The association between shift work disorder and turnover intention among nurses

Data were derived from the ongoing longitudinal cohort study “SUrvey of Shift work, Sleep and Health (SUSSH)” among Norwegian nurses initiated in 2008/2009. In 2008, 6000 nurses were randomly selected from the Norwegian Nurses Organisation’s (NNO) membership roll and invited to participate in the study. Most Norwegian nurses have NNO membership. A total of 2059 nurses participated in the first wave (response rate = 38.1% when excluding 600 questionnaires that were returned due to wrong addresses). In 2009, an additional sample of 905 newly educated nurses were recruited (response rate = 33.1%) to increase the size of the study population. Thus, the total number of included nurses at baseline (2008/2009) was 2964. Of these, 1863 were available for inclusion in 2015.

All nurses included at baseline have since been invited to participate in annual follow-up surveys. Every year, the nurses who responded to the first wave received invitations to participate, including questionnaires by postal mail with prepaid return envelopes. Non-responders received up to two reminders. To increase participation rate, a lottery was conducted for each wave, in which 25 nurses could win a gift card with a value of 500 NOK (~ 50 US $).

The present study reports findings based on data collected in wave 7 (2015) and wave 8 (2016) and includes only persons who (1) were still working as nurses, (2) had a work schedule that was not limited to day-shifts only, and (3) who had completed the Michigan Organizational Assessment Questionnaire assessing turnover intention (TIS) in 2016, which is the dependent variable in the present analyses. Of the 1863 participants available for inclusion, 974 were excluded based on the criteria above. This provided a final sample of 889 participants.

The demographic variables were all collected in 2015, except for sex and age, which were collected at baseline. However, the age variable was recalculated for the year 2015. In addition, we registered the following work-related variables in 2015: Number of nights shifts worked in the last year and number of years working as a nurse. Finally, we collected data on the nurses’ shift work schedules, for which we generated separate dummy variables: 1) evening shifts only, 2) night shifts only, 3) two-shift schedules (day and evening), 4) three-shift schedules (day, evening and night), 5) other schedules that included night work. The latter category served as the reference category in the coding of the dummy variables. The dummy variable for evening shifts only was excluded from the regression analyses, as there was only one person selecting this category. The shift-work schedule data was self-reported. The evening/night shift categories were largely similar for all participants and were defined as follows: Evening shifts typically begin at 14.30 and end at 22.00. Night shifts typically begin at 21.30 and end at 07.00. There is some slight variation, e.g., evening shifts that begin at 15.00 and night shifts that begin at 22.00; however, this variation is limited to such 30-min differences.

SWD was measured in 2015 with three questions based on the criteria from the third edition of the ISCD-3 (American Academy of Sleep Medicine 2014). The questions were formulated as follows: a) “Do you have a work schedule that sometimes overlaps with the time you usually sleep?”; b) “If yes, does this cause insomnia and/or excessive sleepiness due to reduced amount of sleep?”; c) “If yes, has this lasted for at least three months?”. If the nurses answered yes to all three questions, they were classified as having SWD. It is important that this classification of SWD versus non-SWD has not been validated against a clinical diagnosis. Our measure of SWD is thus only a proxy for SWD. Still, we henceforth refer to nurses who have responded yes to all three questions above as having SWD, which is in line with previous studies [19, 44].

Turnover intention was measured in 2016 with a three-item scale adapted from the Michigan Organizational Assessment Questionnaire [45]. The items were formulated as follows: 1) “I will actively look for a new job in the next year,” 2) “I often think about quitting” and 3) “I will probably look for a new job by the next year”. Responses were noted on a 5-point Likert scale from 1 (strongly disagree) to 5 (strongly agree). We added the scores of the three items into one composite score, which ranged from 3 to 15. A higher score reflects a higher degree of turnover intention. The scale had high internal consistency (α = 0.941).

Data on job demands, decision latitude, and social support at the workplace were collected in 2015. The three constructs were measured with subscales of the Swedish Demand-Control-Support Questionnaire [41, 46, 47]. Answers on all three subscales were scored on a 4-point frequency scale (often; sometimes; seldom; never). We made sum scores for all three subscales by adding the item scores for each subscale.

Psychological job demands were measured with five items covering effort, time pressure and conflicts in the work setting. The job demands subscale had good internal consistency (α = 0.778). Decision latitude was measured with six items that assess the degree of control and discretion in the work setting. The job decision latitude subscale had internal consistency below the expected alpha (α = 0.597). For this reason, the results regarding this subscale need to be interpreted prudently. Social support at work was measured with six items that capture aspects of supportive relationships and atmosphere in the work setting. The social support subscale had high internal consistency (α = 0.845).

The study was approved by the Regional Committee for Medical and Health Research Ethics of Western Norway (REK-West, no 088.08). Written informed consent was obtained from all participants.

Descriptive statistics for all relevant variables were calculated. A hierarchical multiple linear regression with nurses’ turnover intention in 2016 as the dependent variable was conducted. The first model included only SWD in 2015 as the independent variable. The second model also included age, sex, number of years working as a nurse, number of night shifts in previous year, shift work schedule, the three sum score variables capturing social support, job demands, and decision latitude, and dummy variables for each of the shift work categories outlined above. Preliminary analyses were conducted to ensure no violation of the assumption of normality, linearity, multicollinearity and homoscedasticity. Multicollinearity diagnostics revealed that two of the dummy variables for work schedules had VIF and tolerance values that were somewhat above and below the recommended thresholds, respectively. However, this threat is mitigated by the fact that these are dummy variables that are used for control purposes. We also ran the regression analyses without these variables and the pattern of results reported below remained the same. We therefore report the full regression analyses. All analyses were conducted using IBM SPSS Statistics 28 for Windows.

This study has some limitations that should be noted. The study was conducted among Norwegian nurses, of whom the vast majority were female. Furthermore, the mean age of the sample was 39.5 years, thus indicating a healthy worker effect, i.e. that turnover may have occurred at earlier ages for nurses who would therefore not be part of the current sample. While our data does not allow us to investigate this, it is likely that turnover intention would have been higher if these nurses had been included. It is also an interesting question whether SWD would have been higher in the sample of nurses who already chose to leave the job, but this is uncertain and needs to be investigated further. Consequently, the sample characteristics need to be taken into account when assessing the possibility to generalize the findings to other populations and settings. Furthermore, while the number of respondents was fairly high, it is difficult to ascertain the degree of representativeness of our sample, since we do not have data on non-participants which would have allowed for a comparison of characteristics of participants versus non-participants.

We should also note that it is a limitation that the data in our study did not allow for causal inference. Although the regression included several relevant variables, we cannot be sure if other variables would potentially affect the relationships under investigation. While the results showed that nurses who experienced low social support at work had higher turnover intention scores, our study cannot shed light on whether an increase in perceived social support (by whichever measure) would actually reduce turnover intention. It is also a limitation of our study that we do not explore moderated relationships. Future studies could explore such relationships (e.g., SWD x stress) as predictors or turnover intention. Another approach could be to conduct latent profile analysis of workers based on both demographic and work-related variables and investigate if turnover intension predicts latent profile membership [56]. Finally, we should note that the explained variance is not very high, which implies that there are unobserved predictors that may be important for the understanding of the relationships under investigation.