Network analysis of depression in prostate cancer patients: Implications for assessment and treatment

Many prostate cancer patients also suffer from depression, which can decrease their life satisfaction and also impede recovery from their cancer. This study described the network structure of depressive symptomatology in prostate cancer patients, with a view to providing suggestions for clinical interventions for depressed patients.


| BACKGROUND
Between 15% and 22% of prostate cancer (PCa) patients suffer from comorbid major depression. 1,2 These depressed PCa patients also have higher likelihoods of emergency room visits, hospitalisations, outpatient visits, and excessive risk of death. 3 Consequently, recommendations for routine screening and treatment of depression in these men have been made, 3 but the heterogeneity of depression, 4 and the limited treatment efficacy for depression in the wider population (about 34% from singular treatments, rising to 74% when treatments are combined) 5 argue for a more 'individualised' method of assessing depression. For example, the difference in treatment efficacy across patients with different depression symptom profiles 6 and subtypes of depression (e.g., depression with anxiety, melancholic features, with atypical features and with mixed features) 7 challenges the 'one size fits all' model of some treatment approaches for these men. 8 Several studies have been made of the nature of depression in PCa patients for example, [9][10][11] using various methods of classifying subgroups of symptoms of depression. One hitherto unused method of identifying the underlying structure of depression symptomatology in PCa patients is by 'network analysis', 12 that describes the causal interplay between symptoms that may also include feedback loops of those symptoms. 12 This information could hold implications for the design and delivery of symptom-focused treatments. Previous network analyses of depression in non-PCa samples have provided valuable insights into the ways that the nine symptoms of Major Depressive Disorder (MDD) are related, with implications for focussed treatment of those symptoms. 13,14 To extend understanding of the nature of depression in PCa patients, with a view to developing more focussed and effective treatments, this study applied network analysis to the depression symptoms experienced by a sample of PCa patients.

| Participants
PCa patients from treatment centres in south-east Queensland participated in the study. All these men had biopsy-proven prostate cancer and were attending either for treatment or follow up after previous treatment. No patients were on active surveillance, and patients with all other stages of prostate cancer were included.
Treatments included radiotherapy and/or surgery and hormone therapy when required. Other inclusion criteria were: (i) the diagnosis of prostate cancer was proven histologically, and (ii) all of the treatment options were properly considered by patients via discussion with their GP, a radiation oncologist and a urologist. Unwillingness to participate in the study was the only exclusion criterion.

| Measures
As well as a questionnaire about their age, PCa status, past and present treatments, relationship status, and date of their first diagnosis, the PCa patients completed the Patient Health Questionnaire-9 (PHQ-9) for how they felt at the time. All these questionnaires were in English.

| Statistical analyses
Network analyses were performed using RStudio. 18 The regularised partial correlation network was estimated using the EBICglasso procedure with a default Extended Bayesian Information Criterion (EBIC) tuning parameter of 0.5. 19 The network structure, node (i.e., PHQ-9 items) centrality estimates, and accuracy and stability of the network and its properties (i.e., centrality, edge weights) were computed using the bootnet package. 20 Node centrality estimates serve as an indicator of the relative importance of nodes (depression symptoms) within the overall network structure; nodes that are more central are those that are more highly interconnected with other nodes. 21 Confidence intervals and significant difference tests for the centrality estimates and for edge-weights were computed with bootstrapping of 2500 samples.
The network structure was visualised using Multidimensional Scaling (MDS), which facilitates interpretation of the distance between nodes (i.e., items, symptoms), such that nodes that are in closer proximity to one another are more closely related, 22 and this was generated using the qgraph package. 23 Communities (i.e., clusters of symptoms within the network structure) were investigated to extract additional information about how depression symptoms relate to one another, as this has seldom been performed in network analyses of depression so far, 24 and was done here with the igraph package 25 using the spinglass community-detection algorithm. 26 In order to detect the presence of possible group differences in overall depression scores, network structure, and associated network properties that might occur due to the treatment stage, patients SHARPLEY ET AL.
-369 undergoing their initial treatment were compared to patients receiving treatment for reocurring PCa, and those in remission (Table 1) via ANOVA. Differences in network structure and properties were examined with the NetworkComparisonTest package (NCT). 27 Finally, node predictability (i.e., a measure of how well a given node can be predicted by its neighbouring nodes) was computed using the mgm package. 28 Reporting standards for network analysis described in Burger at al. 29 were followed. Where appropriate, Bonferroni corrections were made to the acceptable p value to reduce the likelihood of family-wise error.

| RESULTS
The upper section of Table 1 describes the background data for the sample. No data are available regarding the men who did not choose to participate, although the entire sample was recruited from the same patient pool. Shown at the bottom of Table 1, the current sample's distribution of depression severity was very similar to that reported by Kroenke et al. 15 for their sample of 474 participants who did not have a formal diagnosis of depression. The PCa sample data presented here may be said to represent a non-clinically-depressed subsample of the wider population, although with some members who reported severe-moderate, and severe distress, which is consistent with the findings reported by Hinz et al., 17

| Descriptive statistics
Floor effects were observed in PHQ-9 items, in that the means and the standard deviations for each of the 9 items were highly correlated with one another (r = 0.95, p < 0.001). Scores on motor problems and suicidal ideation were highly skewed in this sample (Supplementary Table S1), but means and standard deviations of these two items were similar to those reported in the general population sample by Hartung et al. 14 The relatively lower endorsement of these two items, plus the overall degree of skewness of all items, was consistent with other network analyses on the PHQ-9. 30,31 Figure 1 presents the network structure of the 9 PHQ-9 items for the sample. The low stress-1 value of 0.18 indicates that the distance between nodes was highly interpretable. 22 As such, it can be concluded that: (1) anhedonia was closely linked with depressed

| Node centrality
The strength and expected influence centrality estimates and their associated 95% confidence intervals are presented in Figure 2. The stability of the centrality estimates was acceptable (both the CS-coefficient of the strength centrality estimates [0.52]), and the expected influence centrality estimate (0.60) exceeded the 0.5 cutoff suggested by Epskamp and Fried. 19 By performing difference tests of the centrality estimates, it was found that anhedonia was significantly more central than most other symptoms in the network (see Sup- plementary Figure S1). The next most central symptoms were depressed mood, fatigue/lethargy, and low self-worth.

| Edge weight accuracy
The raw values and associated 95% confidence intervals for edgeweights are presented in Figure 2. The strongest statistical associations between symptoms were between sleep problems and fatigue/lethargy, and between anhedonia and depressed mood (see Supplementary Figure S2), indicating that these were the most reliable and robust symptom associations in the network. These statistical analyses confirm the interpretation of associations between PHQ-9 items that were suggested in Figure 1, (i.e., that patients who had sleep problems frequently also had problems with fatigue or lethargy, and patients who experienced anhedonia also frequently experienced depressed mood). The associations between suicidal ideation and low self-worth/depressed mood suggested in Figure 1 were not found to be robust by this analysis. Associations between other symptoms were not as reliable.

| Group comparisons
A one-way ANOVA indicated that there was a significant difference between the PHQ-9 total scores of the three patient groups defined in Table 1  Overall, all 9 PHQ-9 items were strongly connected with each other.
Second, anhedonia was significantly more central than most other symptoms in the network, followed by depressed mood, fatigue/ lethargy, and low self-worth. This means that if a PCa patient were to experience the symptom of anhedonia, it is likely that he would also suffer from each of the other symptoms of depression. Third, anhedonia was not only the most central (i.e., most highly interconnected) symptom, 52% of its variance was predicted by neighbouring nodes.
Given the relative strength of the anhedonia-depressed mood edge, much of this variance would likely be attributable to the association between these two nodes. As such, because of it being the most highly interconnected node in the network, plus its association with depressed mood, targeting anhedonia for treatment would likely reduce the patients' depressed mood and, subsequently, overall levels of depression. 21 Although anhedonia was found to be the central node in the depression symptom network found here, it is not as well-known as the more global notion of 'depression = sad mood' in the general community, despite being one of the two key symptoms required for a diagnosis of MDD. 7 Defined as loss of interest and/or loss of pleasure in activities which were previously enjoyed by the individual, anhedonia has been conceptualised as a withdrawal response to ongoing uncontrollable stress, 32 which fits with the experiences of a person diagnosed with a life-threatening illness such as PCa. There is some evidence that anhedonia is negatively correlated with activation of the key brain structures that are involved in reward processes, such as the nucleus accumbens, basal forebrain, and hypothalamus, 33 and may result from interference in the dopamine system, which in turn blunts the reinforcing effects of naturally occurring rewarding stimuli such as food, water or sex, leading to behavioural extinction of motor responses designed to access those stimuli-i.e., a loss of interest in pleasurable activities, or anhedonia. 34

| Clinical implications
Because anhedonic patients have difficulty in making the effort to obtain the rewards they desire, the most promising psychotherapy might be Behavioural Activation, 35 which is relatively directive, and therefore may help to overcome the patient's motivation deficits that are a characteristic of anhedonia. Additionally, antidepressant medication designed to re-establish dopamine stores may prove effective, and some recent data suggest that ketamine may prove efficacious with anhedonia. 36 Transcranial magnetic stimulation may also be useful. 37,38 When considering the possible causal factors that may have These data present a different network than reported in some previous studies based on data from a large sample of depressed patients, that did not find anhedonia was central in the network. 13 However, in their study of 4020 cancer patients, Hartung, et al. 14 did find that anhedonia was the most central symptom of MDD on PHQ9 data. These findings suggest that PCa depression may differ to that experienced by the rest of the population, and confirms the need to develop focussed and individualised treatments for these men.

| Study limitations
The usual limitations on generalisability of these findings apply, including the selective and voluntary nature of the sample, geographical and cultural restrictions, temporal limitations (no ongoing data were collected), and the use of a self-report of depression rather than clinician interviews. Some caution must be exerted regarding the lack of significant differences between three subgroups reported under 'Group Comparisons', due to the high network density 372 -SHARPLEY ET AL.
observed and the limited sample size for the treatment reocurring group. 27,39 Future research with larger sample sizes would help in confirming these findings, as would studies that tracked any changes in the depression networks of PCa patients from the time of first diagnosis, through to the end result of treatment, and at followup, particularly in deciding which aspects of depression were most likely to require treatment at which stage of the PCa patient's journey.

| Conclusions
Notwithstanding these limitations, these first data regarding the network of depression symptoms in a sample of PCa patients suggest that anhedonia may be central to the patient's overall depression, and that there are significant associations between some other symptoms of depression in these men. There is also evidence that PCa patients' depression may differ in its network structure to that reported in the general population. As such, the application of 'onesize-fits-all' global treatment models that are focussed upon the total score on a depression inventory may be less efficacious than treatments that are aimed at the specific symptoms exhibited by these men, such as the central symptom of anhedonia.

ACKNOWLEDGEMENT
The authors thank the prostate cancer patients who participated in this study.

CONFLICT OF INTEREST
None of the authors has any conflict of interest to declare.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.