Psychometric properties of the Epworth sleepiness scale in Ethiopian university students

Daytime sleepiness is characterized by the subjective reporting of disturbances in sustaining alertness while awake, especially associated with quick sleep onset in sedentary situations [1]. It is usually seen in many of the sleep disorders including narcolepsy, hypersomnia, obstructive sleep apnea, delayed sleep phase syndrome, REM sleep behavior disorder, Parkinsonism, sleep-related asthma and insomnia [1]. Daytime sleepiness is a fast-growing sleep-related disorder with reported prevalence rates of 10.4–45.6% across the globe [2, 3, 4, 5, 6]. However, the twin problems of frequently underdeveloped sleep health care systems together with the lack of awareness about sleep health matters represent significant barriers to providing patient care in the developing countries of Afro-Asian region [7]. Daytime sleepiness is highly prevalent among university students in Afro-Asian countries including Ethiopia [3, 4]. Moreover, the Ethiopian health care system faces a challenging situation because of limited government resources combined more generally with the psychological and economic difficulties, which populations in developing countries face as part of everyday living [4, 7, 17].

The most objective procedure for assessing daytime sleepiness involves the use of electrophysiological methods, such as those of the Multiple Sleep Latency Test [1]. However, due to the limited health care resources of most healthcare facilities in the developing world, subjective measurements based on comprehensively validated questionnaire tools are the most practical and cost-effective tools for achieving this goal, and thus remain the most important means for screening patients to determine their daytime sleepiness. The Epworth sleepiness scale (ESS) is the most widely used tool for assessing daytime sleepiness [8]. There is considerable evidence showing that the ESS has validated psychometric properties, which make it suitable for screening daytime sleepiness among different populations across the globe [8]. The most important psychometric test properties, such as a test’s construct validity, known-group validity, internal consistency and test-re-test reliability are well established for the ESS in different populations [8]. However, some of the ESS’s validity measures, including the dimensionality and ceiling/floor effects of individual item scores, are not well established [8, 9, 10, 11, 12, 13, 14, 15]. Moreover, there is no comprehensively validated tool for screening daytime sleepiness in Africans, including Ethiopians. In view of this deficiency, the present investigators felt that an investigation would be warranted for providing a valid and easy to use questionnaire tool for screening daytime sleepiness in Ethiopians. Therefore, to address the need for establishing the psychometric properties of the ESS in Ethiopian university students, the present study was undertaken. The cosmin guideline and checklists were followed for assessing structural validity (factor analysis), reliability (internal consistency) and known group: discriminative validity in the study population.

This is the first comprehensive psychometric validation investigation of the original English version of the ESS in the African population in general and among Ethiopian university students in particular. The study assessed item analysis-the ceiling effect and the floor effect, internal consistency, internal homogeneity, known-group validity and factor analysis of the ESS. In accordance with Cosmin checklist, the study got a ‘very good’ score for methodological quality in assessment of internal consistency using Cronbach’s alpha, structural validity employing both EFA and CFA without methodological discrepancies and appropriate statistics for known-group: discriminative validity.

Despite the comparability of psychometric properties following administration of the ESS among the presently studied sample and world samples, most of the ESS item scores showed the floor effect and but only one of them showed the ceiling effect. Moreover, the ESS total score did not have either of these effects [17, 22]. The floor effect and the ceiling effects on individual ESS items is an under-investigated measure [8]. Similar to our findings in this Ethiopian student population, Hagell and Broman 2007 reported no major issues of the floor and ceiling effects for the ESS total score in a study sample of Swedish Parkinson’s patients [10]. The ceiling/floor effects may affect the responsiveness of the ESS item scores [22]. Indeed, previous studies have expressed reservations about the responsiveness of the ESS in a normative population as well as in elderly [23, 24]. One explanation for the presence of the floor effect in all but one of the ESS item scores in this study may be that the study population was non-clinical. This reasoning is supported by the finding of a floor effect for all the item scores of the Hospital anxiety and depression scale reported when tested among a general elderly Swedish population [32]. In summary, the absence of the ceiling/floor effect for the ESS total score does support its applicability in the Ethiopian students [32].

The internal consistency as assessed by the Cronbach’s alpha was good in the study population [33]. The composite reliability of the unidimensional Model-D of the ESS was 0.75, which is reflective of and therefore reinforces its internal consistency in Ethiopian university students. Previous studies have reported the Cronbach’s alpha for the ESS in the range of 0.73–0.86 [8]. In 1992, John reported a slightly lower value of the Cronbach’s alpha of 0.72 among medical students [34]. The item-total correlations for the ESS scores indicate a moderate level of internal homogeneity for the scale in this population. In 2014, Sargento and co-workers reported similar values in the range of 0.43–0.73 among Portuguese adults [9]. The inter-item correlations indicated a weak to moderate level of internal homogeneity. However, this measure of internal validity was relatively higher than that reported for the sample of Portuguese adults (r = 0.05–0.47) [9]. Similarly, Baumgartel and colleagues reported a range of r = 0.07 to r = 0.46 for the inter-item correlations for the ESS scores among an American obstetric population [12].

The diagnostic known-group validity or discriminative validity was strongly supported by the significantly higher values for all the ESS item scores as well as the ESS total score among the subset of Ethiopian students with symptoms of insomnia when compared to those who did not have insomnia symptoms. Insomnia patients usually have increased levels of daytime sleepiness; therefore, the higher values for the ESS item as well as the ESS total score establish the validity of the scale in Ethiopian university students [1]. In comparison with other sleep questionnaires validated among Ethiopians, the ESS is on par with the Leeds sleep evaluation questionnaire and slightly better than the Pittsburgh sleep quality index [17, 18]. For the Leeds sleep evaluation questionnaire, all item scores, as well as the total scores, differed significantly among known-groups, while for the Pittsburgh sleep quality index, two item scores did not differ among known groups [17, 18].

Though, model fit indices showed little difference between the four models evaluated in the study population. The validity of model-C, a 2-Factor model was not favored because, the inter-factor correlation was 0.85, which suggested poor divergent validity of two factors (Table 6, Fig. 1) [19]. Though, performance of model-B according to fit indices was satisfactory, but its application is not recommended because it has 2 items less than the original ESS [11, 16]. This might lead to loss of important information about respondent’s sleepiness [16]. Furthermore, there was no indication to delete these two items in the study population as implied by values of (i) Cronbach’s alpha if item deleted, (ii) Communality and (iii) factor loadings (Table 3, Table 6, and Fig. 1) [19, 30]. Model-D had slightly better fit compared to model-A considering values for GFI, IFI, CFI, RMR and RMSEA. However, only model-D had absolute fit as indicated by a non-significant χ² test and it also had least value for χ²/df (Table 6, Fig. 1). Therefore, the validity of Model-D, in the study population is favored. This is similar to the structural validity reports of the ESS in previous studies of Portuguese adults and Swedish Parkinson’s patients [9, 10]. However, the unidimensionality of the ESS has been debatable because of the disparate findings in different populations [8]. Smith et al. 2008 validated a slightly re-specified 1-Factor model (sans item-6 and item-8) using CFA among Australian obstructive sleep apnea patients [11]. Their model was not found to be valid in our study population. Some studies reported 2-Factor models in different Population [12, 13, 14]. However, unlike our study, Baumgartel et al. 2013 did not implement the modification indices and did not report values for any of the relative fit indices; therefore, a direct comparison is difficult [12]. In 2018, Pilcher and co-workers did not perform CFA [12]; this is contrary to recommended practices, especially when the dimensionality of the scale is debatable [19]. In 2014, Gelaye and co-workers did not incorporate modification indices into their analysis and reported values for only two categories of the fit indices, i.e., non-centrality fit indices (RMSEA and CFI) and a relative fit index (TLI) [13]. The findings of Nguyen et al. (2006) supported the inference of a 3-Factor model in Canadian patients who had complaints of snoring [15]. However, it is difficult to infer the applicability of their 3-Factor model because the study gives little detail about factor analysis, and did not employ CFA [15].

The limitations of the present study included the fact that it did not carry out assessments of the concurrent validity, convergent validity, test-re-test reliability, and had a small number of female participants, thus limiting its generalizability primarily to males. The multiple sleep latency tests were not performed for the assessment of the concurrent validity because polysomnographic testing was not available. A number of the female students did not complete the clinical interview, resulting in a somewhat unbalanced gender proportion in the final study sample. Future studies should employ Rasch analysis as it is a robust measure of item performance. Though, this was a single centric study, but MTU students participating in this study belonged to almost all regions of Ethiopia, therefore the results may be generalizable to students of other universities with some limitations.

Taking all the limitations into consideration, nonetheless, the general overall favorable psychometric properties support the validity of the ESS for the screening of daytime sleepiness among Ethiopian university students.