Identification of subgroups of chemotherapy patients with distinct sleep disturbance profiles and associated co-occurring symptoms

Patients and settings

This study is part of a longitudinal study, funded by the National Cancer Institute, that evaluated the symptom experience of oncology outpatients receiving CTX [21–23]. Patients were eligible if they were at least 18 years of age; had a diagnosis of breast, gastrointestinal (GI), gynecological, or lung cancer; had received CTX within the preceding 4 weeks; were scheduled to receive at least two additional cycles of CTX; were able to read, write, and understand English; and provided written informed consent. Patients were recruited from two Comprehensive Cancer Centers, a Veterans Affairs hospital, and four community-based oncology programs. A total of 2234 patients were approached and 1343 consented to participate (60.1% response rate). The major reason for refusal was being too overwhelmed with their cancer treatment. For this study, 1331 patients completed the General Sleep Disturbance Scale (GSDS).

Instruments

Demographic information was obtained using a questionnaire that included age, gender, ethnicity, marital status, living arrangements, education, employment status, childcare and eldercare responsibilities, exercise regularity, and annual income. Patients’ self-reported their functional status using the Karnofsky Performance Status (KPS) scale that ranged from 30 (I feel severely disabled and need to be hospitalized) to 100 (I feel normal; I have no complaints or symptoms) [24].

The Self-Administered Comorbidity Questionnaire (SCQ) was used to assess comorbidity. The questionnaire consists of 13 common medical conditions that were simplified into language that could be understood without any prior medical knowledge [25]. It allowed patients to note the severity of a comorbidity by indicating if they had the condition, if they received treatment for it, and if it limited their activities. For each condition, a patient can receive a maximum of 3 points and the total SCQ score ranges from 0 to 39. The SCQ has well-established validity and reliability and has been used in studies of patients with a variety of chronic conditions [26].

The GSDS consists of 21 items designed to assess the various aspects of sleep disturbance (i.e. quality, quantity, onset latency, mid and early awakenings, sleep medications, daytime sleepiness). Each item was rated on a 0 (never) to 7 (everyday) numeric rating scale (NRS). The GSDS total score ranges from 0 (no disturbance) to 147 (extreme sleep disturbance). Each mean subscale score ranges from 0 to 7. Subscale scores of at least 3 and a GSDS total score of at least 43 indicate a significant level of sleep disturbance that warrants clinical evaluation and management [27]. The GSDS has well-established validity and reliability [28–30]. In this study, the Cronbach’s α for the GSDS total score was 0.83.

The Spielberger State-Trait Anxiety Inventories (STAI-T and STAI-S) each have 20 items that are rated from 1 to 4. The summed scores for each scale can range from 20 to 80. The STAI-S measures a person’s temporary anxiety response to a specific situation or how anxious or tense a person is “right now” in a specific situation. The STAI-T measures a person’s predisposition to anxiety as part of one’s personality. Cutoff scores of more than 31.8 and more than 32.2 indicate high levels of trait and state anxiety, respectively. The STAI-S and STAI-T inventories have well-established validity and reliability [31–33]. In this study, the Cronbach’s alphas for the STAI-T and STAI-S were 0.92 and 0.96, respectively.

The Attentional Function Index (AFI) consists of 16 items designed to measure attentional function [34]. A higher total mean score on a 0 to 10 NRS indicates greater capacity to direct attention. Total scores are grouped into categories of attentional function (i.e. <5.0 low function, 5.0–7.5 moderate function, >7.5 high function) [35]. The AFI has well-established reliability and validity [34]. In this study, the Cronbach’s α for the total AFI score was 0.93.

The Center for Epidemiological Studies-Depression scale (CES-D) consists of 20 items selected to represent the major symptoms in the clinical syndrome of depression. A total score can range from 0 to 60, with scores of at least 16 indicating the need for individuals to seek clinical evaluation for major depression. The CES-D has well-established validity and reliability [36–38]. In this study, the Cronbach’s α for the CES-D total score was 0.89.

The Lee Fatigue Scale (LFS) consists of 18 items designed to assess physical fatigue and energy [39]. Each item was rated on a 0–10 NRS. Total fatigue and energy scores were calculated as the mean of the 13 fatigue items and the 5 energy items, respectively. Higher scores indicate greater fatigue severity and higher levels of energy. Using separate LFS questionnaires, patients were asked to rate each item based on how they felt within 30 minutes of awakening (i.e. morning fatigue, morning energy) and before going to bed (i.e. evening fatigue, evening energy). The LFS has established cutoff scores for clinically meaningful levels of fatigue (i.e. ≥3.2 for morning fatigue, ≥5.6 for evening fatigue) [27] and energy (i.e. ≤6.2 for morning energy, ≤3.5 for evening energy) [27]. It was chosen for this study because it is relatively short, easy to administer, and has well-established validity and reliability [39–41]. In this study, the Cronbach’s alphas were 0.96 for morning and 0.93 for evening fatigue and 0.95 for morning and 0.93 for evening energy.

Occurrence of pain was evaluated using the Brief Pain Inventory [42]. Patients who responded yes to the question about having pain were asked to indicate if their pain was or was not related to their cancer treatment. Patients were categorized into one of four groups (i.e. no pain, only noncancer pain, only cancer pain, both cancer and noncancer pain).

Study procedures

The study was approved by the committee on human research at the University of California, San Francisco and the institutional review board at each of the study sites. Patients were approached by a research staff member in the infusion unit to discuss participation in the study. Written informed consent was obtained from all patients. Given the challenges associated with the recruitment of patients before their first cycle of CTX, patients were recruited during their second or third cycle of CTX. Depending on the length of their CTX cycle, patients completed questionnaires in their homes, a total of six times over two cycles of CTX (i.e. before CTX administration [i.e. recovery from previous CTX cycle, assessments 1 and 4], approximately 1 week after CTX administration [i.e. acute symptoms, assessments 2 and 5], approximately 2 weeks after CTX administration [i.e. potential nadir, assessments 3 and 6]). Disease and treatment information were collected from medical records.

Data analysis

Descriptive statistics and frequency distributions were calculated for the demographic and clinical characteristics using Statistical Package for the Social Sciences, version 23 (International Business Machines, Armonk, NY). Differences among the latent classes were evaluated using analyses of variance or chi-square analyses. The Bonferroni procedure was used to calculate the post hoc contrasts. A p-value of less than .05 was considered statistically significant.

LPA of sleep disturbance

As previously described for other symptoms [43, 44], unconditional LPA was used to identify the profiles of sleep disturbance means for the total score from the GSDS that characterized unobserved subgroups (i.e. latent classes) of patients over the six assessments. Typically, growth mixture modeling or latent class growth modeling of change trajectories would be used to identify latent classes of individuals who change differently over time. However, the data from this study demonstrated a complex pattern of change because a pretreatment assessment, an immediate posttreatment assessment, and a second posttreatment assessment were conducted over two cycles of CTX (i.e. assessments 1, 2, and 3 and assessments 4, 5, and 6). We expected that the trajectory of change for sleep disturbance, measured six times, over two treatment cycles would have a twin peak pattern that looks like “Λ_Λ”. Therefore, we identified latent classes of patients based on their profiles of means, where the means were estimated from the GSDS total scores measured on six occasions. To incorporate the expected correlations among the repeated measures, we included covariance among GSDS scores that were one or two occasions apart (i.e. a covariance structure with a lag of two). In this way, we retained the within-person correlation among the GSDS scores, at the same time that we focused on the patterns of means that distinguished the latent classes. We limited the covariance structure to a lag of two to accommodate the expected reduction in correlation that would be introduced by two treatments within each set of three measurement occasions and to reduce model complexity.

Estimation was carried out with full information maximum likelihood with standard errors and a chi-square test that are robust to nonnormality and nonindependence of observations (estimator=MLR). Model fit was evaluated to identify the best solution that characterized the observed latent class structure with the Bayesian Information Criterion (BIC), the Vuong–Lo–Mendell–Rubin likelihood ratio test (VLMR) for the K vs. K-1 model, entropy, and latent class percentages that were large enough to be reliable (i.e. likely to replicate in new samples; 15% or about 85 patients) [45, 46]. Missing data were accommodated with the use of the Expectation-Maximization algorithm [47].

Mixture models, such as LPA, are known to produce solutions at local maxima. Therefore, our models were fit with from 1000 to 2400 random starts. This approach ensured that the estimated model was replicated many times and was not due to a local maximum. Estimation was carried out with Mplus, version 7.2 [45].

Characteristics of the sample

The demographic and clinical characteristics of the sample are summarized in Table 1. Of the 1331 patients, the majority were White (69.5%), married or partnered (64.5%) with a mean age of 57.1 (±12.3) years. More than 40% of the patients were diagnosed with breast cancer and 30.4% were diagnosed with GI cancer.

Latent classes for sleep disturbance

Although the BIC for the four-class solution was lower than the three-class solution, the three-class solution was selected because one of the classes in the four-class solution was too small to be reliable and because the profile of means for two of the classes in the four-class solution did not differ in a meaningful way (i.e. either by profile or mean levels of GSDS total scores; Table 2). In addition, the three-class solution fits better than the two-class solution (lower BIC, VLMR test) and the profiles of means were clinically meaningfully different.

As shown in Figure 1, the trajectories for sleep disturbance scores differed among the latent classes. Because a clinically meaningful total GSDS score for sleep disturbance is at least 43 [27], the sleep disturbance classes were named Low, High, and Very High. For both the Low (25.2%) and the Very High (24.0%) classes, sleep disturbance scores remained relatively constant across the six assessments. In contrast, for the High Sleep Disturbance class (50.8%), sleep disturbance scores oscillated over the two cycles of CTX, with slightly higher scores reported at assessments 2 and 5 (i.e. the weeks following the administration of CTX).

Differences in demographic and clinical characteristics among the sleep disturbance classes

Compared to patients in the Low Sleep Disturbance class, patients in the High and Very High classes were significantly younger, had a lower KPS score, and a higher SCQ score (Table 3). Furthermore, compared to the Low Sleep Disturbance class, patients in the High and Very High classes were more likely to be female, less likely to be currently employed, more likely to have childcare responsibilities, and less likely to have a GI cancer diagnosis. In addition, compared to the Low and High Sleep Disturbance classes, patients in the Very High class had a higher body mass index (BMI), were less likely to be married or partnered, were more likely to live alone, and reported a lower annual household income. The remainder of the demographic and clinical characteristics did not differ among the sleep disturbance classes.

Differences in GSDS subscale scores among the sleep disturbance classes

As shown in Table 4, differences were found among the sleep disturbance classes for all of the GSDS subscale scores and total sleep disturbance score at enrollment (all ps < .001). For all of the subscales of the GSDS (i.e. sleep quality, sleep quantity, sleep onset latency, mid-sleep awakenings, early awakenings, medications for sleep, and excessive daytime sleepiness scores [all ps < .001]), as well as for the total GSDS score, these significant differences among the three latent classes followed the expected pattern (i.e. Low < High < Very High).

Severity of co-occurring symptoms among the latent classes

As shown in Table 5, differences were found among the sleep disturbance classes in the severity scores for all of the co-occurring symptoms at enrollment (all ps < .001). Trait and state anxiety, depressive symptoms, and morning and evening fatigue scores followed the expected pattern (i.e. Low < High < Very High). Attentional function, as well as morning and evening energy scores, followed the opposite but expected pattern (i.e. Low > High > Very High). In terms of pain, the post hoc contrasts for no pain (i.e. Low > High > Very High) and for having both cancer and noncancer pain (i.e. Low < High < Very High) were in the expected directions.

Limitations

Several limitations need to be acknowledged. Because patients were not recruited before the initiation of CTX, changes in sleep disturbance before the initiation of treatment were not assessed. Patients in this sample were predominantly White, female, college educated, and had metastatic disease, which suggests that our study sample may not be entirely representative of oncology patients in the United States. Although the sample size was very large, which increases the generalizability of the study findings, these patients received a variety of CTX regimens. As a result, differences in sleep disturbance associated with different CTX regimens cannot be evaluated. Information on the menopausal status of the women in this study is not available. Therefore, we cannot evaluate its influence on sleep disturbance. Finally, our study included only subjective measures of sleep disturbance. Future studies should include both subjective and objective measures of sleep disturbance.