Introduction

Palliative chemotherapy is administered with non-curative intent. The purpose of palliative chemotherapy is to optimize symptom control. Emphasizing palliation rather than toxicity from chemotherapy may improve quality of life (QOL) and, ideally, survival [1, 2]. Symptom management in patients undergoing palliative chemotherapy is especially challenging, as palliative chemotherapy patients often suffer from cancer symptoms as well as symptoms related to chemotherapy. Cancer patients’ concurrent symptoms and the desire for efficient and effective symptom management have driven researchers to identify symptom clusters or groups of interrelated symptoms. A symptom cluster is defined as a stable group of two or more interrelated symptoms occurring simultaneously [3]. The negative impact of symptom clusters has been demonstrated, even affecting survival [4]. Identification of symptom clusters and determination of effective symptom management is crucial for patients receiving palliative chemotherapy.

A systematic review of symptom cluster studies in patients with advanced cancer identified four common symptom clusters: anxiety-depression, nausea-vomiting, nausea-appetite loss, and fatigue-dyspnea-drowsiness-pain [5]. Whether the identified symptom clusters represent the symptom experience of patients receiving palliative chemotherapy is of question. Studies of symptom clusters have often identified symptom clusters among patients receiving various types of palliative care [5]. However, symptoms during the palliative chemotherapy are considered different from those at end-of-life. Symptoms such as dyspnea and tiredness have been found to increase in severity as patients approach end-of-life, whereas symptoms such as pain, nausea, anxiety, and depression have been found to remain relatively stable over time [6]. Studies including patients undergoing palliative radiation therapy alone [7] or in combination with various palliative treatments including chemotherapy [8, 9] have been conducted previously to identify symptom clusters among patients receiving palliative care; however, studies specific to the patients undergoing palliative chemotherapy are lacking. The symptom experiences of patients undergoing palliative chemotherapy may be unique, as symptoms could arise from both advanced cancer and chemotherapy toxicity, and patients may still be far from end-of-life.

Furthermore, symptom cluster studies have often included limited numbers of symptoms. According to a systematic review by Dong et al. [5], the most commonly utilized instrument for measuring multiple symptoms was the Edmonton Symptom Assessment Scale (ESAS), which includes nine symptoms. Limited numbers of symptoms have been included in other symptom scales such as the MD Anderson Symptom Inventory (MDASI; 13 items) and European Organization for Research and Treatment of Cancer Quality of Life-C30 (EORTC QLQ-C30; 10 symptom subscales when emotional and cognitive function subscales are considered as symptom scales). As the numbers of symptoms included in symptom analyses have been limited, comprehensive evaluation of symptom clusters has also been limited. A limited number of studies exist that have utilized scales including a large number of symptoms, such as the Memorial Symptom Assessment Scale (MSAS), which includes 34 symptoms [7]. Identification of symptom clusters among multiple symptoms has also been conducted utilizing author-developed scales and including up to 38 symptoms [10]. However, in the aforementioned studies, 20 symptoms with 20% prevalence [7] and 25 symptoms with >15% prevalence were utilized for cluster analysis [10]. Symptom assessment needs to include comprehensive lists of symptoms, but not burden patients with serial symptom monitoring.

Supportive care for advanced cancer patients undergoing palliative chemotherapy is important, and efficient symptom management needs to be pursued. Understanding symptom experiences, identifying symptom clusters, and understanding the relationships between symptom clusters and functioning and QOL in patients may help guide effective symptom management.

Purpose

The purpose of this study was to identify symptom clusters during palliative chemotherapy and examine the relationship between symptom clusters and functioning and QOL in patients.

Method

Design

A descriptive correlational study was conducted.

Sample

In total, 318 cancer patients were recruited, and 300 cancer patients undergoing palliative chemotherapy participated in the study. Inclusion criteria were adult cancer patients undergoing palliative chemotherapy for lung, breast, colorectal, or gastric cancer treatment. While most patients underwent outpatient chemotherapy, a few underwent inpatient chemotherapy. Exclusion criteria comprised receiving concomitant radiation therapy, as this could influence patients’ symptom experiences. Patients having any cognitive or psychiatric disorders that could hinder symptom assessment were also excluded. Comrey and Lee recommended sample sizes of ≥200 as fair and ≥300 as good for factor analysis [11]. For structural equation modeling (SEM), sample size must also be more than 10 times the number of estimated parameters [12]. Because the current study was designed to include a total of 20 symptoms, three functional scales and one QOL measure, for a total of 24 parameters, at least 240 subjects were required. No imputation was performed for missing values; thus, list-wise analyses were conducted with data from a total of 266 patients.

Measurement

In an attempt to capture relevant symptom experiences but not overburden patients with questions, validated instruments for symptom measurements (MSAS, MDASI, ESAS, CTCAE, FACIT, BSI) and symptom studies were reviewed [13–21], and 20 major symptoms were identified. The identified symptom list was first evaluated among patients undergoing their first two cycles of adjuvant chemotherapy [22], and all 20 symptoms (measured in 0~10 numeric rating scale) demonstrated an incidence higher than 15% (result not previously reported). The 20 symptoms included 12 core symptoms identified by Reeves et al. [20]. A structured questionnaire was utilized to identify patients’ demographic characteristics as well as clinical information regarding cancer diagnosis, stage, Eastern Cooperative Oncology Group (ECOG) status, chemotherapy regimen, and concomitant disease. Participants were approached on the day of chemotherapy and asked to rate their symptom experience during previous chemotherapy cycle.

The EORTC QLQ-C30, a validated instrument measuring cancer-related quality of life, was used to measure patients’ symptoms, functioning and QOL. The Cronbach’s α for the Korean version of the EORTC QLQ-C30 has been reported as >.70 for all subscales except cognitive functioning (.60) [23]. In the current study, subscale reliabilities, except cognitive functioning (α = .49), were >.70. The EORTC QLQ-C30 has no specified recall period for physical functioning items, whereas all the other items require recall within the previous week. The EORTC QLQ-C30 emotional and cognitive functional subscales were regarded as symptom subscales, as they were in a study by Dong et al. [8].

Procedure

The Institutional Review Board of the university hospital approved the study (IRB approval number 4-2014-0700). Research nurses approached patients and explained the purpose and details of the study. Patients who agreed to participate provided written informed consent. Research nurses confirmed inclusion and exclusion criteria. Patients were asked to evaluate the severity of the listed 20 symptoms during previous chemotherapy. Subsequently, patients were asked to evaluate their functional status, and QOL by answering the EORTC QLQ-C30 questionnaires using the previous week as the timeframe except for physical functioning.

Analysis

Statistical software IBM SPSS 22.0, SAS 9.4, and IBM SPSS AMOS 22 were utilized to analyze the data. Descriptive statistics summarized the prevalence and severity of the 20 symptom and EORTC QLQ-C30 subscales. Factor analysis (principal components analysis, promax rotation with Kaiser normalization) was conducted. Factors were identified by an eigenvalue >1. The Kaiser-Meyer-Olkin (KMO) measure and Bartlett’s test of sphericity were used to examine sampling adequacy and appropriateness of the factor analysis. A Cronbach’s α greater than or close to .70 was considered as the criterion for an acceptable symptom cluster. Item exclusion was determined by changes in Cronbach’s α after item deletion. Hierarchical cluster analysis (principal components analysis, oblique rotation) was conducted to confirm identified symptom clusters using a different statistical method. A small 1-R 2 ratio indicated good clustering, which means the items in the cluster have a higher squared correlation score within the cluster when compared to their squared correlation with the next closest cluster. The criteria for item deletion from the identified cluster was R 2 < .40. Structural equation modeling (SEM) with maximum likelihood method was performed to identify relationships between symptom clusters utilizing cluster mean scores [8], functioning and QOL. Multiple criteria were utilized to evaluate the model fit: chi-square test and cutoff values of <.06 for root mean square error of approximation (RMSEA) and .90 for comparative-fit index (CFI) [24]. Optimal model specification was determined by comparing Akaike Information Criterion (AIC).

Results

General characteristics

All participants were undergoing palliative chemotherapy. The median chemotherapy cycle (the number of chemotherapy cycles patients received so far) was 3 (range 1~131). One quarter of participants were undergoing their first cycle of palliative chemotherapy, and 80% of participants had undergone less than 10 chemotherapy cycles. The average age of participants was 54 years (SD = 10.21), and slightly more than half of participants were male (52.7%). Cancer diagnoses were evenly distributed between four types of cancer. Various chemotherapy regimens were reported, and frequently prescribed chemotherapy regimens included cetuximab/FOLlinic acid (leucovorin calcium) + Fluorouracil + IRInotecan hydrochloride (FOLFIRI) (8.7%), FOLlinic acid (leucovorin calcium) + Fluorouracil + OXaliplatin (FOLFOX) (8.7%), FOLFIRI (6.7%), trastuzumab monotherapy (6.7%), bevacizumab/FOLFIRI (5.7%), docetaxel monotherapy (5.0%), XELoda (capecitabine) + OXaliplatin (XELOX) (4.3%), S-1/cisplatin (4.0%), paclitaxel monotherapy (3.7%), trastuzumab/capecitabin/cisplatin (3.7%), and pemetrexed monotherapy (3.7%). Participants’ functional status was rated either ECOG 0 or 1 in most cases (96.6%). Approximately 40% of participants had concomitant disease (Table 1).

Table 1 General characteristics (n = 300)
Full size table

Symptom prevalence and severity during previous chemotherapy cycle

Fatigue was the most prevalent symptom, followed by taste change, loss of appetite, neuropathy, drowsiness, and nausea (>70% prevalence). Sexual problems, diarrhea, vomiting, and urinary problem occurred in less than 50% of participants, but the lowest symptom prevalence was 37.5%.

Fatigue and taste change were ranked as the most severe symptom, followed by loss of appetite, neuropathy, alopecia, skin and nail changes, sleep disturbance, nausea and drowsiness (mean severity rating > 3). Vomiting, diarrhea, and urinary problems were regarded as the least severe symptoms, although patients’ responses ranged from 0 to 10 for all 20 symptoms (Table 2).

Table 2 Symptom prevalence and severity ratings (n = 300)
Full size table

Symptom, functioning, QOL of the past week

Patients reported moderate levels of QOL. Physical functioning was affected the most, followed by role functioning and social functioning. The EORTC QLQ-C30 symptom rankings were similar to the ratings of the 20 symptoms. Fatigue and loss of appetite were rated as the most problematic symptoms (Table 3).

Table 3 EORTC QLQ-C30 score for symptom, functioning and QOL (n = 300)
Full size table

Symptom clusters

Factor analysis of the 20 symptoms identified five factors which explained 62.94% of the total variance: emotional; nausea and vomiting/appetite/taste change, fatigue/cognitive, skin/mucosa; and other (dyspnea, pain, neuropathy, constipation, sleep disturbance). The emotional cluster and nausea and vomiting/appetite/taste change cluster had Cronbach’s α > .80. The fatigue/cognitive and other clusters had Cronbach’s α close to .70. The skin/mucosa cluster consisted of clinically relevant symptoms; however, it was a poorly specified cluster with a Cronbach’s α < .60 and thus was not considered as a symptom cluster and excluded from further analysis using SEM. It is of note that urinary problems, sexual problems, and diarrhea failed to be included in any cluster (determined by changes in Cronbach’s α with item deletion). In the hierarchical cluster analysis, identical symptom clusters were identified, although the explained variance was decreased to 53.23 %. The three symptoms excluded in the factor analysis were also disqualified according to the low R 2 values (<.40) (Table 4).

Table 4 Symptom clusters identified through factor analysis and hierarchical cluster analysis (n = 266)
Full size table

The influence of symptom clusters on functioning and QOL

The SEM included four symptom clusters identified by factor and hierarchical cluster analysis and functioning and QOL subscales from the EORTC QLQ-C30. The SEM including symptom clusters as latent variables failed to meet the criteria for acceptable model fit. Thus, path analysis was conducted utilizing symptom mean scores to represent symptom clusters [8].

The final model demonstrated good fit between the hypothesized model and the data. The absolute fit indices were X 2 = 11.19, df (7) (p = .13) and RMSEA = .05. Incremental fit was CFI = .99, and parsimonious fit was demonstrated by AIC = 85.19. The final model indicated direct negative effects of the emotional cluster on role (ß = −.19, p < .001) and social (ß = −.19, p = .005) functioning. The nausea and vomiting/appetite loss/taste change cluster had a direct negative effect on role functioning (ß = −.12, p = .020). The fatigue/cognitive (ß = −.19, p = .003) and other symptom (ß = −.24, p < .001) clusters had direct negative effect on physical functioning, which demonstrated direct positive impacts on role (ß = .58, p < .001) and social (ß = .22, p < .001) functioning. Among the symptom clusters, only the other symptom cluster had a direct negative influence on QOL (ß = −.18, p = .003). Among the functioning subscales, only role functioning contributed to the QOL of patients (ß = .35, p < .001). The emotional (ß = −.08, p = .002) and nausea and vomiting/appetite/taste change (ß = −.05, p = .011) clusters indirectly affected QOL through role functioning. The fatigue/cognitive (ß = −.04, p = .002) and other symptom (ß = −.05, p = .003) clusters had indirect effects on social functioning through physical functioning. The fatigue/cognitive (ß = −.11, p = .006) and other symptom (ß = −.14, p = .006) clusters had indirect effects on role functioning through physical functioning. Physical functioning had an indirect effect on QOL through role functioning (ß = .22, p = .004). The fatigue/cognitive cluster had an indirect effect on QOL through physical functioning and role functioning (ß = −.05, p = .003). Symptom clusters and physical functioning accounted for a total of 44.1% of the variance in role functioning. The final model accounted for a total of 23.3% of the variance in QOL (Fig. 1).

Fig. 1
figure 1

Structural Equation Model for Symptom Clusters , Functioning and QOL. Emotional emotional cluster,  NVAT nausea and vomiting/appetite loss/taste change cluster, FatigueCog fatigue/cognitive cluster, Other other cluster, Physical physical functioning, Role role functioning, Social social functioning, QOL quality of life, d1 disturbance (combined effect of all other factors influencing the variable) of physical functioning, d2 disturbance of role functioning, d3 disturbance of social functioning, d4 disturbance of QOL. Non-significant relationships are illustrated with dashed lines

Full size image

Discussion

Symptoms experienced by patients receiving palliative chemotherapy

Among the 20 symptoms evaluated, the lowest prevalence was 37.5%, which suggests significant proportions of patients were experiencing symptoms. More importantly, the severity ratings of all 20 symptoms ranged from 0 to 10. Low prevalence does not necessarily mean symptom experience is less severe. Monitoring all 20 symptoms during palliative chemotherapy is strongly recommended. Multiple aspects of symptoms, such as prevalence, severity, and trajectory, need to be considered in symptom management.

Symptom clusters

The current study evaluated symptom clusters using single items representing each symptom’s severity, enabling each symptom to have a non-weighted contribution to the symptom clusters. It is of note that there were symptoms that formed a stable symptom cluster, some of which require further study regarding their cluster membership. Some of the symptoms failed to be included in a cluster.

The emotional cluster, which comprised anxiety and depression, was a robust symptom cluster, although the prevalence and severity of each symptom were not highly ranked. A systematic review of observational studies [5] and longitudinal studies of symptom clusters identified anxiety-depression as a consistent symptom cluster [25, 26]. Comparisons of different analytic techniques have also identified anxiety and depression as constantly clustering symptoms [27]. In the study by Dong et al. [8] using the EORTC QLQ-C30, the tense, worried, irritable, and depressed items formed a robust emotional cluster across five cancer sites. Assessment of symptoms within the emotional cluster and comprehensive interventional approaches targeting these symptoms [28] may enable efficient symptom management.

Studies supported nausea and vomiting [5, 8, 27], nausea and appetite loss [14], as well as nausea and vomiting and appetite loss [8, 29] as symptom clusters, and the current study demonstrated that taste change was clustered with nausea, vomiting and appetite loss. Studies of chemotherapy-related symptoms support the inclusion of taste change in a symptom cluster with nausea [30–32], although the treatments evaluated were not confined to palliative chemotherapy. In comparison, in the Atkas et al. study [27], taste change was included in the fatigue/anorexia-cachexia cluster among patients not undergoing any tumor treatment. Patients’ cancer trajectory status may contribute to the symptom cluster membership of taste change.

Fatigue was clustered with cognitive symptoms, such as difficulty concentrating and drowsiness. Clustering of fatigue/cognitive symptoms has been supported by other studies [5, 33]; however, the stability of the fatigue/cognitive cluster needs further study, as fatigue has also been found to cluster with pain [8, 34], insomnia [35], as well as pain and insomnia [25, 33]. In a systematic review of symptom clusters, fatigue-dyspnea-drowsiness-pain was identified as one of the most common clusters identified in patients with advanced cancer [5]. The observed clustering of fatigue with cognitive symptoms among advanced cancer patients undergoing palliative chemotherapy requires further study.

The other symptom cluster requires special attention. Although it was not easily named, the other cluster, which comprised dyspnea, pain, constipation, neuropathy, and sleep disturbance, demonstrated the most influential direct negative influence on physical functioning and a had direct negative influence on QOL. Clustering of symptoms in the other cluster is not a new finding, as similar cluster membership was demonstrated among older palliative cancer patients with a few differences (taste change, cough, and airway mucus were also included) [36]. Although its membership has not yet been established, the existence of the other cluster and its contribution to functioning and QOL require further attention.

It is also of note that some symptoms failed to be included in the symptom clusters. It is important to acknowledge that symptom clusters may facilitate efficient symptom management; however, some symptoms may require individual attention. Utilization of comprehensive symptom assessment is further supported.

Influence of symptom clusters on functioning and QOL

The study is meaningful, as it explored relationships between symptom clusters and functioning and QOL. Identifying and targeting symptom clusters may be meaningful, as studies have demonstrated symptom clusters, rather than the individual symptoms of the clusters, have significantly higher impacts on functioning [32] and QOL [29, 32].

Discrepancies existed regarding the influence of symptom clusters on functioning and QOL. In the study by Dong et al. [8], the fatigue/pain and emotional clusters were strong predictors of QOL. An indirect effect between symptom clusters and QOL was mediated by social functioning. The observed differences in the relationships between symptom clusters and functioning and QOL might due to the use of different instruments and analytic techniques. The EORTC QLQ-C30 includes multiple items for fatigue, pain, emotional, and cognitive symptoms. Further, characteristics of patients undergoing only palliative chemotherapy might contribute to these different relationships.

Changes in functioning and QOL may contribute to symptom experience; however, limited variance in QOL has been found to be explained by symptom clusters [29]. Factors other than symptoms such as age, gender, and performance status need to be considered as contributors to QOL [37]. Furthermore, the current analysis included only identified symptom clusters. Less well-specified symptom clusters (skin/mucosa cluster) and symptoms failing to be included in symptom clusters (sexual problems, urinary problems, and diarrhea) were excluded from the SEM. The full spectrum of the impact of symptom experience on QOL might not have been captured. Data were collected using a cross-sectional survey. The sample was heterogeneous including various cancer diagnosis and chemotherapy regimens. Empirical 20 symptom lists were utilized for identification of symptom clusters, which might need further validation.

Conclusion

Among the 20 symptoms demonstrating at least 35% prevalence, emotional; nausea and vomiting/appetite loss/taste change, fatigue/cognitive; and other clusters were identified as symptom clusters among patients receiving palliative chemotherapy. Differential contributions of symptom clusters to functioning were identified that eventually contributed to QOL through role functioning. The identified symptom clusters and their relationships with functioning and QOL may help guide approaches to symptom management. Comprehensive symptom assessment and implementation of interventions targeting symptom clusters may help to facilitate effective symptom management and contribute to improving functioning and QOL in patients.