Jordanka Kirkova

Cancer Symptom Clusters: Clinical and Research Methodology

INTRODUCTION Patients with cancer experience multiple symptoms that frequently appear in groups or clusters. We conducted a comprehensive clinical review of cancer symptom cluster studies to identify common symptom clusters (SC), explore their clinical relevance, and examine their research importance. METHODS Published studies and review articles on cancer SC were obtained through a literature search. We identified 65 reports. These varied in assessment instruments, outcomes, design, population characteristics, and study methods. RESULTS Two main approaches to symptom cluster identification were found: clinical and statistical. Clinically determined SC were based upon observations of symptom co-occurrence, associations, or interrelations. These included fatigue-pain, fatigue-insomnia, fatigue-insomnia-pain, depression-fatigue, and depression-pain. They were analyzed by multivariate analysis. They had low to moderate statistical correlations. Disease- or treatment-related SC were influenced by primary cancer site, disease stage, or antitumor treatment. SC determined by statistical analysis were identified by factor and cluster analysis through nonrandom symptom distribution. Nausea-vomiting, anxiety-depression, fatigue-drowsiness, and pain-constipation consistently clustered by either or both of these statistical methods. The individual symptoms of pain, insomnia, and fatigue often appeared in different clusters. A consensus about standard criteria and methodological techniques for cluster analysis should be established. CONCLUSIONS Several important cancer SC have been identified. Nausea-vomiting, anxiety-depression, and dyspnea-cough clusters were consistently reported. The techniques of symptom cluster identification remain a research tool, but one with considerable potential clinical importance. Further research should validate our analytical techniques, and expand our knowledge about SC and their clinical importance.

Cancer symptom clusters—a dynamic construct

Cancer symptoms are influenced by disease progression, treatment, time, patient personality, and demographic factors: age, gender, performance status [1, 2]. Symptom experience may be genetically predetermined [3]. Understanding cancer symptoms as dynamic constructs helps us understand the multifaceted character of the disease process and hopefully improve patient care by improving symptom assessment and management. Cancer patients have multiple symptoms, which often appear in groups or clusters [2, 4]. Symptom research has identified a separate symptom entity, i.e., “a symptom cluster” [4–6]. Symptom clusters consist of co-occurrence of two or more symptoms or symptom correlation based upon a particular symptom domain (prevalence, severity, distress) [5, 7]. The relationship among the symptoms within the cluster is assumed to be stronger than that across different clusters [5, 7]. Symptom clusters may influence disease course, treatment, function, quality of life (QoL), or prognosis, which differs from the sum of individual symptoms [7]. Methodological issues in cluster research include lack of consensus on cluster definition, instruments, study design, and statistical analyses used to identify clusters [7–9]. Further studies in cancer like the one by Chow and colleagues (in this issue) contribute to cluster research and our knowledge of its clinical meaning [4, 5, 10–12]. There are two main approaches to identifying symptom clusters. Firstly, clusters may be clinically predefined, and the association between symptoms within the cluster determined statistically [13]. Predefined clusters use a limited number of targeted symptoms (≤5), for which previous research has suggested clinical associations. Examples of such clusters are [pain, fatigue, and insomnia], or [pain, depression, and fatigue], which have been addressed in multiple studies [5, 13–15]. Clinically predefined clusters find a complex interrelation between symptoms but cannot validate clusters containing more than two symptoms, as they explore symptom interrelations between two symptoms, or correlate a single symptom with function, performance status, and comorbidities. Low to moderate statistical correlations (r<0.70) are accepted as a cluster, which is unconvincing as evidence. It is possible that significant correlations exist between symptoms not included in predefined clusters, or will differ based on symptoms used as a dependent variable in the analysis. Secondly, clusters may be determined by statistical analysis of large symptom datasets without predetermined symptoms [4, 12, 16]. These clusters are usually derived using either factor or hierarchical cluster analysis [9]. Different symptom domains such as prevalence, severity, frequency, or distress can be included in the analysis. Statistically obtained clusters do not presuppose clinical associations, but they may suggest a common pathophysiology between symptoms within the cluster with clinical utility. Cluster or factor analysis defines nonrandom distribution of symptoms, but neither hierarchical cluster nor factor analysis considers assessment bias; both are sensitive to outliers [4]. The instruments, statistical methods and cut-off correlations used to determine a cluster vary between studies, which influences cluster composition and makes comparisons among studies difficult [4, 12, 17]. Patients have also been statistically clustered into groups based upon symptoms [8]. However, this differs significantly from clustering symptoms Support Care Cancer (2007) 15:1011–1013 DOI 10.1007/s00520-007-0259-2

Cancer symptom clusters: old concept but new data.

Individuals with cancer have multiple symptoms, which frequently co-occur. A nonrandom distribution of symptoms suggests a common mechanism. Symptom clusters (SCs) were considered part of various syndromes in the early years of medicine. The SC concept in clinical medicine is old. Symptom clusters were commonly described in the psychology/psychiatry and neurology literature. Symptom cluster may be defined either clinically or statistically. Statistically derived clusters can differ from clinically defined clusters. The clinical importance of statistically derived clusters is unclear. Pain-insomnia-fatigue and pain-depression-fatigue are commonly recognized clinical clusters. Nausea-vomiting and anxiety-depression are also statistically observed clusters. The longitudinal stability of clusters is unknown. Certain SCs, appear to have a greater adverse influence on outcomes (such as performance status and survival) than others. Comorbidities probably influence symptoms at different levels, but their effect on cancer clusters is unknown. Comprehensive symptom assessment is crucial to cluster identification. The potential use of the cluster concept to abbreviate symptom assessment tools needs validation. Symptom cluster can be disease and/or treatment related and may change as individuals undergo antitumor therapies. Polypharmacy in symptom management is frequent but could be minimized if 1 drug could be used to treat cluster symptoms. Symptom cluster appears to vary with the assessment tool, disease stage, symptom domain used to cluster, cluster methodology, and number of symptoms assessed. The validity and reliability of SCs need universally accepted statistical methods, assessment tools, and symptom domains. For now, nausea-vomiting is recognized as a consistent cluster across multiple studies. Pain-depression-fatigue and pain-insomnia-fatigue are also well recognized. Symptom clusters may help in cancer diagnosis, symptom management, and prognostication. However, the cluster method, reliability, and validity need to be established before assessment or treatment guidelines are established. Symptom clusters require further research before becoming part of routine medical symptom assessment and management.

Cancer-related fatigue: central or peripheral?

To evaluate cancer-related fatigue (CRF) by objective measurements to determine if CRF is a more centrally or peripherally mediated disorder, cancer patients and matched noncancer controls completed a Brief Fatigue Inventory (BFI) and underwent neuromuscular testing. Cancer patients had fatigue measured by the BFI, were off chemotherapy and radiation (for more than four weeks), had a hemoglobin level higher than 10 g/dL, and were neither receiving antidepressants nor were depressed on a screening question. The controls were screened for depression and matched by age, gender, and body mass index. Neuromuscular testing involved a sustained submaximal elbow flexion contraction (SC) at 30% maximal level (30% maximum elbow flexion force). Endurance time (ET) was measured from the beginning of the SC to the time when participants could not maintain the SC. Evoked twitch force (TF), a measure of muscle fatigue, and compound action potential (M-wave), an assessment of neuromuscular-junction transmission were performed during the SC. Compared with controls, the CRF group had a higher BFI score (P<0.001), a shorter ET (P<0.001), and a greater TF with the SC (CRF>controls, P<0.05). This indicated less muscle fatigue. There was a greater TF (P<0.05) at the end of the SC, indicating greater central fatigue, in the CRF group, which failed to recruit muscle (to continue the SC), as well as the controls. M-Wave amplitude was lower in the CRF group than in the controls (P<0.01), indicating impaired neuromuscular junction conduction with CRF unrelated to central fatigue (M-wave amplitude did not change with SC). These data demonstrate that CRF patients exhibited greater central fatigue, indicated by shorter ET and less voluntary muscle recruitment during an SC relative to controls.

Symptom severity and distress in advanced cancer

We determined the relationship between symptom severity and distress for multiple cancer symptoms, and examined patient demographic influences on severity and distress in advanced cancer. A Cochran—Armitage trend test determined whether symptom distress increased with severity. Chi-square, Fisher’s exact test and logistic regression analysis examined moderate/severe (‘clinically important’) and distressful symptoms by age (≤65 versus >65), gender, primary site group, and ECOG performance status. Forty-six symptoms were analyzed in 181 individuals. More than 50% of individuals with clinically important symptoms rated them as distressful. The median percentage of individuals with mild but still distressful symptoms was 25%, with a range of 0% (bad dreams) to 73% (sore mouth). In both univariate and multivariate analysis, younger (≤65 years) patients, females, and those with poor performance status had more clinically important and a higher prevalence of distressful symptoms (only anxiety was more frequently distressful to older individuals). Clinically important symptoms and two of those considered distressful varied by primary site group. After control for severity, symptom distress did not differ by primary site group. The prevalence of distress increased with greater symptom severity. Younger individuals, those with poor performance status, and females had greater symptom severity and distress. Mild symptoms were often distressful. After adjustment for severity, age, gender, and performance status all influenced symptom distress.

Symptom prevalence in advanced cancer: age, gender, and performance status interactions.

Age, gender, and performance status (PS) are important patient characteristics which might influence to cancer symptom profile. We conducted a secondary analysis of a symptom database to examine any interaction of these factors on symptom prevalence. 38 symptoms were assessed in 1000 consecutive patients with advanced cancer. The association of the three demographic factors with each symptom was examined using logistic regression analysis. Eight symptoms were associated with more than one of the three factors. Model-based estimates of symptom prevalence were calculated for 30 groups based on combinations of age, gender, and ECOG PS (0-4). Prevalence differences between various groups >10% were empirically classified as clinically relevant. The frequency of all eight symptoms (pain, constipation, sleep problems, nausea, anxiety, vomiting, sedation, and blackouts) was associated with more than one of the demographic characteristics of age, gender, and PS level. The prevalence of all eight decreased with older age. Females had more nausea, anxiety, and vomiting than males; males greater sleep problems. The prevalence of constipation, sedation, and blackouts was higher with worse PS, whereas pain and anxiety became less common with worse PS. Age, gender, and PS appeared to be associated with variations in the prevalence of eight gastrointestinal and neuropsychological symptoms in cancer patients. They should be included as important variables in clinical practice symptom research data.

Consistency of symptom clusters in advanced cancer.

Background: The reproducibility of symptom clusters (SCs) in different populations would support the validity of the cluster concept. Ideal approaches to cluster identification are unknown. The presence of a sentinel (most prevalent) symptom may reduce the number of symptoms in a comprehensive symptom assessment tool. The primary purpose was to assess consistency of SCs between 2 independent data sets. A secondary aim was to evaluate whether use of a sentinel symptom might abbreviate assessment but retain acceptable accuracy. Methods: An agglomerative hierarchical cluster analysis in 922 patients with advanced cancer identified 7 SCs. We conducted the same analysis on an additional 181 cancer patients to assess cluster consistency. The most prevalent symptom within each cluster was defined as the ‘‘sentinel’’ symptom. Positive predictive value (PPV) and negative predictive value (NPV) were calculated to assess ability of the sentinel symptom to predict other symptoms in the cluster. Results: Similar clusters were identified in both data sets, which included nausea/vomiting, neuropsychologic, and aerodigestive clusters. When the sentinel symptom was present, >50% nonsentinel symptoms in a cluster were present; when absent, <50% nonsentinel symptoms were identified. However, the range for PPV and NPV of the sentinel symptom to identify other symptoms in the cluster was 19% to 72% and 41% to 95%, respectively. Conclusions: Consistent SCs were found in 2 separate data sets with the same assessment tool and statistical analysis. These findings support the statistical and clinical validity of the cluster concept through consistency between different populations. The nausea/vomiting, neuropsychologic, and aerodigestive clusters may be reliable for use in assessment. The presence or absence of a sentinel symptom in each cluster did not predict the presence or absence of other symptoms in the cluster. Sentinel symptoms are inadequate to assess symptom burden.

Bioelectrical Impedance Phase Angle Changes During Hydration and Prognosis in Advanced Cancer

Introduction. We wished to determine bioelectrical impedance (BIA) correlates before hydration or changes during hydration and determine if these changes were prognostically important. Methods and Materials. Fifty eligible patients underwent BIA measurements 3 consecutive days. Laboratory studies (electrolytes, creatinine, and hemoglobin) on day 1; weights and vital signs were recorded. Kaplan-Meier survival estimates were made at 30 and 60 days. Hazard ratios (HRs) based on Cox proportional hazards model were calculated. Results. Weight loss was associated with shorter survival. A higher phase angle (PA) on day 1 predicted longer survival. Increased PA during hydration predicted shorter survival: increased weight during hydration predicted longer survival. Discussion. Higher phase angle before hydration predicts poorer survival and, paradoxically, an increase in phase angle during hydration predicted poorer survival and preexisting intracellular dehydration, cachexia, or poor membrane function. Conclusions. Phase angle and weight during hydration predict survival in cancer.

The relationship between symptom prevalence and severity and cancer primary site in 796 patients with advanced cancer.

Knowledge of differences in symptom experience between cancer sites may help better understand symptom pathophysiology. A total of 38 symptoms in 796 consecutive patients with advanced cancer were retrospectively analyzed. Symptom prevalence and severity were compared among the 12 primary site groups (PSGs) by the chi-square test. Pairwise comparisons determined which sites differed. Pain, fatigue, weakness, lack of energy, and anorexia had the highest overall prevalence but did not differ among PSGs. The 3 most common neuropsychological symptoms (insomnia, depression, and anxiety) also did not vary among PSGs. Nineteen (50%) symptoms varied significantly between PSGs, in prevalence (17), severity (14), or both (12). Nine of 17, 6 of 14, and 6 of 12 were gastrointestinal symptoms. Symptoms which varied by PSGs can be included in cancer site-specific symptom assessment instruments.

Components of the anorexia–cachexia syndrome: gastrointestinal symptom correlates of cancer anorexia

IntroductionCancer-related anorexia is traditionally considered part of a complex but ill-defined anorexia–cachexia syndrome in which anorexia is intimately associated with other gastrointestinal (GI) symptoms and weight loss. We surveyed cancer patients with anorexia to learn more about the relationship between anorexia and these symptoms.Materials and methodsA 22-item GI questionnaire assessed the severity of anorexia and the prevalence of concurrent GI symptoms, including taste changes, food aversions, altered sense of smell, and diurnal food intake changes. The relationship between anorexia severity and anticancer therapy and prior menstrual or pregnancy-related appetite changes was also assessed.ResultsNinety-five of 101 patients with anorexia surveyed had complete data. Seventy-eight percent of them had moderate or severe anorexia. Abnormal diurnal appetite variation, taste changes, and food aversions were present in over 50% of all those with anorexia. Judged by the numerical rating scale, the worse the anorexia, the more prevalent were early satiety, constipation, vomiting, and food aversions. Those with more severe anorexia had greater weight loss, and worse performance status. Anorexia severity did not correlate with that during prior menses/pregnancy or antitumor therapy.ConclusionsEvaluation of multiple other GI symptoms is important in understanding the total experience of cancer anorexia. Early satiety, taste changes, food aversions, and altered sense of smell are important accompanying GI symptoms. Most validated anorexia tools do not assess these commonly associated GI symptoms. Future research should develop a comprehensive anorexia symptom questionnaire.