Data are from a mixed-methods study with maternity providers in a rural county in western Kenya. The county is described in detail elsewhere . It has eight sub-counties, each of which has a sub-county hospital, in addition to several health centers. The county population is about one million, with an estimated 40,000 births annually . The county has a low healthcare worker to patient ratio, which is likely a key factor for stress and burnout. Each 100,000 people in the county is served by 32 nurses, 19 clinical officers (non-physician clinicians trained to perform certain duties that usually require a medical doctor), and four doctors . We use the term maternity providers in this study to refer to both clinical staff such as nurses, midwives, doctors, and clinical officers as well support staff, including nurse aides and cleaners working in maternity units (providing antenatal, intrapartum, and postnatal care and support services). We included support staff because they have been shown to play an important role in maternity care and women’s experiences in other studies [42, 43].
The primary data used in the current analysis were collected through structured interviews with 101 maternity providers, along with physiologic assessments of the ANS using a measure of heart rate variability (HRV) and the HPA axis through measurement of hair cortisol. Data was collected between June and September 2019. We purposively recruited providers from 30 health facilities in the county representing those facilities with the highest volume of births: the county hospital, all the sub-county hospitals, and two to three other facilities in each of the eight sub-counties. The goal in each facility was to recruit a minimum of one or two clinical officers (if the facility had any), two or more nurses depending on the number of nurses available, and one or two support staff (nurse aids and cleaners). Two female Kenyan research assistants (RAs) with bachelor’s degrees collected all the data.
The study was approved by the Institutional Review Boards for Protection of Human Subjects of University of California, San Francisco and Kenya Medical Research Institute, and by the Kenya National Commission for Science, Technology & Innovation (NACOSTI). Approval for the study within Migori was granted by the County Commissioner and the County Director of Health. Informed consent was obtained from all participants for each study component (i.e., interview, heart rate variability, hair sample).
Interviews were conducted in English, Swahili, and Dholuo to acquire data on stress and burnout as well as potential individual and situational stressors. All data collection tools were translated and piloted with potential respondents prior to the actual study. Between three and five providers participated in the interviews in most facilities and they lasted about 40 to 60 minutes. Response rate for the interview was 100%. Heart Rate Variability was measured for all respondents using the CorSense monitor by Elite HRV in the middle of the interview—immediately after they responded to a series of questions regarding stress and burnout and before the individual and situational predictors. The sensor was placed on the respondent’s index finger, with readings transmitted via Blue tooth to the Elite HRV app on the tablet. The reading was taken in the seated position for 5 minutes, which is considered acceptable for short term HRV readings [44, 45]. After participants completed the interview, the RA asked for permission to take a sample of their hair from the scalp at the back of the head. The sample consisted of approximately 40-50 strands, 3-4 mm thick, and 3cm long; it was wrapped in aluminum foil and stored in a ziplock bag for later cortisol assay.
Our psychological measures included two self-report questionnaires assessing perceived stress and burnout. Measures of physiologic stress comprised an assessment of two primary stress response systems: the provider’s current ANS state (heart rate variability) and an assessment of the provider’s longer term, HPA axis response in the months preceding data collection (hair cortisol).
We measured perceived stress using the Cohen Perceived Stress Scale (PSS)—a global measure of perceived stress that has been validated in several countries. , including in sub-Saharan Africa [47–49]. It has 10 items asking about feelings and thoughts during the last month to assess the degree of unpredictability, uncontrollability, and overload respondents experience in their lives (Additional file 1). The PSS has undergone substantial testing for its validity and reliability. Internal consistency of the PSS has ranged from alphas of 0.69-0.91 across global studies [46, 50]. A comprehensive psychometric analysis of the measure with an Ethiopian population indicated good factorial validity for the 10 item PSS as well as internal consistency, item discrimination, and convergent validity . Validity and reliability have also been supported in Kenyan populations [49, 52]. The Cronbach alpha for our sample was 0.6. The summative PSS score ranges from 0 to 40, with scores of 0-13 considered low stress, 14-26 considered moderate stress, and 27-40 considered high perceived stress .
The Shirom-Melamed Burnout Measure (SMBM) was also used . This measure assesses the degree of emotional, physical, and mental exhaustion caused by stress. We used the 14 item version which had three subscales for physical fatigue (6 items), emotional exhaustion (3 items) and cognitive weariness (5 items) (Additional file 1). A mean burnout index is calculated for each participant, with scores ranging from 1 to 7 [53, 54]. The SMBM has undergone psychometric testing in various populations with strong evidence for its validity and reliability in different populations [53–56]. Reliability coefficients have exceeded 0.70 in most studies with adult workers in human service professions [53, 56]. Internal reliability testing with the sample in our study found a Cronbach alpha of 0.87. There are no specific cut offs for burnout. But a commonly used cut-off value for high or clinical burnout is ≥3.75, and ≤2.0 as no burnout [57, 58]. We thus considered ≤2.0 as no burnout, 2.1-3.74 as moderate burnout and ≥3.75 as high burnout. Scores for each subscale can be used as well.
Heart rate variability:
HRV is a measure of the variation in beat-to-beat interval between consecutive heart beats. Our HRV assessment specifically acquired an estimate of cardiac vagal control [59, 60]. We evaluated the electrocardiogram (ECG) of each participant to determine the time between R waves (R-R interval) in the QRS complex. We used time domain measures because of their utility and simplicity in short-term assessments including: RMSSD (root mean square of successive differences in RR intervals), lnRMSSD (the natural log of the RMSSD) and SDNN (standard deviation of all normal RR intervals). These indices reflect parasympathetic activity of the ANS, with higher values indicative of higher parasympathetic activity, which is considered more adaptive. Among other functions, parasympathetic activity of the autonomic nervous system elicits a state of relaxation, resting, or calm. Higher HRV is associated with younger age, better physical fitness, and better overall health, while lower levels of HRV have been linked to depression, anxiety, negative affect, high stress, and burnout [58–60]. A meta-analysis of research to date has supported the robust utility of HRV as a measure of stress .
Connectivity issues prevented HRV readings from being recorded for 8 participants, resulting in HRV readings for 93 participants. The reading for each participant was automatically cleaned to eliminate artifact by the Elite HRV software using algorithms they have developed for this purpose [61, 62]. LnRMSSD and SDNN scores were then calculated for each person. We examined the data points for irregularities and dropped one data point that was irregular. Average reading time for the 92 respondents was 5.03 minutes (SD=1.13, range=2.15 to 7.46).
Cortisol is a downstream hormone secreted by the adrenal glands when the HPA axis is stimulated. It can be measured in a variety of specimens, including blood, saliva, urine, and hair . Cortisol is produced primarily in hair follicles and incorporated into the hair as it grows. Levels within a specific hair segment reflect cumulative cortisol secretion within that hair growth period [64, 65]. Conceptually, the accumulation of high levels of cortisol over time may provide an indication of chronic or sustained stress over time, with each 1cm of hair from the scalp assessing stress levels for the prior month [6, 7]. There are no specified cut-offs for cortisol levels and stress but, on average, cortisol levels are higher in people with chronic stress as well as those with various health conditions . One reference range reported for cortisol in hair is 17.7-153.2 pg/mg of hair (median 46.1 pg/mg) . A more recent study  reported a hair cortisol concentration reference interval in healthy individuals with low levels of stress to be 40–128 pg/mg of hair while the range for concentrations in stressed individuals was higher (182–520 pg/mg of hair). Hair specimens in our sample were obtained from 44 respondents, mostly because respondents did not have enough hair to provide a sample. Only one person with enough hair refused to provide a sample. Samples were sent to the Stress Physiology Investigative Team (SPIT) lab at Iowa State University for analysis (details of their analytic process are in Additional file 2). Values for two respondents, with very high cortisol concentrations (i.e. >235.23 pg/mg) were winsorized (transforming extreme values to minimize the influence of outliers) to fall within 3 standard deviations .
The interview included questions regarding factors that have been theorized or found to be associated with stress and burnout in prior studies [38, 68]. These questions addressed both individual and situational or job-related factors.
- Demographic factors: age, gender, marital status, parity
- Socioeconomic factors: education, income, perceived social status, and perceived accomplishments (see wording of questions in box 1)
- Physical Health: self-rated health status, chronic disease, and exercise
Situational factors related to job demands and resources:
- Contextual factor: facility type
- Role and experience: position and years of experience
- Workload: number of workdays per week and work hours per day, overcommitment (from questionnaire described in box1), number of providers (doctors, clinical officers, nurses, and auxiliary staff) at the facility and number usually on duty during the day and at night; and average monthly deliveries.
- Availability of resources: perceived availability of work supplies, availability of essential commodities (based on a composite score of combined responses regarding availability of blood, IV infusions, uterotonics, MgSO4, and general supplies); caesarian section capability; and consistency of water and electricity.
- Experience of traumatic events: personal experience with maternal and neonatal patient deaths, and number of maternal deaths, stillbirths, and neonatal deaths recorded in facility in the last year
- Stressful interpersonal interactions: perceived disrespect from supervisors, colleagues, or patients
- Effort-reward imbalance: balance between efforts spent and rewards received (see Box 1)
Box1: Subjective perception questions
Perceived social status question stem: (Show respondents a drawing of a ladder with 10 rungs and read this to them). This ladder represents where people stand in Kenya. At the top of the ladder are the people who are the best off, those who have the most money, most education, and best jobs. At the bottom are the people who are the worst off, those who have the least money, least education, worst jobs, or no job.
Perceived social status of family growing up: Thinking of when you were growing up (before you had your own family and before you became a health care provider), where will you place your family's social status on this ladder?
Perceived social status now: Thinking of now, where will you place your social status? Select the rung that best represents where you think you stand now on the ladder.
Perceived accomplishments: “Thinking of what you wish you will have accomplished at this stage in your life, would you say you have accomplished less than you hoped, exactly what you hoped, or more than you hoped?”
Perceived availability of work supplies: On a scale of 0 to 10, where 0 means that you don't have any of the things you need to effectively do your work, such as medicines and supplies, and 10 means you have everything you need to work with, where will you place your situation in this facility?
Effort-reward imbalance and overcommitment: Measured with the Effort Reward Imbalance Questionnaire (appendix 1): a validated 16 item measure based on the work stress model to assess the balance between efforts spent (7 items), rewards received (3 items), and commitments (6 items) (Siegrist, Li, & Montano, 2014). Effort Reward Imbalance is calculated as effort score divided by reward score times a correction factor (k) used to adjust for the unequal number of items of the effort and reward scores. Overcommitment is based on the sum of responses to its 6 items, indicating excessive commitment to one’s work.
Data were imported into STATA and merged for quantitative analysis. Preliminary analysis involved factor analysis of the perceived stress and burnout items to assess construct validity, and Cronbach’s alpha to assess internal consistency reliability. We performed these assessments to assure their appropriateness for our sample before generating the summative scores.
We used descriptive statistics (means and proportions) to examine the distribution of the dependent and independent variables. We then examined the bivariate associations between the variables using crosstabulations, correlations, and unadjusted linear regressions. The scores for the psychological measures were approximately normally distributed, so untransformed scores are used for the bivariate and multivariate analysis. For HRV, we used the lnRMSSD which corrects for positive skewness. Cortisol levels were also positively skewed, which was corrected with a log transformation.
Because a number of providers were selected from each facility, we considered multilevel models to account for the clustering. However, the intraclass correlations for the null models were generally low (0.12 for perceived stress, 0.05 for burnout, 0.01 for HRV), except for cortisol which was 0.32. P-values for the Likelihood Ratio tests comparing multilevel vs. single level linear models were not significant at 0.05 or less for any of the outcome measures, suggesting the multilevel model was not significantly different from the ordinary least squares (OLS) model . However, because of differences by facility that emerged in other analyses, we employed a conservative approach, computing multilevel models in the final analyses with facility as level 2. We used restricted maximum likelihood (REML) because of its tendency for less bias in small samples . We also ran the final models as OLS models with robust standard errors to account for clustering within facilities in sensitivity analysis. Only variables that were significant in bivariate models for at least one of the outcome measures were included in the multivariate models. Models were then tested for model fit and collinearity and we removed variables that did not improve the model or were strongly correlated.
Finally, we tested if the relationship between significant situational factors (i.e. overcommitment) and burnout was mediated by perceived stress using the difference of coefficients (c-c’) method. The mediated or indirect effect is the difference between the coefficient in the burnout model without the mediator—perceived stress—(total effect: c) and the coefficient in the model with the mediator (direct effect: c’); and the proportion mediated is ((c-c’)/c) [71, 72]. We did not test the mediated effect of perceived stress on the physiological measures because they were not correlated in bivariate models. We used STATA 15.0 for all analyses .