In the current study, we aimed to examine patterns of associations between patient-reported fatigue, depression, subjective cognitive complaints, brain tumor-related symptoms, and HRQoL by applying symptom network analysis. Additionally, we sought to compare symptom networks of subgroups of glioma patients based on disease characteristics and fatigue status. First, we showed that the 21 studied items and subscales of the questionnaires were highly intercorrelated and could be represented as a network. In particular, fatigue severity, depression, and social functioning were highly connected to other symptoms. Interestingly, we found that how tightly PROMs were connected did not differ between networks based on disease characteristics, but PROMs were more tightly intercorrelated in fatigued patients compared to non-fatigued patients.

In the presented networks, all nodes were connected to each other, signifying how symptoms and HRQoL are highly associated amongst each other. This was the case across all six networks, irrespective of clinical characteristics. Especially, the nodes reflecting fatigue severity, depression, and social functioning were highly connected to other nodes, which clinically implying that, in general, patients do not suffer from isolated symptoms, but instead from a very broad range of symptoms, complaints, and problems. These results add to our clinical understanding of the multidimensionality of symptoms and underline the importance of addressing, assessing, and treating symptoms as multidimensional, and not in isolation. In line with this is the finding that global strength did not differ between networks stratified based on disease characteristics, but global network strength was higher in fatigued patients in comparison to non-fatigued patients. This emphasizes that, importantly. Also, patient-reported variables, and not only clinical characteristics, determine which symptoms patients experience together, and to what extent.

These presented networks corroborate studies investigating the prevalence and burden of symptoms in glioma and are in line with studies identifying fatigue to be highly correlated to symptoms such as depression and physical functioning [1, 12, 13]. Comparably, studies on symptom networks in cancer patients have also found fatigue to be a central node in symptom networks [7, 10, 13, 14]. It has been hypothesized that these central symptoms may be suitable targets for therapeutic interventions, with successful treatment of a central symptom resulting in the simultaneous improvement of connected symptoms [41]. However, this theory has been debated, since such an effect would imply direct causality between symptoms and suggests that it should be possible to design an intervention that targets only one specific symptom, without addressing other symptoms [42, 43]. To date, experimental studies intervening on symptoms from a network perspective are lacking. With regard to the current study, it does stand out that fatigue plays an important role in glioma symptom networks. Unfortunately, there are no effective evidence-based treatments targeting fatigue in brain tumor patients, so developing integrative interventions targeting fatigue should be prioritized [44].

A similarly designed study in patients with gastric cancer before and after surgery and a second study in patients with head and neck cancer before and after radiotherapy showed that the global strength of symptom networks did not change over time [10, 45]. Interestingly, the study in patients with head and neck cancer found higher global strength in patients with higher stress levels [45]. Because global strength did not differ in networks based on disease characteristics, but did differ in networks based on fatigue status, the current results imply that symptom network density and the correlation between symptoms are not solely related to the disease itself, but that symptoms and HRQoL are also highly correlated amongst themselves [3]. To comprehend how network global strength relates to actual symptom burden in patients, it would be of value to investigate whether network global strength indeed decreases after successful symptom management in glioma patients.

As we have shown, all symptoms in the presented networks were highly correlated to each other. Applying symptom networks analysis, instead of more traditional statistical methods, is particularly useful when working with multivariate data, since it allows us to move past understanding or predicting single-outcome measures or symptoms [46]. However, it is important to emphasize that the associations between symptoms in the presented networks are correlative, and do not imply causality. An approach to address this gap would be to compute individual networks from high-density longitudinal individual data. By doing so, we can predict how symptoms influence each other over time [47]. Using this exact approach, a study in cancer survivors with depressive symptoms showed high scores on fatigue and worrying to be strongly predicted by their scores at the previous time point, suggesting self-reinforcing loops [48]. Additionally, individual-level symptom networks can be constructed with these data. In clinical practice, these individual networks have been used as part of psycho-education for fatigued cancer patients, with positive responses from patients [49]. Both these applications would be of great interest in glioma patients to improve our understanding of causal mechanisms behind the emergence of symptoms, and to guide psycho-education and symptom management in individual patients.

Several limitations should be taken into account when interpreting the presented symptom networks. Because the studied sample predominantly consisted of lower-grade tumors, we could not study the symptom network of glioblastoma patients separately [26]. Also, we did not take tumor characteristics like IDH status or tumor location into account, which could be interesting because of their link to functional status [50]. Furthermore, we did include the most common symptoms in glioma, like fatigue, subjective cognitive complaints, and depression, but we did not assess symptoms like anxiety, or chemotherapy-related symptoms. For future studies, we aim to assess a broader range of symptoms, which would also require a larger sample size [51]. Also, in the current study, the sample size of some of the subgroups was relatively small compared to other symptom network studies, reflected by the rather large confidence intervals of the edge weights. Another limitation is that some patients were represented in multiple subgroups as they contributed data on multiple occasions, for example preoperatively and postoperatively. Unfortunately, the paired version of the network comparison test is only available when the entire sample is paired, such as in a pre-post design [39]. Consequently, the assumption of independent observations is violated in the subgroup comparisons. To investigate whether this biased our results, we excluded patients that were represented in multiple subgroups and compared GS between the networks of these groups. The results were similar, namely a significant difference in GS between the non-fatigued and fatigued subgroups. Lastly, the chosen regularization technique affects the number of presented edges in the networks. Because of the exploratory nature of the study, we did not apply a very strict regularization method, since this would result in less spurious edges, but could also result in the removal of true edges [4].

In conclusion, symptom network analysis is a novel approach to uncover the complexity of symptom interactions in glioma. Symptom networks might be specifically valuable in guiding symptom management, finding relevant treatment targets, and personalizing treatment. Interestingly, in this study, we showed that symptom networks in glioma did not differ according to disease status and tumor grade, while we did find that PROMs were more tightly intercorrelated in patients suffering from fatigue. This underlines the need for integrative symptom management targeting fatigue. Our findings add to a growing body of literature underlining how symptoms are not only caused by the disease itself, but are also highly correlated amongst themselves.