Cancer‐related fatigue: Towards a more targeted approach based on classification by biomarkers and psychological factors

Cancer‐related fatigue is a frequent, burdensome and often insufficiently treated symptom. A more targeted treatment of fatigue is urgently needed. Therefore, we examined biomarkers and clinical factors to identify fatigue subtypes with potentially different pathophysiologies. The study population comprised disease‐free breast cancer survivors of a German population‐based case‐control study who were re‐assessed on average 6 (FU1, n = 1871) and 11 years (FU2, n = 1295) after diagnosis. At FU1 and FU2, we assessed fatigue with the 20‐item multidimensional Fatigue Assessment Questionnaire and further factors by structured telephone‐interviews. Serum samples collected at FU1 were analyzed for IL‐1ß, IL‐2, IL‐4, IL‐6, IL‐10, TNF‐a, GM‐CSF, IL‐5, VEGF‐A, SAA, CRP, VCAM‐1, ICAM‐1, leptin, adiponectin and resistin. Exploratory cluster analyses among survivors with fatigue at FU1 and no history of depression yielded three clusters (CL1, CL2 and CL3). CL1 (n = 195) on average had high levels of TNF‐α, IL1‐β, IL‐6, resistin, VEGF‐A and GM‐CSF, and showed high BMI and pain levels. Fatigue in CL1 manifested rather in physical dimensions. Contrarily, CL2 (n = 78) was characterized by high leptin level and had highest cognitive fatigue. CL3 (n = 318) did not show any prominent characteristics. Fatigued survivors with a history of depression (n = 214) had significantly higher physical, emotional and cognitive fatigue and showed significantly less amelioration of fatigue from FU1 to FU2 than survivors without depression. In conclusion, from the broad phenotype “cancer‐related fatigue” we were able to delineate subgroups characterized by biomarkers or history of depression. Future investigations may take these subtypes into account, ultimately enabling a better targeted therapy of fatigue.


What's new?
Fatigue associated with cancer frequently persists years after treatment, often with consequences for quality of life.Moreover, fatigue is commonly treated in an undifferentiated manner, owing to a lack of knowledge of possible fatigue subtypes and related markers.Here, the authors investigated the possibility of delineating fatigue subtypes according to biomarkers and psychological factors.Analyses reveal the existence of three fatigue subgroups: one distinguished by history of depression, another by high levels of inflammatory markers, and the third by high leptin levels.Consideration of potential fatigue subtypes could enable the development of targeted and more effective treatment options.

| INTRODUCTION
Approximately one-quarter to one-third of cancer survivors still suffer from fatigue up to 10 years after therapy, which can significantly impact their quality of life (QoL) and daily functioning. 1In contrast to cancer treatment, for which it is state-of-the-art to customize it to molecular genetic characteristics of the tumor and to host factors, cancer-related fatigue is commonly treated in an undifferentiated manner without further evaluation of the specific fatigue symptomatology.However, fatigue is also a heterogeneous disease that needs to be treated according to its unique characteristics.Fatigue manifests in different dimensions, patterns and temporal courses, and its development and persistence depend on various predisposing, precipitating and perpetuating factors. 2,3Predisposing factors increase the risk for developing fatigue and comprise, for example, pre-existing mental health conditions such as depression or trait anxiety.Precipitating factors can trigger fatigue onset and comprise, for example, cancer treatment, especially chemotherapy, as well as metabolic dysregulation or inflammation.Finally, physical, psychological, behavioral and biologic consequences of cancer diagnosis and treatment, for example physical inactivity, obesity, immune or neuroendocrine alterations, can be perpetuating factors contributing to the persistence of fatigue.Thus, each fatigue case may be determined by a variety of factors, and therefore it is quite conceivable that there might be several different subtypes of fatigue.
To date, however, no well-determined (bio)markers of fatigue subtypes have been defined and established in research and clinical practice.Identifying specific subtypes might enable a more adapted and individual fatigue treatment.5][6] A replication of these findings would further strengthen the evidence for considering fatigue occurring with or without a history of depressive symptoms separately.
Further, fatigue has been associated with inflammatory biomarkers in several studies, and inflammation is the most frequently discussed pathway underlying fatigue. 7,80][11] Yet, patterns and pathophysiology of fatigue are likely more complex. 9Clustering patients with fatigue based on multiple biomarkers and other factors simultaneously might be a more suitable approach to categorize fatigue than considering these factors alone.
Therefore, this study aimed to explore among disease-free breast cancer survivors with long-term fatigue if potential subtypes of fatigue may be distinguished by biomarkers and psychological factors.
Fatigue subgroups of survivors with long-term fatigue but no history of depression, identified by cluster analysis, were compared with each other and with survivors having fatigue in conjunction with a history of depression as well as with the group of fatigue-free survivors, regarding a variety of inflammatory and other biomarkers as well as physical and mental factors.

| METHODS
This is an exploratory, secondary analysis of the MARIE-plus study.

| Study population
Patients were eligible for MARIE if aged 50 to 74 years at diagnosis of histologically confirmed breast cancer, had undergone breast surgery, and resided in one of the study regions.Of the 3813 patients enrolled in MARIE, 510 had deceased, three emigrated and 2542 participated in FU1.For the present analyses we considered only the participants who were cancer-free at FU1 (n = 1871).

| Assessments
At FU1 and FU2, participants completed the multi-dimensional 20-item Fatigue Assessment Questionnaire (FAQ) covering physical, emotional, cognitive as well as a total fatigue in the past 4 weeks. 13ditionally, the validated QoL questionnaire of the European Organization for Research and Treatment of Cancer (EORTC QLQ-C30) 14 was completed.It includes a 3-item score on general fatigue in the past week, for which a threshold of clinical importance has been defined based not only on the level but also the perceived burden of fatigue. 15The EORTC QLQ-C30 includes also a 2-item score on pain and scores for other symptoms such as insomnia, dyspnea, appetite loss and financial difficulties as well as five functional scores and a rating of global health/QoL.Body mass index was calculated from selfreported weight and height (kg/m 2 ).Physical activity was calculated as MET-hours per week (MET = metabolic equivalent task 16 ) from self-reported time and intensity spent walking, cycling and exercising.
Concomitant and previous diseases, including depression, were selfreported and assessed per interview.A normally distributed standard curve was observed for all markers except IL-6.Thus, for IL-6 non-detectable values were set to missing, while non-detectable values for the remaining markers were set to half the detection limit due to highly skewed distributions. 17ssing values above fit curve range as well as extreme outliers were set to missing.

| Statistical analyses
Participants were classified as fatigue cases (n = 1004) if the physical, emotional or total FAQ score was above the third quartile of the ageand sex-specific normative value of the general German population 18 according to Singer et al 19 or if the EORTC QLQ-C30 fatigue score was above 39 (on 0-100 scale), that is, its threshold of clinical importance. 15We used this rather broad approach of classifying fatigue cases to cover the heterogeneity and different patterns in which fatigue can manifest.

Based on our previous studies indicating depression as important
defining marker for a potential separate fatigue phenotype, [4][5][6] we considered participants with fatigue who ever had a diagnosis of depression separately as fatigue-depression subgroup (DEPR).
To compare characteristics of CL1, CL2 and CL3 with each other, as well as with DEPR and with survivors free of fatigue and depression (NONE), we generated radar charts (using EXCEL) and scatter plots considering biomarker levels, fatigue dimensions, as well as factors potentially related to fatigue such as pain, insomnia, depression, age, BMI, physical activity, global health rating and other symptoms and functions assessed by the EORTC QLQ-C30.Additionally, we conducted analyses of covariance (ANCOVA) with biomarkers or fatigue dimensions at FU1 as dependent variables and fatigue subgroup (CL1, CL2, CL3 and DEPR) as independent variable adjusted for age.ANCOVA assumptions were checked by the fit diagnostic panels.
As sensitivity analysis, models were adjusted additionally for pain and/or BMI to check if the cluster differences might be partly explained by these factors.
In addition, we investigated the association of subgroup with change in physical, emotional or cognitive fatigue from FU1 to FU2 using ANCOVA models adjusted for FU1 fatigue levels and age.
All tests were 2-sided using 5% significance level and SAS, Version 9.4.

| RESULTS
Among the 1871 disease-free breast cancer survivors (Figure 1), 1004 (53.9%) had indications of fatigue (790 without and 214 with a history of depression) and 819 reported neither fatigue nor depression (NONE).The population characteristics at FU1 are presented in Table 1.Fatigued survivors were on average 67 years (SD: 6) and 5.7 (1.2) years post-diagnosis, similar to non-fatigued survivors.Those with fatigue had a slightly lower educational level than non-fatigued (Chi 2 P < .001).Regarding comorbidities that might contribute to symptoms of fatigue, survivors with fatigue more often reported a history of depression, thyroid disease, inflammatory joint or spinal diseases and more pain (all P < .001).

| Biomarker profiles of the different subgroups
Among the fatigue-non-depression subgroup three clusters were revealed: CL1 (n = 195), CL2 (n = 78) and CL3 (n = 318).Cancer survivors in CL1 on average had high levels of TNF-α, IL1-β, IL-6, resistin, VEGF-A and GM-CSF (Figure 2).In contrast, CL2 was characterized by low cytokines, low CRP, SAA, ICAM-1 and VCAM-1, but high leptin.The largest cluster CL3 did not show any prominent biomarker characteristics.DEPR showed similar average biomarker levels as NONE.ANCOVA with the different biomarkers as dependent variables and adjusted for age showed significantly higher TNF-α, IL1-β, IL-6, resistin, VEGF-A and GM-CSF for CL1 vs CL2 and CL3, and significantly higher leptin levels for CL2 vs CL1 and CL3 (Data S2).Adjusting for other potential influencing factors such as BMI, pain or concomitant inflammatory diseases did not alter the results.

| Fatigue phenotype and fatigue-related factors of the different subgroups
DEPR had highest physical, emotional and cognitive fatigue levels and lowest emotional function among all fatigue subgroups (Figure 3).Among the fatigue subgroups without depression, CL1 showed highest pain levels and highest BMI.Fatigue in CL1 manifested more in physical dimensions such as low physical and role function (Figure 3) as well as high energy loss, reduced physical performance and feeling heavy limbs (Data S3).CL2 revealed higher cognitive fatigue than CL1 and CL3 (Figure 3), which appeared to involve memory, attention as well as concentration problems (Data S3).CL3 had lower pain and dyspnea than CL1 and CL2 and tended to lower fatigue and better role, social and physical functioning (Figure 3).Differences in fatigue phenotypes were also confirmed by ANCOVA (Table 2).DEPR had significantly higher physical, emotional and cognitive fatigue than CL1, CL2 and CL3.Results did not change when additionally adjusting for BMI and pain.Physical fatigue was significantly higher in CL1 than in CL3.This difference remained statistically significant when additionally adjusting for BMI, but vanished when adjusting for pain (P = .36).CL2 had always significantly higher cognitive fatigue than CL1 and CL3 regardless of adjustment by age, BMI or pain.Emotional fatigue did not differ significantly among the three clusters regardless of adjustment.

| Association of subgroup with fatigue change over time
ANCOVA regarding change in fatigue showed that the DEPR subgroup had significantly less decline in emotional fatigue from FU1 to FU2 compared to the fatigue-non-depression subgroups (Table 3).Similar results were seen also for physical and cognitive fatigue, but reaching statistical significance only for comparison between DEPR and CL3.

| Distributions of biomarkers by subgroup
Distributions of some biomarkers in the subgroups are illustrated in scatterplots (Data S4).The plots show the separation of clusters CL1, CL2 and CL3 by the biomarkers, but also demonstrate that the biomarkers do not clearly separate fatigue cases from non-cases.

| DISCUSSION
This analysis among disease-free breast cancer survivors about 6 years after cancer diagnosis aimed to identify markers to discern potential subtypes of fatigue.We could not classify all fatigue cases with our investigated markers, but delineated three potential fatigue subtypes.First, among the broad phenotype 'cancer-related fatigue' we distinguished fatigue that occurred together with a history of depression.This fatigue type manifested in the physical, emotional, as well as cognitive dimension with high severity and was more persistent than fatigue without depression.Second, among survivors with fatigue without depression, we could discern those with high proinflammatory cytokine profiles.This fatigue was associated with higher pain and BMI and manifested rather in physical aspects.Third, we could discern a subgroup with low pro-inflammatory profile but high leptin levels, which was associated with higher cognitive fatigue.
Cluster analyses generally cannot yield a single "true" solution, but can reveal clustering of cases based on given characteristics.
There is wide agreement that the development of cancer-related fatigue can have multiple causes and that its persistence may depend on various factors.In this regard it is not surprising that the largest cluster among the non-depressed fatigue cases (CL3) remained unspecified, because we only had a limited number of biomarkers and patient characteristics, which could not cover the full range of influencing factors in the cluster analysis.For example, we did not consider cortisol profiles, hormonal factors, vitamin D level or other nutritional factors, nor (trait) anxiety, catastrophizing, loneliness or childhood traumata, nor received therapies or applied coping strategies for fatigue. 2 Yet, the identified subgroups (CL1, CL2 and DEPR) and their associated factors can serve as a starting point for the development and investigations of targeted fatigue treatments.
Our results showing that the fatigue-depression group (DEPR) has higher fatigue levels in all three fatigue dimensions compared to fatigued survivors without depression strengthen previous findings in a large sample (n = 1023) of cancer-free survivors of various cancer types. 4In line with the latter study that investigated a cluster defined by depressive symptoms and anxiety, we observed that the fatiguedepression cluster was predictive for higher fatigue levels also after several years.2][23] Thus, it appears reasonable that future studies examine this subgroup and investigate if the efficacy of fatigue treatment differs between cancer survivors with and without depression/anxiety.It would be of relevance for patients suffering from fatigue along with depressive symptoms if, for example, exercise or a psychological intervention would be more beneficial for them.
The second identified fatigue subgroup had high proinflammatory cytokines (TNF-α, IL-1β, IL-6, IL-2, resistin and GM-CSF), high IL-10 and IL-4, which are pleiotropic cytokines, high CRP and high VEGF-A, which is a key factor in vascular permeability and inflammation.Further, in this inflammation-associated cluster, fatigue appeared to manifest stronger in the physical dimension compared to the other depression-free clusters.5][26][27] Inflammation is one of the most discussed mechanisms for cancer-related fatigue but also for fatigue in patients suffering from multiple sclerosis, rheumatoid arthritis, psoriasis and the chronic fatigue syndrome/myalgic encephalomyelitis (CFS/ME). 2,7Cytokines seem to play a crucial role in the so-called sickness behavior, which is an adaption of the body to infections and characterized by fatigue, reduced activity, altered mood state, changes in cognitive functions and reduced appetite. 28Some previous studies suggest that already inflammation at a low-grade state is sufficient to cause sickness symptoms including fatigue. 2,7Inflammation also may trigger decrease in dopamine synthesis, which may contribute to a lack of motivation that is often associated with fatigue. 29However, investigations of inflammation with respect to the different dimensions of fatigue are scarce, and thus more research is needed to understand the underlying mechanisms. 7Research focused on the fatigue subgroup with higher concentration of inflammatory markers might ultimately lead to targeted treatment approaches.In patients suffering from rheumatoid arthritis or psoriasis, for example, pharmacological treatments aimed at blocking the action of cytokines, like TNF-α inhibitors, have shown to reduce fatigue. 30,31Interestingly, among our cancer survivors with persistent fatigue less than a third was in the subgroup with inflammatory profile.Thus, albeit inflammation is an important possible mechanism, many cases of fatigue probably cannot be attributed to it.
The third fatigue subgroup (with low inflammatory markers) was characterized by high leptin levels.Remarkable in this subgroup was its high level of cognitive fatigue.Leptin, an adipose tissuederived hormone, acts on the hypothalamus to regulate food intake and energy expenditure. 32Leptin can cross the blood-brain barrier and may influence cerebrovascular function and neuroinflammation.It has been primarily studied for its role in regulating appetite and body weight, but in recent years evidence emerged supporting the significant influence leptin exerts on cognitive function.Yet, results are heterogenous possibly due to the large spectrum of cognitive impairments. 33,346][37][38] A study in breast cancer patients undergoing chemotherapy, however, observed an inverse association between leptin and fatigue. 39fortunately, one cannot simply diagnose fatigue by high levels of single biomarkers such as inflammatory cytokines or leptin.They can, but do not necessarily, contribute to fatigue, as indicated by our study, where also several survivors without fatigue had high inflammatory or leptin levels.Previous studies suggest that the effects of inflammatory factors in the central nervous system could be modified by psychological factors such as positive and negative expectations. 40wever, further research in biomarker profiles considering a larger range of parameters as well as machine learning techniques may be promising approaches. 35First attempts to develop fatigue prediction models in cancer patients based on clinical or genetic data are published. 41,42However, as fatigue is a subjective experience that might be best estimated by the patients themselves, prediction of general fatigue might be rather less relevant for the patients than the specification of the fatigue subtype with a subsequent targeted treatment.
Limitations of our study need to be considered.It was explorative in nature using a cluster analysis approach.As many different cluster methods exist depending in addition on the considered characteristics (here circulating biomarkers, pain, BMI), this can lead to very different results.Moreover, not all potential factors underlying specific fatigue subtypes were assessed in our study.We however used this method only to reveal some potential subgroups that we subsequently-as the more relevant step-examined for their characteristics and potential predictive value in detail.Further, biomarkers were derived from non-fasting blood samples and without strict consideration of time of day.This may have increased variation of some biomarkers due to diurnal rhythms or dependency on dietary factors.However, these measurement errors are rather random and unlikely to introduce bias but might blur some associations.Finally, information on diagnosis of depression and other diseases was self-reported, which likely is prone to imprecision and reporting error.Yet, even the self-reported information of previous diagnosis of depression was a strong discriminating factor classifying a clearly emerging fatigue phenotype.

| CONCLUSION
Our data provide evidence for delineating a subgroup of fatigue that occurs along with a history of depression, a fatigue subgroup with high inflammatory markers, and another subgroup with high leptin levels, which might be differentiated from further fatigue types.
Cancer-related fatigue thus needs to be considered in a more differentiated way with regard to different pathophysiologies.Future investigations should take these potential subtypes into account, ultimately enabling an effective targeted therapy of fatigue that is adapted to the specific patient and fatigue characteristics.
enrolled 2002 to 2005 in a population-based case-control study in Germany (MARIE, Mamma Carcinoma Risk Factor Investigation 12 ).In 2009 (FU1), fatigue and other outcomes were assessed through standardized, computer-assisted telephone interviews and non-fasting blood samples were collected.Participants were surveyed again in 2014 (FU2).
Characteristics of the study population at FU1 stratified by fatigue.
Mean standardized values of blood biomarkers of disease-free cancer survivors stratified by different subgroups.CL1-CL3, clusters identified among survivors with fatigue but without depression; DEPR, survivors with fatigue and depression; NONE, survivors with neither fatigue nor depression.Mean standardized values of fatigue and other characteristics of disease-free cancer survivors stratified by different subgroups.CL1-CL3, clusters identified among survivors with fatigue but without depression; DEPR, survivors with fatigue and depression; NONE, survivors with neither fatigue nor depression.
F I G U R E 3