Data from the Klokwerk + study, which is a prospective cohort study, were used (Loef et al. 2016). The aim of Klokwerk + was to study the effects of shift work on infection susceptibility and body weight, and the mechanisms underlying the health effects of shift work. The study population consisted of 611 healthcare workers aged 18–65 years from six different hospitals in the Netherlands. Hospital workers were nurses, physicians, and other (allied) health professionals. The first measurement (T0) consisting of a questionnaire and anthropometric measurements took place in September–December 2016 and the second measurement (T1) after 6 months in April–June 2017. For the current study, only hospital workers with night shifts (n = 467 at T0 and n = 388 at T1) were included. Working night shifts was defined as working shifts between 00:00 and 06.00 AM at the time of the study period.

The institutional review board of the University Medical Center Utrecht in Utrecht, The Netherlands approved the current study on March 15, 2016 (study protocol number 16–044/D, NL56022.041.16). Written informed consent was obtained from all participants.

Sleep quality. Sleep quality was measured using a single item of the Pittsburgh Sleep Quality Index (PSQI) which asks participants to indicate how they rate their overall sleep quality in the past month on a 4-point Likert scale (ranging from very good to very bad) (Buysse et al. 1989). Poor sleep quality was defined as having a very or fairly bad sleep quality.

Sleep duration. Sleep duration was derived from the Medical Outcomes Study (MOS) Sleep Scale, which measures the general usual sleep habits in the past 4 weeks (Hays et al. 2005). The MOS consists of 12 items covering 6 dimensions. One of the items measures duration of sleep by asking to report the amount of hours of sleep per day. Based on the amount of hours slept per day, the measure of sleep duration was dichotomized into recommended sleep duration (7–9 h per day) and non-recommended sleep duration (< 7 or ≥ 9 h per day) (Hirshkowitz et al. 2015).

Sleep disturbances. Sleep disturbances is one of the dimensions of the MOS Sleep Scale, and is based on four items that are scored on a 6-point Likert scale (Hays et al. 2005). The items concern problems falling asleep, how long it takes to fall asleep, not having a quiet sleep, and having problems falling asleep again after waking up during sleep time. The crude scores on the 6-point Likert scale were transformed to a 0–100% range. The percentages of the four items were averaged together into a 0–100 continuous dimension score, and a higher percentage indicated more sleep disturbances. Because of a skewed distribution, the dimension score was then dichotomized so that the upper quartile of sleep disturbances would be compared to the rest. The outcome measures were measured at both T0 and T1.

Age, sex, partner status, level of education, occupation, and chronotype were included as sociodemographic factors. These factors were only measured at T0, except for occupation. Age was a continuous variable, and sex was a dichotomous variable (male/female). Partner status was dichotomized into living together with a partner and not living together with a partner. Level of education was dichotomized into lower (elementary school to vocational education) and higher (higher vocational education/university) educated. Occupation was dichotomized into nurse and other (e.g., physicians, paramedics, caregiver). Self-reported chronotype was divided into three categories based on a self-rated single item of the Munich ChronoType Questionnaire (MCTQ): morning type, evening type, and no specific type (Roenneberg et al. 2003).

Body Mass Index (BMI), physical activity, smoking, alcohol use, and screen use were included as lifestyle factors and were measured at both T0 and T1. BMI was calculated at baseline and follow-up based on weight and height measurements performed by the research team and divided into three categories: normal weight including underweight (BMI < 25 kg/m2), overweight (BMI 25–30 kg/m2), and obesity (BMI ≥ 30 kg/m2) (WHO 2018). Physical activity was measured with the Short Questionnaire to ASsess Health enhancing physical activity (SQUASH) (Wendel-Vos et al. 2003). The number of hours per week was calculated for three categories: sports activities, activity at work, and other activities (including commuting between work and home, leisure time activity excluding sports, and domestic chores). Smoking status was divided into three categories: never smoker, former smoker, and current smoker. Alcohol use was dichotomized into > 7  glasses per week and ≤ 7  glasses per week according to the recommended intake by the Dutch Health Council (Gezondheidsraad 2015). Screen use was based on a self-constructed questionnaire about the amount of times a week devices using lights (such as television, computer, smartphone, tablet) were used in the hour before sleep. Screen use was dichotomized into less than every day of the week and every day of the week.

Working hours, amount of years working in night shifts, and average amount of night shifts per month were included as work characteristics and were measured at both T0 and T1. Working hours was divided into three categories: ≤ 24 h per week, 25–35 h per week and ≥ 36 h per week (fulltime). Amount of years working in night shifts was categorized into < 10 years, 10–19 years, and ≥ 20 years, as was done previously (Loef et al. 2019). The average amount of night shifts per month was categorized into 1–2 per month, 3–4 per month, and ≥ 5 per month (Loef et al. 2019).

Lifestyle factors and working characteristics could change between the two measurements over time (T0 and T1). Therefore, an analysis of the variance (ANOVA) was performed to explore how much variance of time-varying dependent and independent variables were attributed to between- and to within-individuals variation. Given the presence of within-individual variation between T0 and T1, between–within Poisson regression models were fitted to investigate associations between independent variables and three sleep outcomes, while taking into account within-individual variation between T0 and T1. In these models, between-individuals estimates were derived by including the person-specific overall mean of the time-varying variables, and within-individuals estimates by including the deviations from the person-specific mean of the time-varying variables. The focus of the current study was on the differences in associations between individuals, while taking into account the within-individuals variations. Poisson regression models were chosen rather than logistic regression analyses because of the dichotomous outcomes with a high prevalence. Dichotomous outcomes were chosen because of a skewed distribution of the variables.

First, the overlap between poor sleep quality, non-recommended sleep duration, and more sleep disturbances in the study population was visualized with a Venn diagram. Weighted Cohen’s κ were calculated to determine the agreement between poor sleep outcomes, based on the proportion of agreement over and above chance agreement. A weighted Cohen’s κ of less than 0 shows no agreement, 0–0.20 shows slight, 0.21–0.40 fair, 0.41–0.60 moderate, 0.61–0.80 substantial, and 0.81–1.0 perfect agreement (Landis and Koch 1977).

Second, for each of the three sleep outcomes, crude univariate analyses were performed with the sleep outcomes as dependent variables and sociodemographic factors, lifestyle factors, and work characteristics as independent variables. Then, for each of the sleep outcome, multivariable analyses were performed including all sociodemographic factors, lifestyle factors, and work characteristics in the model.

All analyses were performed using IBM SPSS Software version 28.