The RU_SATED as a measure of sleep health: cross-cultural adaptation and validation in Chinese healthcare students | BMC Psychology

Linguistic validation of the Chinese RU_SATED (RU_SATED-C) scale

Using the formal procedure for linguistic validation, the original RU_SATED (v2.0) scale was translated into Chinese following Mapi instructions [20], including translation by two separate translators, qualitative interviews to determine people’s understanding of the questions in the new language (i.e., Chinese), and back-translation by two other translators. The linguistic validation process is essential to ensure that the RU_SATED (v2.0) scale is actually measuring what it was intended to measure in the newly translated language.

Step 1 Preparation: Initial planning and actions carried out before the translation process began included conceptual analysis of the original questionnaire and application for approval to use the original questionnaire. After obtaining permission from the original author (Prof. Daniel J. Buysse, DJB) of the RU_SATED scale, an e-contract was signed with the University of Pittsburgh for the preparation of the Chinese version of the RU_SATED scale.

Step 2 Forward translation: The original RU_SATED scale was translated into Chinese independently by two Chinese native speakers, a psychologist (co-author, MC), and a linguist (BY) with a high level of fluency in both English and Chinese. A panel of five local clinical and research experts (MC, BY, JW, BG, and RM) checked and compared the two translations to create the preliminary initial translated form of the scale.

Step 3 Backward translation: The back-translation into English was undertaken by two independent highly proficient bilingual English-Chinese speakers (i.e., a behavioral scientist and clinical psychologist [LD] and a behavioral scientist and physician [JL]), and was made independently of the forward translation. The original author (DJB) reviewed the two back-translations, which were rated as satisfactory.

Step 4 Pilot Testing: Eight Chinese healthcare students were surveyed to see whether they could understand the meaning of the translated items, instrument instructions, and answer choices. Pilot testing revealed that no explanations were required, with all eight individuals confirming full understanding of the RU_SATED-C scale.

Step 5 Proofreading and finalization: The research team (RM, LD, JL, MC, BY, JW, BG, and DJB) involved in the forward translation, consolidation, and backward translation processes evaluated the pre-final version of the scale and confirmed the equivalence between the Chinese and English versions. The final Chinese RU_SATED scale was delivered to the original author (DJB) and is housed electronically at the University of Pittsburgh.

Participants and procedures

For this validation study, routinely collected data were available from two sample sites (Hangzhou and Ningbo, China) and contained an assessment of sleep using the below three measures from December 2020 until January 2021. The trained investigators were responsible for the conduct of the survey and its onsite quality control. Self-administered paper-and-pencil survey was centralized at recess or evening self-study. Healthcare students were recruited by applying a stratified random sampling approach based on their academic years and majors [21]. Inclusion criteria: individuals who were able to read simplified Chinese and communicate in Mandarin. Exclusion criteria: 1) people who were reluctant to participate; 2) those who had difficulty understanding the study procedures. Given that a retest interval of two to 14 days is usually adequate [22, 23] and reproducibility of health status measures intended for longitudinal use may best be measured at intervals of 1–2 weeks [24]. 976 healthcare students responded to the baseline assessment (Time 1, T1) and 951 completed a follow-up assessment approximately 7 days later (Time 2, T2). A total of 911 questionnaires were matched by student ID at two time points. Each participant received 2 CNY (around 0.30 US dollars) upon completion of the study as compensation for their time.

Measures

RU_SATED scale

Sleep health was assessed using the RU_SATED (Regularity, Satisfaction, Alertness, Timing, Efficiency, Duration) scale, consisting of six key dimensions of sleep health that are consistently associated with various health outcomes [3]. The scale consists of six items/dimensions of sleep health and queries about sleep during the previous month. Each item is scored from 0 to 2 on a three-point Likert scale, with 0 for “never” or “rarely,” 1 for “sometimes,” and 2 for “usually” or “always.” Scoring entails summing the scores of the individual items, with total scores ranging from 0 (poor sleep health) to 12 (good sleep health).

Sleep quality questionnaire (SQQ)

Sleep quality was measured by the Sleep Quality Questionnaire (SQQ) [25]. This questionnaire evaluates two components—daytime sleepiness (four items) and sleep difficulty (six items)—of sleep quality in the last month. Each item is scored from 0 (strongly disagree) to 4 (strongly agree) on a five-point Likert scale. The overall SQQ score ranges from 0 to 40, with higher scores indicating poorer sleep quality. Psychometric data for the Chinese version of the Sleep Quality Questionnaire (SQQ-C) reveal adequate measurement properties in multi-site studies [21, 26,27,28].

Patient health questionnaire-4 (PHQ-4)

A self-report version of the Primary Care Evaluation of Mental Disorders (PRIME-MD) called the Patient Health Questionnaire (PHQ) was developed and validated in two large studies for use with general adult samples [29]. The PHQ-4 is a validated measure of mental health symptoms consisting of the first two items of the PHQ-9 and the GAD-7, respectively [30]. Each item is scored from 0 (not at all) to 3 (nearly every day). The total score ranges from 0 to 6, with a higher score indicating greater severity of anxiety or depression over the last two weeks. The Chinese version of the PHQ-4 (PHQ-4-C) and its instruction manual are publicly available and no permission is required for use [31].

Statistical analysis

Data preparation

Data were checked for data entry errors, missing data, or the presence of extreme outliers. Frequencies (%) were calculated for categorical variables, whereas means and standard deviations were computed for continuous variables. Multivariate normality was assessed via skewness and kurtosis. Data analyses were performed with JASP (v.0.16.1) and R (v.4.1.2). The packages “naniar v 1.0.0” [32], “MVN v.5.9” [33], “lavaan v.0.6-9” [34], “semTools v.0.5-5” [35], “irr v.0.84.1” [36], and “ufs v.0.5.2” [37] under RStudio were utilized to conduct the missing value analysis, multivariate normality tests, confirmatory factor analysis (CFA), longitudinal CFA (LCFA), intraclass correlation coefficient (ICC), and Cronbach’s alpha as well as McDonald’s omega. After missing value analysis, of the 911 participants, 898 (98.6%) had no missing data, while 13 (1.43%) had some missing data. Of the total 12 RU_SATED-C scale items (T1 and T2) missingness ranged from 0.11% to 0.44%. Missingness was therefore considered negligible, and listwise deletion was applied for factor analysis (i.e., structural validity and longitudinal measurement invariance) and reliability analysis (i.e., internal consistency and test–retest reliability). In other analyses (N = 911), convergent and divergent validity and reliability for other measures, missing data was replaced by the mean or median of observed values given that missing data rates did not exceed 10% [38, 39] or 5% [40]. We assessed the below measurement properties of the measures, adhering to the COnsensus-based Standards for the selection of health Measurement INstruments (COSMIN) taxonomy and guideline [41, 42].

Structural validity

Structural validity measures the degree to which the scores of an instrument are an adequate reflection of the dimensionality of the construct measured [42]. The structural validity of the RU_SATED-C scale was assessed by CFAs. Because the six items are supposed to measure one construct (sleep health), we expected that all items would load on a single factor [3], similar to that of findings in the Portuguese, Spanish, and French samples [16,17,18]. The single-factor structure of the RU_SATED-C scale was evaluated across two points in time independently (T1 and T2; a cross-sectional CFA at each time point), as well as through a LCFA approach. We applied the mean and variance adjusted diagonally weighted least squares (DWLS) estimator based on the polychoric correlation matrix to examine unidimensionality, given those responses to items in the RU_SATED-C scale are ordinal [43, 44]. In addition to the one-factor model, we examined the fit of the two-factor models that were found in the English and Japanese samples [9, 19].

Model fit indices include the chi-squared test statistic and its associated degrees-of-freedom (df) and p-value [40]. However, considering that the chi-squared test is known to be very sensitive to large sample sizes, we also included additional relevant fit indices: comparative fit index (CFI), Tucker–Lewis index (TLI), root means square error of approximation (RMSEA) and its corresponding 90% confidence interval. Scaled fit indices instead of unscaled indices were reported in this paper because the former is considered more precise [45]. Following the recommended guidelines, we considered acceptable model fit if CFI ≥ 0.90, TLI ≥ 0.90, and RMSEA ≤ 0.08 [40, 46]; good model fit if CFI ≥ 0.95 or TLI ≥ 0.95, and RMSEA ≤ 0.06 [41, 47].

Longitudinal measurement invariance

Following confirmation of the single-factor and two-factor structure of the RU_SATED-C scale, we explored longitudinal measurement invariance (LMI) in the matched sample (N = 898) across time. LCFA was used to examine four forms of increasingly restrictive invariance: configural invariance (same pattern of free loadings), metric or weak invariance (common loadings over time), scalar or strong invariance (common loadings and intercepts over time), and strict or residual invariance (common loadings, intercepts, and residual variances over time). The fit of two nested models can be compared by taking the difference of the fit indices. However, the scaled chi-square difference suffers from the same issues of significance with large sample sizes as the minimum fit function statistic [48]. Hence, we focused on changes in model fit according to CFI, TLI, and RMSEA when the scaled chi-square difference was significant [48]. Following the recommended cut-off criteria, we considered an acceptable model fit for more restrictive invariant models in the following circumstances: ΔCFI ≤ 0.010, ΔTLI ≤ 0.010, and ΔRMSEA ≤ 0.015 [49]. If at least two out of three changes in fit indices meet the cut-off criteria, we considered that longitudinal measurement invariance held [50].

Convergent and divergent validity

For assessing convergent and divergent validity, we hypothesized that the RU_SATED-C scale total score would have a moderately strong negative correlation (− 0.50 < r < − 0.30, Spearman) with the SQQ-C, given that both instruments measure sleep-related constructs, and a weak negative correlation (− 0.30 < r <  0, Spearman) with the PHQ-4-C, due to the theoretically distinct nature of sleep and mental health constructs [41].

Internal consistency

Internal consistency measures the degree of interrelatedness among measure items [42]. Internal consistency of the RU_SATED-C scale was determined by calculating ordinal Cronbach’s alpha and McDonald’s omega to accommodate categorical data [51]. Values greater or equal to 0.70 was considered sufficient evidence for internal consistency [52].

Test–retest reliability

Test–retest reliability reflects the consistency in measurement taken by the same instrument, on the same subjects, under the same or very similar conditions [53]. ICC estimated by a two-way mixed model was used to evaluate test–retest reliability of the RU_SATED-C scale. An ICC < 0.40 was considered poor, 0.40 ≤ ICC < 0.60 fair, 0.60 ≤ ICC < 0.75 good, and ICC ≥ 0.75 excellent [54].