Mixed Methods Evaluation of the ‘Caring for Providers to Improve Patient Experience’ intervention

Abstract

Objective:

To assess the impact of the “Caring for Providers to Improve Patient Experience (CPIPE)” intervention, which sought to improve person-centered maternal care (PCMC) by addressing two key drivers—provider stress and bias.

Methods:

CPIPE was successfully piloted over 6-months in two health facilities in Migori County, Kenya in 2022. The evaluation employed a mixed-methods pretest-posttest-non-equivalent-control-group design. Data are from surveys with 80 providers (40 intervention; 40 control) at baseline and endline, and in-depth-interviews with 20 intervention providers. We conducted bivariate, multivariate, and difference-in-difference analysis of quantitative data and thematic analysis of qualitative data.

Results:

In the intervention group, average knowledge scores increased from 7.8(SD=2.4) at baseline to 9.5(SD=1.8) at endline for stress(p=0.001), and from 8.9(SD=1.9) to 10.7(SD=1.7) for bias(p=0.001); perceived stress scores decreased from 20.9(SD=3.9) to 18.6(SD=5.3)(p=0.019), and burnout from 3.6(SD=1.0) to 3.0(SD=1.0)(p=0.001); with no significant change in the control group. Qualitative data indicated CPIPE had impact at multiple levels. At the individual level, it improved provider knowledge, skills, self-efficacy, attitudes, behaviors, and experiences. At the interpersonal level, it improved provider-provider and patient-provider relationships leading to a supportive work environment and improved PCMC. At the institutional level, it created a system of accountability for providing PCMC and nondiscriminatory care; and collective action and advocacy to address sources of stress.

Conclusion:

CPIPE impacted multiple outcomes in the theory of change leading to improvements in both provider and patient experience, including for the most vulnerable patients. These findings will contribute to global efforts to prevent burnout and promote PCMC and equity.

INTRODUCTION

Person-centered maternal care (PCMC), care that is responsive, respectful, and compassionate, is an important dimension of quality of care and an indicator of a human rights-based approach to care [1–4]. Yet, research continues to document poor PCMC across settings—manifesting as poor communication and lack of autonomy, and neglectful, disrespectful, and abusive treatment of women during facility-based childbirth [5–7]. This research also highlights inequities in PCMC with the most vulnerable groups having the worst experiences [6–10]. Two key drivers of poor PCMC and inequities in PCMC are provider burnout and bias [11–17].

Burnout, a state of physical, mental, and emotional exhaustion, is an important predictor of both poor patient and provider experiences [18,19]. A key symptom of burnout, depersonalization, manifests as negativism, cynicism, and poor attitudes towards patients, leading to poor patient experiences [18,19]. Further, burnout is characterized by decreased professional efficacy, resulting in poor job satisfaction [20,21]. Burnout is caused by prolonged stress from exposure to stressors—environmental factors over which people have no control [22]. For maternal healthcare workers in low-resource settings, these stressors are numerous: high workload, lack of basic resources, unsupportive work environments, limited competency to manage obstetric and newborn emergencies, and repeated trauma from avoidable morbidity and mortality, among others [23–25]. Provider burnout, while previously high, has risen to crisis levels since the COVID-19 pandemic [26,27], with high rates among providers in Sub-Saharan Africa (SSA) [28–30]. In Kenya, over 80% of maternal health providers had moderate to high perceived stress and low to high burnout levels [31,32]. Further studies among maternal health providers in Kenya highlight provider stress and burnout as key drivers of poor PCMC [11,13,14,16].

Bias—perceiving and treating people differently according to their social group—on the other hand, leads to differential patient experiences [33,34]. Bias can be implicit (unconscious or unintentional) or explicit (conscious or intentional), and manifests as differing care based on patients’ attributes such as socio-economic status (SES), racial identity, ethnicity, and age. For example, lower SES patients often have more negative healthcare experiences than higher SES patients due to providers perceiving them to be less intelligent, compliant, and uninterested in promoting their health [35–37]. Providers also spend less time with low SES patients and provide insufficient information, contributing to distrust, poor adherence to treatment recommendations, and health disparities [38–40]. Studies in SSA have highlighted the role of provider bias in PCMC, demonstrating differential care provision based on women’s appearances, perceived compliance, aptitude, and self-advocacy[13–15,17].

Patient experience is intrinsically linked to provider experience. Provider stress and burnout contribute to negative patient experiences [11–13,16], while provider bias contributes to inequities in PCMC [13–15]. Further, research shows that people are more likely to be biased when stressed [41]. Yet, there are no documented evidence-based interventions to improve PCMC that address these underlying drivers. To address this gap, we developed the “Caring for Providers to Improve Patient Experience (CPIPE)” intervention, focused on addressing provider stress and bias. CPIPE was designed iteratively guided by literature, behavior change theory—Social Cognitive Theory [42], Trauma-Informed System framework [43], and the Ecological Perspective [44]—formative research, and continuous feedback from key stakeholders in Kenya. CPIPE has five strategies—provider training, peer support, mentorship, embedded champions, and leadership engagement. The training, which includes content on PCMC, stress management, dealing with difficult situations, and implicit and explicit bias, aims to increase provider knowledge, skills, and self-efficacy to impact attitudes and behaviors toward preventing burnout, mitigating bias, and improving PCMC. The remaining intervention strategies create an enabling environment for behavior change [45]. We piloted CPIPE to assess its feasibility, acceptability, and preliminary effectiveness. The intervention development and the implementation process are described elsewhere [46,47]. In this manuscript, we assess the effect of the pilot intervention on key study outcomes. We hypothesized that CPIPE would enable providers to better manage stress to prevent burnout and increase providers’ awareness of their biases and motivation to avoid biased behaviors, which would improve both providers’ and patients’ experiences and mitigate the effects of their biases, thus reducing inequities in PCMC.

MATERIALS AND METHODS

Study design and setting

We evaluated CPIPE using a pretest-posttest non-equivalent-control group design embedded within a mixed methods approach [48,49]. The study was conducted in Migori County in western Kenya from August 2021-July 2022. Migori County has a population of about 1.1 million, with a crude birth rate of 30.8 per 1000 [50,51]. It has eight sub-counties, each with a sub-county hospital and several health centers; provider/patient ratio is estimated at 32 nurses and four doctors per 100,000 people, respectively [52]. Most births in the county (89.2%) occur in health facilities [53]. The study setting is previously described [3,54].

In consultation with our community advisory board (CAB), we selected two high volume delivery facilities (more than 500 births per year) [55]—Migori County Referral Hospital (MCRH) and Kuria West Sub County Hospital (Kehancha)—as intervention sites. We then selected four other high volume delivery facilities as control sites (Rongo, Ntimaru, Awendo, and Isibania sub-county hospitals). Allocation was not random, and more control sites were included because of the relatively fewer providers there compared to MCRH and Kehancha. The characteristics of study facilities are provided in table 1. All maternity providers—inclusive of clinical staff (nurses, midwives, doctors, clinical officers (non-physician clinicians trained to perform certain physician duties)) and support staff (nurse aides, cleaners), from antenatal, intrapartum, and postnatal care units were eligible to participate if they did not plan to relocate during the study period.

Table 1:

Characteristics of Study facilities

	Maternity bed capacity	Deliveries per year	Number of providers in maternity unit
Facility			Doctors	Nurses/ midwives	Clinical officers	Support staff
MCRH	100	4200	3	23	2	4
Kehancha	24	1300	1	7	0	3
Rongo	13	1440	1	9	2	3
Ntimaru	20	1200	0	9	2	1
Awendo	12	600	1	8	2	4
Isbania	12	600	0	7	2	3

Intervention description

The CPIPE intervention strategies are summarized in figure 1. The training content was first delivered through short didactic and interactive sessions over two days, with some content integrated into an emergency obstetric and neonatal care (EmONC) simulation to enable providers to apply concepts in the context of an emergency. A team of two midwives trained in simulation facilitation delivered the training, supported by our study team comprising of a psychiatrist and three staff members with social science training. This was followed by monthly refreshers facilitated by embedded champions, providers selected by their peers to lead activities in their facility. We also created a WhatsApp group where resources were shared weekly. Peer support groups were cadre-specific in-person groups based on the preferences of the providers, facilitated monthly by a peer leader. Discussion topics included debriefing on events at the maternity unit and brainstorming solutions; and engaging in stress management activities such as mindfulness exercises, singing, and dancing. The in-person and onsite mentorship utilized an intentional pairing based on a survey of mentor and mentee needs and preferences. Embedded champions in each facility organized intervention activities, delivered training content during monthly refreshers, modeled behavior change, and provided support and feedback. A multi-level leadership engagement started during the planning phase in the form of a CAB with representation at the County, facility, and unit levels, as well as provider and patient representatives. The CAB, which met quarterly, provided input on the intervention strategies as well as the implementation and served as a place for high level advocacy for issues at the facility level. The intervention was delivered from November 2021-May 2022. The control group did not receive any intervention during the study period.

Figure 1:

CPIPE Intervention Strategies

Data collection

We collected quantitative data through close-ended surveys with providers in the intervention and control groups before the launch of the intervention (baseline) and after the end of the intervention (endline); qualitative data were collected through in-depth interviews (IDIs) at endline. The intervention timing and data collection are shown in figure 2.

Figure 2:

Flowchart of study activities

Quantitative survey:

We first obtained a list of all providers in the maternity unit at each facility. Two female Kenyan research staff (BO and EN), trained in qualitative and quantitative research data collection, approached providers, explained the study, obtained informed consent and collected survey data and biomarkers (hair samples and heart rates). Sample size was based on feasibility given this was a pilot study. All participants who were approached agreed to participate. Baseline surveys were conducted with 83 providers from August to October 2021, and follow-up surveys with the 80 participants from May to June 2022. Three participants from the baseline cohort were unavailable at endline (two moved outside the County and one on maternity leave). One-on-one structured surveys were conducted by BO and EN in English, at a private location in the facility, using tablets programmed with REDCap [56]. Interviews lasted 40 to 60 minutes.

Qualitative:

IDIs were conducted with a sub-set of 20 surveyed providers from the intervention sites, from July-September 2022, purposively sampled based on facility and position. BO and EN conducted the IDIs in the participant’s preferred language (English, Swahili, or Luo), using a semi-structured interview guide. Interviews were audio recorded and transcribed.

Study ethics

Ethical approval was obtained from the Institutional Review Boards of the Kenya Medical Research Institute (SERU 3682) and the University of California, San Francisco (IRB number 17–22783), with additional approvals from the Kenya National Commission for Science, Technology and Innovation and the Migori County Commissioner and Director of Health. All participants provided written informed consent and received an incentive of 300 Kenyan shillings (~$3) for each interview.

Quantitative Measures

The primary outcomes were (1) stress knowledge, (2) implicit bias knowledge, (3) perceived stress, and (4) burnout. Secondary outcomes were (1) explicit bias, (2) implicit bias, (3) PCMC provision, and (4) physiologic measures of stress: heart rate variability (HRV) and hair cortisol levels. We used preexisting validated or new tools developed by our team to measure all outcomes described in box 1. Additionally, we collected socio-demographic and job-related data to control for other factors which may explain the intervention outcomes including age, gender, marital status, parity, education, perceived social status, religion, position, years of work experience, workdays and hours, and prior training on stress management, bias, and patient-provider interactions.

Box 1: Study outcome measures.

Stress knowledge	A 13-question knowledge of stress and stress management assessment adapted from existing questions [57]. Each multiple-choice question had one correct answer; responses were coded into binary (right or wrong) and summed to create a composite stress knowledge score. Scores range from 0 to 13; higher scores indicate higher stress knowledge. Cronbach alpha for the 13 items in the study sample at baseline was 0.69.
Implicit bias knowledge	A 15-question assessment of unconscious bias adapted from existing questions [58]. Each multiple-choice question had one correct answer; responses were coded into binary (right or wrong) and summed to create a composite implicit bias knowledge score. Scores range from 0 to 15; higher scores indicate higher implicit bias knowledge. Cronbach alpha for the 15 items in the study sample at baseline was 0.75.
Perceived Stress	The 10-item Cohen Perceived Stress Scale (PSS) on people’s feelings and thoughts in the past month [59]. The PSS has undergone substantial testing for validity and reliability in different populations including in the study setting [31,29,30,60]. Questions capture how nervous or stressed, unpredictable, uncontrollable, and overloaded respondents find their lives; response scales are 0 (never) to 4 (very often). Summative cores are created after recoding all response to the same direction. Scores range from 0 to 40; higher scores indicate higher perceived stress. Cronbach alpha for the 10 items in the study sample at baseline was 0.62.
Burnout	The14-item Shirom-Melamed Burnout Measure (SMBM) on feelings at work in the past month [61]. The SMBM has undergone psychometric testing in various populations with strong evidence for its validity and reliability in the study setting [31,29,30,61–63]. Questions capture three domains: physical fatigue, emotional exhaustion, and cognitive weariness, with responses ranging from 1 (never or almost never) to 7 (always or almost always). Mean scores are generated which range from 1 to 7; higher scores indicate higher burnout. Cronbach alpha for the 14 items in the study sample at baseline was 0.92.
PCMC	The 9-item provider reported PCMC scale measures care provision in the last month [16]. This scale was developed based on the patient reported PCMC scale and mixed-methods research with providers [3,7,13,14,64]. Questions gauge routine PCMC practices including introductions to patients, calling patients by name, explaining purpose of examinations, procedures, and treatments, asking questions; allowing preferred birth companions and birth positions; and negative behaviors such as verbal and physical abuse. Each question has 4 frequency response options: no, never (0); yes, a few times (1); yes, most of the time (2); yes, all the time (3). Summative scores are created after recoding all response to the same direction; higher scores indicate higher PCMC. Scores range from 0 to 27. Cronbach alpha for the 9 items in the study sample at baseline was 0.75.
Explicit bias	Two situationally specific vignettes assessing providers’ perceptions of women’s PCMC expectations and behaviors based on SES [15,17]. One vignette describes a woman with low SES and the other vignette a high SES woman; vignettes were read in counter-balanced orders to providers and followed by 8-questions assessing providers’ perceptions of the woman’s expectations for introductions, consenting, companionship, potential to cooperate, understand explanations, and exaggerate pain; and provider behavior needed to convey seriousness and gain patient cooperation. The vignettes and questions were informed by measurement of explicit bias in prior literature [65,66] and prior research in this setting [14,13,64–66]. Response options range from strongly disagree to strongly agree on a 4-point scale. Summative scores are created after recoding all responses to the same direction; higher scores indicate higher bias towards the woman in the vignette. Scores range from 0 to 27. Cronbach alpha for the 8 items in the study sample at baseline was 0.70 for the low SES vignette and 0.80 for the high SES vignette.
Implicit bias	The SES-PCMC Implicit Association Test (IAT) measures associations between women’s SES and providers’ perceptions of women as ‘difficult’ or ‘good.’ [15], programmed in Inquisit Lab version 5 [67]. The IAT is a cognitive-behavioral test that measures the strength of automatic associations between concepts in people’s minds based on a sorting task that has been used to measure implicit bias based on factors such as race, gender, SES, religion [68–70]. The IAT used for this study was first developed with providers in the study setting [15]. Attributes of ‘good’ patients were likable, cooperative, respectful, intelligent, and responsible; and attributes of ‘difficult’ patients were irresponsible, uncooperative, rude, annoying, and stupid. High SES descriptors were wealthy, well-educated, well-dressed, and a banker; and low SES descriptors were poor, uneducated, old/torn clothes, and a cleaner. The IAT score represents the difference in the average length of time it takes to sort words during various sections of the test and range from −2 and +2. Increasing positive scores indicate a stronger association of high status and good patient and low status and difficult patient; increasing negative scores indicate a stronger association of high status and difficult patient and low status and good patient.
Physiologic measures of stress	We included heart rate variability (HRV) and hair cortisol as measures of acute and chronic stress respectively [71,72]. HRV is a measure of the beat-to-beat interval between consecutive heart beats to estimate cardiac vagal tone [71,73]. HRV was measured using the CorSense finger monitor by Elite HRV [74], and the natural log of the root mean square of successive differences in RR intervals (lnRMSSD) calculated [71]. The LnRMSSD reflect parasympathetic nervous system activity [71]. Higher LnRMSSD is thus more adaptive and associated with relaxation and calm states, while low levels are associated with anxiety, high stress, and burnout [73,75]. Hair cortisol reflect cumulative secretion of cortisol within the hair growth period, with high levels indicative of chronic or sustained stress [72,76]. We obtained hair samples from providers who had enough hair (at least 2cm long) and willing to provide hair samples. Hair samples were processed at the Stress Physiology Investigative Team lab at the University of Oregon (analytic processes are described elsewhere) [31].

Statistical analysis

We used descriptive statistics (means and proportions) to examine characteristics of the sample and distributions of the outcome variables. We then conducted factor analysis of the perceived stress, burnout, explicit bias, and PCMC items to assess construct validity and Cronbach’s alpha to assess internal consistency reliability, before generating summative scores used in subsequent analysis. For respondents missing responses, we used item-mean substitution to impute scale scores [77]. This was done for only two respondents with incomplete data on the explicit bias items. We tested whether there was a change in outcome measure scores from baseline to endline using Wilcoxon paired signed-rank test for both the intervention and control groups. Multivariate linear regression models were used to determine the effect of CPIPE on the outcome measures, controlling for relevant confounders. In the model building process, we first included facility-level random components, but the ICC values were negligible. We thus present the results of the multivariate linear regression models with robust standard errors. A difference-in-difference analysis was used to estimate the attribution effect of CPIPE on outcome measures. For a given outcome, the intervention effect is the outcome change from baseline to endline for the intervention group minus that of the control group [48]. We used an intent-to-treat design, thus did not account for movement of providers. A p-value of 0.05 was considered statistically significant. Data were analyzed using STATA 15 [78].

Qualitative analysis

IDI transcripts were analyzed using a codebook thematic analysis approach [79–81]. A coding framework, developed deductively based on the interview guide, was iteratively refined with the addition of inductive codes following coding of initial transcripts. A team of four researchers (MG, BO, EN, JK) coded using the coding framework in Dedoose version 8.3.19 [82], with discrepancies addressed by discussion. Codes were queried and analytical memos drafted to identify themes with illustrative quotes.

Mixed methods integration

We used a convergent mixed-methods methods design, with the interview guide developed to help illustrate, expand, and explain the quantitative results [49]. IDI respondents were drawn from the survey sample to connect the findings. We present the quantitative and qualitative findings separately in the results section and then draw on both to interpret the findings in the discussion [83].

RESULTS

Participant characteristics at baseline

The analytic sample is the 80 providers who completed both baseline and endline surveys. There were few differences in the demographic characteristics of participants in the intervention and control groups at baseline; but all, except marital status, were not statistically significant (Table 2). Most participants were female (26 (65%) of control and 32 (80%) of intervention), between 30 and 40 years of age (20 (50%) of control and 16 (40%) of intervention) and married (34 (85%) of control and 29 (72%) of intervention). Most were nurses or midwives (25 (62.5%) of control and 31 (77.5%) of intervention), and about a third had been providers for less than 5 years. Less than 3% in both groups reported ever participating in a formal training on stress management; 1 (2.6%) and 2 (5%) in the control group reported ever having a training in unconscious bias and interpersonal interactions with patients, respectively, compared to 6 (15%) for both trainings in the intervention group.

Table 2:

Baseline characteristic of the study participants, N=80

Characteristics	Category	Control (N=40)	Intervention (N=40)	P
		Frequency (%)	Frequency (%)
Gender	Male	14 (35.0)	8 (20.0)	0.133
	Female	26 (65.0)	32 (80.0)
Age	30 or below years	9 (22.5)	12 (30.0)	0.632
	31–40	20 (50.0)	16 (40.0)
	41 or above	11 (27.5)	12 (30.0)
Current marital status	Single	2 (5.0)	11 (27.5)	0.003
	Currently Married	34 (85.0)	29 (72.5)
	Divorced/Widow/separated	4 (10.0)	0 (0.0)
Number of children	No children	1 (2.5)	5 (12.5)	0.149
	1 to 3	26 (65.0)	27 (67.5)
	4 or more	13 (32.5)	8 (20.0)
Position	Doctor	2 (5.0)	1(2.5)	0.287
	Clinical officer	3(7.5)	0(0.0)
	Nurse/Midwife	25(62.5)	31(77.5)
	Support staff	10(25)	8(20.0)
Years of work as a health provider	0 to 5 years	12 (30.0)	15 (37.5)	0.169
	6 to 10 years	19 (47.5)	11 (27.5)
	More than 10 years	9 (22.5)	14 (35.0)
Years in current facility	2 years or below	22 (55.0)	20 (50.0)
	3 or more years	18 (45.0)	20 (50.0)	0.654
Days of work per week	5 of fewer	31 (77.5)	29 (72.5)	0.606
	More than 5 days	9 (22.5)	11 (27.5)
Training on how to deal with stress	No	39 (97.5)	39 (97.5)	1.000
	Yes	1 (2.5)	1 (2.5)
Ever had training on unconscious bias (n=79)	No	38 (97.4)	34 (85.0)	0.108
	Yes	1 (2.6)	6 (15.0)
Training on interpersonal interactions	No	38 (95.0)	34 (85.0)	0.263
	Yes	2 (5.0)	6 (15.0)
Highest level of education	Secondary or below	8 (20.0)	10 (25.0)	0.847
	College (middle level)	26 (65.0)	25 (62.5)
	University or above	6 (15.0)	5 (12.5)
Perceived social status of family growing up	Bottom half	25 (62.5)	28 (70.0)	0.478
	Upper half	15 (37.5)	12 (30.0)
Perceived social status of self now	Bottom half	18 (45.0)	11 (27.5)	0.104
	Upper half	22 (55.0)	29 (72.5)
Social mobility	Upward mobility	27 (67.5)	33 (82.5)	0.163
	No change	7 (17.5)	6 (15.0)
	Downward mobility	6 (15.0)	1 (2.5)
From the County	No	10(25.0)	18(45.0)	0.061
	Yes	30(75.0)	22(55.0)
Religion	Catholics	12 (30.0)	12 (30.0)	0.597
	Methodist/Presby/Anglican	2 (5.0)	3 (7.5)
	Protestant/Pentecostal	6 (15.0)	10 (25.0)
	Seventh Day Adventist	20 (50.0)	15 (37.5)

The full sample consists of 80 providers (40 each in the intervention and control groups). Total less than 80 due to missing data are noted.

Quantitative results on impact of intervention on study outcomes

There were no statistically significant differences between the intervention and control groups in any of the key outcomes at baseline. Bivariate analysis showed slight increases from baseline to endline in stress and unconscious bias knowledge scores, as well as decreases in perceived stress and burnout scores in the control group, but these changes were not statistically significant (Table 3). There were, however, statistically significant increases in stress and implicit bias knowledge scores and decreases in stress and burnout levels in the intervention group (Table 3). The average stress knowledge score increased from 7.8 to 9.5 (p<0.001), while implicit bias knowledge increased from 8.9 to 10.7 (p<0.001) for the intervention group. On the other hand, perceived stress scores decreased from 20.9 to 18.6 (P=0.019), while burnout scores decreased from 3.6 to 3.0 for the intervention group (p=0.001). There were decreases in the explicit bias scores for both groups, but the magnitude of change was descriptively, though not significantly, larger in the intervention group. The mean score on the low SES vignette decreased from 17.4 to 15.7 (p=0.051) in the intervention group compared to 16.30 to 14.9 (p=0.081) in the control group. Mean scores for the high SES vignette also decreased from 17.9 to 15.3 (p=0.006) in the intervention group compared to 17.0 to 15.0 (p=0.011) in the control group. IAT scores decreased for both groups, with a larger and statistically significant difference in the control group (p=0.017). Self-reported provision of PCMC increased from 57.8 to 65.5 (p=0.026) in the intervention with no statistically significant change in the control group (p=0.270). There were no significant changes in HRV from baseline to endline for both intervention (p=0.871) and control groups (0.231). The average hair cortisol level however increased in the control group (p=0.043) but decreased in the intervention group (p=0.460).

Table 3:

Estimates of primary outcomes at baseline and endline for control and intervention groups

Outcome measure	Statistic	Overall estimates		P*
		Baseline	End line
Control group (N=40)
Stress knowledge	Mean (SD)	7.95 (2.77)	8.53 (2.01)
	Median (Range)	8.50 (0.00,11.00)	9.00 (3.00,12.00)	0.399
Implicit bias knowledge	Mean (SD)	9.13 (2.14)	9.45 (2.16)
	Median (Range)	9.00 (4.00,13.00)	9.50 (4.00,13.00)	0.604
Perceived Stress	Mean (SD)	20.58 (4.51)	20.05 (5.57)	0.399
	Median (Range)	20.50 (9.00,30.00)	20.00 (8.00,31.00)
Burnout	Mean (SD)	3.70 (1.10)	3.42 (1.08)	0.330
	Median (Range)	3.96 (1.14,5.29)	3.57 (1.23,5.21)
IAT (n=34)	Mean (SD)	0.67 (0.46)	0.43 (0.55)	0.017
	Median (Range)	0.81 (−0.41,1.54)	0.44 (−1.48,1.47)
Low SES vignette score	Mean (SD)	16.30 (3.27)	14.91 (3.12)	0.081
	Median (Range)	15.90 (10.00,25.00)	15.00 (7.00,20.00)
High SES vignette score	Mean (SD)	17.85 (3.22)	15.25 (4.66)	0.006
	Median (Range)	17.50 (11.00,24.15)	15.00 (7.00,26.00)
Self-reported PCMC provision	Mean (SD)	59.83 (18.76)	62.80 (16.48)	0.270
	Median (Range)	58.93 (28.57,96.43)	64.29 (28.57,96.43)
HRV (n=32)	Mean (SD)	3.71 (0.61)	3.84 (0.63)
	Median (Range)	3.68 (2.01,4.58)	3.78 (2.03,5.46)	0.231
Cortisol (N=7)	Mean (SD)	8.43 (3.03)	36.05 (58.29)
	Median (Range)	6.90 (5.55,13.35)	15.00 (5.25,167.18)	0.043
Intervention group (N=40)
Stress knowledge	Mean (SD)	7.78 (2.41)	9.50 (1.80)
	Median (Range)	8.00 (1.00,11.00)	9.50 (5,13)	<0.001
Implicit bias knowledge	Mean (SD)	8.93 (1.94)	10.70 (1.73)
	Median (Range)	9.00 (5.00,13.00)	10.00 (7.00,14.00)	<0.001
Perceived stress	Mean (SD)	20.89 (3.86)	18.63 (5.34)	0.019
	Median (Range)	21.50 (11.00,29.00)	19.00 (3.00,31.00)
Burnout	Mean (SD)	3.64 (1.05)	2.95 (0.97)	0.001
	Median (Range)	3.93 (1.07,5.21)	3.21 (1.00,4.64)
IAT (n=34)	Mean (SD)	0.75 (0.37)	0.62 (0.52)	0.209
	Median (Range)	0.79 (−0.18,1.35)	0.63 (−0.50,1.61)
Low SES vignette score	Mean (SD)	17.38 (3.06)	15.65 (3.76)	0.051
	Median (Range)	17.00 (11.00,25.00)	15.50 (7.00,23.00)
High SES vignette score	Mean (SD)	18.14 (4.38)	15.60 (4.31)	0.011
	Median (Range)	17.00 (9.00,28.00)	15.00 (7.00,23.00)
Self-reported PCMC provision	Mean (SD)	57.75 (17.16)	65.51 (13.61)
	Median (Range)	53.57 (28.57,96.43)	64.29 (35.71,92.86)	0.026
HRV (n=34)	Mean (SD)	3.70 (0.59)	3.71 (0.43)
	Median (Range)	3.76 (2.33,5.18)	3.72 (2.92,4.62)	0.871
Cortisol (N=15)	Mean (SD)	74.13 (147.38)	41.50 (72.19)
	Median (Range)	14.25 (6.30,455.70)	13.35 (4.50,278.78)	0.460

^*Wilcoxon paired signed-rank test

IAT: Implicit Association Test; SES: Socioeconomic status; PCMC: Person-centred maternity care; HRV: Heart rate variability.

Missing data on IAT and HRV are due to some providers not able to complete the IAT and loss of internet connectivity during HRV measurement resulting in data not saved, respectively. Missing data on hair cortisol are due to some providers having their hair cut too short or unwilling to provide a hair sample with many providing samples at only one time point resulting in very few paired data points (only 7 paired for control and 15 for intervention)

In multivariate analysis, controlling for relevant confounders (Table 4), these relationships persisted with greater and statistically significant increases in knowledge scores, decreases in stress, burnout, and explicit bias scores, and increases in self-reported PCMC from the baseline to endline in the intervention group, with smaller or no statistically significant changes in the control group. For example, controlling for other factors, perceived stress scores decreased by 2.5 points from baseline to endline in the intervention group (β=−2.50; CI:−4.70, −0.30), with no statistically significant change in the control group (β=−0.76; CI:−2.99,1.47). Similarly, self-reported PCMC provision increased by 8 points in the intervention group (β=8.07; CI:1.29,14.84), with no significant change in the control group (β=1.71; CI:−5.40,8.82). The change in IAT scores also remained marginally significant in the control group (β=−0.25; CI:−0.49, −0.01), but not significant in the intervention group (β=−0.05; CI:−0.27,0.17).

Table 4:

Multivariate analysis of primary outcomes stratified by study group

Study period	Control		Intervention
	Coefficient (95 % CI)	P	Coefficient (95 % CI)	P
Stress knowledge
Baseline	Reference	-	Reference
End line	0.66 (−0.23,1.55)	0.144	1.59 (0.75,2.43)	<0.001
Implicit bias knowledge
Baseline	Reference	-	Reference
End line	0.28 (−0.63,1.18)	0.543	1.64 (0.82,2.45)	<0.001
Perceived stress
Baseline	Reference	-	Reference
End line	−0.76 (−2.99,1.47)	0.498	−2.50 (−4.70, −0.30)	0.026
Burnout
Baseline	Reference	-	Reference
End line	−0.48 (−0.95,−0.02)	0.041	−0.82 (−1.24, −0.41)	<0.001
IAT
Baseline	Reference	-	Reference
End line	−0.25 (−0.49,−0.01)	0.042	−0.05 (−0.27,0.17)	0.652
Low SES vignette score
Baseline	Reference	-	Reference
End line	−1.55 (−3.01,−0.08)	0.039	−2.09 (−3.59,−0.58)	0.007
High SES vignette score
Baseline	Reference	-	Reference
End line	−2.79 (−4.78,−0.79)	0.007	−3.14 (−5.15,−1.12)	0.003
PCMC
Baseline	Reference	-	Reference
End line	1.71 (−5.40,8.82)	0.633	8.07 (1.29,14.84)	0.020
HRV
Baseline	Reference	-	Reference
End line	0.21 (−0.07,0.49)	0.146	0.01 (−0.21,0.23)	0.922
Cortisol
Baseline	Reference	-	Reference
End line	25.52 (−6.97,58.01)	0.101	−39.66 (−98.24,18.93)	0.178

Notes:

^1.Stress knowledge model controls for sex, social status, length of provider and being a clinician

^2.Implicit biased knowledge model controls for education level and length of provider

^3.Perceived stress model controls for marital status, role, and work hours

^4.Burnout model controls for sex, age, religion, social status, and work hours

^5.IAT model controls for marital status, days per week worked, and being a clinician

^6.Low SES model controls for hours worked and being a clinician

^7.High SES model controls for hours worked and being a clinician

^8.PCMC model controls for education level, religion and being a clinician

^9.HRV model controls for sex and length of provider

^10.Cortisol model controls for length of provider, hours worked and being a clinician

Differences in HRV and hair cortisol levels were not statistically significant. In difference-in-difference analysis (Table 5), only the percentage difference in implicit bias knowledge was significant, with 36% of the change attributable to the intervention. Although not statistically significant, the intervention accounted for about 81% of the decrease in perceived stress and about a third of the decrease in burnout and explicit bias and increase in PCMC provision.

Table 5:

Difference in difference analysis of CPIPE effect on different outcomes

Outcome measure	Coefficient (95% CI)	Percentage difference	P
Stress knowledge	0.90 (−0.27,2.06)	10.83%	0.130
Implicit bias knowledge	1.31 (0.13,2.49)	36.43%	0.030
Perceived Stress	−1.68 (−4.59, 1.23)	−81.35%	0.256
Burn out	−0.42 (−1.02,0.17)	−34.54%	0.160
Low SES	−0.39 (−2.34,1.57)	−32.02%	0.696
High SES	−0.43 (−2.94,2.07)	−35.25%	0.732
PCMC	6.29 (−3.34,15.92)	33.60%	0.199

Notes: DID presented for only variables with statistically significant differences in multivariate analysis

Qualitative results

We present themes based on the Social Cognitive Theory constructs—impact on knowledge, skills, self-efficacy, behavior, experience, and environment related to the intervention outcomes—organized by the Ecological Framework’s multiple levels of influence—individual, interpersonal, and institutional levels. Provider accounts of the intervention impacts, however, usually cut across different constructs as reflected in the quotations below and in table 6.

Table 6:

Intervention Impact themes and representative quotes

Level	Theme	Representative quotes
Individual	Increased knowledge and competence in stress management	“…during the training, we were taught that when you want to release the stress you sing, dance and this will help you to release stress. I have been doing the exercise whenever I am stressed.” FA-SS-0223 “Stress just comes suddenly but the techniques like grounding and the breathing in exercise helped me to calm my nerves unlike before when I could just be angry and project it to the client.” FB-CP-0102 “I now know how to calm down am not the bitter nurse I. was before, I know how to cope with the stresses that we are going through in the maternity. I can handle difficult cases, the way I use to freeze when I got emergencies, I no longer freeze I just breath in and out then I quickly do what is expected of me” FB-CP-0107
	Improved management of clinical complications and difficult situations	“Actually, we have discussed four topics, about the PPH, but for PPH am not the one who took her (mentee) through. I am the mentor, but I came to realize that there are some areas that my mentee is so good [that] I learned from her. The areas where I had weakness, she also helped me, and this is because we set the ground at the beginning that we can help each other. As a result of this we mentored each other very well.” FB-CP-0104 “Like today we had a patient who was critical in our ward and the patient husband was somehow difficult to deal with and we had the language barrier, am not a Luo and the husband is a Luo and speaking English: and Kiswahili was difficult to him. So, he was talking some nasty things in Luo about the nurses who were on duty, so somebody translated them to me, and apparently, I was the one who was handling that patient and it reached a point where I told myself like should I leave what I’m doing because of the husband. But then again. I thought of the mindful tactics, and it really helped me because I just relaxed for a moment and continued to help the patient because had I put all the stress caused by the husband, then the mother would have suffered, and at the end of it all, I managed to help the patient. FA-CP-0205 “[in a prior case a patient] she refused to be stitched until she became anaemic because of the bleeding, so she got referred and was transfused. … After I had attended the training, I learnt that when I get such a client, I don’t need to get angry with her the way I was, but I should calm down and look for a language to talk to her…I learnt the technique of being calm, good approach, not being angry and breathing in and out. At least in fact it helped me. FB-CP-0107
	Improved provider wellbeing and experience	“Previously we were having a lot of stress and we did not know how to manage such stress, so through CPIPE we have managed stress and I personally I have learnt a lot and it has helped me because I used to have high blood pressure and since the training and monthly refreshers, I was having the [blood] pressure and it stopped. I no longer take medication for pressure.” FB-CP-0204 “…sharing my stresses with my colleagues now made me to enjoy coming to work and my anger towards the patients was also delt with. Before, I could shout at the mothers so much in the maternity without bothering that the mothers are in pain and that is why maybe they are not cooperating. But nowadays, it’s not easy to find me shouting at them. Even when one [a patient] is messing around the linen, I just ask the caretaker to talk to the mothers quietly to make them cooperate. Nowadays, I no longer insult the mothers.” FB-CP-0106 “.. it helped me to understand myself better and I realized how best I can handle situations in both my workplace and my social life. FB-CP-0101
	Increased bias awareness and mitigation behaviours	“…this is because before we attended this training, we had discrimination when we looked at an individual, like, maybe this is a nurse or even a doctor then we want to give them the best attention compared to the other class of mothers who are dirty without anything. At least we can now give them equal service (with) no discrimination. If everybody can get to the seminar, we can change like me. Ever since I went to the training, I have changed.” FB-CP-0107 “In my a profession especially as a nurse when a patient come we need her to be well equipped because we know that at antenatal clinic they are usually taught how to prepare for delivery and many a time women come for delivery when they have not planned, some come without towels so initially we were evaluating them by using unconscious bias and a mother who come very organized we would attend to them very nicely and those who came unkept we would not attend to them well but now we have realized our mistakes and we no longer treat them based on their appearance.” FA-CP-0208 “There is a time, I had a girl who was fourteen years old, and she came to deliver and with the assumptions that we had that these young girls are difficult […]and we were judging them that they are not supposed to be mothers, they should be in school, and that they should cooperate the way they cooperated when they got pregnant. So, I have learnt that we should handle the girls just like any mother because pregnancy comes with pain; I explained to the girl on what I expect her to do during labor, encouraging her and talking to her and getting consent from her.” FB-CP-0114
Interpersonal	Supportive work environment	“…previously, we the providers never had time to laugh but after the training, and when we were going on with our refresher training, we would vent our stress and share your experience and you feel good about your work. At least nowadays, you will find that after we have gone through the shift report, people nowadays spare ten to fifteen minutes to vent out and share how the shift was […] well it’s kind of a debrief.” FA-CP-0201 “They [peer groups] were awesome because you get to understand what others are going through, so what others are going through is common and so you get to come up with mechanisms on how to cope with them and we shared our issues with the facility management, some were sorted, but some are still in the process. But you find that burnout which was making us to be irritable is no longer there. So, it has helped us understand that there are those things which you can change and those that you cannot change, but life will continue.” FB-CP-0109 “…[before] we as support staff could be seen as very low people socially. Nurses could just call us that you helper can you come here, but now that they have been taught about unconscious biasness, at least we now work with them like colleagues, the issues of unconscious biasness has stopped….Also, the issue of salary which I have already told you that we were paid and the remaining two months we have told today that it has been signed and we will be paid anytime and also relation between us and patients has improved unlike before. FB-SS-0116
	Improved provision of PCMC	“Like when a mother on labor comes to the maternity, I greet a patient and introduce myself to them to make them want to free with me. Like there was a day when a mother came to labor and we were on duty with a colleague nurse, he wanted to do for her VE, but the lady refused, so I went and talked to the lady and I explained for her the reason why we examine her is to know how far the baby is, and so she accepted that I do her the VE… She declined [because] the other nurse was a male one and that made her to be more difficult. So just proper communication after I introduced myself to her and explained why we are doing the examination to her and that the male nurse is a colleague who is also working in the maternity to do the same then she accepted. FB-CP-0106 “…through the training they really refer to CPIPE as being of great help, at least the idea of providers shouting and slapping the women has really changed. These trainings have really made me to know how to deal with different types of patients and their challenges.” FA-CP-0206 “Providers had fun and I can say even their attitude changed, in as much as it is so hard to gauge the attitude. But nowadays even the reception of a client is good, I can attest to that because there is an organization that call mothers and to meet quarterly and they give feedback. They even have a voice record of mothers who come to deliver in our facility, they are saying that the services at [facility A] are just wonderful in as much as we are overwhelmed with work, we usually update the mothers of their stages; like there is feedback of information and that is because of the trainings FA-CP-0201
Institutional	Increased accountability	“.. Like we have another nurse here, who like me, was so rough on patients and she has calmed down because everybody has changed. We [nurses] are all on her neck, so she definitely has to stop being rough.” FB-CP-0106 “Before when a mother entered the facility, a nurse would shout can you give me your clinic card, when did you start feeling pain? But nowadays they welcome them and even introduce themselves to the patients, they ask the patients name, how we handle patients has changed like pinching the mothers has stopped. When I first came to work here, I didn’t like it as a nurse could shout at a mother from the nursing desk, you mother I want a baby not anything else, until when I see the head is when I will come but nowadays, the nurse would stand next to the mother, talk to the mother on what to do when she is in pain. People who had bad characters here in maternity have changed…” FB-SS-0116 “It is good to train all of us because there is always rotation in departments so if we are all trained then whoever is working in OPD when brought to maternity, if trained then will work well without bringing the old nursing where we could insult and beat mothers in maternity without anyone questioning.” FB-CP-0106
	Collective action and advocacy to address sources of stress.	“If it were not for us sitting in this peer support group meetings and putting pressure on our nursing officer as a team who is also putting pressure on the top leadership[…]we saw miracles happen […]I think you did engage them well, because even if we were discussing issues like lack of toilets, we could just discuss and give the reports to the nursing officer. He could take the information to the county because the toilets have been constructed by the county. So, I feel that our complaints reached them [leadership]. The essential commodities we have complained about them for a long time. Nothing was done about it until when you [CPIPE] came in. After our peer support meetings, I don’t know how you did it, but the county team brought us the commodities which we have not been having for so many years.” FA-CP-0104 “Some of the commodities that we did not have in the maternity like sometimes we could be dealing with mothers with preeclampsia and you do not know her pressures, you find her convulsing all over sudden so when we raised a complain on the same during our peer support meetings, we were brought for a BP machine….Before you could find a BP machine without batteries but nowadays the management ensures that the BP machines have working batteries in them all the time and they are functioning…Things like gloves, strappings after we complained nowadays, we have them. FB-CP-0107 R: There was a day you came for the training, and we really complained about the problem with the toilets and the bathrooms. After you left at least we saw a change like sanitation at least has improved, now patients are having their meals…unlike before when they had to buy food from outside and also some commodities have been bought, even though they are not enough but it is better… yes like sanitation for maternity has really improved like the bad smell in the maternity is no longer there, the county unblocked the toilets, one toilet was also constructed outside here, the casuals got paid and so now they are working and doing their cleanings well and so we have seen a lot of benefits. FB-CP-0109

Notes: FA: Facility A; FB: Facility B; CP: clinical provider; SS: Support staff.

Individual level Impacts

At the individual level, participants credited the CPIPE intervention for increasing their knowledge and competency in stress management, bias mitigation, and PCMC provision, which in turn increased their confidence, attitudes, and behaviors, resulting in positive impacts on their personal wellbeing and experiences and patient experience. IDs for quotations represents the facility (FA and FB: facility A and B) and provider type (CP: Clinical Provider and SS: Support staff). These are used to avoid identifying providers.

Increased knowledge and competence in stress management

The training content on stress management was particularly salient and credited for increasing providers’ ability to cope with and manage stress. Participants discussed how increased knowledge of different stress management techniques contributed to reductions in their stress level.

“…whatever I was taught during training has been of great help to me, because initially I had a lot of stress, and when I attended the training, I was taught how to manage the stress that I may encounter; as we talk now, the stress level has reduced and now I am good.”

FA-SS-0222

Increased knowledge of stress management was reflected in participants’ descriptions of the different techniques they were using to manage stress. These included talking to a trusted person, singing, dancing, self-care activities, breathing, and other mindfulness-based techniques.

“…the topic of stress really helped me because sometimes I could get so stressed, but I could not tell anyone. So, I came to realize that [in] sharing with a colleague or a person whom I trust, then I can get relieved.”

FB-CP-0106

Improved management of clinical complications and difficult situations.

Further, training and mentorship were credited with increasing competence in the clinical management of obstetric emergencies such as post-partum hemorrhage (PPH) and birth asphyxia, both previously reported major stressors. This in turn helped improve providers’ confidence when managing emergencies, leading to decreased fear and anxiety, which improved both clinical and person-centered care during emergencies.

“It helped me a lot because whenever I have any stress in the wards, then I know how to cope with, especially PPH. The anxiety that I had when I got a PPH at least I can now handle it so well and attend to it properly without fear.”

FB-CP-0107

Providers also reported how the training enabled them to better manage difficult situations, which previously were cited as reason for disrespect and abuse of patients, contributing to improvements in their interactions with patients in such situations. Increased confidence to manage complications and difficult situations were credited for calming both the provider and the patient.

“I have really learnt a lot; before attending the trainings, I did not know how to handle difficult patients and mothers in labor. And when I attended the training, I was trained how to handle such difficult cases.”

FA-SS-0223

Improved provider wellbeing and experience

Participants expressed a positive change in their wellbeing due to a reduction in stress, which they credited to the intervention activities. Some even noted changes in their health following the intervention.

“I benefited because I was so stressed, I could sit down and just get rumbles in my stomach, but when I attended the training, I got to know how to control my stresses and it really helped me.”

FB-SS-0115

Others recounted how stress reduction techniques helped with managing their anger, which would have been directed towards patients. Providers reported higher job satisfaction because of the intervention which made them enjoy coming to work and improved their interactions with patients.

“… people like me we opened up and we got help from colleagues, something which I did not know that was so easy, that I could be feeling much lighter and happier coming to work.…Sharing my stresses with my colleagues now made me to enjoy coming to work and my anger towards the patients was also delt with.”

FB-CP-0106

Increased bias awareness and mitigation behaviors

The intervention was also credited for increasing providers’ awareness of implicit and explicit bias. This encouraged providers to reflect on their assumptions, which enabled them to control the impact of their biases and avoid discriminatory practices. For example, providers noted avoiding discrimination based on their perceptions of a patient’s social status or appearance, a common practice before the intervention.

“So, initially we were evaluating [patients] by using unconscious bias and a mother who come very organized we would attend to them very nicely; and those who came unkept, we would not attend to them well. But now we have realized our mistakes and we no longer treat them based on their appearance.”

FA-CP-0208

Others reported how the training on bias had enabled them to be more open-minded and empathize with their patients, without discriminating based on patient attributes such as ethnicity and age.

“Mostly understanding my clients because when I came to work in this facility, I was so negative. I just already had an opinion with the Kurias (an ethnic group) that they do behave like this or that. So, now am getting to understand them, and it has also helped me to open an avenue where I can begin a conversation; I can understand my clients and my colleagues other than sticking to my opinions that these people are like this, primitive and violent, which may not be so.”

FB-CP-0109

Interpersonal level Impact

At the interpersonal level, the intervention was credited for improving relationships among providers and between providers and patients leading to a supportive work environment and improved PCMC.

Supportive work environment

Participants felt that the intervention made them more collaborative and open with colleagues. They also felt the intervention led to more friendly interactions with colleagues. These facilitated a more supportive work environment, which improved experiences at work.

“It made me more open to my colleagues, I listen more and when somebody is talking, I let them talk even if I was rushing somewhere, unless it was an emergency. I give them a listening ear. I spare more time for my colleagues.”

FA-CP-0201

In addition, mentorship facilitated a non-hierarchical cross-cadre learning which enabled providers to enhance their clinical skills in a supportive environment. The peer support groups also provided an opportunity to debrief, share experiences at work and beyond, and discuss solutions together. These activities enabled providers to better understand each other.

“We could discuss the challenges that we experience both at work and out of work then we could brainstorm on the solution to these challenges, we could do some activities like singing and dancing and this I saw to have helped us in change of mind set, you become relieved.”

FB-CP −0110

Improved relationships occurred within and across cadres. Support staff noted how clinical providers now treated them with more respect. Participants also reported improved relationships with their supervisors, who also noted improvements in their staff. For instance, one maternity-in charge discussed how the training improved her understanding of staff issues and improved interactions.

“I have had staff who have had issues sometimes, but due to such a training, I have been able to give them time to explain where they have been having issues and looking on sharing basis on how best to handle certain situations. Like in a situation where someone failed to report on duty, I ask them to explain what happened and after explanation now I am able to understand that someone can be in such a situation and I am able to tell them that they are able to communicate when caught in such situations and so she appreciated because we did not take any disciplinary measures against her. We understood the situation and… I placed someone to work for her and the issue ended well. Unlike before when I could forward the case for a disciplinary action before listening to the staff.

FB-CP-0101

Improved provision of PCMC

The intervention was credited for influencing providers’ approach to patient care. Providers noted a shift in the general patient-provider interactions by improving providers’ knowledge in communicating with patients, friendly service provision, and compassionate care.

“After the trainings that I received, I was able to handle and talk to women in a friendly way without being biased, and with that the women were getting quality care from me. And during delivery, I could talk to women in a soothing way until the baby is delivered and they all go alive.”

FA-CP-0222

As seen in previous quotes, improved individual provider experience was reported to translate into better interactions with patients, thus improving PCMC. Providers shared that these positive changes were appreciated by their patients, leading to more trust, positive perceptions from the community, and increased service utilization.

“[…] it has come to a point where community people have started saying that [Facility B] is good, and they no longer go to private facilities. They are saying that ‘here we are treated well, nobody is becoming rough’.”

FB-CP-0106

Institutional level Impacts

At the institutional levels, the intervention was credited for improving accountability and advocacy with leadership which served as reinforcement for behavior change and addressed some sources of stress. Impacts on all levels led to a more positive facility culture which created an enabling environment for ongoing behavior change.

Increased accountability

Because most providers in the maternity unit participated in the training, they were holding each other accountable in many waysInitial changes in someproviders motivated even potentially resistant providers to change their own behaviors, androviders reminded each other of their biases to avoid discriminatory care in their facilities.

“[The] training on unconscious bias, the way somebody appear, you could attend to somebody based on how they look but with time, now that everybody was trained, now when unconscious (comes-up), then your colleagues are able to remind you of the unconscious [bias] training and then you get back to your conscious [self]; you think that I should treat everybody the same. The training has really helped.”

FA-CP-0201

Collective action and advocacy to address sources of stress.

Providers felt the intervention strategy to engage leadership made it easier to implement the project activities and enhanced support for the intervention. Leadership engagement, together with the peer support group strategies, were credited for facilitating collective action and advocacy to address health systems issues including lack of supplies, poor infrastructure, and delayed salaries.

“You saw it here in our meeting, we had so many challenges but so many of the challenges had been sorted. Like we had a challenge of salaries for casuals, had it been that we did not engage the county then we could have not paid them… it’s because of the study. The issues of supplies you had it mentioned in the meeting. The issue of pharmaceutical and non-pharm came out loudly in our last meeting and out of that meeting, it never took like one month before we could see the supplies trickling into our facility, so I think that leadership engagement has really brought out things to….be done in the right way.”

FA-CP-0101

DISCUSSION

The findings, consistent with our hypothesis, show that the CPIPE intervention contributed to increases in provider knowledge and competency in stress management and bias mitigation; decreases in stress, burnout, and explicit bias; and increase in self-reported PCMC. Although there were small changes in the control group, likely due to social desirability bias, repeated testing, and maturation [48], the greater changes in the intervention group provide evidence for the intervention impact. Further, the difference-in-difference analysis suggests that about a third or more of the changes may be due to the intervention, but the small sample size limited our ability to achieve significance levels for attribution effects. The decrease in average hair cortisol levels in the intervention group compared to an increase in the control group (although not statistically significant likely due to the small sample size), also suggests a potential physiologic impact of the intervention on provider stress levels. Finally, although there were no significant changes in IAT scores in the intervention group, the qualitative data highlights providers’ awareness of their biases and adoption of strategies to mitigate the effects of those biases. Qualitative data also illustrate mechanisms by which the intervention impacted study outcomes.

The reduction in provider stress and burnout levels is a critical finding, given the intervention did not directly address stressors. One potential pathway for this effect is that the intervention influenced how providers perceived these stressors. It is recognized that people exposed to the same stressors may have different physiological and psychological responses, depending on how they perceive the stressors [22]. Further, the training provided participants with knowledge and tangible skills to cope with stress and prevent burnout. Thus, although stress is unavoidable in the life of healthcare workers due to the nature of their work, the impact of stressors can be diminished by equipping them with the tools necessary to change their perception, manage the stressors, and promote positive coping mechanisms. Qualitative data also show that the intervention contributed indirectly to addressing some of the stressors, lessening the frequency of exposure to stressors. Prior interventions to prevent burnout among providers have leveraged similar mechanisms including individual-focused strategies (such as stress management training, self-care workshops, and mindfulness.) as well as structural approaches (workload or schedule changes and improving workplace environment) [84].

The decreased explicit bias scores are likely due to the increased knowledge of bias, which qualitative data show was being translated into action to avoid discrimination and provide PCMC.. The non-significant decrease in IAT scores in the intervention group, on the other hand, reflects the implicit bias training goal of increasing awareness and mitigation behaviors, rather than removing the underlying associative process [85]. Additionally, short term implicit bias trainings are insufficient to root out deeply held biases. Instead, the goal is to raise people’s awareness of implicit bias to facilitate the development of structural changes to mitigate the effects of individual biases [41,86]. The significant reduction in IAT scores in the control group was, however, unexpected; although likely due to the low test-retest reliability of the IAT [87,88]. Prior studies have shown lower subsequent IAT scores regardless of intervention exposure, calling for caution with repeated IAT measures [87,88]. The CPIPE intervention included several components recommended for integrating implicit bias recognition and management into health professions curricula, including, creating a safe and nonthreatening learning context, increasing knowledge about the science of implicit bias, emphasizing how implicit bias influences behaviors and patient outcomes, increasing self-awareness of existing biases, improving conscious efforts to overcome implicit bias, and enhancing awareness of how implicit bias influences others [89]. These likely contributed to the positive effects of the intervention with evidence that providers were more aware of their biases and adopting strategies to mitigate their effects. Increased bias awareness is a first step towards mitigation. Further, providers were learning how to replace biased responses with responses more consistent with their goals, which can help mitigate the effects of bias [86]. Emerging research also suggest that it is possible to reduce the effects of people’s bias through activities that elevate the alternate selves and goals that people endorse, without actually removing their deep-seated associative biases—referred to as sidelining implicit bias—due to the inherently complex and often contradictory nature of people [90,91].

The CPIPE outcomes are a result of the synergistic effects of the different intervention strategies. As posited in the social cognitive theory, people can change their behavior despite obstacles if they feel capable and confident in their ability to engage in the behavior (self-efficacy), believe the change benefits them (outcome expectancies), and if there are tangible goals, positive role models, and reinforcement [45]. Further, learning and behavior change occurs in a social context with a dynamic and reciprocal interaction of the person, environment, and behavior [42]. Training contributes to self-efficacy and outcome expectancies, and is necessary, though often insufficient to solely achieve and sustain behavior change. Yet, most interventions on PCMC continue to focus on only training, with most content focused on only respectful maternity care [92]. CPIPE uniquely leverages training content and delivery, as well as theory and evidence-based multilevel integrated strategies, which were critical for its success.

Peer support programs, while quite common among patient groups, are not as widely utilized among healthcare workers in SSA [93–98]. The cadre-specific in-person peer support groups provided an opportunity for maternity providers to offer emotional and social support, share knowledge and experience, practice coping mechanisms to reduce stress, and discuss ways to address bias and provide PCMC. Mentoring is an accepted approach to ongoing professional development [99–101]. The CPIPE onsite mentorship program created new relationships between providers contributing to a positive environment for compassionate care, while improving providers’ competencies to manage clinical emergencies. Embedded champions are recognized as central to the success of implementation and behavior change [102–104]. CPIPE embedded champions served as trainers, mentors, conveners, liaisons, role models, and advocates, which contributed to the intervention success, increased self-efficacy of other providers, and provided reinforcement for behavior change [42]. Finally, leadership engagement is critical for successful implementation, ownership, and sustainability of any intervention [105]. In CPIPE, leadership engagement facilitated County buy-in, implementation, and leadership access to address facility or health systems level challenges. The reciprocal interactions of individual behaviors, peers, mentors, mentees, embedded champions, and leaders all created an enabling environment for learning and reinforcing behavior change.

Our findings have several implications. First, given increasing stress and burnout among healthcare workers [27,29,31,106], CPIPE is a timely theory and evidence-based approach to prevent burnout, promote the wellbeing of providers, and improve quality of care. Second, given the role of bias in healthcare inequities and the dearth of evidence on effective ways to address it [15,17,33], CPIPE demonstrates efforts to promote health equity through bias awareness and mitigation. Third, there is a major gap in theory and evidence-based interventions to promote responsive, compassionate, and respectful maternity care [92,107]. This work extends this limited research on interventions to improve PCMC, with the added strength of addressing equity and provider wellbeing and experience. Finally, CPIPE demonstrates how to facilitate a workplace culture that both motivates and holds providers accountable. Planning for and creating policies within health systems for the implementation of similar interventions could have multiplicative effects beyond the outcomes of this study.

Limitations and strengths

First, given this was a pilot feasibility study with a small sample in one setting, generalizability is limited. The small sample size also likely contributed to the non-significant difference-in-difference results. Second, individual provider randomization was not appropriate given the facility level activities and high potential for contamination. Thus, we utilized a cluster design; but exact matching of facilities was not possible due the highly variable size of the facilities in the County. This resulted in some differences between the intervention and control sites. Third, self-reported measures are prone to social desirability bias. We included the IAT and physiologic measures of stress to address this limitation, but the low test-retest reliability of the IAT and acute nature of HRV (influenced by what is happening in the moment) limited their utility. Providers’ reluctance to have their hair cut also significantly reduced the analytic sample for hair cortisol measurement. Finally, although the goal is to improve patient experience, patient interviews were not included in this pilot phase and the one-time assessment period at the end of the intervention limits our ability to assess sustained effects.

Nonetheless, the study has several strengths. To our knowledge, this is the first study to show the effect of an integrated provider targeted intervention on perceived stress, burnout, and bias—key drivers of inequities in PCMC. The findings address a gap in efforts to improve both patient and provider experience as well as advance health equity. Second, the pretest-posttest control group design allowed us to address common threats to internal validity, thus increasing the internal validity of the results [48]. We also used validated measures to assess our outcomes further increasing internal validity. Despite our inability to include patient interviews, self-reported PCMC results are promising as prior research has shown that provider reports in this setting are concordant with patient reports, even given social desirability bias [13,14,16]. Finally, the convergent mixed methods approach provided a more nuanced understanding of the impact of the intervention. The success of this pilot sets it up for testing in a larger sample with patient reported experience measures and outcomes and with longer term follow-up to strengthen the evidence for effectiveness, generalizability, and sustainability.

Conclusions

CPIPE is a theory and evidence-based integrated provider intervention that addresses key interrelated drivers of poor PCMC—provider stress, burnout, and implicit and explicit bias—as a way of improving and achieving equity in PCMC. Through this pilot, we have demonstrated the preliminary effectiveness of CPIPE. The evaluation results show strong promise of its ability to reduce provider stress and burnout, as well as to mitigate the effects of explicit and implicit bias, to improve and achieve equity in PCMC. Further, the multilevel intervention strategies contributed to individual provider and facility level changes that created an environment conducive for providing responsive, respectful, and compassionate maternity care. CPIPE holds great promise for other settings seeking to improve both provider and patient experience, while centering the experience of the most vulnerable patients. Future studies in more diverse settings will strengthen the evidence to inform programs and policies on such interventions.

Synopsis:

CPIPE had impacts at the individual, interpersonal, and institutional level leading to improved provider experience, a supportive work environment, and accountability for equity in PCMC

Abbreviations

CPIPE

Caring for providers to improve patient experience

PCMC

Person-centered maternity care

SES

Socioeconomic status

IAT

Implicit association test

HRV

Heart rate variability

FA

Facility A

FB

Facility B

CP

clinical provider

SS

Support staff