620 Journal of Pain and Symptom Management Vol. 45 No. 3 March 2013

Brief Methodological Report

Factor Analysis of the Pittsburgh SleepQuality Index in Breast Cancer SurvivorsJulie L. Otte, PhD, RN, Kevin L. Rand, PhD, Janet S. Carpenter, PhD, RN, FAAN,Kathleen M. Russell, PhD, RN, and Victoria L. Champion, PhD, RN, FAANSchool of Nursing (J.L.O., J.S.C., K.M.R., V.L.C.), Indiana University, Indianapolis, and

Department of Psychology (K.L.R.), Indiana University-Purdue University, Indianapolis,

Indiana, USA

Abstract

Context. Sleep is a significant problem in breast cancer survivors (BCS) and

measured frequently using the Pittsburgh Sleep Quality Index (PSQI). Thus, it isimportant to evaluate its factor structure. The two-process model of sleepregulation was the theoretical framework for this study.

Objectives. To perform a confirmatory factor analysis of the PSQI in BCS andcompare results between African-American and Caucasian BCS.

Methods. This was a secondary analysis of cross-sectional data using local andregional health care facilities and Eastern Cooperative Oncology Group referrals.The study included 1174 nondepressed BCS (90% Caucasian), with a mean age of57 years andmedian PSQI global scores at the cutoff for poor sleep (median¼ 6.00,interquartile range¼ 4.00e9.00). Measurements included self-reporteddemographics, medical history, depression, and sleep.

Results. Acceptable fit was not reached for the traditional one-factor model thatwould be consistent with current PSQI scoring or for alternative models in thepublished literature from other populations. A new two-factor model (i.e., sleepefficiency and perceived sleep quality) best fit the data but nested-modelcomparisons by race showed different relationships by race for 1) sleep quality-sleep latency and 2) sleep efficiency-sleep quality.

Conclusion. Results were inconsistent with current PSQI scoring that assumesa single global factor and with previously published literature. Although a newtwo-factor model best fit the data, further quantitative and qualitative analyses arewarranted to validate our results in other populations before revising PSQIscoring recommendations. Additional recommendations are described forresearch. J Pain Symptom Manage 2013;45:620e627. � 2013 U.S. Cancer PainRelief Committee. Published by Elsevier Inc. All rights reserved.

Key Words

Sleep, breast cancer survivor, measurement of sleep, factor analysis

Address correspondence to: Julie L. Otte, PhD, RN,School of Nursing, 1111 Middle Drive, NU W401,Indianapolis, IN 46202, USA. E-mail: jlelam@iupui.edu

Accepted for publication: March 9, 2012.

� 2013 U.S. Cancer Pain Relief Committee.Published by Elsevier Inc. All rights reserved.

0885-3924/$ - see front matterhttp://dx.doi.org/10.1016/j.jpainsymman.2012.03.008

Vol. 45 No. 3 March 2013 621Factor Analysis of the PSQI

Introduction need to be revised to capture the essential7e9

Current evidence suggests that 48%e65% ofbreast cancer survivors (BCS) report poorsleep,1,2 a higher incidence compared withage-matched midlife women.1,3,4 Significantcorrelates of sleep disturbances in BCS includeestrogen depletion, poor physical functioning,depression, and/or distress.1 Additionally, ra-cial disparities exist, with African-AmericanBCS being three times more likely to reportpoor sleep.1 Sleep disturbances are importantto address because, if left untreated, they cannegatively affect quality of life.

Most evidence about sleep in BCS has beenbased on the Pittsburgh Sleep Quality Index(PSQI) questionnaire that captures both clini-cal and behavioral information about sleepthrough a single global score,5,6 which is typi-cally the outcome variable reported. ThePSQI is widely accepted because of its estab-lished psychometrics in BCS and otherpopulations.6

It remains unclear if the PSQI capturesall attributes of sleep. Within the past five years,several studies have shown that multifactormodels for the PSQI might be moreaccurate.7e9 Cole et al.9 conducted an explor-atory factor analysis (EFA) and a confirmatoryfactor analysis (CFA) of the PSQI in a sampleof depressed and nondepressed men (45%)and women (55%) aged 60 years or older. TheirEFA showed that a two-factor model (i.e., sleepefficiency and perceived sleep quality) best fitthe data. In contrast, their CFA found thata three-factor model of sleep efficiency, per-ceived sleep quality, and daily disturbancesbest fit the data.9 The authors recommendedthe use of a three-factor structure to betterreflect sleep disturbances. This three-factorstructure was supported in two additional stud-ies: one used principal components analysis ofdata from 520 Nigerian university students7

and the other was done in 135 postrenal trans-plant patients.8 These studies were based onrelatively small samples (n¼ 135 and 520) thatincluded both men and women, with a widerange of ages.

Evaluating the factor structure of the PSQIhas important implications for its use in de-scriptive and intervention research. If theglobal score is not the most accurate depictionof sleep, scoring for the questionnaire might

attributes of sleep and treatment efficacy.The current scoring of the PSQI produces

a single summed score that does not reflectthe multiple dimensions of poor sleep as sug-gested by the two-process model of sleep regu-lation.10 This model postulates that there aremultiple physiological processes that drivesleep and wakefulness; if one process is disrup-ted, sleep and/or wakefulness are affected. Be-cause sleep is a significant problem in BCS andfrequently measured using the PSQI, it isimportant to evaluate its factor structure to de-termine which of the commonly used or previ-ously reported factor structures (one-, two-, orthree-factor model) is most relevant for thispopulation and in line with the theoreticalframework.

The previously reported one-, two-, andthree-factor models are the driving constructand basis for this study. In addition, becausesleep in BCS varies by race,1 it is importantto know if the factor structure is differentbetween African-American and CaucasianBCS. This information would maximize knowl-edge of sleep disturbances in both Caucasianand African-American BCS. Therefore, thepurpose of this study was to 1) perform aCFA of the PSQI in a large sample of BCSand 2) compare the factor structure betweenAfrican-American and Caucasian BCS. Thecorresponding research questions were 1)will the CFA of the PSQI in a large sample ofBCS support the use of a one-, two-, or three-factor model and 2) is the resulting factorstructure consistent between the African-American and Caucasian BCS? The hypothesiswas that the factor structure will differ fromprevious research, which did not includewomen with cancer. It also was hypothesizedthat the factor structure should be similar be-tween African-American and Caucasian BCS.

MethodsProcedures

PSQI responses from two separate cross-sectional, descriptive, comparative, parentstudies were evaluated. Data were collected ata single point in time for both studies. Subjectswho completed the PSQI and other question-naires were included in the analyses. The

622 Vol. 45 No. 3 March 2013Otte et al.

parent studies were reviewed and approved bythe institutional review board at the universityand the cancer center scientific review commit-tee. Both studies used similar local recruitmentstrategies by trained research assistants. One ofthe two parent studies used regional referralsfrom the Eastern Cooperative OncologyGroup and a southeastern university. Eligibleand willing BCS who signed consents weregiven a questionnaire packet in person orthrough the mail. The eligibility criteria foreach of the two studies were similar and in-cluded women who were aged at least 18 years;Stages IeIII at diagnosis; cancer free at thetime of study enrollment; no history of othercancers; able to read, write, and speak English;and at least two years but no more than 10years post-completion of surgery, radiation,and/or chemotherapy. A total of 1174 BCSwere identified from the two studies (Study 1,n¼ 130; Study 2, n¼ 1044) for this analysis.

MeasuresDemographic and other comparative vari-

ables were assessed. Variables thought to im-pact sleep were evaluated, such as age, bodymass index (BMI), marital status, years of edu-cation, menopausal status, and hot flashes.Sample characteristics were collected usingan adapted multi-item questionnaire11 thatcaptured race, age, marital status, educationaland income levels, employment status, height,and weight. Information about breast cancerand treatment was obtained from abstractingmedical records using a standardized form.Data included date and stage at diagnosis,types and dates of treatment (surgery, chemo-therapy, and radiation), and endocrine ther-apy. Frequency and severity of menopausalsymptoms such as the frequency of hot flashwere measured using a generic 14-itemquestionnaire.

Depressive symptoms were assessed usingthe Center for Epidemiologic Studies Depres-sion Scale (CES-D), a 20-item self-report instru-ment assessing the presence and severity ofdepressive symptoms occurring over the pastweek.12 Scores of 16 or greater are indicativeof high depressive symptoms but are not con-sidered a diagnosis of clinical depression.13,14

This cutoff point of 16 has been used exten-sively in other studies, including studies ofBCS.6 Internal consistency reliability has been

shown to be adequate in cancer and non-cancer populations, with Cronbach’s alpha co-efficients of >0.85.15

Sleep disturbances were assessed using thePSQI,5 a 19-item scale producing a global sleepquality index score based on seven componentscores: sleep quality, sleep latency, sleep dura-tion, habitual sleep efficiency, sleep distur-bance, use of sleep medications, and daytimedysfunction. Global scores range from 0 to21, and scores above 5 reflect poor global sleepquality.5 Psychometrics of the PSQI have beensupported in a variety of populations,16,17 in-cluding noncancer populations (n¼ 52;a¼ 0.83)5 and BCS (n¼ 102; a¼ 0.80).6 Con-struct validity also has been demonstrated.6

Statistical AnalysisFrequency anddescriptive statistics were used

to analyze demographic variables using SPSSversion 17.0 (SPSS, Inc., Chicago, IL). Chi-square and t-tests were used to examine poten-tial differences between African-American andCaucasian BCS on the following variables: age,marital status, years of education, cancer treat-ment received (e.g., chemotherapy), use of hor-monemodulators, number of hot flashes in thepast month, sleep disturbance (PSQI globalscore), and depressive symptoms (CES-D).Responses to the PSQI were scored into

seven components,5 and 95.7% of the samplehad complete data on all composite scores.The greatest missingness was for habitual sleepefficiency, with 3.52% of the sample missingdata on this item. Because of the small propor-tion of missing data, PRELIS 2.818 (ScientificSoftware International, Inc., Chicago, IL) wasused to impute missing values by matchingwith other variables on the scale.19 Listwise de-letion was used to eliminate cases for whichmissing values could not be imputed, resultingin a final sample size of 1172.A CFA using LISREL 8.818 (Scientific Soft-

ware International, Inc.) was conducted. Be-cause the single global score of the PSQI isfrequently used in the literature, we examinedthe fit of a one-factor model. Also, because therecent study by Cole et al.9 found support forboth a two-factor model (in their EFA) anda three-factor model (in their CFA) of thePSQI, we decided to examine the fit of boththese models in the present sample. Giventhat parsimonious models are preferred to

Vol. 45 No. 3 March 2013 623Factor Analysis of the PSQI

more complex ones, we examined thesemodels in the order of increasing complexity.

The data from the PSQI are ordinal ratherthan continuous and should not be analyzedusing methods that assume they have metricproperties.19 Therefore, PRELIS 2.818 wasused to estimate the polychoric correlationsamong the indicators and their asymptotic co-variance matrix.19 This matrix was then used toestimate factor models using weighted leastsquares.

Several indices were used to evaluate modelfit: 1) Chi-square corrected for non-normality(non-normal c2), 2) standardized root meansquared residual (SRMR), 3) root mean squareerror of approximation (RMSEA), and 4) com-parative fit index (CFI). The non-normal c2

and the SRMR are indices of absolute fit. Anonsignificant c2 is evidence of acceptablefit. Smaller values of SRMR indicate bettermodel fit, with values #0.08 indicating accept-able fit.20 The RMSEA is a commonly used par-simonious index, with values #0.06 indicatingacceptable fit.20 The CFI is an incremental fitindex, with values $0.95 indicating acceptablefit.20 Models are identified according to the au-thors of origin to provide clarification.

ResultsSample Characteristics

Most BCS were non-Hispanic (98%), Cauca-sian (90%), partnered (73%), well educated(39% some college), and postmenopausal(85%). Most had had some type of cancer sur-gery (99%), chemotherapy (96%), radiotherapy

Table 1Correlation Matrix and Descriptive S

Indicators 1 2

1. Subjective sleep quality d 0.502. Sleep latency d3. Sleep duration4. Habitual sleep efficiency5. Sleep disturbances6. Sleep medication use7. Daytime dysfunctionMean 0.97 1.14SD .75 1.02Median 1.00 1.00Interquartile range (75&e25&) 1.00 2.00

PSQI¼ Pittsburgh Sleep Quality Index.All correlations are significant at P< 0.05. Data were based on final imputed

(93%), andhotflashes in thepastmonth(52%),and 35% were taking hormone modulators.Mean (SD) age was 57 (11.42) years (range28e79 years), and mean (SD) BMI was 28.32(6.08). BCS were not depressed (CES-D scoresmean [SD] 9.74 [8.92]), but the median PSQIglobal score was at the cutoff for poor sleep (me-dian 6.00; interquartile range 9.00e4.00; mean[SD] 6.46 [3.75]). African-American BCS hadsignificantly higher BMI (mean 31.42 vs.27.97), were significantly more depressed(mean 14.42 vs. 9.18), reported significantlypoorer sleep (mean 8.25 vs. 6.26; medians8.00 vs. 6.00; interquartile ranges 12.00e5.00vs. 9.00e3.00), and were not partnered (34%vs. 77%) compared with Caucasian BCS. Thecorrelation matrix and descriptive statistics forthe PSQI indicators are shown in Table 1.

Confirmatory Factor AnalysisFirst, the one-factor model fit was tested

(global score). This single-factor model didnot show acceptable fit, non-normal c2

ðdf¼14Þ ¼199.88, P< 0.05, SRMR¼ 0.071, RMSEA¼0.105, and CFI¼ 0.96. Second, a two-factormodel was tested consistent with the EFA resultsof Cole et al.9 This model did not show accept-able fit to the data, non-normal c2

ðdf¼13Þ ¼89.70, P< 0.05, SRMR¼ 0.0048, RMSEA¼0.075, and CFI¼ 0.98. Third, the fit of theCole et al. three-factormodel consisting of sleepefficiency, perceived sleep quality, anddaily disturbances was tested. This model didnot show acceptable fit, non-normal c2

ðdf¼11Þ ¼80.32, P< 0.05, SRMR¼ 0.0044, RMSEA¼0.076, and CFI¼ 0.98.

tatistics of PSQI Indicators

Correlations

3 4 5 6 7

0.39 0.44 0.45 0.29 0.390.25 0.45 0.36 0.34 0.21d 0.51 0.21 0.07 0.20

d 0.29 0.26 0.25d 0.20 0.32

d 0.18d

0.87 0.62 1.38 0.75 0.84.77 .94 .58 1.16 .74

1.00 0.00 1.00 0.00 1.001.00 1.00 1.00 1.00 1.00

data set. N¼ 1172.

624 Vol. 45 No. 3 March 2013Otte et al.

Next, modification indices were examined todetermine how acceptable model fit might bereached. Based on the modification indices,the error terms were correlated among the fol-lowing pairs of indicators: 1) sleep duration-subjective sleep quality, 2) sleep duration-sleepmedication use, 3) sleep latency-habitual sleepefficiency, 4) sleep latency-sleep medicationuse, and 5) sleep latency-daytime dysfunction.The resulting Otte et al. three-factor model(sleep efficiency, perceived sleep quality,and daily disturbances) showed acceptablefit to the data, non-normal c2

ðdf¼6Þ ¼ 3.34,P¼ 0.76, SRMR¼ 0.012, RMSEA¼ 0.000, andCFI¼ 1.00.

There was a high correlation between dailydisturbances and perceived sleep quality(r¼ 0.93). Hence, we conducted a nested-model comparison with this correlation fixedto 1.0 to see if a new Otte et al. two-factor solu-tion fit the data as well as the Otte et al. three-

Sleep

Efficiency

Habitual

Sleep

Efficiency

Sleep

Duration

.06 n.s. .58 .26

.66†

(.37A

, .71C

)

.64 .97 .86

Subjective

Sleep

Quality

Sleep

Latency

.51

.13

.08

.10

.70†

(.82 A

, .66 C

)

Fig. 1. Otte et al. two-factor model of the Pittsburgh Sleep QAll coefficients are standardized and significant at P< 0.05,ables. Rectangles represent measured variables. Single-headarrows represent correlations; italicized coefficients are errin the variable not accounted for by the model. yparameterAmerican and Caucasian subsamples; Aparameter estimate fofor Caucasian subsample.

factor model. This nested-model comparisonshowed no significant decrement in model fit,Dc2

ðdf¼1Þ ¼ 1.13, P¼ 0.29. Therefore, we exam-ined the overall fit of the new Otte et al.two-factor model with the correlated errorterms. This model showed acceptable fit, non-normal c2

ðdf¼8Þ ¼ 8.85, P¼ 0.36, SRMR¼ 0.019,RMSEA¼ 0.016, and CFI¼ 1.00. Hence, theOtte et al. two-factor model (sleep efficiencyand perceived sleep quality) best fit the dataamong BCS (Fig. 1). A one-factor model withthe correlated error terms did not showacceptable fit to the data, c2

ðdf¼10Þ ¼ 127.72,P< 0.001, SRMR¼ 0.058, RMSEA¼ 0.110, andCFI¼ 0.97.To test whether the same factor structure

was constant between African-American andCaucasian subsamples, several two-group com-parisons were conducted. First, all parameterswere constrained to be equal across bothgroups. This model showed mixed evidence

Perceived

Sleep Quality

Sleep

Disturbances

.56 .79

.66 .46 .55

Sleep

Medication

Use

Daytime

Dysfunction

.70

-.11

.11

uality Index sleep quality in breast cancer survivors.unless otherwise noted. Ovals represent latent vari-ed arrows represent factor loadings. Double-headedor terms, representing the proportion of variations were found to differ significantly between African-r African-American subsample; Cparameter estimate

Vol. 45 No. 3 March 2013 625Factor Analysis of the PSQI

for fit, non-normal c2ðdf¼37Þ ¼ 68.03, P¼ 0.003,

SRMR¼ 0.027, RMSEA¼ 0.039, and CFI¼0.99. Next, the factor loadings were freed, andthe model showed significantly improved fit, Dnon-normal c2

ðdf¼7Þ ¼ 14.42, P¼ 0.04, indicat-ing that the factor loadings differed betweenAfrican-American and Caucasian BCS.

A series of nested-model comparisons wereconducted freeing individual factor loadingsto determine which ones significantly differedbetween African-American and Caucasian BCS(Table 2). The factor loading of perceivedsleep quality onto sleep latency differed signif-icantly, with a factor loading of 0.82 and 0.66for African-American and Caucasian BCS(Fig. 1), respectively, indicating that the timerequired to fall asleep is a stronger indicatorof perceived sleep quality among African-American BCS compared with Caucasian BCS.

Next, the correlation between sleep effi-ciency and sleep quality was freed to varybetween groups, and the model showed signif-icantly improved fit, D non-normal c2

ðdf¼1Þ ¼6.64, P¼ 0.0099, indicating that the relation-ship between sleep efficiency and sleep qualitydiffered between African-American and Cauca-sian BCS. The relationship between sleep effi-ciency and perceived sleep quality was moremodest among African-American BCS (r¼0.37) than among Caucasian BCS (r¼ 0.71)(Fig. 1).

Finally, the correlations among residualswere freed to vary across the two groups. Thisdid not result in improved model fit, D non-normal c2

ðdf¼5Þ ¼ 9.91, P¼ 0.078. Hence, thecorrelated residuals were held invariantacross groups. The final Otte et al. two-groupmodel with the factor loading from perceivedsleep quality to sleep latency freed andthe correlation between sleep efficiency and

Table 2Chi-Square Difference Tests Comparing Factor Loadin

Caucasian Breast Can

Factors Indicators

Sleep efficiency Sleep duration 0Habitual sleep efficiency 0

Perceived sleep quality Subjective sleep quality 0Sleep latency 0Sleep disturbances 0Sleep medication use 0Daytime dysfunction 0

PSQI¼ Pittsburgh Sleep Quality Index; lA¼ factor loading in African AmeriBold indicates significance of P < 0.05.

perceived sleep quality freed showed good fitto the data, non-normal c2

ðdf¼35Þ ¼ 51.25,P¼ 0.037, SRMR¼ 0.028, RMSEA¼ 0.026,and CFI¼ 1.00.

DiscussionThree important findings emerged. First,

the study supports previous results that thePSQI single global score derived by the sevencomponents may not be consistent with thefactor structure of this questionnaire. Thisfinding is supported by the theoretical frame-work because sleep is not a unidimensionalproblem. The single-factor model from thisstudy produced an average standardized load-ing of 0.56 for each of the component scores,which was lower compared with that in previ-ous studies.8,9

Second, Cole et al.’s three-factor model fitthe data of BCS better than a single-factormodel, but the correlation between the latentvariables of perceived sleep quality anddaily disturbances was 0.93, which is highercompared with two previous studies (r¼0.75e0.81).8,9 Because this strong correlationbetween two variables suggests a single factor,the Otte et al. two-factor model of sleep effi-ciency and perceived sleep quality fit the dataas well as the Otte et al. three-factor model.These findings suggest that, before alteringPSQI scoring, further exploration of the factorstructure is needed to determine if the factorstructure varies by the gender, race, or othercharacteristics of various populations.

Third, the overall factor models were consis-tent between African-American and CaucasianBCS with two important differences. First,there was a weaker relationship between the

gs on PSQI Indicators in African-American vs.cer Survivors

lA lC Dc2 df P

.57 0.62 0.09 1 0.764

.99 0.96 0.68 1 0.410

.83 0.87 0.01 1 0.920

.82 0.66 7.00 1 0.008

.73 0.63 0.43 1 0.512

.53 0.46 1.76 1 0.185

.66 0.53 2.32 1 0.128

can subsample; lC¼ factor loading in Caucasian subsample.

626 Vol. 45 No. 3 March 2013Otte et al.

latent variables of sleep efficiency and per-ceived sleep quality in African-American BCS.Second, sleep latency was a stronger indicatorof perceived sleep quality among African-American BCS than among Caucasian BCS.It has been noted in previous studies thatAfrican-American BCS perceive aspects ofsurvivorship and quality of life differentlythan Caucasian BCS.21 This difference in per-ception has been reflected in how African-American BCS respond to questionnaire items.African-American BCS could be answeringPSQI items based on a different understand-ing of the meaning of ‘‘time in bed’’ or ‘‘timeto fall asleep.’’ For example, African-AmericanBCS might spend time in bed relaxing beforefalling asleep. This might result in calculationsof sleep efficiency and sleep latency that donot match perceptions of sleep quality amongCaucasian BCS. It also is important to notethat African-American BCS reported more de-pressive symptoms in this study, which couldhave influenced question responses.

In science, parsimony is desirable; the Otteet al. two-factor model was used for the secondpart of the study comparing the factor struc-ture of African-American and Caucasian BCS.The sample size of African-American BCS wassmaller than that of Caucasian BCS. However,because African-American BCS are at higherrisk for sleep problems than Caucasian BCSand there were similar numbers of African-American BCS subjects compared with previ-ous studies, the benefit of this informationwould outweigh the limitation of the smallerAfrican-American BCS sample size and we pro-ceeded with caution in testing racial differ-ences in the factor structure.

Limitations of the study include the follow-ing: 1) this was a cross-sectional study; 2) BCSmight differ from newly diagnosed, in treat-ment, or completing treatment BCS; 3) thiswas a heterogeneous sample of BCS; and 4) re-sults could be related to being female orhaving a history of breast cancer, limiting gen-eralization of findings to males or other cancersurvivors.

In sum, this study provides new informationabout the factor structure of the PSQI in BCSwith results that do not match the previouslypublished results and do not reflect the multi-dimensionality of sleep problems as suggestedin the two-process model of sleep regulation.

Further quantitative and qualitative analysesare needed of individual sleep items withinthe questionnaire to examine the existing fac-tor structure of the PSQI. Alternative scoringwould include revisiting the content of eachitem and matching with the specific constructof sleep problems, revising the scoring to re-duce the complexity and subscales produced,and performing additional reliability and val-idity testing to ensure the overall measure is re-flective of poor sleep.

Disclosures and AcknowledgmentsFunding was provided by the American Can-

cer Society grant no. RSGPB-04-089-01-PBP(V. L. C.), National Institutes of Health/National Cancer Institute no. R03CA97737(K. M. R.), and the Walther Cancer InstituteFoundation (K. M. R.). One of the parent stud-ies (V. L. C.) was conducted by the Eastern Co-operative Oncology Group (Robert L. Comis,MD) and supported in part by Public HealthService grants CA21115 and CA37604 andfunding from the National Cancer Institute Di-vision of Cancer Prevention, National Institutesof Health, and the Department of Health andHuman Services. Study contents are solely theresponsibility of the authors and do not neces-sarily represent the official views of theNationalCancer Institute. The authors have no disclo-sures or conflicts of interest to report.The authors thank Dr. Phyllis Dexter for her

editorial support.

References1. Otte JL, Carpenter JS, Russell KM, Bigatti S,Champion VL. Prevalence, severity, and correlatesof sleep-wake disturbances in long-term breast can-cer survivors. J Pain Symptom Manage 2010;39:535e547.

2. Carpenter JS, Andrykowski MA. Menopausalsymptoms in breast cancer survivors. Oncol Nurs Fo-rum 1999;26:1311e1317.

3. Owens JF, Matthews KA. Sleep disturbance inhealthy middle-aged women. Maturitas 1998;30:41e50.

4. Owen DC, Parker KP, McGuire DB. Comparisonof subjective sleep quality in patients with cancerand healthy subjects. Oncol Nurs Forum 1999;26:1649e1651.

Vol. 45 No. 3 March 2013 627Factor Analysis of the PSQI

5. Buysse DJ, Reynolds CF 3rd, Monk TH,Berman SR, Kupfer DJ. The Pittsburgh Sleep Qual-ity Index: a new instrument for psychiatric practiceand research. Psychiatry Res 1989;28:193e213.

6. Carpenter JS, Andrykowski MA. Psychometricevaluation of the Pittsburgh Sleep Quality Index.J Psychosom Res 1998;45:5e13.

7. Aloba OO, Adewuya AO, Ola BA, Mapayi BM.Validity of the Pittsburgh Sleep Quality Index(PSQI) among Nigerian university students. SleepMed 2007;8:266e270.

8. Burkhalter H, Sereika SM, Engberg S, et al.Structure validity of the Pittsburgh Sleep Quality In-dex in renal transplant recipients: a confirmatoryfactor analysis. Sleep Biol Rhythms 2010;8:274e281.

9. Cole JC, Motivala SJ, Buysse DJ, et al. Validationof a 3-factor scoring model for the Pittsburgh SleepQuality Index in older adults. Sleep 2006;29:112e116.

10. Borbely AA. A two process model of sleep regu-lation. Hum Neurobiol 1982;1:195e204.

11. Giesler RB, Given B, Given CW, et al. Improvingthe quality of life of patients with prostate carcinoma:a randomized trial testing the efficacy of a nurse-driven intervention. Cancer 2005;104:752e762.

12. Radloff LS. The CES-D Scale: a self-reportdepression scale for research in general population.Appl Psychol Meas 1977;1:385e401.

13. Comstock GW, Helsing KJ. Symptoms of depres-sion in two communities. Psychol Med 1976;6:551e563.

14. Berkman LF, Berkman CS, Kasl S, et al. Depres-sive symptoms in relation to physical health andfunctioning in the elderly. Am J Epidemiol 1986;124:372e388.

15. Hann D, Winter K, Jacobsen P. Measurement ofdepressive symptoms in cancer patients: evaluationof the Center for Epidemiological Studies Depres-sion Scale (CES-D). J Psychosom Res 1999;46:437e443.

16. Tiffin P, Ashton H, Marsh R, Kamali F. Pharma-cokinetic and pharmacodynamic responses tocaffeine in poor and normal sleepers. Psychophar-macology (Berl) 1995;121:494e502.

17. Gentili A, Weiner DK, Kuchibhatla M,Edinger JD. Test-retest reliability of the PittsburghSleep Quality Index in nursing home residents.J Am Geriatr Soc 1995;43:1317e1318.

18. Joreskog KG, Sorbom D. LISREL (version 8.8)[computer software]. Chicago, IL: Scientific Soft-ware International, 2006.

19. Joreskog K. Structural equation with ordinalvariables using LISREL. Chicago, IL: Scientific Soft-ware International, 2005:1e77.

20. Hu L, Bentler PM. Cutoff criteria for fit indexesin covariance structure analysis: conventional crite-ria versus new alternatives. Struct Equ Model 1999;6:1e55.

21. Kooken WC, Haase JE, Russell KM. ‘‘I’ve beenthrough something’’: poetic explorations of AfricanAmerican women’s cancer survivorship. West J NursRes 2007;29:896e919; discussion 920e899.