cancer survivors

Objective Breast cancer survivors (BCS) may experience problems to adjust to their situation after cancer treatment completion. In case of severe distress, an adjustment disorder (AD) might develop. This study investigates the course of AD symptoms during 1 year and its predictors in BCS up to 5 years post-treatment. Methods BCS completed the Hospital Anxiety and Depression Scale (HADS) at baseline, 3, 6, and 12 months. HADS total scores were defined as no mental disorder (MD) symptoms (≤ 10), AD symptoms (11–14), and any other MD symptoms (≥ 15). Over the course of four assessments, symptom trajectories were a priori defined as no MD symptoms, AD symptoms, fluctuating AD symptoms below and above cut-offs, or any other MD symptoms. Complementary, latent class growth analysis (LCGA) was used to identify data-driven trajectories. Results Among 293 BCS with complete data, the majority was classified as no MD symptoms (54.4%), followed by 37.5% in the fluctuating AD symptoms trajectory. Only 1.4% had AD symptoms, and 6.8% had any other MD symptoms. With LCGA (N = 459), three trajectories were found: stable no MD symptoms (58.6%), stable AD symptoms (32.9%), and high increasing any other MD symptoms (8.5%). Compared to BCS with no MD symptoms, BCS with fluctuating AD symptoms or any other MD symptoms were younger, less able to handle daily activities, and showed more social support discrepancy, neuroticism, and less optimism. Conclusions Results of our study showed that AD symptoms in BCS up to 5 years post-treatment fluctuate over 1 year. It is thus important to appropriately assess AD over the course of 5 years post-treatment as AD symptoms can fluctuate.


Background
Breast cancer is the most prevalent cancer type in women. Improved methods for early cancer detection and innovations in cancer treatment have increased the 5-year survival rate in breast cancer survivors (BCS), which is currently 91% in The Netherlands [2,32]. As a consequence, more BCS are dealing with the long-term complications of their cancer treatment, including the psychological burden [5,14,24,38]. When cancer-related distress is severe, an adjustment disorder (AD) can be diagnosed [10,39], if symptoms are not solely an exacerbation of a pre-existing mental disorder (MD) and the criteria of another MD are not met. In The Netherlands, reimbursement of psychological interventions for cancer survivors is available if a MD such as an anxiety disorder or major depression is diagnosed. It is currently being investigated whether AD can be added to the reimbursement scheme.
In the DSM-V, AD has been defined as the presence of emotional and behavioural symptoms in response to an identifiable stressor(s) occurring within 3 months of the onset of the stressor(s). The accompanied distress is out of proportion to normal reactions to the stressor in social or cultural context. After ceasing of the stressor or its consequences, symptoms of AD resolve within 6 months (Criterion E); however, if stressors or its consequences continue, this may result in persistent AD [3]. After curative cancer treatment, continuous confrontation with stressors is possible due to [18,27], optimism [18,26], neuroticism [18], and benefit finding [27]. Most studies [17,18,22,26] used a growth mixture modelling approach to determine trajectories from a data-driven point of view, and one study used cut-off scores from a clinical point of view [27].
The primary aim of this study was to detect trajectories of AD symptoms during 1 year using the HADS in BCS, using both clinically relevant cut-off scores and a data-driven growth modelling approach. The secondary aim was to identify predictors for distinct trajectories.

Participants and procedure
Physicians invited 1205 BCS from three hospitals in The Netherlands to participate with an information letter. Eligible participants were cancer-free, ≥ 18 years old, with stages I-III breast cancer treated with curative intent, who finished primary cancer treatment in the past 5 years, and were able to complete questionnaires in Dutch. BCS currently on hormonal therapy or treatment with specific antibodies (trastuzumab) were also eligible. After informed consent, participants received a questionnaire booklet (paper-and-pencil) or email with a link to a secured online system. Questionnaires were sent upon enrolment and after 3, 6, and 12 months.

Demographic, clinical, and psychosocial measures
Participants completed socio-demographic variables (e.g. age and employment status) and self-reported clinical variables (e.g. type of treatment and time since diagnosis) in the baseline questionnaire.
Social support was measured with the Social Support List-Discrepancies (SSL-D), a 34-item questionnaire (4-point Likert scale). The SSL-D measures the perceived discrepancy between the amount of received social support and the desired amount of social support [42], further referred to as social support discrepancy. A higher total score of SSL-D (range 34-102) indicates more social support discrepancy. The test-retest reliability is 0.85 and a Cronbach's alpha of 0.95 [42].
Neuroticism was measured using the Big Five Inventory (BFI) [16], a 44-item questionnaire designed to measure the Big Five factor structure of personality (5-point Likert scale). The Neuroticism subscale (BFI-N) measures the trait of neuroticism opposed to emotional stability, with increasing scores indicating a larger tendency to experience negative emotions. This version of the BFI has a Cronbach's alpha of 0.86.

AD symptoms and trajectories
Our definition of "AD symptoms" was theoretically derived from the DSM-V definition of AD, which describes that AD symptoms are characterized by marked distress, while the distress should not meet criteria for another MD. Cutoff thresholds for the HADS have been established that are sensitive to detect any MD (score ≥ 11) and depression (score ≥ 15) [43]. As such, we have assumed that a HADS score of 11 to 14 (i.e. marked distress but not depression) are indicative of AD symptoms. This is in line with previous studies reporting that the optimal HADS total score for screening for AD is 10 or 11 [1,25,33]. Thus, we predefined categories on HADS total: (1) ≤ 10 was defined as "no MD symptoms", (2) 11 to 14 as "AD symptoms", and (3) ≥ 15 as "any other MD symptoms". Trajectories were created based on HADS total over four assessments and defined as (a) no MD symptoms at all four assessments, (b) AD symptoms at all four assessments, (c) any other MD symptoms at all four assessments, and (d) fluctuating AD symptoms, i.e. an increase, decrease, or irregular pattern of HADS total.

Data processing and statistical analyses
Missing item scores on the HADS were replaced by the participants' subscale mean if at least four subscale items were answered [8]. Participants who completed the HADS all four assessments were considered completers, and participants who had a missing HADS or did not report date of birth or time since diagnosis were considered non-completers. Completers and non-completers were compared on demographic and clinical variables using t-tests for continuous variables and chi-square tests for categorical variables. Completers were assigned into our a priori defined trajectories based on their score above or below cut-off scores, and trajectories were compared on demographic and psychosocial variables with univariate testing (oneway ANOVA, chi-square tests, and post hoc analysis).
Variables that were significantly associated with trajectories membership were entered in a final multinominal regression analysis. Analyses were performed with SPSS version 25.
Latent class growth analysis (LCGA) was conducted using MPlus version 7 to identify data-based trajectories (classes) over time for HADS total, following the guidelines described by Jung and Wickrama [21]. By estimating individual differences (variability) in parameters reflecting participants' change in outcome over time, individuals are classified into latent classes based upon similar patterns in the outcome of interest (HADS). MPlus' full information maximum likelihood estimation for handling missing data was applied.
Following the guidelines, a single-class growth curve model was specified, as well as a three-class model. To determine the number of classes, the three-class model was compared with a two-class and four-class model, and the four-class model was compared with a three-class and five-class model. In total, the fit of five unconditional latent class models (i.e. models with no covariates) were estimated, with one to five linear classes. The number of classes was determined based on fit indices, model parsimony, and clinical interpretability. The model with the best fit has the smallest Bayesian Information Criterion (BIC) and significant p-values (p < 0.05) for the Vuong-Lo-Mendell Rubin Likelihood Ratio Test (LMR-LRT) and the Bootstrap Likelihood Ratio Test (BLRT), which indicate that a model with a k number of classes has a better fit than a model with k-1 number of classes. Other considerations were a higher entropy statistic (near 1.0), indicating the degree to which latent trajectories may be clearly distinguished, and higher posterior probabilities of group membership (near 1.0), indicating the degree to which individuals have been correctly classified into a class. For clinical interpretability, we also considered the number of participants (not less than 5% of total sample (n ≥ 23)) of the identified classes. For each individual patient in the database, the predicted class of the best fitting model (i.e., with the optimal number of subgroups) was obtained.

Sample characteristics
Of the 1205 eligible BCS who were invited, 459 participants (38.1%) consented and completed the HADS at least once. Demographic and clinical variables of completers, non-completers, and the full sample are shown in Table 1. Compared to non-completers, completers were older (p = 0.002) and had a lower education level (p = 0.026).

Trajectories based on cut-off scores
At group level, the average of all four HADS-assessments was 72.1% with no MD symptoms, 12.9% AD symptoms, and 15.0% any other MD symptoms. Classification in trajectories resulted in 157 BCS (53.6%) in the trajectory no MD symptoms, 4 BCS (1.4%) in the trajectory AD symptoms, 20 BCS (6.8%) in the trajectory any other MD symptoms, and 112 BCS (38.2%) in the trajectory fluctuating AD symptoms.

Predictors of trajectories based on cut-off scores
Given the low number of BCS in the trajectory AD symptoms, these BCS were merged with the trajectory fluctuating AD symptoms (stable&fluc-AD symptoms) and compared to the trajectory no MD symptoms and trajectory any other MD symptoms on demographic and psychosocial characteristics at baseline (Table 2).
The seven significant predictors were entered simultaneously in the multinominal logistic regression analysis with the trajectory no MD symptoms as the reference group ( Table 2). The final model was statistically significant (Χ 2 = 166.9, df = 14, p ≤ 0.001, Cox & Snell R 2 = 0.45, Nagelkerke = 0.54, McFadden = 0.34). Experiencing a recent life event and previous psychological counselling did not contribute significantly to the overall statistical model ( Table 2). Compared to BCS in the reference group, BCS in the trajectories stable&fluc-AD symptoms and any other MD symptoms were less able to handle daily activities, perceived a larger social support discrepancy, and showed less optimism and more neuroticism. Additionally, BCS in the trajectory stable&fluc-AD symptoms were younger compared to BCS in the reference group.

Predictors of trajectories based on LCGA
Univariate analyses (Table 4) showed differences between trajectories for age (p = 0.015), satisfaction with medical treatment (p < 0.004), being able to handle daily activities (p < 0.003), previous psychological counselling (p ≤ 0.001), experiencing a recent life event (p = 0.010), social support discrepancy (p ≤ 0.001), optimism (p ≤ 0.001) and neuroticism (p ≤ 0.001). These eight predictors were included in the final model with the low trajectory as reference group. The final model was statistically significant (χ 2 = 264.5, df = 16, p ≤ 0.001, Cox & Snell R 2 = 0.46, Nagelkerke = 0.55, McFadden = 0.35). Age, ability to handle daily activities, social support discrepancy, neuroticism, and optimism contributed significantly to the statistical model. BCS in the AD symptoms trajectory and the high increasing trajectory were less able to handle daily activities, perceived a larger social support discrepancy, and showed more neuroticism compared to BCS in the low trajectory. BCS in the AD symptoms trajectory were younger compared to BCS in the low trajectory, and BCS in the high increasing trajectory reported less optimism compared to BCS in the low trajectory.

Discussion
In this study, three distinct 1-year trajectories in HADS scores were found in BCS using two different statistical approaches: one approach with clinical cut-off scores to indicate AD symptoms or MD symptoms and one data-driven approach to predict classes of BCS with a similar course of AD symptoms or MD symptoms. The "low" trajectory was found in more than half of the BCS. The second trajectory with (fluctuating) AD symptoms was found in about one-third of the BCS. The trajectory with (high increasing) any other MD symptoms was found in fewer than one in ten BCS. Furthermore, the approach based on cut-off scores showed a very low (1.4%) percentage of BCS with stable AD symptoms and fluctuating scores below and above cutoff scores in almost 40% of the participants. With the latent modelling approach, we found a trajectory AD symptoms in one-third of the BCS, with a wide confidence interval of HADS scores per assessment and no significant change over time. Thus, both statistical approaches showed that AD symptoms can fluctuate in a significant proportion of BCS over time and that a pattern of stable AD symptoms was   not present in this sample. This questions the validity of the diagnosis AD in BCS. The detection of the low trajectories and (high increasing) any other MD symptoms is in line with previous trajectory studies [17,18,22,26,27]. These studies all reported trajectories as "resilient" and "chronic", with a stable course of few and high symptoms post-treatment up to 6 months [18,22,27], 8 months [26], and 4 years [17]. Non-stable trajectories were observed in all previous studies as well. Our study provided additional detailed observations by means of multiple assessments within a 1-year period, indicating more individual fluctuations in AD symptoms than was expected based on earlier findings. Fluctuations in AD symptoms were found independent of time since diagnosis, which is not in line with the DSM-V definition of AD [3] which assumes that AD diminishes over time, implying a self-healing process [11]. The discrepancies between this study and the established AD criteria stress the debate of AD diminishing after 6 months after AD symptom occurrence or becoming persistent in case of ongoing stressors in the cancer survivor context. Therefore, future research could be directed towards exploring acute and persistent AD immediately post-diagnosis and whether symptoms might fluctuate over time.
Compared to BCS in the trajectories with no MD symptoms, characteristics of BCS in the trajectories of AD symptoms or MD symptoms were a larger social support discrepancy, less optimism, and more neuroticism. These findings are in line with previous trajectory studies, where less social support [17,27], less optimism [18,26], and higher scores on neuroticism [18,26] were observed in "chronic distress" or "lower mental functioning" trajectories. Lower ability to handle daily activities is in line to the criteria of AD [3], where poorer functioning in social relations, work, or study is observed in people who are diagnosed with AD. Lastly, with the exception of Lam et al. [26], previous trajectory studies did not find age differences between trajectories, which is contradictory to our study. BCS with a trajectory of (fluctuating) AD symptoms were almost 3 years younger compared to BCS belonging to the trajectory no MD symptoms. A systematic review including cross-sectional and longitudinal studies found that a younger age increased the risk of distress [40]. These previous findings regarding predictors combined with the results of our study emphasize the relevance for clinicians to monitor these predictors to detect vulnerable BCS showing AD symptoms.

Study limitations
The results should be interpreted carefully because of selection bias in the study sample. Participants who completed all questionnaires were older and lower educated compared to participants who did not complete all questionnaires, although the sample used for the cut-off score analysis was comparable to the sample in the LCGA analysis. Furthermore, analysis of the predictive value of education was not possible due to too small cells, resulting in inconclusive findings on education. In our study, we assessed whether participants had previously received psychological counselling. We did not, however, assess whether participants had a history of mental illness, which could have been an important predictor to developing AD. This study was an additional analysis of a dataset on the course of fear of cancer recurrence (FCR) over time in BCS [15]. Secondary analyses reduce research participation burden, but results might be less generalizable to the overall BCS population, since BCS signed up for research investigating FCR instead of AD related to cancer.
For research purposes, the analyses of the HADS are of great value to gain insight in which BCS are at risk for AD. The HADS, however, does not assess impairments in social or occupational functioning, which is a limitation. While not assessed thoroughly in our study, BCS with a trajectory of (fluctuating) AD symptoms reported less ability to handle daily activities and had a larger perceived social support discrepancy. A diagnostic interview, use of an AD-specific questionnaire, or combining measures would capture AD more accurately. Finally, due to the small number of participants with AD symptoms, we were not able to further categorize subtypes of AD.

Clinical implications
This study used two different approaches to analyse the data, combining methodologies used in previous studies to observe the course of AD symptoms: a clinical point of view in using a cut-off score to screen for a possible AD or MD and a statistical point of view to predict latent classes based on scores over time. Both methodologies detected fluctuating symptoms over time. This would imply that conclusions based on single assessment HADS scores in clinical practice would not be sensitive enough to detect those patients with AD symptoms and for whom a diagnosis of AD might be applicable.

Conclusion
A substantial proportion of BCS up to 5 years post-diagnosis showed fluctuating AD symptoms, and only a negligible percentage of the cases had a stable course of AD symptoms. We suggest handling single assessment cut-off scores with caution.