Cancer‐related fatigue: Identification of hallmarks to enable refined treatment approaches

Recommendations for fatigue management are commonly given in an undifferentiated manner without further evaluation of patient's specific symptomatology. Thus, we aimed to identify hallmarks of potential fatigue subgroups which might guide more refined treatment.


| BACKGROUND
Cancer-related fatigue is one of the most common and burdensome symptoms during and after cancer treatment. 1 Although its pathophysiology is still not well understood there is wide agreement that fatigue is multi-causal as well as multi-dimensional in nature. 2 Yet, while tailoring therapies individually to patient and cancer characteristics is state-of-the-art for cancer treatment, fatigue treatment is commonly still very undifferentiated without further evaluation of the symptomatology. The existing treatment approaches for fatigue, which vary widely such as physical exercise interventions or cognitive-behavioral therapies, are usually recommended to patients in a nonspecific manner. 3,4 To date, no well-determined hallmarks of potential subtypes of fatigue have been defined and established in research and cancer care. Singling out specific subtypes might enable a more adapted and individual patient treatment. 5 Although associations of fatigue with some biological markers, such as pro-inflammatory cytokines, 6,7 cortisol dysregulations, 6,8 reduced energy metabolism or neuroendocrine changes 9 have been shown, research seems still far from developing laboratory diagnostics to classify specific fatigue types. Thus, to advance better targeted treatment approaches it may be more promising to consider previously identified determinants of fatigue including psychological, physiological, socio-environmental, and cognitive-behavioral factors. 6,10,11 Besides sex and age as components influencing the experience of fatigue, depressive symptoms, anxiety, insomnia, and pain have frequently been observed to cluster or be associated with fatigue. 10,12,13 These symptoms might share common etiologies with fatigue, but could also be predisposing or perpetuating factors. 2 For example, dysfunctional cognitive appraisals or illness cognitions might be components to contribute to both, the development of emotional distress (depression/anxiety) and fatigue. 12 Depressive disorders are also a risk factor for poor psychological adjustment to the stress caused by cancer diagnosis and treatment, and thus may increase vulnerability to fatigue. Longitudinal studies have shown that depressed mood predicts elevated fatigue levels during and after treatment. 14,15 Moreover, depressive symptoms and anxiety can increase inflammatory response to stress and hereby might increase the risk for fatigue or influence its severity and persistence over time.
Besides, insomnia can have multiple causes, manifestations and might share common etiologies with fatigue and can also sometimes be a consequence of fatigue. 16 The association of pain with fatigue might be just as complex. Causes of pain and insomnia might include inflammatory processes, sleep problems, stress as well as inactivity and subsequent physical deconditioning. 16 Further determinants of fatigue comprise obesity, lack of physical activity as well as psychosocial risk factors like childhood adversity or loneliness. 2,17 Obesity is thought to contribute indirectly to fatigue due to inflammatory processes in fatty tissue which weaken the immune system. 10 However, it is to-date unclear which of the above described associated factors might be the most useful hallmark(s) to break down fatigue cases into subgroups that have distinct phenotypes and that might benefit from different treatments. Therefore, we performed a hierarchical cluster analysis to determine which of the most common correlates of fatigue, that is, emotional distress (depressive symptoms and anxiety), pain, insomnia, and obesity, may be the most relevant hallmark of a potential fatigue subtype. Further, we explored how this subtype describes current and can predict future fatigue patterns.

| Study population
The FiX study enrolled 2508 patients about 2 years after diagnosis of cancer (T0) including 15 different cancer entities. Patients were randomly sampled from the Epidemiological Cancer Registry of Baden-Württemberg, Germany, and have been described in detail previously. 18 After 2 years, a follow-up survey (T1) was conducted. In the cluster analysis all patients were included with prevalent fatigue at T0 (n = 1023), that is, with a total fatigue score above the age and sex specific 75% percentile of the general population in Germany. 19 Of the 1023 survivors with fatigue at T0, 128 were deceased before T1, for 55 contact addresses were no longer valid, and 129 did not respond, resulting in 711 participants at T1 of whom 467 were apparently cancer-free survivors. For the detailed number on patient flow from T0 to T1 by cluster see Supplement 1. All patients have given written informed consent.

| Assessments
Fatigue at T0 and T1 was assessed multidimensionally with the 12item EORTC QLQ-FA12, which addresses the physical, emotional and cognitive dimension of fatigue and also provides a total fatigue score. It was shown that it can detect clinically significant fatigue in cancer patients. 20 Additionally, to gain insights into change of patients' fatigue burden during the cancer continuum, patients were asked to retrospectively rate their pre-diagnostic fatigue level and their highest fatigue level since diagnosis in addition to their current fatigue level on a 0-10 scale based on items of the Brief Fatigue Inventory. 21 Emotional distress was assessed with the PHQ-4 consisting of two items measuring depressive symptoms and two items measuring anxiety symptoms (each scale 0-3). 22  According to the World Health Organization, a BMI ≥30 kg/m 2 was considered as obesity.

| Statistical methods
A hierarchical cluster analysis was performed based on T0 data (n = 1023) using SAS PROC CLUSTER with the Ward method and included the dichotomized factors emotional distress, insomnia, pain, and obesity. The number of clusters was based on the pseudo-Fstatistic. Clusters were subsequently calculated with PROC TREE.
Multicollinearity was checked in advance. Fatigue and patient characteristics were compared across the identified fatigue clusters using Kruskal-Wallis tests.
Analysis of covariance (ANCOVA) was conducted to investigate the association of the baseline fatigue clusters with subsequent fatigue at T1 adjusting for baseline fatigue levels, sex, age, and cancer entity. Hereby, separate ANCOVA models were calculated for physical, emotional, and cognitive fatigue. As cancer progression may have substantial impact on fatigue, this analysis was additionally conducted within the subgroup including only survivors who were apparently cancer-free at T1. Further explorative analyses using ANCOVA models investigated the association of the fatigue clusters at T0 with different statements concerning psychosocial factors and attitudes towards fatigue.

| RESULTS
The characteristics of the 1023 fatigue cases are summarized in Table 1

| Clinical implications
The Note: All ANCOVA models were adjusted for baseline fatigue, sex, age, and cancer entity. Only cancer survivors who were apparently cancer-free at T1 are included to avoid confounding of results by cancer progression. nobs=number of non-missing observations. *p < 0.05, **p < 0.01, ***p < 0.001: least square mean significantly different to CL1. Our results underline the importance of these symptoms in relation to fatigue and show that they contribute differently to the experienced fatigue burden. Obesity and pain may be further distinguishing markers for the treatment of fatigue. Further research into these hallmarks of potential fatigue subtypes and correspondingly tailored treatments is warranted.