psychometric evaluation of the pittsburgh sleep quality index in cancer patients

9
140 Journal of Pain and Symptom Management Vol. 27 No. 2 February 2004 Original Article Psychometric Evaluation of the Pittsburgh Sleep Quality Index in Cancer Patients Susan L. Beck, PhD, APRN, FAAN, Anna L. Schwartz, PhD, FNP, FAAN, Gail Towsley, MS, William Dudley, PhD, and Andrea Barsevick, DNSc, RN University of Utah College of Nursing (S.L.B., G.T., W.D.), Salt Lake City, Utah; University of Washington School of Nursing (A.L.S.), Seattle, Washington; and Fox Chase Cancer Center (A.B.), Philadelphia, Pennsylvania, USA Abstract This report summarizes findings related to the psychometric properties (internal consistency and construct validity) of the Pittsburgh Sleep Quality Index (PSQI) and discusses issues related to its use based on data from two clinical studies with diverse samples of cancer patients. Subjects completed a questionnaire that included the PSQI, the Schwartz Cancer Fatigue Scale, and specific demographic, disease, and treatment variables. There were complete data on 170 (of 214) cases in Study 1 and 249 (of 259) cases in Study 2. The Cronbach’s alpha for the Global Sleep Quality scale was 0.81 in Study 1 and 0.77 in Study 2. A comparison of Global Sleep Quality in two contrasting groups with low and high fatigue yielded statistically significant differences in both samples. Psychometric evaluation supports its internal consistency reliability and construct validity. However, the scoring is rather cumbersome and raises questions regarding level of measurement and appropriate analysis techniques. J Pain Symptom Manage 2004;27:140–148. 2004 U.S. Cancer Pain Relief Committee. Published by Elsevier Inc. All rights reserved. Key Words Cancer, sleep, insomnia, measurement, psychometrics, medical oncology, neoplasm Introduction Sleep disturbances can seriously influence physical and mental well-being as well as quality of life. These effects are often more pro- nounced in individuals who are facing the multiple consequences of a serious and life- threatening illness such as cancer. There is some evidence that cancer patients have more Address reprint requests to: Susan L. Beck, PhD, Univer- sity of Utah College of Nursing, 10 South 2000 East, Salt Lake City, UT 84112-5880, USA. Accepted for publication: June 30, 2003. 2004 U.S. Cancer Pain Relief Committee 0885-3924/04/$–see front matter Published by Elsevier Inc. All rights reserved. doi:10.1016/j.jpainsymman.2003.12.002 problems sleeping than healthy individuals and that the degree of difficulty in initiating and maintaining sleep for cancer patients may even be as high as in suicidal patients or known insomniacs. 1 Sleep deprivation can have pro- found physical effects including fatigability, pain intolerance, and decreased immune func- tioning as well as emotional consequences such as irritability, depression, and decreased plea- sure in work and social activities. 2 Research on sleep in cancer patients has been limited by the lack of an effective measurement tool for clinical trials. There are few empirical data using psychometrically sound measures about the real prevalence and nature of sleep disturbances in cancer patients. 3,4 It is not

Upload: susan-l-beck

Post on 05-Sep-2016

213 views

Category:

Documents


0 download

TRANSCRIPT

140 Journal of Pain and Symptom Management Vol. 27 No. 2 February 2004

Original Article

Psychometric Evaluation of the PittsburghSleep Quality Index in Cancer PatientsSusan L. Beck, PhD, APRN, FAAN, Anna L. Schwartz, PhD, FNP, FAAN,Gail Towsley, MS, William Dudley, PhD, and Andrea Barsevick, DNSc, RNUniversity of Utah College of Nursing (S.L.B., G.T., W.D.), Salt Lake City, Utah; Universityof Washington School of Nursing (A.L.S.), Seattle, Washington; and Fox Chase Cancer Center(A.B.), Philadelphia, Pennsylvania, USA

AbstractThis report summarizes findings related to the psychometric properties (internal consistencyand construct validity) of the Pittsburgh Sleep Quality Index (PSQI) and discusses issuesrelated to its use based on data from two clinical studies with diverse samples of cancerpatients. Subjects completed a questionnaire that included the PSQI, the Schwartz CancerFatigue Scale, and specific demographic, disease, and treatment variables. There werecomplete data on 170 (of 214) cases in Study 1 and 249 (of 259) cases in Study 2. TheCronbach’s alpha for the Global Sleep Quality scale was 0.81 in Study 1 and 0.77 inStudy 2. A comparison of Global Sleep Quality in two contrasting groups with low andhigh fatigue yielded statistically significant differences in both samples. Psychometricevaluation supports its internal consistency reliability and construct validity. However, thescoring is rather cumbersome and raises questions regarding level of measurement andappropriate analysis techniques. J Pain Symptom Manage 2004;27:140–148. � 2004U.S. Cancer Pain Relief Committee. Published by Elsevier Inc. All rights reserved.

Key WordsCancer, sleep, insomnia, measurement, psychometrics, medical oncology, neoplasm

Address reprint requests to: Susan L. Beck, PhD, Univer-sity of Utah College of Nursing, 10 South 2000 East,Salt Lake City, UT 84112-5880, USA.Accepted for publication: June 30, 2003.

ltdad

IntroductionSleep disturbances can seriously influence

physical and mental well-being as well as qualityof life. These effects are often more pro-nounced in individuals who are facing themultiple consequences of a serious and life-threatening illness such as cancer. There issome evidence that cancer patients have more

ptmbifpt

� 2004 U.S. Cancer Pain Relief CommitteePublished by Elsevier Inc. All rights reserved.

roblems sleeping than healthy individuals andhat the degree of difficulty in initiating and

aintaining sleep for cancer patients may evene as high as in suicidal patients or known

nsomniacs.1 Sleep deprivation can have pro-ound physical effects including fatigability,ain intolerance, and decreased immune func-ioning as well as emotional consequences such

as irritability, depression, and decreased plea-sure in work and social activities.2

Research on sleep in cancer patients has beenimited by the lack of an effective measurementool for clinical trials. There are few empiricalata using psychometrically sound measuresbout the real prevalence and nature of sleepisturbances in cancer patients.3,4 It is not

0885-3924/04/$–see front matterdoi:10.1016/j.jpainsymman.2003.12.002

Vol. 27 No. 2 February 2004 141Evaluation of the PSQI in Cancer Patients

known which types of patients are more likelyto have significant problems and whether thereis a relationship between sleep disturbances anddemographic characteristics, type of cancer,extent of disease, or type of treatment. It hasbeen proposed that other common symptomsin cancer patients such as pain, fatigue, anxiety,and depression may be associated with sleepdisturbances but the strength and significanceof these relationships is not known. Advancingthe science in this area, including clinical trialsof interventions to improve sleep, will requireconsistent use of measures with establishedreliability and validity in cancer patients. Thepurpose of this report is to summarize findingsrelated to the psychometric properties of a self-report tool, the Pittsburgh Sleep Quality Index(PSQI), and to discuss issues related to its usebased on data from two clinical studies withdiverse samples of cancer patients.

Sleep Measurement: Polysomnographyand Actigraphy

There are numerous approaches to sleepmeasurement that range in expense and easeof use. Tools that measure various aspects ofsleep include polysomnography, actigraphy,and self-reports including sleep logs and ques-tionnaires. Polysomnography, or all-night sleeprecordings, is considered the most accuratemeasure of sleep and yields measures of spe-cific sleep stages. Polysomnography monitorssleep-related physiologic parameters such asrespiratory, neuromuscular, cardiac, gastroin-testinal, and endocrine functions.5 Parametersmay be measured by electroencephalogram(EEG) (the core of polysomnography), electro-oculogram (EOG), electrocardiogram (ECG),or electromyogram (EMG) readings. Althoughpolysomnography is the physiologic “gold stan-dard” to monitor sleep, it is most often utilizedin a sleep lab that can be expensive, inconve-nient, cumbersome, or uncomfortable. Thesereasons may explain why no clinical studies incancer patients have used polysomnography.

An alternative but less inexpensive approachis the use of actigraphy, an instrument that re-cords a patient’s movement pattern over aperiod of time. The wrist actigraph is a smalldevice that can be worn on the wrist or leg.6

Movement data from internal motion sensorsis transferred to a computer for an analysis ofstandard descriptors of sleep and wake periods.

E1ti

actbttspani

S

adnpsauabattropsajwit

clIdiormdlrd

ach actigraphy wristwatch costs about US$,000 and data management and interpreta-ion can be time-intensive, thus its applicabilityn large-scale clinical trials may be limited.

Recent research with cancer patients has usedctigraphy. Miaskowski and Lee reported thatancer patients (n � 24) receiving radiationherapy experienced significant sleep distur-ances as measured by actigraphy. Furthermore,hose who were given a higher proportion ofheir radiation treatment dose reported moreleep problems.7 Berger and Farr used actigra-hy to study 72 women receiving chemother-py. They reported an increased number ofighttime awakenings that were associated with

ncreased fatigue.8

leep Measurement: Self-ReportSelf-report measures are the most common

pproach to measuring sleep and include sleepiaries, sleep logs, and questionnaires. Theseon-physiologic measures of sleep have multi-le utility and can allow comparison betweenleep parameters, monitor adherence to ther-py, facilitate longitudinal data collection, eval-ate treatment progress, monitor symptoms,nd promote self management.9 Sleep logs cane used to give a day-to-day account of sleepctivities 24 hours per day over a period ofime. The advantages of self reports includeheir ease of use, convenience, low expense,eflection of the natural setting, relative non-btrusiveness, and recording of the person’serceived sleep experience. Disadvantages ofelf-reports consist of their subjectivity; they maylso be burdensome if used daily. They are sub-ect to reporting-bias and may yield missing datahen respondents do not answer all items or

naccurate data if participants fail to completehem in a timely way.9,10

Numerous clinical studies on side effects ofancer treatment, cancer pain, and quality ofife in cancer patients use self-report measures.n a recent review of 15 prevalence studies iniverse cancer populations, most reported find-

ngs from a single question embedded in an-ther instrument. Between 30–50% of patientseported sleep difficulties of some type.3 In aore comprehensive, 82-item, investigator-

esigned survey of 150 patients with breast andung cancer, Engstrom et al. found that 44%eported “a problem with sleep disturbances”uring the past month. In a follow-up interview

142 Vol. 27 No. 2 February 2004Beck et al.

with those experiencing problems, 57% re-ported the severity of the sleep disturbance asat least moderate in intensity.11 In a small com-parative study of subjective reports of sleep pat-terns of cancer patients to a control group,cancer patients were three times more likely toexperience sleep difficulties; specifically, manyfaced increased difficulty with sleep onset.12

The Pittsburgh Sleep Quality IndexThe Pittsburgh Sleep Quality Index (PSQI)

is a self-report questionnaire that assesses sleepquality and quantity. The original version wasdesigned to measure sleep reports over a one-month interval.13 The19-item self-report ques-tionnaire yields 7 component scores: subjectivesleep quality, sleep latency, duration, habitualsleep efficiency, sleep disturbances, use ofsleeping medication, and daytime dysfunction.There are five additional questions that arecompleted by a bed partner if there is one.These are not used in the scoring. A Cronbach’salpha of 0.83 was reported for the Global SleepQuality scale. A Global Sleep Quality score greaterthan 5 discriminated between good and poorsleepers and yielded a diagnostic sensitivity of89.6% and specificity of 86.5%.13,14 There isevidence of the reliability and validity of thePSQI in the elderly,14,15 bereaved spouses,16 pa-tients with AIDS,17 panic disorder,18 and pho-bias.19 Additionally, the PSQI has demonstratedsensitivity in measuring the effectiveness ofdrugs20 and exercise.21

There is limited research on the use of thePSQI with cancer patients: one study with 15patients and a historical control and one studylimited to women with breast cancer followingtreatment. Owen et al. used the PSQI to com-pare a small sample of 15 cancer patients to52 healthy individuals.22 They found that thecancer patients had significantly poorer overallsleep quality and scored significantly worse on5 of the 7 component scores. Carpenter andAndrykowski found in their research of thepsychometric evaluation of the PSQI that itdemonstrated utility for self-administration, in-ternal consistency reliability (alpha � 0.80 acrossall diagnostic groups), and construct validityacross a variety of clinical populations, includ-ing a group of 102 women with breast cancerreceiving routine follow-up care.23 They recom-mended that a cut-off score of �8 (vs. 5 asrecommended by the tool developers) may be

mc

avtustoautasotcpiabs

MS

etPasaitUdfqnTtPSsmancpo

ore appropriate to determine poor sleep inlinical populations.

Evidence to support the reliability, validity,nd sensitivity of the PSQI in patients with di-erse types of cancer and undergoing activereatment is lacking and essential to support itsse as an outcome measure in future clinicaltudies. This report summarizes an analysis ofwo clinical studies and includes: 1) discussionf missing data and issues related to scoringnd feasibility; 2) an evaluation of reliabilitysing an item analysis and reliability analysis

o determine internal consistency; and 3) anssessment of construct validity by contrastingleep quality in two groups—one with low andne with high fatigue using independent group

-tests. Data for these analyses come from twolinical studies with diverse samples of canceratients with a variety of primary diagnoses and

ncluding patients receiving radiation therapynd chemotherapy. Both studies were approvedy the Institutional Review Boards at their re-pective institutions.

ethodstudy OneThis cross-sectional study was designed to

valuate the psychometric properties (i.e., in-ernal consistency and construct validity) of theSQI (adapted for a one-week time interval) inheterogeneous sample of cancer patients. The

tudy used a prospective, consecutive samplingpproach. All patients who were receiving caren three settings (outpatient oncology, radia-ion therapy, and inpatient oncology) at theniversity of Utah (UU) in Salt Lake City, Utah,uring a designated time period were screened

or study eligibility and invited to complete auestionnaire. Completion of the question-aire implied voluntary consent to participate.he symptom questionnaire included several

ools but those reported here include theSQI13 and the Schwartz Cancer Fatiguecale (SCFS).24 The SCFS is a short, 6-item in-trument that asks the respondent to rate “howuch has fatigue made you feel (e.g. tired)?” onfive-point verbal numeric scale with 1 beingot at all and 5 being extremely. Two subscalesan be computed that measure physical anderceptual fatigue. The reliability and validityf this tool have been established.24,25 Both

Vol. 27 No. 2 February 2004 143Evaluation of the PSQI in Cancer Patients

tools were framed within the context of “thepast week.”

Of those who were eligible, 214 patients con-sented to participate (radiation therapy n � 81,outpatient oncology clinic n � 86, and inpa-tient oncology unit n � 47). The study demo-graphic characteristics are summarized inTable 1. Forty-nine percent were female andage ranged from 14 to 88 years, with a mean ageof 53 years. Racial/ethnic diversity was limited;92.3% were Caucasian, reflective of the state’spopulation at that time. The patients had multi-ple types of cancer as a primary diagnosis (seeTable 2); 19.6% had breast cancer and 47.1%of the patients had advanced disease.

Study 2The second longitudinal study addressed

similar aims by a secondary analysis using datafrom a randomized clinical trial to comparean energy conservation/activity management

S

BLPLMCC

(ctcecfTt

Table 1Demographic Characteristics of the Two

Study Samples

Study 1 Study 2(n � 214) (n � 259)

Characteristic n % n %

SexFemale 103 49.0 231 89.2Male 107 51.0 28 10.8

Race/EthnicityWhite 193 92.3 235 90.7African American 1 0.5 15 5.8Hispanic 7 3.3 5 1.9Native American 5 2.4 – –Asian/Pacific Islander 3 1.4 – –Other 2 �1

Marital StatusMarried/Living 142 67.9 182 70.8

with PartnerSingle 28 13.4 20 7.8Separated or Divorced 21 10.1 29 19.1Widowed 18 8.6 26 29.2

EducationLess than High School 19 9.2 7 2.7High School or GED 57 27.4 74 28.4Some college/technical 57 27.4 89 34.6Associate Degree plus 75 36.1 88 34.2

IncomeLess than 15,000/year 42 22.9 – –15,000 to 35,000/year 46 25.1 – –35,000 to 50,000/year 31 16.9 – –50,000 to 70,000/year 28 15.3 – –More than 70,000 36 19.7 – –

movprtqa

alacctamCavImSqtwltca

ECAM) intervention with an attentionalontrol group for the management of cancerreatment-related fatigue.26 The interventiononsisted of structured information aboutnergy conservation and activity managementombined with training in skills for coping withatigue in an emotionally supportive context.he attentional control group received nutri-

ion information unrelated to fatigue or treat-ent and was designed to provide the amount

f instruction equivalent to the ECAM inter-ention. The intervention was delivered by tele-hone by a nurse counselor. Participantseceived three phone calls to discuss educa-ional materials sent in the mail and conse-uently developed and implemented anction plan.

The sample included men and women, overge 18, who were initiating treatment for breast,ung, colorectal, cervical, testicular, prostate,nd lymphoma cancers. The treatment was forure or local control and involved at least threeycles of chemotherapy, 5–6 weeks of radiationherapy treatments, or concurrent radiationnd chemotherapy. This study was imple-ented at two clinical sites; Fox Chase Cancerenter (FCCC) in Philadelphia, Pennsylvaniand the Huntsman Cancer Institute at the Uni-ersity of Utah (UU) in Salt Lake City, Utah.n addition to the PSQI, eight other instru-ents were utilized in this study, including the

chwartz Cancer Fatigue Scale.24 The PSQIuestions were framed within the context ofhe sleep experience during the past month,hich is the original version. Data were col-

ected at three time points, which varied byreatment regimen. In the chemotherapy andoncurrent group, measures were at baselinend 48 hours following the second and third

Table 2Number and Percent of Individuals with

a Cancer Primary Site in Each Sample

Study 1 Study 2(n � 214) (n � 259)

ite n % n %

reast 41 19.6 204 78.8ymphoma/Leukemia 41 19.6 16 6.2rostate/Testicular 25 11.9 – –ung 14 6.5 30 11.6elanoma 11 5.3 – –ervical/Ovarian 9 4.3 5 1.9olorectal 4 1.9 4 1.5

144 Vol. 27 No. 2 February 2004Beck et al.

chemotherapy administrations. In the radia-tion therapy group, measures were at baseline,during the last week of treatment, and onemonth following treatment. Time 3 data wereutilized in this analysis.

Research staff initially screened individualswho were diagnosed with eligible cancers andscheduled for initial evaluations. They then ex-plained the study objectives and procedures toeligible individuals; those interested providedwritten consent to participate. The individualthen completed the baseline questionnaires. Ifcontact was made by phone, questionnairesand consent forms were mailed to participants.After reviewing returned questionnaires, re-search staff called participants if responseswere not complete.

The sample at Time 3 included 259 cancerpatients aged 26 to 83 years, with a mean age of56.6 years. One hundred twenty-nine (49.8%)were receiving radiation therapy and 130(50.2%) were receiving chemotherapy or con-current therapy. The study demographic char-acteristics are summarized in Table 1. More(89.2%) were female and 90.7% were Cauca-sian. Although all target cancer diagnoses wererepresented (see Table 2), the majority (78.8%)had breast cancer.

ResultsThere are 19 individual questions in the

PSQI. The scoring of these questions transformsthem into seven components, each ranging from0 to 3, with a higher score representing poorersleep quality. Some questions are simply re-coded and others combine 2–9 questions andthen recode responses to yield a 0 to 3 scale.13

The sum of these 7 components yields a GlobalSleep Quality score ranging from 0 to 21. Thescoring is prescribed in the original method-ological paper but has several glitches, suchas overlapping categories. Clarifications to thescoring as used in these analyses are detailedin Appendix 1.

In order to compare the psychometric prop-erties of the PSQI with reports in other popula-tions, each of the seven component scores istreated as an “item” in the analysis. In addi-tion, the results from several of the originalquestionnaire variables are reported as theymay be more clinically relevant and sensitive tochange in the cancer patient population.

A

imeilrTplS24t

I

airrifTmst00wtodcrtnw

C

1234567G

nalysis of Missing DataThe first step in the analysis was to evaluate

ndividual questions to identify the pattern ofissing data. Unless corrected in the analysis,

ach missing question can result in a missingtem and a missing Global Sleep Quality score. Aarge amount of missing data can influence theeliability of the tool and bias the findings.he number and percent missing for each com-onent are summarized in Table 3. The cumu-

ative effect in the ability to compute the Globalleep Quality score was marked in Study 1, as1% of cases were lost. This was decreased to.2% in Study 2, as participants were calledo follow-up on missing questions.

tem and Reliability AnalysisThe second step in the analysis was to evalu-

te the range and variance of responses to eachtem, that is, component score, following theecoding to 0–3. Responses included theange of all possible scores for each of the seventems. The Global Sleep Quality score rangedrom 0 to 21 in Study 1 and 1 to 19 in Study 2.he highest possible score would be 21. Theeans and standard deviations for each

ample are summarized in Table 4. In Study 1,he mean component scores ranged from.99–1.48; in Study 2, the means ranged from.85–1.47. The item-to-item correlation matrixas examined to identify items with low (less

han 0.30, indicating minimal contribution)r high (greater than 0.70, indicating redun-ancy) correlations. There were no item–itemorrelations greater than 0.70, indicating littleedundancy in the items. In Study 1, six correla-ions were less than 0.30; three were for compo-ent 6, Use of Sleeping Medication, and three wereith component 7, Daytime Dysfunction. These

Table 3Percent of Missing Data for Each

Component Score

Study 1 Study 2(n � 214) (n � 259)

omponent Scores (0–3) n % n %

. Subjective sleep quality 4 2 1 �1

. Sleep latency 11 5 3 1

. Hours of actual sleep 6 3 0 0

. Sleep efficiency 21 10 6 2

. Sleep disturbances 14 7 8 3

. Use of sleeping medication 4 2 0 0

. Daytime dysfunction 7 3 0 0lobal Sleep Quality (1–21) 44 21 11 4

Vol. 27 No. 2 February 2004 145Evaluation of the PSQI in Cancer Patients

Table 4Mean and Standard Deviation for Each Component Score and Other Key Variables

Study 1 (n � 214) Study 2 (n � 259)

Component Scores (0–3) Mean SD Mean SD

1. Subjective sleep quality 1.18 0.83 1.10 0.732. Sleep latency 1.25 1.05 1.06 0.953. Sleep duration 1.00 1.05 0.90 0.914. Habitual sleep efficiency 0.99 1.18 0.87 1.115. Sleep disturbances 1.48 0.58 1.47 1.626. Use of sleeping medication 1.07 1.35 0.85 1.237. Daytime dysfunction 1.18 0.85 1.04 0.70

Global Sleep Quality (1–21) 8.15 4.70 7.31 4.03Total sleep disturbance (0–30 possible) 10.60 5.31 8.33 4.28Minutes to fall asleep 27.76 30.60 22.44 25.48Sleep efficiency in percent 0.81 0.17 0.82 0.14

may represent distinct concepts, but it wouldbe difficult to factor analyze as these are onlysingle questions. In Study 2, the pattern wassimilar except there were also three low correla-tions with component 5, Sleep Disturbances.

A reliability analysis of the 7 componentscore items was conducted using SPSS 11.0. Thecorrected item-to-total (i.e., Component-to-Global) correlations and Cronbach alpha coef-ficients if the item were deleted are summarizedfor each sample in Table 5. In Study 1, theComponent-to-Global correlations ranged from0.38–0.64. In Study 2, the Component-to-Globalcorrelations ranged from 0.32–0.63. In both,components 6 and 7 were lowest. The overallscale alpha was 0.804 in Study 1 and 0.770 inStudy 2. The alpha increased minimally by de-leting component 6 (Use of Sleeping Medication)in Study 2.

Validity AnalysisThe construct validity was examined by com-

paring the Global Sleep Quality scores in two con-trasting groups: low fatigue and high fatigue.

T5TcScTipws

D

lemiadI

Table 5Reliability Analysis of PSQI in Two Samples

Study 1 (n � 214) Study 2 (n � 259)

Corrected Item-Total Alpha if Corrected Item-Total Alpha ifComponent Scores Correlation Item Deleted Correlation Item Deleted

1. Subjective sleep quality 0.63 0.744 0.63 0.6952. Sleep latency 0.55 0.753 0.56 0.6993. Sleep duration 0.62 0.739 0.53 0.7074. Habitual sleep efficiency 0.64 0.735 0.57 0.6955. Sleep disturbances 0.53 0.771 0.40 0.7366. Use of sleeping medication 0.39 0.801 0.32 0.7727. Daytime dysfunction 0.40 0.781 0.37 0.741

Total Standardized Alpha 0.808 0.770

t takes 5–10 minutes to complete.

he mean score on the SCFS (possible 1 low tohigh) was categorized into two groups.

hose with a fatigue score from 1 to 3 wereategorized as low and from 3.1 to 5 as high.cores on the PSQI Global Sleep Quality wereompared using independent t-tests (Table 6).he findings indicate that the Global Sleep Qual-

ty score was significantly higher (indicatingoorer sleep) in the high fatigue groups. Thisas true in both samples and supports the con-

truct validity of the PSQI.

iscussionIn order to effectively study the sleep prob-

ems that cancer patients experience, it is nec-ssary to identify valid and reliable tools toeasure sleep that can be easily administered

n the clinical or home setting. The PSQI isself-administered questionnaire that collects

ata regarding multiple facets of sleep quality.

146 Vol. 27 No. 2 February 2004Beck et al.

Table 6Comparison of Global PSQI in Groups with

Low and High Fatigue

Study 1

Global SleepGroup Quality Mean SD

Low fatigue (n � 104) 6.82 4.57High Fatigue (n � 56) 10.20 4.28t �4.556Significance P � 0.001

Study 2

Low fatigue (n � 214) 6.88 3.77High Fatigue (n � 42) 9.50 4.52t �3.52Significance P � 0.001

afd

tifdwatsiiws

When self-administered, there can be a signif-icant amount of missing data unless the ques-tionnaire is reviewed for completion andmissing items are revisited with the respondent.In Study 1, the question about “usual bed time”had a high percent missing. Many people wouldwrite in yes versus indicating a time. This re-sulted in an inability to accurately compute SleepEfficiency. Several analytic strategies can beapplied to not lose cases in the scoring. Forexample, in computing the Sleep Disturbances,the sum was computed by recoding missingitems as 0 so all subjects with at least one re-sponse would have a score. In computing theGlobal Sleep Quality, several approaches can beapplied. It is possible to impute a mean score fora missing component based on the other com-ponents. An alternate approach was used inStudy 2. A Global Sleep Quality computation wasallowed if at least 5 of the 7 components werepresent. A mean of the non-missing compo-nents was computed and the result multipliedby 7 to give a comparable score. The mean sumin using this approach was 7.308 with 259 cases,as compared to 7.297 with 249 cases when onlycomplete cases were included.

There are a few other practical points of inter-est. Because patients report time data (e.g.,time that you usually go to bed), it is wise todetermine the type of time and format that willbe used to enter the data a priori. Military timeis recommended. In addition, it is useful toestablish a system to cross check the multiplecomputerized calculations. For example, im-porting data into an Excel spread sheet canallow for cross-checking the hours in bed andsleep efficiency calculations. This approach

Mlilrfr

pasGhsqasacrdmanpo

iatiEhAcTp

llows easier detection of problems that may beormulaic or unusual cases in which originalata should be verified.The evidence from these two studies supports

he internal consistency (reliability) of the PSQIn these two samples of cancer patients. Theseindings are similar to those reported in the toolevelopment in which the Cronbach’s alphaas 0.8313 and in two other studies in which thelpha ranged from 0.80 to 0.83.22,23 Each ofhe seven components contributes to the mea-urement of the overall construct of sleep qual-ty. Another similar finding is that the largesttem-total (Component-to-Global) coefficientsere found for habitual sleep efficiency and

leep quality. In these studies, Use of Sleepingedication and Daytime Dysfunction had the

owest Component-to-Global correlations, whilen the original study, Sleep Disturbances had theowest item-total (Component-to-Global) cor-elation. Additional research to replicate theseindings and to test sensitivity and test–retesteliability in cancer patients is recommended.

The findings from these two studies also sup-ort the construct validity of the PSQI. Using

contrasting groups approach, there wereignificant and clinically relevant differences inlobal Sleep Quality between groups with low andigh fatigue in both samples. This finding isimilar to the significant difference in sleepuality found by Buysse et al between controlsnd patients with depression.13 Additional re-earch might consider factor analysis to evalu-te whether the PSQI only measures oneonstruct or has subscales. This approach seemseasonable conceptually (e.g., is Sleep Efficiencyifferent than Sleep Latency?), but is limitedathematically by the 7 component scores that

re built into the scoring of tool and limitedumber of total questions (n � 19). Most com-onents, except Sleep Disturbances, are based onne or two questions.One limitation of the scoring approach used

n the PSQI is that the level of measurementnd potential variance are decreased as ques-ions are combined and rescored. This is truen particular with Minutes to Fall Asleep and Sleepfficiency (a computed percent: hours asleep/ours in bed), which are interval level data.nother problem with Sleep Efficiency is that theomputed percent may be greater than 100%.his was true for 10 cases (3.9%) in Study 2. Thishenomena is explainable as subjects answer

Vol. 27 No. 2 February 2004 147Evaluation of the PSQI in Cancer Patients

individual questions (e.g., what time do youusually go to bed) that are used to computesleep efficiency with an average for a week ormonth. This does not really pose a problem inthe scoring as all Sleep Efficiency ratings greaterthan 85% received a score of “0”. However, ifSleep Efficiency is used as a percentage variable,these scores will inflate the mean. One ap-proach is to recode values greater than 100% toequal to 100%. The scoring scheme also yields aloss of data in regards to Sleep Disturbances. Thisitem could range from 0 to 27 and is reducedto 0–3 in the scoring, thus losing sensitivity. Incancer patients, specific causes of sleep distur-bance, such as pain, are relevant clinically andas a confounding variable in research. An alter-native approach is to evaluate these items asoutcomes independent of the components orscoring instructions.

In conclusion, the PSQI is a relatively easy-to-administer tool that can measure sleep qualityover a period of time. Psychometric evaluationsupports its internal consistency (reliability) andconstruct validity in cancer patients. The scoringis rather cumbersome and raises questions re-garding level of measurement and appropriateanalysis techniques. Additional research shouldexamine the test-retest reliability of the PSQIand sensitivity to change, and further evaluateits construct validity. Research in cancer pa-tients with greater ethnic and racial diversity isrecommended to further support its use as anoutcome measure in clinical studies.

AcknowledgmentsStudy 1 was supported by an intramural award

from the University of Utah College of Nursing.Study 2 was supported by the National Institutefor Nursing Research and National Cancer In-stitute RO1 NR04573-04. The support of WeiChen Tung, Jose Benuzillo, and the ResearchStaff at Fox Chase Cancer Center and the Uni-versity of Utah College of Nursing in data entryand management is gratefully acknowledged.

References1. Silberfarb PM, Hauri PJ, Oxman TE, et al. Assess-

ment of sleep patients with lung cancer and breastcancer. J Clin Oncol 1993;11:997–1004.

t

c2

p

ld8

S

dtS

ir

ec1

1ai2

1a1

1tJ

1Pf1

1fmi

1ri1

1as

1it1

1pat

2. Sheey LC. Sleep disturbances in hospitalized pa-ients with cancer. Oncol Nurs For 1996;23:109–111.

3. Savard J, Morin CM. Insomnia in the context ofancer: a review of a neglected problem. J Clin Oncol001;19(3):895–908.

4. Hu D-S, Silberfarb PM. Management of sleeproblems in cancer patients. Oncology 1991;5:23–27.

5. American Electrocephalographic Society. Guide-ines for polygraphic assessment of sleep-related disor-ers (polysomnography). J Clin Neurophysiol 1992;9:8–96.

6. Hauri PJ, Wisbey J. Wrist actigraphy in insomnia.leep 1992;15:293–301.

7. Miaskowski C, Lee KA. Pain, fatigue, and sleepisturbances in oncology outpatients receiving radia-

ion therapy for bone metastasis: a pilot study. J Painymptom Manage 1999;17:320–331.

8. Berger AM, Farr L. The influence of daytimenactivity and nighttime restlessness on cancer-elated fatigue. Oncol Nurs For 1999;26:1663–1671.

9. Cohen FL. Measuring sleep. In: Frank-Stromb-rg M, Olsen SJ, eds. Instruments for clinical health-are research. 2nd ed Sudbury, MA: Jones & Bartlett,992:264–279.

0. Richardson SJ. A comparison of tools for thessessment of sleep pattern disturbance in criticallyll adults. Dimensions Crit Care Nurs 1997;16(5):26–239.

1. Engstrom CA, Strohl RA, Rose L, et al. Sleeplterations in cancer patients. Cancer Nurs 1999;22:43–148.

2. Kaye J, Kaye K, Madow L. Sleep patterns in pa-ients with cancer and patients with cardiac disease.Psychol 1983;114:107–113.

3. Buysse DJ, Reynolds III CF, Monk TH, et al. Theittsburgh Sleep Quality Index: a new instrument

or psychiatric practice and research. Psychiatry Res989;28:193–213.

4. Buysse DJ, Reynolds CF, Monk TH, et al. Quanti-ication of subjective sleep quality in healthy elderly

en and women using the Pittsburgh Sleep Qual-ty Index (PSQI). Sleep 1991;14:331–338.

5. Gentili A, Weiner D, Kubchibhatla M, et al. Test-etest reliability of the Pittsburgh Sleep Quality Indexn nursing home residents [Letter]. J Am Geriatr Soc995;43:1317–1318.

6. Reynolds CF, Hoch CC, Buysse DJ, et al. Sleepfter spousal bereavement: a study of recovery fromtress. Biol Psychiatry 1993;34:791–797.

7. Rubinstein ML, Selwyn PA. High prevalence ofnsomnia in an outpatient population with HIV infec-ion. J Acquir Immune Defic Syndr Hum Retrovirol998;19(3):260–265.

8. Stein MB, Chartier M, Walker JR. Sleep in nonde-ressed patients with panic disorder: I. Systematicssessment of subjective sleep quality and sleep dis-urbance. Sleep 1993;16:724–726.

148 Vol. 27 No. 2 February 2004Beck et al.

19. Stein MB, Kroft CD, Walker JR. Sleep impair-ment in patients with social phobia. Psychiatry Res1993;49:251–256.

20. Nierenberg AA, Adler LA, Peselow E, et al. Trazo-done for antidepressant-associated insomnia. Am JPsychiatry 1994;151:1069–1072.

21. Singh NA, Clements KM, Fiatarone MA. A ran-domized controlled trial of the effect of exerciseon sleep. Sleep 1997;20(2):95–101.

22. Owen DC, Parker DP, McGuire DB. Comparisonof subjective sleep quality in patients with cancer andhealthy subjects. ONF 1999;26:1649–1651.

2eP2t12o12pt3

3. Carpenter JS, Andrykowski MA. Psychometricvaluation of the Pittsburgh Sleep Quality Index. Jsychosom Res 1998;45:5–13.4. Schwartz AL. The Schwartz Cancer Fatigue Scale:esting reliability and validity. Oncol Nurs Forum998;25:711.5. Schwartz A, Meek P. Additional construct validityf the Schwartz Cancer Fatigue Scale. J Nurs Measure999;7(1):35–45.6. Barsevick AM, Whitmer K, Sweeney C, et al. Ailot study examining energy conservation for cancerreatment-related fatigue. Ca Nurs 2002;25(5):333–34.

Appendix 1Scoring the PSQI

Original Scoring Revised Scoring

Component 3 Sleep Duration Component 3 Sleep Duration�7 hours 0 �7 hours 06–7 hours 1 �6 and �7 hours 15–6 hours 2 �5 and �6 hours 2�5 hours 3 �5 hours 3

Component 4 Sleep Efficiency Component 4 Sleep Efficiency�85% 0 �85% 075–84% 1 75–85% 165–74% 2 65–74% 2�65% 3 �65% 3