We adhered to Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines and methodology.
Participants
In November 2016 we conducted a cross-sectional exploratory study [
2]. obtained a random sample of 3150 U.S. registered nurses’ provided by Redi-Data, a company that maintains over 5.8 million postal addresses and over 1.8 million e-mail addresses for U.S. nurses obtained from state licensing data (more information available:
http://www.redidata.com/healthcare-lists/mailing-email-lists/state-licensed-nurses-rns-mailing-email-lists). There were 3 duplicates, resulting in emails being sent to 3147 nurses. The e-mail informed the nurses of the purpose of the study (e.g., to better understand the factors that contribute to satisfaction among U.S. nurses) and provided a link to the survey. Non-responders to the web-survey received a paper survey in the mail. From the sample of 3147 nurses, we were unable to reach 47 (no functional e-mail or address) and were notified 2 were deceased, resulting in 3098 nurses having received an invitation to participate in the study. Participation was voluntary and all responses were anonymous. Nurses who indicated they had an associate degree or higher (e.g., baccalaureate degree in nursing, masters of science in nursing, doctorate of nursing practice, or doctorate of nursing) and were not advance practice providers (i.e., certified nurse practitioners, certified registered nurse anesthetist, certified clinical nurse specialists, certified nurse midwife) were included in this analysis. We excluded advance practice providers as contributors and consequences of their work stress likely vary from other nurses given their broader scope of practice.
Study measures
The survey items can be found in the Additional file
1. Items on the survey inquired about personal characteristics and professional characteristics. The survey included questions about demographics (age, gender, relationship status [single, married, partnered, widowed], parental status [yes/no]), practice characteristics (work hours, current practice setting, years working as a nurse, highest academic degree related to nursing, advanced practice certification), satisfaction with work-life balance, and standardized instruments to measure absenteeism, work performance, burnout, depression, and fatigue.
To measure absenteeism (i.e., work days missed due to mental or physical illness) and self-rated work performance we used the World Health Organization Health and Work Performance Questionnaire (HPQ), an instrument used by the WHO in 25 countries, that has excellent reliability and validity, and has been validated in multiple occupation samples in the U.S. and abroad and in samples of individuals employed in the health care sector [
36‐
40]. Data obtained from this instrument on self-reported absenteeism and work performance has good concordance with employee archival measures of absenteeism, daily diary reports, and worker performance in a variety of professions [
36‐
38,
41]. For absenteeism, respondents were asked to indicate the number of entire work days they missed due to personal physical or mental health problems in the last month. In samples of U.S. workers, good concordance has been found between HPQ self-reported absenteeism and employer payroll records in multiple occupations (Pearson correlations of 0.66 to 0.71 for 28 day recall) [
37,
38]. We dichotomized responses into those who had been absent one or more days due to a personal health problem in the last month versus those who had not.
For work performance, the HPQ has a series of three questions where the respondent uses a 0 (worse performance) to 10 (top performance) scale to rate their own work performance. First, respondents are asked to rate the usual performance of most workers in a similar job to their own. Then, they are asked to rate their own usual job performance over the past year or two. Lastly, the respondent is asked to rate their own overall job performance on the days they worked during the past 4 weeks. These questions are general so that they apply to all occupations, but focused enough to allow for individual reflection. The first and second questions are for memory priming only, and response to the third question is used for analysis. The lower end of the scale is truncated at 0–7 as only a small percentage of respondents rate themselves less than 7 [
37,
38].
We categorized respondents into low performers (self-ratings of 7 or lower), medium performers (self-ratings of 8) and high performers (self-ratings of 9 or higher) as previous studies of U.S. workers have reported that individuals who rate themselves 7 or lower have statistically significantly lower supervisor work performance ratings than do individuals with self-ratings of 8, and that individuals who rate themselves at an ‘8’ have significantly worse supervisor work performance ratings than individuals with self-ratings of 9 and above [
37,
38,
42]. For example, in a study of reservation agents, in comparison to individuals with a HPQ work performance rating of 9 or higher, those with HPQ work performance ratings of 7 or lower had 3.2-times greater odds of poor supervisor ratings and individuals with a HPQ work performance rating of 8 had a 2.4-times greater odds of poor supervisor ratings [
38]. We further dichotomized individuals as having poor work performance or not based on if their self-rating score was less than or equal to 8 or not.
Previous validation studies in US workers have demonstrated significant associations between HPQ scores and payroll records and job performance assessments by supervisors and other records (receiver operating characteristic curves of 0.58–0.72 in US workers) [
37,
38]. The HPQ has been used widely in samples of workers [
39,
40,
43], although not specifically in nurses.
We used the full 22-item Maslach Burnout Inventory (MBI) Human Services Survey to measure burnout [
44]. The MBI includes three subscales: emotional exhaustion, depersonalization, and low sense of personal accomplishment. Individuals are asked to indicate how often they have experience various job-related feelings (response options: never, a few times a year or less, once a month or less, a few times a month, once a week, a few times a week, every day). Psychometric properties of the MBI (i.e., reliability coefficients, test re-test reliability, convergent validity, and discriminant validity) among human service professionals can be found in the manual [
1] and has recently been summarized [
45]. Previous studies showing relationships between burnout, as measured by the MBI, and health care outcomes provide additional validity data [
3,
46]. Consistent with other studies, nurses were considered to have symptoms of burnout if they scored high on the emotional exhaustion (score ≥ 27) and/or depersonalization (score ≥ 10) subscale [
47,
48].
We identified symptoms of depression by using the 2-item Primary Care Evaluation of Mental Disorders (PRIME MD) [
49], a screening tool that performs as well as longer instruments [
50]. The PRIME MD inquiries about symptoms over the past month and has a sensitivity of 86 to 96% and a specificity of 57 to 75% for major depressive disorder [
49,
50]. Similar to the approach described by West et al. [
51], we assessed fatigue on a standardized linear analog scale (0 = “As bad as it can be”; 10 = “As good as it can be”) where lower score indicates a greater degree of fatigue [
52]. Standardized linear analog scales have been widely validated across medical conditions and populations [
53‐
57].
Statistical analysis
We calculated standard descriptive statistics. Associations between variables were evaluated using Fisher exact or chi-square tests, as appropriate. We conducted multivariable analysis (forward stepping logistic regression with backwards stepping confirmation) to identify personal and professional characteristics independently associated with the dependent variables absenteeism (1 or more work days missed due to personal mental or physical health) and self-rated poor work performance (HPQ self-rated job performance of 8 or below). Variables included in the multivariable models were: relationship [not dichotomized] and parental status, work hours in the past 7 days, academic degree, practice setting, burnout, depression, fatigue, and satisfaction with work-life integration. Age and sex were kept in the models because are traditional confounders; burnout was also kept in all models. All variables entered into the models were chosen a priori. We used a 5% type I error rate and a two-sided alternative. All analysis was conducted using SAS version 9 (SAS Institute, Cary, NC).