Abstract
The aim of this study was translating and exploring psychometric properties of Serbian Pittsburgh Sleep Quality Index (PSQI) in a sample of “good” and “bad” sleepers suffering from depression or obstructive sleep apnea (OSA). Formal translation and validation were performed on a sample of healthy controls, patients with untreated OSA, and with diagnosed major depressive disorder with evaluation of internal consistency, test–retest reliability, and construct and criterion validity. Controls and OSA subgroups were recruited from a larger sample of commercial drivers. One hundred and forty subjects, 84.3% male, 22–67 years old, were included. OSA subgroup had 59 subjects and depression subgroup had 40 subjects (22 females). Mean ± SD total PSQI was 3.5 ± 2.2 in controls, 4.9 ± 3.6 in OSA subjects, and 9.0 ± 4.9 in patients with depression. Cronbach’s α for total PSQI was 0.791. Subscale scores were significantly correlated to global PSQI in all subgroups. Intraclass correlation coefficient for global PSQI was 0.997 (p < .001). Epworth Sleepiness Scale score was significantly correlated to global PSQI (ρ = 0.333, p < .001). Three subgroups differed significantly in total PSQI and PSQI ≥ 5, even after adjustments for age and gender (p < .001). OSA patients had higher mean PSQI than controls but not significantly (p = .272). PSQI-reported sleep latency did not correlate with PSG-measured sleep latency (r = .130, p = .204). Total PSQI was significantly correlated to OSA severity (ρ = 0.261, p < .05). Serbian PSQI showed good internal consistency, test–retest reproducibility, and adequate construct and criterion validity, which supports further exploration of its use as a sleep quality screening tool in different target populations.
Sleep quality is recognized as one of the most important determinants of healthy sleep (Buysse, 2014), but the indicators of “good” or “bad” sleep are based more on the individual perception than the objective characteristics of this behavioral state. Prevalence of qualitative sleep disturbances is very high in general population, ranging from 8% (Ohayon, 2002) to over 35%, as reported in the U.S. National Health and Nutrition Examination Survey (Bansil, Kuklina, Merritt, & Yoon, 2011).
Inadequate sleep quality is one of the major defining elements of several sleep disorders (insomnia, obstructive sleep apnea [OSA], and circadian rhythm disorders) (American Academy of Sleep Medicine [AASM], 2005). Poor sleep is correlated with suboptimal daytime functioning (excessive sleepiness, motor vehicle accidents, and work-related injuries) (Braeckman, Verpraet, Van Risseghem, Pevernagie, & De Bacquer, 2011) and higher incidence of cardiovascular and metabolic disorders (Bruno et al., 2013; Cappuccio, D’Elia, Strazzullo, & Miller, 2010; Hoevenaar-Blom, Spijkerman, Kromhout, van den Berg, & Verschuren, 2011; Van Cauter, Spiegel, Tasali, & Leproult, 2008).
Sleep quality can be depicted as a function of measurable parameters of sleep (polysomnography [PSG], nonrapid eye movement electroencephalogram frequency spectral analysis) (Buysse, 2014; Krystal & Edinger, 2008) or through determining its effect on daily activities, work ability, and general health (AASM, 2005; Braeckman et al., 2011; Bruno et al., 2013; Cappuccio et al., 2010; Hoevenaar-Blom et al., 2011; Van Cauter et al., 2008). Some authors (Johns, 1991; Kecklund, Akerstedt, & Axelsson, 2003; Yi, Shin, & Shin, 2006) believe that the best way to perceive adequacy of someone’s sleep is through sleep diaries or standardized questionnaires, such as Pittsburgh Sleep Quality Index (PSQI).
Developed as a tool for recognition of and discrimination between good and bad sleepers, PSQI contains 19 self-reported items and five questions covering seven dimensions of perceived sleep quality over a 1-month interval (sleep duration, disturbances, latency, daytime dysfunction, habitual sleep efficiency, sleep quality, and use of sleeping medications) (Buysse, Reynolds, Monk, Berman, & Kupfer, 1989). PSQI has shown high sensitivity and specificity and has been validated in different population samples, ranging from children to older adults (Mesquita & Reimão, 2010; Spira et al., 2012; Yuksel et al., 2007), from healthy subjects (Buysse et al., 2008) to patients with sleep disorders (Mondal, Gjevre, Taylor-Gjevre, & Lim, 2013; Shochat, Tzischinsky, Oksenberg, & Peled, 2007), psychiatric disorders (Doi et al., 2000), and chronic conditions (Kotronoulas, Papadopoulou, Papapetrou, & Patiraki, 2011; Mariman et al., 2012; Sabbatini et al., 2008). Also, PSQI has been used in various occupational settings (Braeckman et al., 2011; Loft & Cameron, 2014; Mahdi, Mohamed, & Shafe, 2014). It is by far the most explored and most translated sleep quality questionnaire (Mollayeva et al., 2016).
Psychometric properties of PSQI have never been formally tested in Serbia. The only validated sleep health questionnaire is Serbian version of Epworth Sleepiness Scale (ESS; Kopitovic et al., 2011) that measures excessive sleepiness. The aim of the present study was translating PSQI into Serbian and exploring its psychometric properties in a sample of “healthy” and bad sleepers suffering from depression or OSA.
Subjects and Methods
Translation Study
We performed translation and cross-cultural adaptation of PSQI to Serbian, with author’s permission, according to MAPI Research Institute (Acquadro, Conway, Girourdet, & Mear, 2004) and Beaton, Bombardier, Guillemin, and Ferraz (2000) guidelines. Two forward translations were produced by accredited medical English translators, native in Serbian, not involved in the study, discussed with the research team, until a final translation was agreed. Backward translation was performed by a professional bilingual translator, native in English, compared and discussed by an expert panel. Prefinal version was tested for equivalence of terms with the original PSQI on 10 bilingual subjects (health-care workers, undergraduate medical students). Since there were no differences in their answers, PSQI was tested on 21 patients with diagnosed major depressive disorder. After cognitive debriefing, a final version of PSQI-Serbian was created and a full report on the procedure was sent to the PSQI author.
Validation Study
One hundred commercial drivers working in several public transportation companies in Belgrade, Serbia, and 40 patients with depression, treated in the Clinic for Psychiatry, Clinical Centre of Serbia, were included. Commercial drivers were part of a larger study examining sleepiness, OSA, and traffic accidents. Subjects were classified into three subgroups: “Healthy sleepers”—commercial drivers with no regular sleep complaints, ESS < 10, no previous or current diagnosis of OSA or any other sleep disorder, no sleep altering medication use. “Bad sleepers—OSA”—commercial drivers with one or more sleep complaints or ESS ≥ 10, OSA diagnosis confirmed by PSG, no previous or current diagnosis of other sleep or psychiatric disorder, no sleep altering medication use. “Bad sleepers—depression”—one or more sleep complaints or ESS ≥ 10, diagnosis of major depression disorder according to Diagnostic and Statistical Manual of Mental Disorders, fourth edition criteria (American Psychiatric Association, 1994), no previous or current diagnosis of OSA or other sleep disorder, regular medication use.
After detailed explanation, written and oral informed consent was obtained from all individual participants included in the study. Serbian was the native language for all participants.
Ethical Approval
The study has received ethical approval from the Ethical board of Faculty of Medicine, University of Belgrade, Serbia and the Ethical board of Clinical Centre of Serbia. The study has been performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments.
Study Design
Subjects were approached at their workplace (drivers) or during their visit to a physician (depression group). They completed a generic sleep and health questionnaire, Serbian version of PSQI and ESS. In the second phase, drivers were admitted to Serbian Institute of occupational health, retested with the same questionnaires, and underwent an overnight attended complete PSG (Type I PSG) or cardiorespiratory polygraphy (Type III PSG).
Instruments
Generic sleep and health questionnaire comprised questions on demographics, driving and sleep-related habits, sleepiness, current health, personal and family medical history, allergies, and medication use.
ESS is a self-administered questionnaire measuring the presence of excessive daytime sleepiness (Johns, 1991). The subject rates his or her chances of dozing off in eight common situations (reading, watching TV, etc.) on a scale of 0–3. If the sum of scores equals 10 or more, patient is considered having excessive sleepiness. ESS has been validated in Serbian (Kopitovic et al., 2011).
PSQI (Buysse et al., 1989) is a standardized self-administered questionnaire for assessment of sleep quality over the past month. It comprises 19 items forming seven subscales: subjective sleep quality (C1, 1 item), sleep latency (C2, 2 items), sleep duration (C3, 1 item), habitual sleep efficiency (C 4, 3 items), sleep disturbances (C5, 9 items), use of sleep medication (C6, 1 item), and daytime dysfunction (C7, 2 items). Scoring model is suggested by the author (Buysse et al., 1989). Subscales are scored on a 0–3 scale. Sum of subscale scores generates a global PSQI score, which, when 5 and more, is considered indicative of poor sleep quality. PSQI also comprises five questions, not included in the scoring model, answered by the subject’s bed partner.
PSG
All drivers underwent a single night attended complete PSG using a Type I system (Alice-5, Philips Respironics Inc., the Netherlands) or cardiorespiratory PSG using a Type 3 polygraph (MS-310, Müller & Sebastiani Elektronik GmbH, Germany). Complete PSG was performed according to standard techniques, with simultaneous monitoring of the electroencephalogram (frontal, central, and occipital leads), electrooculogram, chin electromyogram, flow (oronasal thermistor, nasal air pressure transducer), thoracic and abdominal respiratory effort, oximetry, body position, snoring, and leg electromyogram. Cardiorespiratory polygraphy included monitoring of the respiratory flow (oronasal thermistor), thoracic and abdominal respiratory effort, oximetry, body position, and snoring. PSGs were performed in hospital sleep lab settings, with continuous video and technician monitoring. Recordings were scored according to the AASM manual of scoring sleep and sleep-associated events (AASM & Iber, 2007).
Apnea was defined as a drop in the peak thermal sensor excursion by ≥90% of baseline, lasting minimum 10 s. Hypopnea was defined as a nasal pressure signal excursion drop by ≥50% of baseline, with a ≥3% oxygen desaturation or an arousal (AASM & Iber, 2007). Apnea–Hypopnea Index (AHI), the number of apneas and hypopneas per hour of sleep, was used for establishing OSA diagnosis and severity. OSA was considered mild if AHI was ≥5 and <15 per hour, moderate if AHI was ≥15 and <30, and severe if AHI was ≥30 per hour.
Statistical Analysis
Validation included evaluation of internal consistency, test–retest reliability, and construct and discriminant validity. Internal consistency was analyzed with Cronbach’s α (0.7 or higher was significant). Test–retest reliability was assessed with intraclass correlation coefficient (ICC). We examined discriminant validity by exploring differences in PSQI scores between subject subgroups and by exploring relationship between PSQI scores with sleep parameters obtained through PSG. Construct validity was evaluated by comparing PSQI with ESS scores.
Data are presented as mean ± SD for continuous or n (%) for categorical data. Subgroup differences were tested with t-test and Mann–Whitney U-test. Kruskal–Wallis one-way analysis of variance was used to test subgroup PSQI differences. Spearman correlation was used to assess relationship between PSQI, ESS scores, and PSG-derived parameters. In order to test age and gender effects, analysis of covariance (ANCOVA) was used for global PSQI comparisons between control and diagnostic groups. All data were analyzed in SPSS 20.0 (International Business Machines corporation) statistical package. All p values less than .05 were considered significant.
Results
A total of 140 subjects, 84.3% male, 22–67 years old, completed all questions in the aforementioned questionnaires. None of the subjects had reported any issues with PSQI items comprehension. After PSG, OSA (AHI ≥ 5) was found in 59 subjects, and they were included in Bad sleepers—OSA subgroup. Eleven had moderate and 13 had severe OSA. Bad sleepers—depression subgroup included 40 subjects (22 females).
Three groups differed significantly considering their age (p = .004) and body mass index (BMI; p < .001). Healthy sleepers were significantly younger than bad sleepers (p = .014 for OSA; p = .011 for depression), while OSA and depression subjects were of similar age (p > .05). OSA subgroup had significantly higher BMI than other two groups (p < .001).
Excessive sleepiness (ESS ≥ 10) was recognized in 27 subjects, mean ESS score was 5.9 (min 0, max 19). The mean ± SD total PSQI score was 3.5 ± 2.2 in healthy sleepers, 4.9 ± 3.6 in OSA subjects, while patients with depression had an overall PSQI of 9.0 ± 4.9 (Table 1).
Demographic and Sleep-Related Characteristics of the Study Population According to Risk Group.
Note. Data are presented as mean ± SD. BMI = body mass index; ESS = Epworth Sleepiness Scale; sleep latency = self-reported time from going to sleep and falling asleep; PSG = polysomnography; PSQI = Pittsburgh Sleep Quality Index; sleep efficiency = total sleep time/time in bed; sleep onset = time from lights out to beginning of the first epoch of N1, N2, N3, R; WASO = wake after sleep onset; AHI = Apnea–Hypopnea Index; arousal index = total number of registered arousals and awakenings per hour of sleep; NA = not applicable.
There were no significant differences considering daytime naps, but controls had higher frequency of prolonged, night and shift work in comparison to two other groups (p < .001). OSA subjects had almost 85% of regular snorers in the subgroup. Around 10% of healthy sleepers, 38% in OSA group, and 72.5% in depression group sometimes felt sleepy. The latter group had the highest frequency of reported nightmares (57.5%) and head injuries (35.9%). One half of the depression subgroup had a family member with depression.
PSQI Performance
Internal consistency
Homogeneity of items was tested by calculating Cronbach’s α for total PSQI and after the removal of every subscale score (Table 2). Overall reliability coefficient for total PSQI was 0.791, ranging from 0.408 in healthy sleepers to 0.794 in bad sleepers with depression. Subscale scores were significantly correlated to global PSQI score in all three subgroups (Table 3).
Cronbach’s α Coefficient If Subscale Deleted for the Pittsburgh Sleep Quality Index Total Score.
Correlation Coefficients Between the PSQI Total and Component Scores.
Note. PSQI = Pittsburgh Sleep Quality Index.
*p value < .001 between the PSQI total and component score.
**p value < .05 between the PSQI total and component score.
Test–retest reliability
Healthy sleepers and OSA subjects answered PSQI on two occasions 3–9 months apart: during the initial contact and before PSG. ICC for global PSQI score was 0.997 (p < .001). ICCs for PSQI subscales ranged from 0.989 (daytime dysfunction and sleep quality) to 1.000 (sleep disturbance) (p < .001).
Construct validity
In the total sample, ESS score significantly correlated with global PSQI score (ρ = 0.333, p < .001) and subscale scores for sleep disturbances (ρ = 0.270, p < .001), daytime dysfunction (ρ = 0.385, p < .001), and subjective sleep quality (ρ = 0.272, p < .001).
The relationship between ESS and PSQI was most prominent in OSA subgroup including total PSQI, daytime dysfunction, habitual sleep efficiency, subjective sleep quality, and sleep medication use. In depression group, ESS score correlated with sleep medication use (ρ = 0.320, p < .05), while in healthy sleepers, there was no relationship between ESS and PSQI. We found no significant difference in excessive sleepiness (Pearson χ2 = 3.517, p > .05) or in mean ESS scores (Kruskal–Wallis χ2 = 4.194, p > .05) between three groups.
Criterion validity
Kruskal–Wallis test showed significant differences in total PSQI and PSQI ≥ 5 between three groups (p < .001). Post hoc tests revealed significantly higher total PSQI score in depression subgroup than in controls and OSA patients (p < .001). OSA patients had higher mean PSQI scores compared to healthy sleepers, but the difference did not reach statistical significance (p = .272). The ANCOVA showed that significant difference in PSQI total scores persisted after the adjustments for age and gender (F = 9.202, p < .001).
We explored the relationship between PSQI scores and PSG sleep parameters (sleep efficiency, sleep onset, time wake after sleep onset, and arousal index) in healthy and OSA subgroup and found no significant PSQI differences. Self-reported sleep latency did not correlate well with PSG measured sleep latency (r = .130, p = .204). Total PSQI correlated to OSA severity (ρ = 0.261, p < .05).
Discussion
The main aim of this study was to explore psychometric properties of PSQI—formal Serbian translation.
We found that PSQI-Serbian had high degree of internal consistency (Cronbach’s α 0.791 for overall sample), which is somewhat lower, but still in the line with the original study of Buysse, Reynolds, Monk, Berman, and Kupfer (1989) and similar validation studies (Bertolazi et al., 2011; Curcio et al., 2013; Shochat et al., 2007; Sohn, Kim, Lee, & Cho, 2012). Largest component to total PSQI correlation coefficients were found for sleep latency and subjective sleep quality, comparable to Buysse et al. study (1989) (habitual sleep efficiency and subjective sleep quality) and the validation study of Korean PSQI (Sohn et al., 2012) (sleep latency, habitual sleep efficiency, and subjective sleep quality).
Mean PSQI values for healthy sleepers were similar, while for subjects with OSA (4.9 ± 3.6) or depression disorder (9.0 ± 4.9) were lower than presented in other studies. For example, average score for OSA patients in the validation of the Brazilian Portuguese PSQI (Bertolazi et al., 2011) was 8.1, whereas in the Italian study (Curcio et al., 2013), it reached 11.2. Similarly, subjects with major depressive disorder had mean PSQI from 11.1 to 14.5 (Bertolazi et al., 2011; Buysse et al., 1989; Curcio et al., 2013). We recognized significant distinctions in total PSQI and PSQI ≥ 5 between the three subgroups that persisted after the age and gender adjustments, but post hoc analyses showed that the PSQI difference between healthy and OSA subjects failed to reach statistical significance. Potential explanation for this misalignment is that both subgroups belonged to a specific occupational group (commercial drivers) and thus shared a lot of similarities concerning sleep-related habits (hours of sleep, sleep latency, and work schedule). Several studies have confirmed that commercial drivers underreport sleep disorder symptoms, like excessive daytime sleepiness (Talmage, Hudson, Hegmann, & Thiese, 2008), in order to avoid potential economic and occupational consequences (Parks, Durand, Tsismenakis, Vela-Bueno, & Kales, 2009). As an indirect support for our presumption, mean ESS scores in both subgroups were also lower than expected for general population but comparable to ESS scores in a similar sample (Braeckman et al., 2011).
Our subjects with major depressive disorder answered the PSQI while actively receiving medical treatment, which is in contrast with circumstances created by Buysse et al. (1989). This was the main reason why we did not use depression subgroup for test–retest PSQI comparison and could be the reason for lower mean PSQI and ESS scores in this subgroup.
As expected, there was a significant disparity in PSG parameters between healthy sleepers and OSA subgroup, but PSQI scores and derived sleep latency did not reflect those discrepancies. This could be evidence in favor of the conclusion made by several researchers, including the authors of the questionnaire (Bertolazi et al., 2011; Buysse et al., 1989, 2008; Curcio et al., 2013) that PSQI and PSG approach the problem of sleep unevenly considering time frames and measures of sleep.
Total PSQI cutoff value of five is recommended by the authors for differentiating good and bad sleepers, with 89.6% sensitivity and 86.5% specificity (Buysse et al., 1989). In some validation studies, this high predictive power of PSQI was found at higher cutoffs (Kotronoulas et al., 2011; Sohn et al., 2012). Our best sensitivity to specificity ratio (69.7–58.5%) was at PSQI of four, probably due to similarities between two aforementioned subgroups.
Our study has several limitations. Healthy and OSA study subsamples consisted of commercial drivers only, so there was potential underreporting of subjective sleep problems due to fear of losing their job. PSG was not performed in the depression subgroup, but the diagnostic procedure excluded presence of OSA symptoms or signs. Also, previous studies do not support significant correlation between subjective and objective sleep features. We addressed the influence of covariates such as age and gender, while several other differences were identified, but not analyzed in details, since they were mostly subjective reports that could not be objectively confirmed (frequency of nightmares, head injuries, snoring, and family history of depression) and were considered as expected subgroup features.
Conclusions
Serbian version of Pittsburgh sleep quality index showed good internal consistency, test–retest reproducibility, and adequate construct and criterion validity, which supports further exploration of its efficiency as a screening tool for prediction of good or bad sleep quality in different target populations.
Footnotes
Authors’ Note
All authors have completed the Unified Competing Interest form at 
and declare that (1) some authors (1–4) had received support from Ministry of Education and Science of Serbia through the project ON 175081 (2011–2014) for the submitted work; (2) authors have no relationships with companies or other competing interests in the past 3 years that could be perceived to constitute a conflict of interest; (3) spouses, partners, or children of authors have no financial relationships that may be relevant to the submitted work; and (4) authors have no nonfinancial interests that may be relevant to the submitted work. Ministry of Education and Science of Serbia through the project ON 175081 (2011–2014) provided financial support but had no involvement concerning any part of the research.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work is supported by Ministry of Education and Science of Serbia through the project ON 175081 (2011–2014).