Supportive Care in Cancer

, Volume 20, Issue 10, pp 2611–2619

Identification of distinct subgroups of breast cancer patients based on self-reported changes in sleep disturbance

Authors

  • Christina Van Onselen
    • Department of Physiological Nursing, School of NursingUniversity of California
  • Bruce A. Cooper
    • Department of Physiological Nursing, School of NursingUniversity of California
  • Kathryn Lee
    • Department of Physiological Nursing, School of NursingUniversity of California
  • Laura Dunn
    • School of MedicineUniversity of California
  • Bradley E. Aouizerat
    • Department of Physiological Nursing, School of NursingUniversity of California
    • Institute for Human GeneticsUniversity of California
  • Claudia West
    • Department of Physiological Nursing, School of NursingUniversity of California
  • Marylin Dodd
    • Department of Physiological Nursing, School of NursingUniversity of California
  • Steven Paul
    • Department of Physiological Nursing, School of NursingUniversity of California
    • Department of Physiological Nursing, School of NursingUniversity of California
Original Article

DOI: 10.1007/s00520-012-1381-3

Cite this article as:
Van Onselen, C., Cooper, B.A., Lee, K. et al. Support Care Cancer (2012) 20: 2611. doi:10.1007/s00520-012-1381-3

Abstract

Purpose

The purposes of this study were to identify distinct subgroups of patients based on self-reported sleep disturbance prior to through 6 months after breast cancer surgery and evaluate for differences in demographic, clinical, and symptom characteristics among these latent classes.

Methods

Women (n = 398) who underwent unilateral breast cancer surgery were enrolled prior to surgery. Patients completed measures of functional status, sleep disturbance (i.e., General Sleep Disturbance Scale (GSDS); higher scores indicate higher levels of sleep disturbance), fatigue, attentional fatigue, depressive symptoms, and anxiety prior to surgery and monthly for 6 months.

Results

Three distinct classes of sleep disturbance trajectories were identified using growth mixture modeling. The high sustained class (55.0%) had high and the low sustained class (39.7%) had low GSDS scores prior to surgery that persisted for 6 months. The decreasing class (5.3%) had high GSDS score prior to surgery that decreased over time. Women in the high sustained class were significantly younger, had more comorbidity and poorer function, and were more likely to report hot flashes compared to the low sustained class. More women who underwent mastectomy or breast reconstruction were in the decreasing class. Decreasing and high sustained classes reported higher levels of physical fatigue, attentional fatigue, depressive symptoms, and anxiety compared to the low sustained class.

Conclusions

A high percentage of women has significant sleep disturbance prior to surgery that persists during subsequent treatments (i.e., radiation therapy and chemotherapy). Clinicians need to perform routine assessments and initiate appropriate interventions to improve sleep prior to and following surgery.

Keywords

Breast cancerSleep disturbanceGrowth mixture modelingLatent class analysisFatigueDepressionAnxietyAttentional fatigue

Introduction

Occurrence of sleep disturbance in women with breast cancer ranges from 20% to 70% [6, 7, 13, 22, 25, 46]. Some of this variability may be related to the timing of the assessments in relationship to the patients' treatment trajectory. For example, in studies that evaluated sleep disturbance prior to breast cancer surgery, 70% [48] to 88% [12] of patients reported sleep disturbance. In two studies that evaluated patients prior to adjuvant treatment, between 59% [30] and 66% [33] reported sleep disturbance. Of note, 50% [30] to 58% [4] reported sleep disturbance 2 months after treatment ended.

In terms of severity, in a cross-sectional study [41], breast cancer patients undergoing chemotherapy (CTX) slept fewer hours than healthy controls. In two longitudinal studies [33, 47], total sleep time decreased during the first two CTX cycles and returned to pretreatment levels during the third cycle. In contrast, in breast cancer patients who underwent radiation therapy (RT) [53], the severity of sleep disturbance and sleep onset latency decreased over the course of RT. In a recent longitudinal study that evaluated the natural course of insomnia [46], approximately 37.6% of patients diagnosed with insomnia syndrome prior to cancer surgery kept that status throughout the 18-month follow-up period.

Attempts to characterize specific types of sleep disturbance (e.g., problems with sleep onset latency, sleep maintenance, or early awakening) in breast cancer patients have produced inconsistent results [3, 4, 47, 48, 53]. These studies relied on an evaluation of mean scores or clinically meaningful cutpoints to categorize various types of sleep disturbance. However, no longitudinal studies were found that evaluated for distinct subgroups of breast cancer patients (i.e., latent classes) based on changes in sleep disturbance over time. This type of evaluation using more sophisticated methods of data analysis may assist with the identification of patients who are at higher risk for more severe sleep disturbance.

Only one population-based study of patients enrolled in a Midwest health plan was found that used latent class analysis (LCA) to identify distinct subgroups of patients using their self-reports of sleep disturbance and functional impairments associated with sleep disturbance [24]. Four distinct latent classes of individuals with insomnia were identified, namely distressed (33.2%; individuals who had a single sleep complaint that occurred weekly and emotional distress), transient (25.5%; individuals who had a variety of sleep-related symptoms that occurred with different frequencies), difficulty maintaining sleep (21.9%; individuals who had insomnia symptoms every night), and comorbid with non-restorative sleep (19.4%; individuals who had sleep problems every night and daytime dysfunction). These four subgroups differed on the number of sleep disturbance symptoms, presence of non-restorative sleep and comorbidities, degree of daytime impairment, and insomnia severity. This study demonstrates the usefulness of using LCA to identify distinct subgroups of patients. However, the cross-sectional design did not allow for an evaluation of distinct subgroups of patients whose sleep disturbance might persist over a period of months or years.

Therefore, the purposes of this study were: to identify distinct latent classes of breast cancer patients based on self-reported sleep disturbance from the time prior to surgery through 6 months after surgery and to evaluate for differences in baseline demographic and clinical characteristics, as well as differences in the preoperative severity of other common symptoms among these latent classes. In addition, differences in the severity of a number of sleep parameters prior to surgery were evaluated among the latent classes.

Methods

Patients and settings

This study is part of a larger study that evaluated neuropathic pain and lymphedema in women who underwent breast cancer surgery. Patients were recruited from breast care centers located in a Comprehensive Cancer Center, two public hospitals, and four community practices. Women were eligible to participate if they were >18 years of age, underwent breast cancer surgery on one breast, and were able to read, write, and understand English. Patients were excluded if they were: having bilateral breast cancer surgery and/or had distant metastasis at the time of diagnosis. A total of 516 patients were approached, 410 enrolled in the study (response rate, 79.4%), and 398 completed questionnaire booklets. The major reasons for refusal were: too busy, overwhelmed with the cancer diagnosis, or insufficient time available to complete the baseline assessment prior to surgery.

Instruments

At enrollment, demographic and clinical information were obtained. At each subsequent assessment, patients provided information on current treatments for breast cancer. The occurrence of hot flashes was evaluated by asking women if they experienced hot flashes (0 = no and 1 = yes) during the past week. Medical records were reviewed to obtain clinical information. The Karnofsky Performance Status (KPS) scale was used to evaluate functional status [28].

Self-Administered Comorbidity Questionnaire (SCQ) was used to assess comorbidities. It consists of 13 common medical conditions. Patients were asked to indicate if they currently had the condition (“yes/no”) and, if “yes,” to indicate whether they received treatment for it and whether it limited their activities. Each condition yields a maximum of three points with a total score of 39 points [8, 44].

The 21-item General Sleep Disturbance Scale (GSDS) was used to evaluate overall sleep disturbance during the past week. Each item is rated on a scale that ranges from 0 (never) to 7 (everyday). The GSDS consists of seven subscales (i.e., quality of sleep, quantity of sleep, sleep onset latency, mid-sleep awakenings, early awakenings, medications for sleep, excessive daytime sleepiness) that can range from 0 to 7 and a total score that can range from 0 (no disturbance) to 147 (extreme sleep disturbance). A total GSDS score of ≥43 indicates a clinically meaningful level of sleep disturbance [23]. Subscale scores represent the number of days a week a patient finds the sleep parameter problematic, and a score ≥3 indicates a clinically meaningful level of disturbance [15, 35, 42]. Cronbach's alpha for the GSDS total score was 0.86.

The Lee Fatigue Scale consists of 18 items designed to assess physical fatigue and energy. Patients were asked to rate each item on a 0 (not at all) to 10 (extremely) numeric rating scale (NRS) based on how they felt right now. Separate 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 [32]. Cutoff scores of ≥4.4 and ≤4.8 indicate high levels of fatigue and low levels of energy, respectively [15, 36]. Cronbach's alphas for the fatigue and energy subscales were 0.96 and 0.93, respectively.

Attentional Function Index (AFI) consists of 16 items designed to measure attentional fatigue in patients with cancer. Each item is rated on a 0 to 10 NRS. A mean AFI score was calculated, with higher scores indicating greater capacity to direct attention (i.e., lower levels of attentional fatigue) [11]. Cronbach's alpha for the AFI was 0.95.

The Center for Epidemiological Studies-Depression Scale (CES-D) scale consists of 20 items selected to represent the major symptoms in the clinical syndrome of depression. Scores can range from 0 to 60, with scores of ≥16 indicating the need for individuals to seek clinical evaluation for major depression [43, 50]. Cronbach's alpha for the CES-D was 0.90.

Spielberg State-Trait Anxiety Inventories (STAI-T and STAI-S) consist of 20 items each that are rated from 1 to 4. The scores for each scale are summed and can range from 20 to 80. A higher score indicates greater anxiety. Cutoff scores of ≥31.8 and ≥32.2 indicate high levels of trait and state anxiety, respectively [5, 29, 51]. Cronbach's alphas for the STAI-T and STAI-S were 0.88 and 0.95, respectively.

Study procedures

The Committee on Human Research at the University of California, San Francisco, and at each of the study sites approved the study. During the patient's preoperative visit, a staff member explained the study to the patient. For women who were willing to participate, the staff member introduced the patient to the research nurse, who met with the women, determined eligibility, obtained written informed consent, and had patients complete the study questionnaires prior to surgery (assessment 0). Patients were contacted 2 weeks after surgery to schedule the first postsurgical appointment. The research nurse met with the patients in their home, the Clinical Research Center, or the clinic at 1, 2, 3, 4, 5, and 6 months after surgery to complete the study instruments.

Statistical analyses

Data were analyzed using SPSS version 18.0 [52] and Mplus version 5.21 [39]. Descriptive statistics and frequency distributions were calculated for the patients' demographic and clinical characteristics and symptom severity scores.

Unconditional growth mixture modeling (GMM) with robust maximum likelihood estimation was carried out to identify latent classes with distinct sleep disturbance trajectories. These methods are described in detail elsewhere [19]. In brief, a single growth curve that represented the “average” change trajectory was estimated for the whole sample. Then, the number of latent growth classes that best fit the data was identified using guidelines recommended by a number of experts [27, 40, 54].

Model fit for the GMM was assessed statistically by identifying the model with the lowest Bayesian Information Criterion (BIC). The parametric bootstrapped likelihood ratio test (BLRT) was used to evaluate whether a model with K classes fit the data better than a model with K-1 classes. In addition to using the BLRT to compare models, we examined the Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (VLMR) for the “K” versus “K-1” class models. When the VLMR test is non-significant, it does provide evidence that the K-class model is not better than the K-1-class model. The fourth index used to evaluate model fit was entropy, with >0.80 being preferred [9, 39]. Finally, the best-fitting model was visually inspected by plotting observed against model-predicted values to determine whether the predicted trajectories followed the empiric trajectories for the classes and to evaluate whether the predicted plots “made sense” theoretically and clinically [38].

Intercepts and linear and quadratic slopes for each latent class were estimated for each model. Intercept variances were estimated for each class and were allowed to differ across classes. Given the relatively small sample sizes, the within-class linear and quadratic slope variances were fixed at zero for two classes, because estimation failed when they were free to vary. Without setting these slope variances to zero, the model could not be estimated due to non-positive definite covariance matrices. Mixture models are known to produce solutions at local maxima, so each model was fit with random starts to be sure that the solution for the model with the maximum log likelihood values was replicated [39]. Missing data for the sleep disturbance scores were accommodated by Mplus version 5.21 through the use of Full Information Maximum Likelihood and the use of the Expectation-Maximization algorithm. This method assumes that any missing data are missing at random [37, 49].

Analysis of variance and chi-square analyses were used to assess for differences in baseline demographic and clinical characteristics and symptom severity scores among the GMM latent classes. Post hoc contrasts were done using the Bonferroni procedure to control the overall familywise alpha level of the three pairwise contrasts for the three GMM classes at 0.05. For any of the three pairwise contrasts, a p value of <0.017 (0.05/3) was deemed statistically significant.

Results

Patient characteristics

As summarized in Table 1, the majority of the sample was White (64.1%) and well educated (15.7 ± 2.7 years). Approximately, 24% lived alone, and 42% were married or partnered.
Table 1

Differences in baseline demographic and clinical characteristics among the three latent classes

Characteristic

Total sample (n = 398)

Low sustained (1)

Decreasing (2)

High sustained (3)

Omnibus statistics and post hoc comparisons

n = 158 (39.7%)

n = 21 (5.3%)

n = 219 (55.0%)

Mean (SD)

Mean (SD)

Mean (SD)

Mean (SD)

Age (years)

54.9 (11.6)

57.7 (12.1)

53.8 (9.8)

53.0 (10.9)

F(2, 395) = 7.95; p = 0.0004; 1 > 3

Education (years)

15.7 (2.7)

15.5 (2.6)

15.5 (2.1)

15.9 (2.7)

F(2, 390) = 0.98; p = 0.38

KPS score

93.2 (10.3)

96.5 (6.8)

92.9 (10.1)

90.9 (11.7)

F(2, 388) = 14.20; p ≤ 0.0001; 1 > 3

SCQ score

4.3 (2.8)

3.7 (2.4)

3.9 (2.7)

4.8 (3.1)

F(2, 394) = 7.11; p = 0.001; 3 > 1

Body mass index (kg/m2)

26.8 (6.2)

26.5 (5.8)

24.5 (4.6)

27.2 (6.5)

F(2, 389) = 2.15; p = 0.12

 

n (%)

n (%)

n (%)

n (%)

 

White

255 (64.1)

102 (65.0)

17 (81.0)

136 (62.4)

χ2 = 2.92; p = 0.23

Married/partnered

165 (41.5)

62 (39.5)

11 (52.4)

92 (42.6)

χ2 = 1.37; p = 0.51

Work for pay

189 (47.5)

86 (54.4)

10 (47.6)

93 (43.1)

χ2 = 4.73; p = 0.09

Lives alone

95 (23.9)

37 (23.7)

8 (38.1)

50 (23.1)

χ2 = 2.36; p = 0.31

Gone through menopause

248 (62.3)

104 (68.0)

14 (66.7)

130 (61.0)

χ2 = 1.93; p = 0.38

Experiencing hot flashes

127 (31.9)

37 (23.4)

4 (19.0)

86 (39.3)

χ2 = 12.30; p = 0.002; 3 > 1

Stage of disease

0

73 (18.3)

25 (15.8)

8 (38.1)

40 (18.3)

χ2 = 11.83; p = 0.07

I

151 (37.9)

72 (45.6)

5 (23.8)

74 (33.8)

IIA, IIB

141 (35.4)

50 (31.6)

7 (33.3)

84 (38.4)

IIIA, IIIB, IIIC, IV

33 (8.3)

11 (7.0)

1 (4.8)

21 (9.6)

Surgical treatment

Breast-conserving

318 (79.9)

131 (82.9)

10 (47.6)

177 (80.8)

χ2 = 14.63; p = 0.001; 2 > 1, 3

Mastectomy

80 (20.1)

27 (17.1)

11 (52.4)

42 (19.2)

Sentinel node biopsy

328 (82.4)

138 (87.3)

16 (76.2)

174 (79.5)

χ2 = 4.53; p = 0.10

Axillary lymph node dissection

149 (37.4)

52 (32.9)

5 (23.8)

92 (42.2)

χ2=5.15; p = 0.08

Breast reconstruction at the time of surgery

86 (21.6)

30 (19.1)

11 (52.4)

45 (20.5)

χ2 = 12.44; p = 0.002; 2 > 1, 3

Neoadjuvant chemotherapy

79 (19.8)

27 (17.1)

4 (19.0)

48 (22.0)

χ2 = 1.41; p = 0.50

Adjuvant chemotherapya

133 (33.4)

43 (27.2)

4 (19.0)

86 (39.3)

χ2 = 8.05; p < 0.02; 3 > 1

Adjuvant radiation therapyb

224 (56.3)

99 (62.7)

8 (38.1)

117 (53.4)

χ2 = 6.16; p < 0.05

KPS Karnofsky Performance Status, SCQ Self-Administered Comorbidity Questionnaire, SD standard deviation

aPercentage of patients in each latent class (i.e., 1, 2, or 3) who were receiving adjuvant chemotherapy at each month (percent of patients (latent class)—month 1 (3.3% (1), 0.0% (2), 5.8% (3)), month 2 (31.8% (1), 23.8% (2), 25.5% (3)), month 3 (44.7% (1), 33.3% (2), 35.6% (3)), month 4 (32.9% (1), 19.0% (2), 30.5% (3)), month 5 (17.% (1), 5.3% (2), 19.2% (3)), and month 6 (16.2% (1), 5.9% (2), 17.6% (3))

bPercentage of patients in each latent class (i.e., 1, 2, or 3) who were receiving adjuvant radiation therapy at each month (percent of patients (latent class)—month 1 (4.6% (1), 4.8% (2), 8.2% (3)), month 2 (21.9% (1), 9.5% (2), 24.0% (3)), month 3 (24.0% (1), 14.3% (2), 35.1% (3)), month 4 (24.5% (1), 19.0% (2), 33.2% (3)), month 5 (21.5% (1), 21.1% (2), 33.5% (3)), and month 6 (15.5% (1), 11.8% (2), 24.2% (3))

GMM analysis

Three distinct latent classes of sleep disturbance trajectories were identified using GMM (Fig. 1). A three-class model was selected because its BIC was smaller than the two-class and four-class models. In addition, comparisons of the other fit indices supported the choice of the three-class model (Table 2).
https://static-content.springer.com/image/art%3A10.1007%2Fs00520-012-1381-3/MediaObjects/520_2012_1381_Fig1_HTML.gif
Fig. 1

GSDS trajectories for observed (actual) scores and estimated (predicted) scores for patients in each of the three predicted classes, as well as the mean GSDS scores for the total sample

Table 2

Fit indices for the GMM class solutions for general sleep disturbance scale total scores

GMM solution

LL

AIC

BIC

Entropy

BLRT (df)

VLMR (df)

1-Classa

−4,522.03

9,076.05

9,139.84

N/A

N/A

N/A

2-Class

−4,488.48

9,018.95

9,102.67

0.56

67.10* (5)

67.10** (5)

3-Classb

−4,473.49

8,998.99

9,102.63

0.71

29.97* (5)

29.97** (5)

4-Class

−4,471.97

9,007.93

9,135.50

0.56

5.63ns (5)

5.63ns (5)

Note that these models were estimated from GSDS scores that were linearly transformed into 10-point intervals (y/10) to reduce the size of variance components and improve estimation

ns not significant, GMM growth mixture model, LL log likelihood, AIC Akaike Information Criterion, BIC Bayesian Information Criterion, BLRT parametric bootstrapped likelihood ratio test for K-1 (H0) vs K classes, VLMR Vuong-Lo-Mendell-Rubin likelihood ratio test for K-1 (H0) vs K classes

*p < 0.00005; **p ≥ 0.01

aLatent growth curve model with linear and quadratic components; chi2 = 43.72, 19 df, p < 0.001, CFI = 0.99, RMSEA = 0.057

bThe 3-class model was selected

As shown in Table 3, the majority of the patients were classified into the high sustained class (55.0%). These patients had total GSDS scores that were high at baseline (58.0 ± 2.7) and maintained a similar level of sleep disturbance throughout the study. Patients in the second largest class, the low sustained class (39.7%), had total GSDS scores that were low at baseline (32.5 ± 3.6) and maintained a similar level of sleep disturbance throughout the study. Finally, patients in the third class, the decreasing class (5.3%), had high preoperative GSDS scores (62.4 ± 4.2) that decreased during months 1 through 3 and then stabilized during months 4 through 6 after surgery.
Table 3

Parameter estimates for predicted growth mixture model latent classes from seven assessments of the GSDS

Parameter estimatesa

Low sustained

Decreasing

High sustained

n = 158 (39.7%)

n = 21 (5.3%)

n = 219 (55.0%)

Mean (S.E.)

Mean (S.E.)

Mean (S.E.)

Intercept

32.5 (3.6)*

62.4 (4.2)*

58.0 (2.7)*

Linear slope

0.70 (0.8)ns

−14.8 (2.2)*

1.6 (0.7)**

Quadratic slope

−0.19 (0.1)ns

1.4 (0.3)*

−0.28 (0.1)**

Variancesb

Intercept

104.4 (47.8)**

28.9 (13.3)**

318.4 (42.9)*

Linear slope

0b

0b

47.7 (18.4)***

Quadratic slope covariance

0b

0b

1.4 (0.5)***

Trajectory group sizes are for classification of individuals based on their most likely latent class probabilities

ns not significant, S.E. standard error

*p ≤ .005; **p < 0.05; ***p ≤ 0.01

aGrowth mixture model estimates were obtained with robust maximum likelihood

bFixed at zero to aid in model convergence

Differences in demographic and clinical characteristics

As shown in Table 1, patients in the high sustained class were significantly younger, had lower KPS scores, higher SCQ scores, and were more likely to report hot flashes than those in the low sustained class. In addition, a significantly higher proportion of women who underwent a mastectomy or breast reconstruction at the time of surgery were in the decreasing class compared to both the low and high sustained classes. While significant differences were found among the latent classes in the proportion of patients who received adjuvant CTX and RT, post hoc contrasts only revealed a significant difference for adjuvant CTX. A significantly higher proportion of women who underwent adjuvant CTX were in the high sustained class compared to the low sustained class.

Differences in preoperative sleep disturbance scores

As shown in Fig. 2, differences among the latent classes were found for all of the GSDS subscale scores, as well as for the total sleep disturbance score (all p ≤ 0.001). Post hoc contrasts demonstrated that the decreasing and high sustained classes had significantly higher sleep quality, sleep onset latency, sleep quantity, mid-sleep awakenings, early awakenings, and excessive daytime sleepiness scores compared to the low sustained class (all p ≤ 0.001). In addition, the decreasing class had a significantly higher sleep onset latency score than the high sustained class (p = 0.005). The high sustained class had higher medications for sleep scores than the low sustained class (p ≤ 0.001). Finally, the decreasing and high sustained classes had higher total sleep disturbance scores than the low sustained class (both p ≤ 0.001).
https://static-content.springer.com/image/art%3A10.1007%2Fs00520-012-1381-3/MediaObjects/520_2012_1381_Fig2_HTML.gif
Fig. 2

Differences in total and subscale scores on the GSDS, prior to surgery, among the three predicted classes. All values are plotted as means ± standard deviations. For the quality, quantity, mid-sleep awakenings, early awakenings, and excessive daytime sleepiness subscales scores, as well as the total GSDS scores, post hoc contrasts demonstrated that low sustained class < decreasing and high sustained classes (all p ≤ 0.001). For sleep medication subscale scores, post hoc contrasts demonstrated that low sustained class < high sustained class (p < 0.001). For sleep onset latency subscale scores, post hoc contrasts demonstrated that low sustained class < high sustained class < decreasing class (both p ≤ 0.005)

Differences in preoperative fatigue, energy, and attentional fatigue scores

Significant differences in physical fatigue, energy, and attentional fatigue scores (all p < 0.001) among the latent classes are shown in Fig. 3a. The decreasing and high sustained classes had higher mean fatigue scores compared to the low sustained class (both p ≤ 0.01). The high sustained class had lower mean energy scores compared to the low sustained class (p ≤ 0.001). The decreasing and high sustained classes had lower mean AFI scores (i.e., higher levels of attentional fatigue) compared to the low sustained class (both p ≤ 0.01).
https://static-content.springer.com/image/art%3A10.1007%2Fs00520-012-1381-3/MediaObjects/520_2012_1381_Fig3_HTML.gif
Fig. 3

a Differences in physical fatigue, energy, and attentional fatigue scores prior to surgery, among the three predicted classes. All values are plotted as means ± standard deviations. For physical fatigue, post hoc contrasts demonstrated that low sustained class < decreasing and high sustained GSDS classes (both p < 0.007). For energy, post hoc contrasts demonstrated that low sustained class > high sustained class (p ≤ 0.001). For attentional fatigue, post hoc contrasts demonstrated that low sustained class > decreasing and high sustained classes (i.e., higher attentional function scores indicate lower levels of attentional fatigue, both p ≤ 0.002). b Differences in trait anxiety, state anxiety, and depressive symptoms, prior to surgery, among the three predicted classes. All values are plotted as means ± standard deviations. For trait anxiety, post hoc contrasts demonstrated that low sustained class < high sustained class (p < 0.001). For state anxiety and depression, post hoc contrasts demonstrated that low sustained class < decreasing and high sustained classes (both p < 0.001)

Differences in preoperative depression and anxiety

Significant differences in levels of trait and state anxiety, as well as depressive symptoms (all <0.001), among the latent classes are shown in Fig. 3b. The decreasing and high sustained classes had higher trait anxiety scores compared to the low sustained class. Only the high sustained class had higher state anxiety scores compared to the low sustained class (all p < 0.001). The decreasing and high sustained classes had higher CES-D scores compared to the low sustained class (both p < 0.001).

Discussion

To our knowledge, this study is the first to use GMM to identify three subgroups of breast cancer patients with distinct sleep disturbance trajectories prior to through 6 months after surgery. Although specific sleep disorders (e.g., insomnia, delayed sleep phase disorder) could not be diagnosed, 55% of these women reported total GSDS scores that were well above the clinically meaningful cutoff score of ≥43 [23] for over 6 months. In addition, women in the decreasing class had an elevated sleep disturbance score prior to surgery that persisted through the first month after surgery. Equally important, ∼40% of these patients had relatively low levels of sleep disturbance prior to and following surgery. If more traditional approaches for longitudinal data analysis were used, these distinct sleep disturbance phenotypes would have been missed.

Women who were younger, had poorer functional status and a worse level of comorbidity, and who reported hot flashes were more likely to report higher and persistent levels of sleep disturbance. These associations are consistent with previous reports. For example, in two studies [35, 42], younger oncology patients reported higher levels of sleep disturbance. In addition, studies of heterogeneous samples of cancer patients found that poorer functional status was associated with higher levels of sleep disturbance [2, 10, 13, 16, 17]. Finally, an association between hot flashes and sleep disturbance was reported in breast cancer patients during [30] and after treatment [45].

The finding that women who underwent a mastectomy and/or had reconstruction to the affected breast were more likely to be in the decreasing class suggests that specific surgical procedures have an influence on women's sleep patterns prior to and in the immediate postsurgical period. One possible explanation for this finding is that this group of women were more anxious about the surgical intervention they are about to undergo. This hypothesis is supported by the finding that the decreasing class reported significantly higher state anxiety scores than the low sustained class. In addition, based on an analysis of a single item from the quality of life questionnaire that was used in this study [18, 21], women in the decreasing class reported more difficulty coping with their disease and treatment (5.7 ± 2.9; 0 = not at all difficult to 10 = extremely difficult) than the low sustained (2.3 ± 2.4) and high sustained (3.8 ± 2.7; both p ≤ 0.006) classes. However, once the surgery and initial recovery were completed, sleep disturbance in these women resolved, unlike the women in the high sustained class.

Both groups of women who reported higher levels of sleep disturbance experienced higher levels of physical fatigue, attentional fatigue, depression, and anxiety and lower levels of energy than women in the low sustained class. Severity scores were above the cutoff scores for clinically meaningful levels of these co-occurring symptoms and are consistent with previous reports. For example, higher levels of physical fatigue and sleep disturbance were reported in breast cancer patients before CTX [3, 26, 30, 55] and RT [30]. Higher levels of attentional fatigue were associated with higher levels of sleep disturbance in patients prior to the initiation of RT for breast cancer [34]. The relationship between depressive symptoms and sleep disturbance is consistent with reports of breast cancer patients before [1], during [13, 30, 53], and after adjuvant treatment [14, 53]. In one of the few studies that evaluated anxiety and sleep disturbance in women prior to surgery for breast cancer [56], a positive association was found between these two symptoms. Findings from this study suggest that high levels of sleep disturbance co-occur with a number of symptoms in over 60% of women who are about to undergo breast cancer surgery. These women may represent a particularly vulnerable group who require interventions to deal with multiple symptoms before and following their initial treatment.

In terms of the sample's mean total GSDS score prior to surgery, this score was comparable to scores reported by patients prior to the initiation of RT for breast cancer (i.e., 44) [15], as well as by outpatients undergoing a variety of cancer treatments (i.e., 54.7 [35] and 51.2 [42]). However, patients in this study had higher GSDS scores than women prior to (i.e., 42.3) and 6 weeks after hysterectomy (i.e., 39.1) [31].

Women with higher levels of sleep disturbance had problematic sleep disturbance parameters prior to surgery compared to women with lower levels of sleep disturbance (see Fig. 2). These findings suggest that women in the high sustained class had problems with sleep initiation, as well as maintenance. The reports of increases in sleep onset latency and mid-sleep awakenings are similar to findings reported by women undergoing CTX [3, 4, 47] and RT [15] for breast cancer. In addition, women in the decreasing class had worse sleep quality, quantity, onset latency, mid-sleep awakenings, early awakenings, excessive daytime sleepiness, and overall sleep disturbance prior to surgery than women in the low sustained class. Interestingly, women in the decreasing class had even worse sleep onset latency scores prior to surgery than women in the high sustained class. This finding suggests that these women had a particularly difficult time with the initiation of sleep that may warrant an intervention.

Several study limitations need to be acknowledged. Although a valid and reliable measure was used to evaluate sleep disturbance, no objective evaluation was done to corroborate the patients' self-reports. Although this study had an ample sample size to obtain meaningful, reliable estimates of the latent classes, the decreasing class was quite small (n = 21). While significant differences were found for some of the variables between the decreasing class and the other two classes, this small sample size may have led to an underestimation of differences among the classes. In addition, with a larger sample size, additional latent classes may be identified (e.g., a class where initial levels of sleep disturbance are low and then increase over time). Therefore, findings warrant replication in future studies. Finally, patients provided self-report information at each assessment on the receipt of adjuvant treatments. Therefore, the specific effects of each treatment on sleep disturbance cannot be evaluated.

Despite these limitations, these findings have implications for clinical practice and research. Clinicians need to be aware that a relatively high percentage of women with sleep disturbance prior to surgery for breast cancer will continue to have sleep problems during subsequent treatments. Women who will undergo a mastectomy, undergo reconstruction, are younger, and are experiencing hot flashes may represent a particularly high-risk group of women who require a sleep intervention prior to surgery. Clinicians need to perform routine assessments of sleep disturbance prior to surgery and institute appropriate interventions [20].

Additional studies are needed to determine how long sleep disturbance persists in the women in the high sustained class. Future studies need to determine additional phenotypic and genotypic predictors that distinguish among the latent classes. Additional research is warranted to determine the mechanisms that underlie the high as well as the low levels of sleep disturbance. Finally, interventions need to be developed and tested to alleviate sleep disturbance and its associated symptoms.

Acknowledgments

This study was funded by grants from the National Cancer Institute (CA107091 and CA118658). Dr. Bradley Aouizerat was funded through the National Institutes of Health (NIH) Roadmap for Medical Research Grant (KL2 RR624130). Dr. Dunn received funding from the Mount Zion Health Fund. Dr. Christine Miaskowski is an American Cancer Society Clinical Research Professor. This project is supported by NIH/NCRR UCSF-CTSI Grant Number UL1 RR024131. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NIH.

Conflicts of interest

None.

Copyright information

© Springer-Verlag 2012