For the last decade, the study of the symptom cluster, defined as a group of coexisting interrelated symptoms, has been recognized as a priority in oncology research because symptom assessment/management is thought to be improved by targeting a cluster, rather than a single symptom.1–3 The main focus in this research field has been the identification of symptom clusters at a specific time point in a course of cancer treatment4–6 or in disease progress.7 Empirical studies have found that psychoneurological symptoms such as fatigue, depressed mood, cognitive disturbance, sleep disturbance, and pain form a symptom cluster.4–7 For instance, fatigue, cognitive impairment, and mood problems consistently formed a cluster with some other psychoneurological symptoms in 3 different samples of breast cancer patients (40 women with early-stage cancer following primary surgery and prior to adjuvant therapy; 88 with stage I, II, and III who completed surgery and adjuvant chemotherapy [CTX]; 26 with metastatic cancer).4 In another study with breast cancer patients, depressed mood, cognitive disturbance, fatigue, insomnia, and pain consistently formed a cluster at 3 time points across the cancer treatment trajectory.6 A psychoneurological symptom cluster (ie, pain, fatigue, sleep disturbance, lack of appetite, and drowsiness) was also found in a heterogeneous sample of oncology patients.5
An important new line of research is to examine intraindividual change in the symptom cluster experience over time and interindividual variations in change. Findings from such studies could help clinicians predict patients at risk of poor long-term outcomes, determine the best timing of assessment/management, understand the potential mechanisms of symptom clustering, and develop potential therapies of a cluster.8–10
Few longitudinal studies have examined the patterns of change in symptom cluster intensity over time in cancer patients, and thus little is known about whether subgroups of patients with distinct trajectories of symptom cluster intensity exist and what leads to those trajectories. However, research has examined the pattern of change in individual symptom intensity, such as fatigue intensity. Those studies indicated that symptom experience in cancer patients changed along the course of cancer treatment. For instance, Berger and Higginbotham11 reported that fatigue in breast cancer patients reached its peak 1 to 4 days after treatment; after then, it decreased gradually. Their measurement points were (a) 2 days before CTX (doxorubicin + cyclophosphamide) cycle 3; (b) at days 1 to 4, at day 5 to 10, and at day 11 to 21 after treatment cycle 3; and (c) 3 weeks after and 2 months after treatment cycle 4. Fatigue during and after radiation therapy (4- to 5-week treatment duration) was highest in the last week of treatment and returned to the baseline at 3 months after treatment.12 In addition, studies also indicated that interindividual variations existed in symptom experience over time along the course of cancer treatment. For instance, Miaskowski et al13 reported a large amount of interindividual variability in fatigue intensity from the time of simulation to 4 months after the completion of radiation treatment (RTX). Furthermore, various clinical/demographic factors (eg, treatment modality, treatment protocol, sequence of different treatment modes, cancer type, disease stage, age, and gender) appeared to influence the patterns of change in fatigue intensity.10,13–15 As in individual symptoms, intraindividual and interindividual variations may exist in a symptom cluster experience over time, and many variables may influence such variations. These variations should be considered when developing strategies to assess and manage a symptom cluster.
The present study builds upon previous published work,6 which empirically identified a psychoneurological cluster (ie, depressed mood, cognitive disturbance, fatigue, insomnia, and pain) in a breast cancer population. The present study expands the authors’ previous work by investigating the temporal aspects of this particular psychoneurological cluster. The purposes of this study were to identify subgroups of breast cancer patients with different patterns of change in psychoneurological symptom cluster intensity (PSCI) during treatment and to examine the differences in subgroups with regard to clinical and demographical characteristics (antecedents) as well as patient outcomes. Comparing subgroups with different patterns of change in symptom intensity would assist in sorting out individuals at risk of more severe symptoms and worse functional outcomes.
Brant and her colleagues8 proposed a new symptom management model, which describes the temporal aspect of symptom experience. The components of this model include antecedents of symptom experience (eg, demographics, psychological factors, physiological factors, and/or situational factors), symptom trajectory (ie, the individual symptom experience over time), consequences (eg, function, quality of life, and survival), and symptom management intervention. This model comprehensively incorporates existing symptom management theories/models, such as the theory of symptom management,16 the theory of unpleasant symptoms,17 the symptoms experience model,18 and the symptoms experience in time model.19 The analyses executed in the present study used this model to conceptualize relationships between variables.
Theory-driven measurement time points are fundamental for the study of temporal changes.20 Symptoms in breast cancer patients during cancer treatments are most likely to be treatment-related, and thus, symptoms in the present study were selected to be measured at 3 time points across the treatment cycle: one prior to CTX or RTX and 2 at follow-ups after initiating treatment.
The specific research questions addressed in this study were as follows:
1. For breast cancer patients, what are the subgroups of patients with different patterns of change in PSCI during cancer treatment?
2. Which demographic and/or clinical antecedents (such as age, education level, previous cancer treatment experience, current treatment mode, comorbid conditions, baseline physical performance status, or symptom management intervention) predict membership for subgroups with different patterns of change in PSCI?
3. Do subgroups with different patterns of PSCI differ in functional performance outcome at a selected time point?
Participants and Procedures
A secondary analysis was performed using the breast cancer subset (n = 160) from a primary data set with diverse cancer types (n = 396). The primary study examined the effectiveness of a cognitive behavioral intervention on fatigue in cancer patients in a randomized clinical trial.21 The control group received general information on nutrition. The experimental group received the interventions (ie, education about fatigue and energy conservation strategies) over the course of 3 interactive telephone sessions (a total of 75 minutes) during CTX or RTX. Patients who planned to receive at least 3 cycles of CTX, 6 weeks of RTX, or concurrent RTX and CTX for various cancer types were recruited from 2 American cancer centers from 1999 to 2002. They had not received prior treatment other than surgery for at least 1 month prior to enrollment and had either cure or local control as a goal of treatment. Patients were not recruited if they had planned to receive stem cell transplantation, interleukins, interferon, or tumor necrosis factor; had a diagnosis of chronic fatigue syndrome or evidence of a psychiatric disorder; had received treatment for anemia or depression during the prior 3 weeks; and/or were enrolled in another psychoeducational intervention study.
Three different time points were selected for data collection: one at baseline data (time 1: prior to CTX or RTX) and 2 at follow-up time points (times 2 and 3: after treatment given). For the CTX treatment arm, times 2 and 3 were 48 hours after the second and third CTX treatments, respectively. For the RTX treatment arm, times 2 and 3 were in the last week of RTX and 1 month after completion of RTX, respectively. The follow-up time points were approximated according to when maximum fatigue levels were found to occur in cancer patients.12,22 Note that CTX patients had CTX cycles of at least 21 or 28 days; RTX patients had at least 6 weeks of RTX. By this sampling design, patients within each treatment arm (eg, patients receiving CTX) were to be in a similar time point in relation to the treatment administered during the primary study. However, patients were not at the same time points in relation to their initial diagnosis.
Data for the present study consisted of only the breast cancer subset that provided information for key symptom variables across all 3 selected data collection points. Patients with missing information were not included in the study because the selected main analysis (ie, cluster analysis) does not allow cases with missing data. Otherwise, no additional inclusion or exclusion criteria were applied. The approvals of the institutional review boards were obtained for this secondary analysis.
Participants in this secondary analysis (Table 1) were all female breast cancer patients. The majority were middle-aged (mean age, 54 years), married (73%), and white (92%). The majority had early-stage cancer (stages 0–2, 89%) and a post–high-school education (66%). About half of the patients had at least 1 comorbid condition and received a cognitive behavioral intervention for fatigue management during the primary study. Half of the patients received CTX, whereas the rest received RTX. No patients received concurrent treatment of CTX and RTX. About 12% had previous cancer treatment experience either for recurrent cancer or for a different cycle of treatment. The time lapse since the previous treatment varied, but it was at least 1 month.
Antecedents of temporal change in the symptom cluster included demographic/clinical data sheets and the Eastern Cooperative Oncology Group (ECOG) performance status measured at baseline. The ECOG performance status was used to measure baseline physical performance status. It consists of a single item: a 0- to 4-point scale, where 0 = normal activity without symptoms and 4 = unable to get out of bed.23 It is a widely used validated measure.24
Each symptom in the psychoneurological cluster was measured separately. The intensities of depressive mood and cognitive disturbance were measured by the depression and confusion subscales of the Profile of Mood States–Short Form.25 The measurement time frame was the prior 2 to 3 days. Each subscale consists of 5 items with the 1- to 5-point scale range, where 1 = not at all and 5 = extremely. Test developers provided evidence for the validity (content, predictive, concurrent, and divergent validities) and reliability (internal consistency and test-retest reliability) of the Profile of Mood States–Short Form based on correlation studies in diverse populations.25 In this study, Cronbach’s α for the depression and confusion subscales were .81 and .75, respectively.
The intensity of fatigue was measured by 1 item from the General Fatigue Scale.26 General Fatigue Scale measures the level of distress caused by fatigue, the impact of fatigue on daily activities, and the fatigue intensity during various time frames. Construct validity was assessed through factor analysis, and reliability was tested with internal consistency.26 For this analysis, fatigue (quantified by a scale range of 1–10, where 1 = no fatigue and 10 = the greatest possible fatigue) within the past week was assessed in order to create consistency regarding the time frame/dimension across symptoms.
Insomnia was measured by the Pittsburgh Sleep Quality Index (PSQI).27 Although it generally includes 5 questions that are bed-partner or roommate-rated, only the 19 self-rated questions for sleep quality over the past month were used for scoring, where response options were various. Internal homogeneity (α = .83), stability (test-retest reliability = 0.85), and validity reported by the developers were deemed acceptable.27 Cronbach α for the global score of the PSQI was .83 in this study. Pain was measured by 1 item asking about pain intensity for the past week using the Likert-type scale, where 0 = no symptoms, 1 = not at all severe, and 4 = quite a bit severe.
The consequence of symptoms was the functional limitations for various daily activities measured at the second follow-up time points, and these were measured by the Functional Performance Inventory.28 This measure includes 65 items with 6 different activity domains (body care, household maintenance, physical exercise, recreation, spiritual activities, and social activities). The scale ranged from 0 to 3, where 0 = do not do because of heath reasons and 3 = do with no difficulty. Acceptable validity (ie, significant correlations of the Functional Performance Inventory with existing functional status measures) and reliability (α = .96; intraclass correlation = 0.85) were reported.29 In the conceptual framework, symptom management intervention can influence changes in symptom intensity over time; thus, the present study included cognitive behavioral intervention, given for fatigue management during the primary study, as an influencing factor. Note that the same recall time frame was applied to both CTX and RTX patients.
To identify the subgroups of patients undergoing different patterns of change in the PSCI, cluster analysis of the subjects was conducted with the PSCI scores across 3 time points (times 1, 2, and 3). Because cluster analysis is a statistical method grouping patients with a similar profile, patients were grouped based on their PSCI score profile across the 3 time points.
In order to obtain the PSCI score, the intensity of each symptom in the psychoneurological cluster (depressed mood, cognitive disturbance, fatigue, insomnia, and pain) was first standardized because the scale differed across symptom measures. It should be noted that data on each symptom were standardized using the grand mean and the corresponding SD. The grand mean is the pooled mean across the time points; it is also known as the marginal mean and is used for statistical tests in repeated-measures analyses, such as in analyses of variance (ANOVAs). This standardization procedure allows various symptom measures to have the same scale without restricting the variances of individual symptoms across the time points. Individual PSCI scores were computed by combining the standardized individual symptom intensity scores for each individual at each time point.
For clusters (ie, the subgroups in this study) extraction, Ward’s30 minimum-variance method was chosen because of its popularity and strength in population recovery.31–32 Criteria for selecting the number of subgroups with distinct patterns of PSCI include (a) error variance (marked increase without subsequent plateauing of increments); (b) simultaneous elevation of the pseudo-F statistic over the pseudo-t2 statistic;33 and (c) Mojena’s34 stopping criterion (within the feasible range).
Multinomial logistic regressions were conducted to determine which antecedent variables were predictive of subgroup membership. Initial analyses with 11 variables were first done to prevent problems caused by the large number of variables in the logistic model that had a small sample size. From these analyses, only 7 variables were selected: age, education level, previous cancer treatment experience, current treatment modality (ie, CTX vs RTX), number of comorbidities, baseline physical performance status, and symptom management intervention (ie, education for fatigue management). Four variables (ie, race, disease stage, employment status, and marital status) were excluded from the final logistic model because they did not have variance in some subgroups (eg, race) and/or did not significantly influence the model (eg, disease stage, marital status, and employment status; all P ≥ .05 in type 3 analyses). The likelihood ratio tests indicated no difference between the models with and without these 4 variables (all P > .05). The influence of each variable was examined in a separate model.
Also, it should be noted that the primary study was an intervention study of fatigue. The experimental and control groups were neither statistically nor clinically different regarding fatigue levels for the first 2 time points (3.3 ± 1.8 vs 3.3 ± 1.8 at time 1; 4.6 ± 2.2 vs 4.6 ± 2.0 at time 2). The 2 groups statistically significantly differed at the third time point (4.1 ± 2.2 vs 4.7 ± 2.1 on a 10-point scale). Although the difference was not clinically significant, the primary study demonstrated a statistically significant decrease in fatigue over time for the experimental group. Thus, the effect of the intervention on the pattern subgroups was examined in the present study.
One-way ANOVAs were conducted to examine the differences of subgroups within a selected consequence, that is, within a functional limitation at time 3.There was no sample size requirement for the cluster analysis as it explores the patterns in a given data without inferential tests. Also, a minimum of 10 cases per variable is the rule of thumb for logistic regression.35 When determining models and the number of predictors in the final model, extremely large estimators were examined as they indicated inadequate sample size.36
Subgroups With Different Patterns of Change in PSCI Across Treatment Trajectory
As shown in the Figure, the mean of PSCI scores changed only slightly over time. However, cluster analysis indicated the existence of 5 subgroups with distinct patterns of change in PSCI across treatment trajectory. The PSCI score at each time point was the sum of the standardized scores of the 5 psychoneurological symptoms, where the mean approximated to 0 and the SD to 4. The PSCI score does not have absolute clinical meaning, but rather relative meaning in comparing symptom intensity across time points and subgroups. In the present study, having a value that is half of the SD (ie, SD = 4, half of SD = 2), above or below the mean ( ie, 0) was considered as high or low, respectively, in symptom intensity.37 Moderate symptom intensity in the PSCI scores encompassed values between −2 and +2.
The PSCI in group 1 (26% of the total sample) was moderate at baseline and gradually increased over time. This group was named “the gradually increasing pattern subgroup.” Group 2 (37%), in which the PSCI was low at baseline and slightly decreased over time, was designated as “the constantly low pattern subgroup.” The PSCI in group 3 (14%) was low at baseline, increased to a moderate level at time 2, and then decreased to baseline level at time 3; hence, it was deemed “the start low with dramatic increase and decrease pattern subgroup.” The PSCI in group 4 (13%) was high at baseline and constantly increased over time; it was named as “the constantly high pattern subgroup.” Lastly, group 5’s PSCI (10%) was high at baseline and dramatically decreased after treatment was given; therefore, it was named as “the start high with dramatic decrease and leveling pattern subgroup.”
Table 2 presents the scores of the 5 individual psychoneurological symptoms in the original metric. One can observe that most patients in this study experienced low or moderate intensity of psychoneurological symptoms across the time points. In addition, in each subgroup, all 5 individual symptoms tended to concurrently follow a similar clinical course of the overall PSCI score. In other words, group 4 with the constantly high pattern had all 5 symptoms at high intensity compared with other subgroups across all time points, whereas group 2 with the constantly low pattern experienced all 5 symptoms at a very low intensity across all time points.
Distinguishing Antecedents of the Subgroups
The subgroups were compared with regard to the 7 antecedent variables using multinomial logistic regression analyses. These analyses were conducted to determine if subgroups could be identified early on the basis of clinical characteristics or by being in the symptom intervention group. In these analyses, the reference group was group 4 (the constantly high pattern), which appeared to be the important target for symptom assessment/management (Table 3). Group 4 was best distinguished from group 2, which demonstrated the constantly low pattern; specifically, group 4 was less likely to have had previous cancer treatment experience and more likely to have a poor baseline performance status. Group 4 differed from group 5 (the start high with dramatic decrease and leveling pattern) only with respect to education, in that patients with a higher education were more likely to be in group 4. Group 4 was different from group 3 (the start low with dramatic increase and decrease pattern) with regard to the type of cancer treatment, that is, group 4 was more likely to have had CTX. Group 4 was different from group 1 (the gradually increasing pattern) only concerning baseline performance status, where group 4 was more likely to have a poor baseline performance status. Age, receipt of the symptom intervention, and comorbid conditions (ie, the number of comorbid conditions and the absence vs presence of comorbid conditions) did not distinguish group 4 from any other groups. Because the types of comorbid conditions in our data were very diverse, their systematic influence on symptom experience cannot be examined. Also, initial analysis indicated that disease stage was not associated with symptom cluster experience over time. Because the total study period was relatively short (2–3 months) and most patients had early-stage cancer, it is not expected that patients had serious disease progress during the study.
Differences in Consequence of the Subgroups
Table 4 shows differences in functional limitations by subgroup at the second follow-up (time 3). A lower score indicates higher limitations. Although patients in our sample did not have serious limitations in performing daily activities (the range of scores were 2–3 on a 0- to 3-point scale), some patients experienced relatively more limitations compared with other patients. One-way ANOVA demonstrated that pattern subgroups differed from each other significantly in functional limitation (F4,155 = 19.23, P < .0001). Follow-up analyses showed that patients in group 4 (the constantly high pattern) differed from all other groups and had the most serious limitations in daily activities, whereas patients in group 2 (the constantly low pattern) experienced the least amount of limitations.
This study clustered individuals with a similar PSCI score profile and found 5 distinct subgroups representing different patterns of symptom cluster intensity during breast cancer treatment. Groups 2 and 4 consistently had different patterns of PSCI over time; group 4 was consistently high in symptom intensity, and group 2 was consistently low. These 2 groups also differed on 2 antecedent factors that could partially explain the difference in patterns. Fewer group 4 members had received previous cancer treatment. It is possible that previous treatment represented a learning effect for group 2, resulting in less uncertainty about what symptoms to expect and/or how to deal with them. More group 4 members also had poor performance status prior to treatment, which could have contributed to their overall symptom burden and intensified their experience of PSCI. Both of these factors (ie, previous cancer treatment and poor performance status) could be useful in identifying individuals who, like group 4, are at risk for continuously high symptoms over time.
Group 1 had a pattern of PSCI that was similar to group 4, although the intensity of symptoms was lower in group 1. This group also differed from group 4 concerning the antecedent variable of performance status, with group 4 having a worse pretreatment performance status. This suggests that the ability to function prior to treatment could be an important factor that is predictive of symptom intensity and functional limitations during and after treatment.
Groups 5 and 3 showed dramatic changes at time 2. Group 5 started at baseline with a level of high symptom intensity similar to group 4. However, group 5 experienced a dramatic improvement in the PSCI after the initiation of treatment, whereas group 4 maintained intense symptoms over time. Patients with a higher education were more likely to be in group 4. Group 3 had low symptom intensity at the baseline and experienced increased symptoms after treatment initiation and lowered intensity at the second follow-up. Patients receiving RTX were more likely to be in group 3 rather than group 4. The pattern of group 3 was expected for RTX patients as overall symptom intensity is generally lower during RTX compared with CTX. More importantly, these patients were in the last week of treatment at time 2 and at 1 month after the completion of treatment at time 3. The pattern indicates that, for some RTX patients, psychoneurological symptoms are likely to occur only during the treatment and fade away after the completion of treatment.
The importance of distinguishing individuals with patterns of high or low symptom burden during treatment is underscored by the finding that group 4, with consistently high PSCI, had more functional limitations during and after treatment than all the other groups. This result indicates that symptom burden could have a detrimental effect on an individual’s ability to function in usual daily activities. Decline in functioning and quality of life after cancer diagnosis is most likely due to symptoms.38–40 In particular, psychoneurological symptoms have been inversely associated with patient outcomes.1,38,40–42 The effective management of psychoneurological symptoms could ultimately improve the quality of life in cancer patients. Maintaining a good quality of life becomes an important goal in caring for cancer patients as cancer becomes a chronic disease, thus requiring long-term care. Patients at risk of deteriorating symptom experience over time should be the highest priority for symptom assessment and management because they have the worst outcome.
The findings of this research have demonstrated the heterogeneity of symptom cluster intensity patterns during cancer treatment. Sorting out these differences is an important first step in determining the mechanisms underlying the symptom cluster. Parker and colleagues43 have suggested that symptom mechanisms could be psychological as well as biological. One could speculate that the underlying psychological or physiological mechanisms are responsible for different trajectories of PSCI. High PSCI prior to cancer treatment could be due to situational distress related to uncertainty or it could represent an ongoing style of coping with distress. Two subgroups, groups 4 and 5, which had a similarly high level of symptom intensity at the baseline, had different levels of symptoms after treatment initiation. Such differences may be due to discrepancies in an individual’s psychological adjustment to cancer treatment. In fact, individual variations in the course of psychological adjustment after stressful events are known to exist.44,45 For example, some cancer patients can unexpectedly experience reduction in psychological distress after treatment begins.12 Furthermore, it has been reported that social and personal resources (such as social support, self-image, optimism, or perceived control) can influence the trajectories of psychological as well as physical adjustment to breast cancer over a long period.44
High PSCI during treatment could be attributed to physiological changes that occur during cancer treatment. Several investigators postulated that psychoneurological symptoms in cancer patients may be due to the hyperactivated proinflammatory cytokines produced in response to cancer treatment or cancer itself.46–48
It will be important for future research to examine the contributions of both psychological and biological mechanisms to the experience of PSCI. In particular, examining patterns of PSCI and their characteristics can provide important information regarding the underlying mechanisms of symptoms. For instance, Thornton et al45 provided powerful evidence for the negative influence of subjective stress on immunity (specifically natural killer cell toxicity and blastogenesis) in cancer patients by associating trajectories of subjective stress with change of immunity after breast cancer diagnosis.
The impact of antecedent variables on the temporal aspect of symptom experience cannot be concluded from this study alone because of the correlational approach. The study results should be viewed as hypothesis-generating rather than conclusive. The antecedent variables should be examined further as likely predictors of PSCI.
As a secondary analysis, there are several limitations to this research: the small number of late stage cancer patients; exclusion of cases with missing information; the lack of information regarding specific treatment regimens, doses, and medications given for symptom management; the slight inconsistency in time frame over measures; and the use of old data. These impose a threat to the external validity of the findings, that is, the findings may only reflect phenomena in a population with certain characteristics, such as having early stages of cancer.
At baseline, patients who missed a data point had more severe symptom intensities regarding fatigue (P = .01) and pain (P = .08). Also, there was a trend for missing data related to late stages of disease (P = .059). Likewise, the missing pattern was not missing at random, and thus, imputations were not appropriate and cases with missing information were excluded. By doing so, the present findings may underrepresent the symptom patterns over time for those with severe symptoms or with more advanced disease, as indicated by the smaller number of subjects in the constantly high pattern subgroup (n = 21).
In addition, patients were asked to recall symptoms for slightly different time frames: the past week for pain and fatigue, the past 2 to 3 days for depressed mood and cognitive disturbance, and the past month for insomnia. Although patients were asked to recall slightly different time frames, patients would generally remember their symptoms from the past week. A 1-month recall time frame for insomnia (PSQI) was chosen by the measurement developer because a consistent pattern of poor sleep quality may capture insomnia better than a transient sleep problem. However, it should be noted that each item of the PSQI measures the frequency of events over a week. Patients are more likely to report the sleep problems that occurred during the past week. Thus, it is most likely that the time frame of recall would be around a week, and it is unlikely to have biased results.
Although the present study’s approach provides some understanding of the temporal aspect of psychoneurological symptom cluster during treatment, it is limited in showing what led to such distinguishing patterns. Future studies need to investigate the extensive mechanistic associates of temporal changes of symptoms, such as the possible biological and psychological variables. The inclusion of more data points, a larger sample size, and/or a special sampling design will also allow for more complex modeling of the PSCI trajectories.
The present study findings are preliminary using old data and need to be replicated in prospective studies. However, these results are provocative and point to the need for the early detection of risk for a deteriorating trajectory of a psychoneurological cluster. These results also suggest the need for specified patient preparation for an upcoming cancer treatment. Uncertainty for the unknown consequences of cancer and its treatment causes emotional stress and further lowers quality of life in cancer patients and their families. Patients and their families will be empowered by receiving adequate information on the symptom experience when dealing with stressful events of cancer treatments. This study’s findings can also be considered in personalizing intervention in future symptom cluster research. For instance, the results suggest the need to evaluate specific interventions for specific subgroups and to examine the causal mechanisms (biological/psychological) underlying multiple symptoms.
The authors acknowledge the contribution of Drs Susan Beck and William Bill Dudley, who provided methodological consultations for this work.
1. Dodd MJ, Miaskowski C, Paul SM. Symptom clusters and their effect on the functional status of patients with cancer. Oncol Nurs Forum. 2001; 28 (3): 465–470.
2. Kim HJ, McGuire DB, Tulman L, Barsevick AM. Symptom clusters: concept analysis and clinical implications for cancer nursing. Cancer Nurs. 2005; 28 (4): 270–282.
4. Bender CM, Ergun FS, Rosenzweig MQ, Cohen SM, Sereika SM. Symptom clusters in breast cancer across 3 phases of the disease. Cancer Nurs. 2005; 28 (3): 219–225.
5. Chen ML, Tseng HC. Symptom clusters in cancer patients. Support Care Cancer. 2006; 14 (8): 825–830.
6. Kim H-J, Barsevick AM, Tulman L, Mc Dermott P. Treatment-related symptom clusters in breast cancer: a secondary analysis. J Pain Symptom Manage. 2008; 36 (5): 468–478.
7. Walsh D, Rybicki L. Symptom clustering in advanced cancer. Support Care Cancer. 2006; 14 (8): 831–836.
8. Brant JM, Beck S, Miaskowski C. Building dynamic models and theories to advance the science of symptom management research. J Adv Nurs. 2009; 66 (1): 228–240.
9. Richardson A, Ream E, Wilson-Barnett J. Fatigue in patients receiving chemotherapy: patterns of change. Cancer Nurs. 1998; 21 (1): 17–30.
10. Schwartz AL, Nail LM, Chen S, et al. Fatigue patterns observed in patients receiving chemotherapy and radiotherapy. Cancer Invest. 2000; 18 (1): 11–19.
11. Berger AM, Higginbotham P. Correlates of fatigue during and following adjuvant breast cancer chemotherapy: a pilot study. Oncol Nurs Forum. 2000; 27 (9): 1443–1448.
12. Irvine DM, Vincent L, Graydon JE, Bubela N. Fatigue in women with breast cancer receiving radiation therapy. Cancer Nurs. 1998; 21 (2): 127–135.
13. Miaskowski C, Paul SM, Cooper BA, et al. Trajectories of fatigue in men with prostate cancer before, during, and after radiation therapy. J Pain Symptom Manage. 2008; 35 (6): 632–643.
14. Donovan KA, Jacobsen PB, Andrykowski MA, et al. Course of fatigue in women receiving chemotherapy and/or radiotherapy for early stage breast cancer. J Pain Symptom Manage. 2004; 28 (4): 373–380.
15. Kozachik SL, Bandeen-Roche K. Predictors of patterns of pain, fatigue, and insomnia during the first year after a cancer diagnosis in the elderly. Cancer Nurs. 2008; 31 (5): 334–344.
16. Larson P, Carrieri-Kohlman V, Dodd M, et al. A model for symptom management. The University of California, San Francisco School of Nursing Symptom Management Faculty Group. Image J Nurs Scholarsh. 1994; 26 (4): 272–276.
17. Lenz ER, Pugh LC, Milligan RA, Gift A, Suppe F. The middle-range theory of unpleasant symptoms: an update. ANS Adv Nurs Sci. 1997; 19 (3): 14–27.
18. Armstrong TS. Symptoms experience: a concept analysis. Oncol Nurs Forum. 2003; 30 (4): 601–606.
19. Henly SJ, Kallas KD, Klatt CM. The notion of time in symptom experiences. Nurs Res. 2003; 52 (6): 410–417.
20. Collins LM. Analysis of longitudinal data: the integration of theoretical model, temporal design, and statistical model. Annu Rev Psychol. 2006; 57: 505–528.
21. Barsevick AM, Dudley W, Beck S, Sweeney C, Whitmer K, Nail L A randomized clinical trial of energy conservation for patients with cancer-related fatigue. Cancer. 2004; 100 (6): 1302–1310.
22. Meek PM, Nail LM, Barsevick A, et al. Psychometric testing of fatigue instruments for use with cancer patients. Nurs Res. 2000; 49 (4): 181–190.
23. Oken MM, Creech RH, Tormey DC, et al. Toxicity and response criteria of the Eastern Cooperative Oncology Group. Am J Clin Oncol. 1982; 5 (6): 649–655.
24. Conill C, Verger E, Salamero M. Performance status assessment in cancer patients. Cancer. 1990; 65 (8): 1864–1866.
25. McNair DM, Lorr M, Droppleman LF. Profile of Mood States, Revised (POMS). San Diego, CA: EdITS/Educational and Industrial Testing Service; 1981
26. Meek PM, Nail LM, Jones LS. Internal consistency reliability and construct validity of a new measure of cancer treatment related fatigue: the General Fatigue Scale (GFS). Oncol Nurs Forum. 1997; 24: 334.
27. Buysse DJ, Reynolds CF 3rd, Monk TH, Berman SR, Kupfer DJ. The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research. Psychiatry Res. 1989; 28 (2): 193–213.
28. Leidy NK. Functional status and forward progress of merry-go rounds: toward a coherent analytical framework. Nurs Res. 1994; 43: 196–202.
29. Leidy NK. Psychometric properties of the functional performance inventory in patients with chronic obstructive pulmonary disease. Nurs Res. 1999; 48 (1): 20–28.
30. Ward JH. Hierarchical grouping to optimize an objective function. J Am Stat Assoc. 1963; 58 (301): 236–244.
31. Finch H. Comparison of distance measures in cluster analysis with dichotomous data. J Data Sci. 2005; 3: 85–100.
32. Romesburg CH. Cluster Analysis for Researchers. Raleigh, NC: Lulu Press; 2004
33. Copper MC, Milligan GW. The effect of error on determining the number of clusters. In: Proceedings of the International Workshop on Data Analysis, Decision Support, and Expert Knowledge Representation in Marketing and Related Areas of Research. London, England: Springer-Verlag; 1988: 319–328.
34. Mojena R. Hierarchical grouping methods and stopping rules—an evaluation. Comput J. 1977; 20: 359–363.
35. Hosmer DW, Lemeshow S. Applied Logistic Regression. New York: Wiley; 1989
36. Tabachnick BG, Fidell LS. Using Multivariate Statistics. 4th ed. Needham Heights, MA: Allyn & Bacon; 2001
37. Norman GR, Sloan JA, Wyrwich KW. Interpretation of changes in health-related quality of life: the remarkable universality of half a standard deviation. Med Care. 2003; 41 (5): 582–592.
38. Deimling GT, Bowman KF, Wagner LJ. The effects of cancer-related pain and fatigue on functioning of older adult, long-term cancer survivors. Cancer Nurs. 2007; 30 (6): 421–433.
39. Monga U, Kerrigan AJ, Thornby J, Monga TN, Zimmermann KP. Longitudinal study of quality of life in patients with localized prostate cancer undergoing radiotherapy. J Rehabil Res Dev. 2005; 42 (3): 391–399.
40. Portenoy RK, Thaler HT, Kornblith AB, et al. Symptom prevalence, characteristics and distress in a cancer population. Qual Life Res. 1994; 3 (3): 183–189.
41. Given B, Given C, Azzouz F, Stommel M. Physical functioning of elderly cancer patients prior to diagnosis and following initial treatment. Nurs Res. 2001; 50 (4): 222–232.
42. Miaskowski C, Cooper BA, Paul SM, et al. Subgroups of patients with cancer with different symptom experiences and quality-of-life outcomes: a cluster analysis. Oncol Nurs Forum. 2006; 33 (5): E79–E89.
43. Parker KP, Kimble LP, Dunbar SB, Clark PC. Symptom interactions a mechanisms underlying symptom pairs and clusters. J Nurs Scholarsh. 2005; 37 (3): 209–215.
44. Helgeson VS, Snyder P, Seltman H. Psychological and physical adjustment to breast cancer over 4 years: identifying distinct trajectories of change. Health Psychol. 2004; 23 (1): 3–15.
45. Thornton LM, Andersen BL, Crespin TR, Carson WE. Individual trajectories in stress covary with immunity during recovery from cancer diagnosis and treatments. Brain Behav Immun. 2007; 21 (2): 185–194.
46. Bower JE, Ganz PA, Aziz N, Fahey JL. Fatigue and proinflammatory cytokine activity in breast cancer survivors. Psychosom Med. 2002; 64 (4): 604–611.
47. Lee BN, Dantzer R, Langley KE, et al. A cytokine-based neuroimmunologic mechanism of cancer-related symptoms. Neuroimmunomodulation. 2004; 11 (5): 279–292.
48. Kim HJ, Barsevick AM, Fang CY, Miaskowski C. Common biological pathways underlying the psychoneurological symptom cluster in cancer patients [published online ahead of print]. Cancer Nurs. 2012; 35 (6): E1–E20. doi: 10.1097/NCC.0b013e318233a811.