Effect of eHealth intervention (ADHERE) on adherence to partograph and WHO safe childbirth checklist in intrapartum care: implementation research in Ethiopia

Background

The partograph and WHO safe childbirth checklist (SCC), which include on-admission, before-birth, after-birth and discharge components, are quality improvement (QI) tools in intrapartum care that have long been recommended for use. However, poor-quality intrapartum care remains a significant barrier to improving maternal and newborn health outcomes in countries like Ethiopia, largely due to total non-use or poor adherence to these QI tools. eHealth-based interventions present promising strategies to enhance healthcare quality. Therefore, we conducted implementation research to examine the effect of ADHERE (Antenatal care, chilDbirtH carE and postnatal ca.RE), an innovative eHealth system with a clinical decision support element, on adherence to QI tools during intrapartum care in health facilities in Ethiopia.

Methods

A quasi-experimental study was conducted in 3 hospitals and 6 health centers (5 intervention and 4 control) in Ethiopia, guided by the Implementation Research Logic Model. Data were collected from labor and delivery charts by trained data collectors. A total of 2,190 charts (1,076 baseline [538 per arm] and 1,114 end-line [554 control and 560 intervention arms] charts) were included in the analysis. Difference-in-difference (DiD) analysis was used to estimate the effect of the intervention. The partograph adherence score and SCC completion rates were compared between the two arms. Statistically significant differences were determined using a p-value < 0.05 cutoff.

Results

Baseline mean partograph scores in the control and intervention arms were 4.11 and 4.23, respectively. Baseline SCC adherence rates at admission, before-birth, after-birth, and discharge were 25.8%, 24.9%, 24.9%, and 24.3% in the control arm, respectively, and 49.6%, 47.8%, 45.9%, and 45.5% in the intervention arm, respectively. The ADHERE intervention increased partograph adherence score by 4.27 units (DiD = 4.27, 95%CI [1.80, 5.09]). The ADHERE intervention increased SCC adherence by 9.3%-points (DiD = 0.093, 95%CI [0.036, 0.151]) at admission, 22.3%-points (DiD = 0.223, 95%CI [0.121, 0.325]) before-birth, and 15.2%-points (DiD = 0.152, 95%CI [0.072, 0.231]) after-birth. Conversely, the ADHERE intervention did not significantly change SCC adherence at discharge (DiD = -0.025, 95%CI [-0.071, 0.021]). The lack of significant improvement may be due to providers less attention to discharge evaluation because of work overload.

Conclusion

The ADHERE intervention significantly improved adherence to partograph and SCC at admission, before-birth, and immediately after-birth during intrapartum care. However, it did not significantly improve adherence to the discharge checklist. Therefore, ADHERE-assisted implementation of QI tools can potentially prevent adverse birth outcomes related to low-quality intrapartum care and could be considered for scale up across health facilities in Ethiopia.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12978-026-02279-z.

Plain language summary

Poor-quality childbirth care remains a significant barrier to improving health outcomes in countries like Ethiopia. eHealth interventions present promising strategies to enhance healthcare quality. We conducted implementation research to see the effect of a clinical decision support intervention on childbirth care in health facilities in Ethiopia. The study was conducted in three hospitals and six health centers: five received the new intervention and four continued with the usual care. We followed a step-by-step guide during project implementation. The data were gathered from 2,190 patient charts. We compared how childbirth care practices changed between the two groups over the two time periods (before- and after-intervention). We followed scientific procedure to declare that the differences were real. The intervention increased the correct use of partograph by 4.27 units as compared to without intervention. The intervention increased essential childbirth practices at admission by 9.3% points, practices during labor by 22.3% points, and practices immediately after birth by 15.2% points. Conversely, the intervention did not change essential childbirth practices at discharge from the facility. The intervention clearly helped health workers use the partograph correctly and follow key childbirth care steps when the woman arrived, during labor, and right after birth. Therefore, eHealth-assisted provision of key childbirth care can potentially prevent bad birth outcomes related to low-quality childbirth care and could be considered for scale-up across health facilities in Ethiopia.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12978-026-02279-z.

Introduction

The partograph (also called partogram) and the safe childbirth checklist (SCC) are quality improvement (QI) tools recommended by the World Health Organization (WHO) for intrapartum care [1–3]. The partograph, a graphic record that tracks the progress of labor and the health of mother and fetus, was introduced nearly three decades ago [4]. The WHO SCC, a list of essential practices that help health workers provide high-quality care to women giving birth, was first publicly released in 2015 [3, 5]. The partograph has effectively reduced prolonged labor (6.4% to 3.4%), labor augmentation (20.7% to 9.1%), emergency cesarean Sect. (9.9% to 8.3%), and stillbirths (0.5% to 0.3%) [6, 7]. The WHO SCC has effectively reduced stillbirths and improved maternal morbidity management, partograph utilization, and postpartum danger sign counseling [8–10]. For instance, a coaching-based implementation of SCC was associated with a 37.3% reduction in perinatal mortality rates [11]. Despite clear evidence supporting the utility of intrapartum quality maintaining tools like partograph and WHO SCC [5, 12] and the significant burden from maternal and child morbidity and mortality, poor-quality intrapartum care and limited use of quality maintain tools remain a public health concern in low- and middle-income countries (LMICs) including Ethiopia [12–17]. In Ethiopia, maternal mortality and neonatal mortality have remained high over the last two decades. According to the 2025 WHO estimate, the maternal mortality ratio was 195 deaths per 100,000 live births in 2023 [18] and, according to the Ethiopian Demographic and Health surveys further analysis report, early neonatal mortality was 27 deaths per 1000 live births in 2019 [19].

In Ethiopia also non-use (40% – 45%) [20, 21] or substandard adherence to partograph were common problems as reported by previous studies [22–24]. Similarly, the rate at which facility-based birth attendants failed to complete the WHO SCC in a typical birth was as high as 60% [15, 25]. Therefore, strategies like eHealth interventions and coaching are quite important to alleviate the challenges related to total non-use or substandard completion of the QI tools during intrapartum care [4, 11, 12, 25–30]. For instance, clinical decision support systems (CDSS) based on electronic medical records (EMRs) are promising eHealth interventions for enhancing the quality of intrapartum care in LMICs [31–33].

To overcome these challenges, our multidisciplinary team developed an innovative eHealth intervention, the ADHERE system, and deployed it at the point-of-care (POC) in public health facilities. The system consists of an electronic clinical decision support (eCDS) system integrated with an electronic partograph (ePartograph) and an electronic WHO SCC. The eCDS system allows the health care provider to review scheduled clients for each day, provides popups for out-of-range values, provides alerts to intervene with suggested interventions and provides a platform to reduce the missing of interventions necessary for the mother and their newborn. The details of the ADHERE system, including its development and implementation, were described in our former publications [34–36]. The ADHERE intervention implementation was guided by the Implementation Research Logic Model (IRLM), including the Consolidated Framework for Implementation Research (CFIR) as a determinant framework. We conducted implementation research (IR) to examine the effect of the ADHERE intervention on adherence to QI tools (partograph and the WHO SCC) during intrapartum care in Ethiopia. The study findings can provide practitioners, policy makers and development partners with evidence-based approaches for scaling up the QI initiative across health facilities in Ethiopia. This can have an impact on reducing adverse maternal and fetal outcomes related to low-quality intrapartum care.

Materials and methods

Study design

The study design was quasi-experimental. Specifically, we used a pre- and post-intervention design with a non-equivalent control group, a commonly used quasi-experimental design. This design uses a control group in place of a randomized comparator population [37–39]. We attempted to balance the two groups by selecting health facilities based on healthcare level, case-load, location, and distance from the capital of Amhara region. Three health facilities (two intervention and one control) were located in a city, two (one intervention and one control) in towns, two (one intervention and one control) in villages near Bahir Dar city, and two (one intervention and one control) in remote villages. The current implementation research report was prepared following the Standards for Reporting Implementation Studies (StaRI) checklist [40, 41] (Additional File 1).

Study setting and population

The study was conducted in three public hospitals and six health centers (five in the intervention arm and four in the control arm) in the Amhara region of Ethiopia. The health facilities represented urban, peri-urban and rural settings. The urban health facilities (three intervention and two control) were in Bahir Dar city and two district towns (Adet and Merawi). The peri-urban health facilities (one intervention and one control) were in Bahir Dar Zuria district while rural health facilities (one intervention and one control) were in remote villages (Rim and Dagi). The labor and delivery charts of the mothers were reviewed to compare changes in the implementation of the QI tools between the intervention and control health facilities over time.

Eligibility criteria

Labor and delivery mothers’ charts with recorded medical registration number (MRN) on the labor and delivery logbook were included in the study. We excluded medical charts from review when the chart lacked MCH (maternal and child health)-related records or if there was evidence that the patient was not labor and delivery client, such as charts for male patients, children, elderly women (i.e., those outside the reproductive age group), or those documented for other medical conditions. Labor and delivery charts that either lacked a partograph form or contained a form marked as non-eligible were also excluded.

Intervention

In the intervention arm, we implemented the ADHERE system for 12 months, from June 01, 2022, to May 30, 2023. The ADHERE intervention implementation was guided by the IRLM model. The implementation challenges and facilitators were also assembled following the CFIR framework. A detailed description of the IRLM structure used was included in a previous publication [34]. The primary intervention was ADHERE-assisted implementation of labor and delivery care QI tools (i.e., partograph and WHO SCC) at a POC. The ADHERE system had an electronic clinical decision support (eCDS) as a QI component. The eCDS system had features that facilitate health care providers’ (HCPs) decisions at POC. The ePartograph, a part of the ADHERE system, can ensure automated plotting or drawing of each measurement value upon entry instead of manual partograph drawing. During clinical evaluation, when HCPs enter out-of-range values for the partograph attributes that require swift intervention, the eCDS system can automatically detect and provide reminders to intervene. It can also provide a list of suggested interventions for HCPs. During WHO SCC completion, the system can display potential danger signs and other issues to be considered when HCPs hover the mouse pointer over appropriate items in each subtype of the checklist. Besides, implementation strategies were developed and executed progressively via iterative processes to tackle implementation challenges at the individual, health facility, and health system levels. The individual-level implementation strategies were training all HCPs working in MCH units on the ADHERE system and performance review. The strategies at the health facility-level were equipping health facilities with basic amenities to run the ADHERE system, providing technical support, health facilities performance review, and provision of a troubleshooting guide. We set the ADHERE system to function with a local area network (LAN) and an uninterruptible power supply (UPS) provided for the server computer set at each facility. The health system-level strategy was engaging key stakeholders, including health system managers at different levels, via co-creation workshops, iterative discussions, supportive supervisions, and review meetings. We conducted two co-creation workshops during the preparation phase of the project. We also conducted iterative discussions at each intervention health facility that enabled us to identify contextual challenges and apply implementation strategies to overcome the challenges.

During the ADHERE project implementation phase, we conducted both on-site and off-site meetings with key implementers. We also engaged multiple stakeholders (including health system managers, partners and regional research institute managers) in supportive supervision. Besides, we established an interdisciplinary implementation research team (study coordination, learning and support, and data management sub-teams) to facilitate project implementation. Overall, the strategies were tailored to the specific health facilities context following the IRLM and the implementation was strictly monitored by the implementation research team. The supervisory team has used various checklists, such as the system functionality assessment checklist and FRAME-IS checklist, to monitor implementation fidelity. Detailed information about the strategies, including information about the implementation teams, was described in a previous publication [34].

In the control arm, standard labor and delivery care continued in the health facilities. This involved paper-based modality and routine supervision through the regular system. The usual/standard labor and delivery care includes care following the paper-partograph and the paper-based WHO SCC. The routine supportive supervision includes a monthly supervision by district health office and a quarterly supervision by regional health bureau. The investigator team had also conducted monthly visits to ensure proper data capture by onsite data collectors, and identify and document any concurrent interventions. There was no evidence of concurrent intervention differentially deployed in the control health facilities during the intervention period.

Study outcomes and measurement

The study outcomes were adherence to the partograph and the WHO SCC. Adherence to the partograph was measured based on nine parameters (Table 1). The nine parameters were selected because they are universally practiced in public health facilities in Ethiopia due to their clinical significance [42]. These were fetal heart rate (FHR), amniotic fluid status, molding of fetal skull, cervical dilation, descent of fetal head, uterine contraction, maternal body temperature, maternal pulse rate, and maternal blood pressure (BP). A score of 1 was assigned to each parameter correctly plotted on the partograph, and a score of 0 was assigned otherwise. A total score, ranging from 0 to 9, was calculated by summing the scores of individual parameters. Adherence was classified as either full or not full based on this total score. Full adherence was defined as all parameters being correctly recorded, resulting in a total score of 9.

Table 1.

Description of the nine partograph parameters

Fetal condition
1.	Fetal heart rate plotted every 30 mimutes
2.	Color of liquor (am niotic fluid) plotted per every vaginal examination (atleast 4 hours a part)
3.	Molding of the fetal skull plotted at every 4 hours
Labor progress
4.	Cervical dilation plotted every 4 hours
5.	Decent of fetal head plotted every 4 hours
6.	Frequency and strength of uterine contractions plotted every 30 minutes
Maternal condition
7.	Maternal pulse rate plotted every 30 minutes
8.	Maternal blood pressure plotted every 4 hours
9.	Maternal body temperature plotted every 2 hours

In this study, nine partograph parameters were used, representing three distinct conditions of labour: three for maternal condition, three for fetal condition and three for labor progress

The other outcome was the completion status of WHO SCC. There are four categories of SCC based on the timing of the application: on-admission, before childbirth (before push), after childbirth (within 1 h after birth), and on-discharge. The admission checklist consisted of 8 essential practices. The before-childbirth and after-childbirth checklists had 6 and 9 essential practices, respectively. The discharge checklist also consisted of 8 essential practices (Additional File 2). We created a binary outcome for each category using an all-or-none approach full adherence versus non-adherence. Full adherence to each SCC category was defined as having all items within each checklist category completely addressed.

Sample size

The sample size was determined for baseline and end-line assessments. The sample size was determined using the two-population proportion formula by Epi-Info version 7.2.4.0 STAT calc. The sample size was calculated by taking a 21.5% proportion of full adherence to partograph in a study done in Ethiopia (P₁ = 0.215) [42], 80% power, 95% confidence level, 5% marginal error, 10% difference due to the intervention (P₂ = 0.315), 1.5 design effect and 15% non-response rate. The minimum sample required was 1,076 (538 participants to each arm).

Data collection

A checklist was prepared to extract data from the client’s medical record. The survey tool was developed following the national guideline for labor and delivery care [43–45]. We conducted baseline and end-line surveys at control and intervention arms. For the baseline survey, medical record data were accessed between May 01, 2022, and May 30, 2022, in health facilities of both arms. Data were collected by BSc midwives using tablet computers and the Epicollect5 software, under the supervision of two members of the investigative team. Data collectors uploaded the data daily to the server computer. Data collectors received two days of training, including pretesting of the data extraction checklist on the second day. The checklist was amended according to the feedback obtained from the pretest. The amendment includes adjustments to the flow, skip rules and clarity of the questions. The use of data collectors with a BSc midwifery background also facilitated data extraction from labor and delivery charts due to their familiarity with MCH forms and guidelines. For the end-line survey, data were extracted daily from each control health facility for the whole project implementation period by trained data collectors deployed for this purpose (one data collector per health center and two data collectors per hospital). For the intervention group, data were extracted from the ADHERE database. The data anonymity was maintained during data extraction.

Data analysis

Descriptive statistics (frequencies, proportions, percentages, mean, and standard deviations) were computed and presented in tables or graphs with appropriate textual descriptions. The chi-square test was used to compare characteristics of women in the control and intervention arms across study time. Age, gravidity, parity, and marital status had missing values, ranging between 3.2% and 11.7%. We applied median and mode imputation approaches for continuous and categorical variables, respectively. The sensitivity of the imputed versus non-imputed was checked by comparing the outputs of complete-case analysis (non-imputed) and imputed analysis after running each separately. The results were found insensitive because the imputation did not significantly change the DiD model estimates magnitude and direction of effect as compared with the results of the model with no imputation.

We applied difference-in-differences (DiD) analysis. The DiD was used to evaluate the effect of ADHERE-assisted implementation of QI tools during intrapartum care on providers’ adherence to partograph and WHO SCC. DiD compares changes in an outcome over time between an intervention and a control arm. The DiD estimates the average effect of an intervention on outcomes in the treated group, known as the average treatment effect on the treated (ATET) [46–48]. In this paper, we used two sets of DiD models: linear-based DiD for the continuous outcome (partograph adherence score) and non-linear-based DiD for binary outcomes (WHO SCC).

In the linear-based DiD models, we used DiD in a standard linear regression Eq. (1) given below.

Where; Y_it is the outcome of interest (i.e., partograph adherence score), group_i is a binary variable showing the intervention group, post_t is a binary variable for the post-intervention period, group_i *post_t represents the interaction term, and ε_it is the residual error term.

Based on the regression Eq. 1, the unadjusted DiD estimator is derived as indicated in Eq. (2) below.

Where G indicates treatment group assignment (G = 1 for the intervention group and G = 0 for the control group). Y represents the outcome such that Y indicates the observed FQI score or ICSI score, Y^G indicates the FQI score or ICSI score counterfactual on intervention time. ΔY & ΔY^G are the change in outcome and counterfactual between the pre- and post-intervention time points.

The equation given above (2) is extended to include covariates to generate an adjusted DiD estimate. In addition to the treatment effect across periods, we considered women’s characteristics (such as age, gravidity, and parity) and health facility-related factors (such as health facility type and health facility location) as covariates in the model.

In estimating the DiD, the standard cluster-robust variance estimator (CRVE) is recommended to account for clustering, especially in cases where an intervention is assigned at the group level, such as a hospital or department [46]. In this study, we considered clustering at the health facility level since the ADHERE system was deployed at the health facility unit. However, the use of CRVE in a study with few clusters (i.e., less than 40) leads to a downwards-biased CRVE and, upon bias-correction, even leads to over-rejection if the critical values are from the standard normal distribution [49, 50]. A wild cluster bootstrap (WCB) is a recommended method to adjust the variance estimate for a small number of clusters [50–52]. Therefore, we applied the WCB method to correct the variance estimates since the number of clusters involved was fewer (i.e., 9 clusters) than the recommended level. Furthermore, Roodman et al. (2019) recommend the WCB with the Webb six-point distribution when the cluster number is less than 12 [52]. We used boottest command in Stata to analyze the WCB test statistics. The boottest package performs wild bootstrap tests of linear hypotheses with fast speed as a postestimation command. It reports the t-statistic from the Wald test and its bootstrapped p-value. It can also invert these tests to construct confidence sets for the default confidence level, which is normally 95% [52].

Besides the DiD based on the standard linear model, we fitted mixed-effects linear regression models to account for group (i.e., health facility) level variations in addition to individual-level variations. In linear-based models, the DiD is the coefficient for the interaction term between the intervention group and post-intervention time [46, 53].

Where; Y_ijt is for the outcome of interest (i.e., partograph adherence score), group_j is a binary variable showing the intervention group, post_t is a binary variable for the post-intervention period, group_j *post_t represents the interaction term, u_j is a random intercept for health facility-level variation and ε_ijt is the individual-level residual error term. Equation 3 is extended to include additional covariates.

In the nonlinear-based DiD, we used a mixed-effect logistic regression model. The mixed-effect logistic regression model DiD was used to evaluate the effect of the ADHERE intervention on adherence to each subtype of WHO SCC. In nonlinear-based models, the coefficient for the interaction term between the intervention group and post-intervention time cannot be directly interpreted as DiD effect [54]. Instead, scholars recommend using the marginal effects analysis to estimate the average marginal effects (AMEs) [46, 53–56]. The basic DiD in the mixed-effects logistic regression Eq. (4) is given below.

Where; Logit (P(Y_ijt =1) is the probability for the outcome of interest (i.e., adherence to WHO SCC), group_j is a binary variable showing the intervention group, post_t is a binary variable for the post-intervention period, group_j *post_t represents the interaction term, u_j is random intercept for health facility-level variation and ε_ijt is individual-level residual error term.

Equation 4 was extended to include additional covariates on which the adjusted DiD model is based. Three models were fitted: model-0 (a model with random-effect only), model-1 (simple DiD model) and mode-2 (adjusted DiD model). Model-0 is a null model used to evaluate the importance of accounting for variation between health facilities. We checked it using the intraclass correlation coefficient (ICC) and the likelihood ratio (LR) test. Besides, we used model comparison statistics like Log-likelihood (LL), Akaike’s Information Criterion (AIC) and Bayesian Information Criterion (BIC) [55, 57] to choose the best-fitted model. The LL value provides information about how well the model fits the data. The higher (less negative) the LL values, the better the fit is the model. The AIC and BIC are also the most commonly applied criteria to choose among different models by balancing model fit and complexity. A model with lower AIC and BIC is a better fit model [57]. In the null models, the ICC for the on-admission checklist, before-birth checklist, after-birth checklist, and on-discharge checklist were 66.36%, 67.23%, 68.85% and 66.61%, respectively. The ICC values from the null models were significantly high, supporting the use of mixed-effects models to account for variation between health facilities. The LL for on-admission, before-birth, after-birth, and on-discharge checklists consistently increased from model-0 to model-2, indicating model-2 was a better fit. The AIC for the discharge checklist also consistently reduced from model-0 to model-2, whereas for other WHO SCC subtypes showed reduction from model-0 to model-1 and relatively stable from model-1 to model-2. The details of the model comparison statistics for each outcome are given as supplementary information (Additional File 3). Based on the mixed-effects logistic regression Eq. 4, the unadjusted DiD estimator is derived as indicated in Eq. (5) below.

Where; Y = 1 is the event of interest (i.e., adherence to WHO SCC = 1), G is study groups (1 for intervention group and 0 for control group), T is time period (1 for end-line and 0 for baseline), P(Y = 1|G = 1, T = 1) is the probability of adherence to WHO SCC in the intervention group at end-line, P(Y = 1|G = 1, T = 0) is the probability of adherence to WHO SCC in the intervention group at baseline, P(Y = 1|G = 0, T = 1) is the probability of adherence to WHO SCC in the control group at end-line, and P(Y = 1|G = 0, T = 0) is the probability of adherence to WHO SCC in the control group at baseline.

Equation 5 was extended to include covariates to generate an adjusted DiD estimate. In this study, the margins command in Stata was used to estimate the DiD effect after executing the mixed-effects logistic regression model. A significant ADHERE intervention effect was declared at a p-value less than 0.05. Data analyses were performed by using Stata version 17.

Results

A total of 2,190 labor and delivery charts were retrieved during the baseline and end-line evaluation periods. Of which, 1,076 (538 from each arm) charts were retrieved at baseline evaluation and 1,114 (554 from control and 560 intervention arms) charts were reviewed at end-line evaluation (Fig. 1).

Fig. 1.

Study sample size distribution flow chart, Amhara region, Ethiopia

At the baseline survey, 86.4% of the mothers in the control arm and 86.6% of the mothers in the intervention arm were aged between 20 and 34 years old. 62% of the mothers in the control and 60.2% of the mothers in the intervention group were multigravida. 36% of the control and 36.8% of the intervention arm mothers were nulliparous. Chi-square test results showed no significant differences in maternal characteristics at the baseline assessment.

In the end-line survey, 90.1% of the mothers in the control and 85.7% the mothers in the intervention arm were aged between 20 and 34 years old. Nearly 58% of the mothers in the control and 50% of the mothers in the intervention group were multigravida. 37% of the control and 41.6% of the intervention arm mothers were nulliparous. Chi-square test results showed no significant differences in maternal characteristics at the end-line assessment, except for gravidity (Table 2).

Table 2.

Participants characteristics at baseline and end-line by intervention arms, Amhara region, Ethiopia

Variable	Baseline (n = 1,076)		X2(p-value)	End-line (n = 1,114)		X2(p-value
	Control (n = 538)	Intervention (n = 538)		Control (n = 554)	Intervention (n = 560)
	№ (%)	№ (%)		№ (%)	№ (%)
Age of women
15–19	29 (5.4)	31 (5.8)	0.17(0.917)	21 (3.8)	34 (6.1)	5.21(0.074)
20–34	465 (86.4)	466 (86.6)		499 (90.1)	480 (85.7)
35+	44 (8.2)	41 (7.6)		34 (6.1)	46 (8.2)
Gravida
Primigravida	148 (27.5)	171 (31.8)	2.96(0.228)	177 (32.0)	220 (39.3)	7.77(0.020)*
Multigravida	337 (62.6)	324 (60.2)		321 (57.9)	279 (49.8)
G/multigravida	53 (9.9)	43 (8.0)		56 (10.1)	61 (10.9)
Para
Nullipara	166 (30.8)	196 (36.4)	5.94(0.115)	204 (36.8)	233 (41.6)	2.91(0.407)
Primipara	197 (36.6)	171 (31.8)		174 (31.4)	166 (29.7)
Multipara	144 (26.8)	149 (27.7)		144 (26.0)	129 (23.0)
G/multipara	31 (5.8)	22 (4.1)		32 (5.8)	32 (5.7)
Marital status
On marriage	537 (99.8)	535 (99.4)	2.00(0.367)	551 (99.5)	546 (97.5)	7.11(0.008)*
Never married	0 (0.0)	2 (0.4)		3 (0.5)	14 (2.5)
Divorced	1 (0.2)	1 (0.2)		0 (0.0)	0 (0.0)

^*statistically significant at p-value < 0.05

Partograph adherence indices at baseline by study arms

At the baseline survey, the top two properly plotted partograph indices were uterine contraction and FHR in both arms. The uterine contraction was properly plotted in 64.7% the control and 68.4% of the intervention. Fetal heart rate was also properly plotted in 64.5% of the control and 68.0% of the intervention. Maternal pulse rate was the third most properly plotted condition, 53.0% in the control arm and 55.2% in the intervention arm. On the other hand, maternal temperature and molding of the fetal skull were the two least properly plotted conditions in both arms. The maternal body temperature was properly plotted in 33.6% of the control and 29.7% of the intervention arms. The molding of the fetal skull was properly plotted in 31.4% of the control and 31.0% of the intervention arms. Cervical dilation was properly plotted in 44% of the control and 50% of the intervention arms (Fig. 2). The chi-square test showed no significant differences in adherence to partograph adherence indices at baseline assessment, except for cervical dilation. The detailed comparisons for the nine partograph adherence indices are given as supplementary information (Additional File 4).

Fig. 2.

Proportion of partograph adherence indices at baseline by study arms, Amhara region, Ethiopia

Adherence to partograph and WHO safe childbirth checklists

The baseline mean partograph scores in the control arm and intervention arm were 4.11 and 4.23, respectively. The corresponding end-line mean partograph scores in the control arm and intervention arm were 4.14 and 8.47, respectively. The mean partograph adherence score difference between the pre- and post-intervention measures among the intervention arm was 4.24 (95% CI: [4.12, 4.36]). The mean difference between the pre- and post-intervention measures among the control group was 0.03 (95% CI: [−0.09, 0.15]), which is not significant. The DiD in partograph score was 4.21 (95% CI: [3.68, 4.73]) (Table 3).

Table 3.

Comparison of baseline and end-line mean partograph adherence score by study arms, Amhara region, Ethiopia

Outcomes	Intervention time		Difference [95% CI]	DiD [95% CI]
	Baseline	End-line
Mean score
Control	4.11 [4.03, 4.19]	4.14 [4.06, 4.22]	0.03 [−0.09, 0.15]	4.21 [3.68, 4.73]*
Intervention	4.23 [4.14, 4.31]	8.47 [8.38, 8.55]	4.24 [4.12, 4.36]*

DiD Difference-in-difference

^*statistically significant at p-value < 0.001

The difference in the proportion of full adherence to partograph between pre-and post-intervention in the intervention arm was 75.1% (95% CI: [71.2, 79.0]) whereas in the proportion of full adherence to partograph between pre-and post-intervention in the control arm was 1.6% (95% CI: [−3.1, 6.3]). The DiD in proportion of full adherence to partograph was 73.5% (Table 4).

Table 4.

Proportion of baseline and end-line partograph adherence by study arms, Amhara region, Ethiopia

Outcomes	Intervention time		Difference [95% CI]	DiDp
	Baseline	End-line
Full adherence
Control	18.9% [15.6, 22.3]	20.5% [17.2, 23.9]	1.6% [−3.1, 6.3]	73.5%
Intervention	15.6% [12.5, 18.7]	90.7% [88.3, 93.1]	75.1% [71.2, 79.0]*

DiDp  difference-in-difference in proportion

^*statistically significant at p-value < 0.001

The results in Table 5 below were obtained by a two-sample test of proportions. At the baseline survey, the proportion of adherence to admission SCC in the control and intervention arms was 25.8% and 49.6%, respectively. The proportion of adherence to before-birth SCC in the control and intervention arms was 24.9% and 47.8%, respectively. The proportion of adherence to after-birth SCC in the control and intervention arms was 24.9% and 45.9%, respectively. The proportion of adherence to discharge SCC in the control and intervention arms was 24.3% and 45.5%, respectively. The difference in proportion of full adherence to the admission SCC between pre-and post-intervention in the intervention arm was 13.8% (95% CI: [8.0, 19.6]), which was statistically significant, whereas the difference in proportion of full adherence to the admission SCC between pre-and post-intervention in the control arm was − 0.6% (95% CI: [−5.7, 4.6]), which was statistically significant. The difference in proportion of full adherence to before-birth SCC between pre-and post-intervention in the intervention arm was 31.2% (95% CI: [25.8, 36.6]), which was statistically significant, while in the control arm, it was 0.2% (95% CI: [−5.0, 5.3]), which was not statistically significant. The difference in proportion of full adherence to the immediately after-birth SCC between pre- and post-intervention in the intervention arm was 23.0% (95% CI: [17.3, 28.7]), which was statistically significant, whereas in the control arm, it was − 0.9% (95% CI: [−6.0, 4.2]) and not statistically significant. The difference in proportion of full adherence to the discharge SCC between pre- and post-intervention in the intervention arm and control arm was − 2.7% (95% CI −8.6 to 3.2) and − 1.1% (95% CI −6.1 to 4.0), respectively, and both estimates were not statistically significant (Table 5).

Table 5.

Proportion of baseline and end-line adherence to WHO SCC by study arms, Amhara region, Ethiopia

Outcomes	Intervention time		Difference [95% CI]	DiD@
	Baseline	End-line
Admission checklist
Control	25.8% [22.1, 29.5]	25.3% [21.7, 28.9]	−0.6% [−5.7, 4.6]	14.4%
Intervention	49.6% [45.4, 53.9]	63.4% [59.4, 67.4]	13.8% [8.0, 19.6]*
Before-birth checklist
Control	24.9% [21.3, 28.6]	25.1% [21.5, 28.7]	0.2% [−5.0, 5.3]	31.0%
Intervention	47.8% [43.5, 52.0]	78.9% [75.6, 82.3]	31.2% [25.8, 36.6]*
After-birth checklist
Control	24.9% [21.3, 28.6]	24.0% [20.5, 27.6]	−0.9% [−6.0, 4.2]	23.9%
Intervention	45.9% [41.7, 50.1]	68.9% [65.1, 72.8]	23.0% [17.3, 28.7]*
Discharge checklist
Control	24.3% [20.7, 28.0]	23.3% [19.8, 26.8]	−1.1% [−6.1, 4.0]	−1.6%
Intervention	45.5% [41.3, 49.7]	42.9% [38.8, 47.0]	−2.7% [−8.6, 3.2]

DiD difference-in-difference

^@the DiD reported here is the difference in proportion obtained by manual calculation

^*statistically significant at p-value < 0.001

Effect of ADHERE intervention on adherence to partograph and WHO SCC

The DiD result showed that the estimated effect of ADHERE intervention on the partograph adherence score among the intervention arm was 4.27. On average, the ADHERE intervention increased partograph adherence score by 4.27 units as compared to the level it would be without intervention (DiD = 4.27, 95% CI = [1.80, 5.09]) (Fig. 3).

Fig. 3.

The effect of ADHERE intervention on partograph adherence score, Amhara region, Ethiopia. DiD = difference-in-difference; the corresponding simple DiD with 95% CI was 4.21 [1.71, 5.10] and adjusted DiD with 95% CI was 4.27 [1.80, 5.09]; reported 95% CI was obtained by wild cluster bootstrap method

In addition, the adjusted mixed-effects linear-based DiD for partograph adherence score was 4.24 (95% CI = [3.75, 4.74]) (Fig. 4). The standard linear-based DiD and mixed-effect linear-based DiD estimates were nearly the same (4.27 vs. 4.24). However, the standard linear-based 95% CI for DiD estimate was wider than the mixed-effects model since it was adjusted for a small number of clusters using the WCB analysis.

Fig. 4.

Mixed-effect linear-based DiD estimate for partograph adherence score, Amhara region, Ethiopia. DiD = difference-in-difference; the corresponding simple DiD with 95% CI was 4.22 [3.72, 4.72] and adjusted DiD with 95% CI was 4.24 [3.75, 4.74]

The DiD result showed that the ADHERE intervention increased (from baseline to end-line) adherence to admission SCC by 9.3% points in the intervention arm relative to the control arm (DiD = 0.093, 95% CI = [0.036, 0.151]). Similarly, the ADHERE intervention resulted in a 22.3% points increase (from baseline to end-line) in adherence to before-birth SCC during labor and delivery care in the intervention arm compared to the control arm (DiD = 0.223, 95% CI = [0.121, 0.325]). The adherence to the SCC within an hour after birth was also increased (from baseline to end-line) 15.2% points in the intervention arm compared to the control due to the ADHERE intervention effect (DiD = 0.152, 95% CI = [0.072, 0.231]). On the other hand, the ADHERE intervention did not result in a significant change (from baseline to end-line) in adherence to the discharge SCC among the intervention arm relative to the control arm (DiD = −0.025, 95% CI = [−0.071, 0.021]) (Table 6). The lack of significant improvement in adherence to discharge SCC may be due to providers’ less attention to discharge evaluation because of work overload, or some clients may self-discharge against medical advice.

Table 6.

Effect of ADHERE intervention on adherence to each WHO SCC component, Amhara region, Ethiopia

Outcomes	Time		Difference	Simple DiD (95% CI)	Adjusted DiD (95% CI)
	Pre	Post
Admission checklist
Control	0.025	0.025	0.000	0.135 [0.072, 0.198]	0.093 [0.036, 0.151]*
Intervention	0.487	0.622	0.135
Before-birth checklist
Control	0.024	0.025	0.001	0.342 [0.252, 0.431]	0.223 [0.121, 0.325]*
Intervention	0.400	0.742	0.342
After-birth checklist
Control	0.019	0.020	0.001	0.247 [0.184, 0.309]	0.152 [0.072, 0.231]*
Intervention	0.386	0.632	0.246
Discharge checklist
Control	0.018	0.017	−0.001	−0.044 [−0.107, 0.019]	−0.025 [−0.071, 0.021]
Intervention	0.460	0.415	−0.045

CI  confidence interval, DiD difference-in-difference, WHO SCC WHO safe childbirth checklist

^*statistically significant at p-value < 0.001

Discussion

The current implementation research found that the ADHERE-assisted implementation of labor and delivery care significantly improved adherence to QI tools in intrapartum care. Specifically, the ADHERE-assisted implementation significantly increased HCPs’ adherence to partograph parameters and WHO SCC, for all time periods except for the on-discharge SCC.

In the current study, the ADHERE intervention increased HCPs’ adherence to partograph by 4.27 units as compared to the level it would be without intervention. A quasi-experimental study in Kenya also reported greater adherence to intrapartum care in the electronic partograph group than the paper-partograph group due to the clinical decision support intervention embedded in the ePartograph. Skilled birth attendants using the digital partograph were more likely than those using the paper version to take eCDS-assisted actions (such as ambulation, feeding, and fluid intake) to maintain normal labor [29]. Studies in South Asia countries also reported a similar effect of ePartograph on correct use of partograph as a result of the clinical decision support system that helped the providers to adhere [28, 58]. A cross-over study done in Bangladesh showed that the partograph user rate had significantly improved with the ePartograph compared to the paper-partograph (38% vs. 21.3% on 1 st phase and 38% vs. 2.8% on 2nd phase) due to the advantage of eCDS and automated plotting of the parameters [58]. A scoping review also remarked that the digital partograph produces superior results to the paper version due to its clinical decision support system, including visual and/or auditory alerts, simplicity to adopt and use by obstetric caregivers, and having the potential to save time [4]. Conversely, a study in Kenya indicated that the ePartograph group had lower rates of clinical rules triggered, significantly so for duration and frequency of uterine contractions. The rate of interventions to address labor abnormalities was not significantly higher in ePartograph group than the paper-partograph group [29].

In the current study, the ADHERE intervention significantly increased adherence to WHO SCC at admission (9% points), before birth (22.3% points), and immediately after birth (15.2% points) among the intervention group, compared with the control group. Earlier studies in Africa also indicated that QI interventions led to improved adherence to WHO SCC [11, 59]. In Namibia, the implementation of SCC led by a facility champion, with support from a QI team, reported an increase in average essential birth practices (EBPs) delivered from 68% to 95% [11]. A multi-country study in Kenya and Uganda also reported that the SCC completion at admission had the highest rates of completion in both countries. In Kenya SCC completion was greater than 70% for all pause points, while in Uganda, it ranged from 39% to 75%. Overall, intervention facilities exposed to QI initiative demonstrated higher completion rates than control sites [59]. A cluster randomized controlled trial in India found that a coaching-based SCC implementation demonstrated a greater adherence to EBPs. Birth attendants’ average adherence to SCC increased by 24% points when a coach was present. Safe childbirth checklist use increased from the first to final coaching month, including checklist use at admission (84% to 98%), before birth (66% to 94%), immediately after birth (75% to 95%), and discharge (90% to 99%) [60]. A study conducted in Turkey also concluded that the eCDS system integrated with the ministry information management system significantly improved the proper use rate of the WHO SCC (61% to 97.4%) [61].

Conversely, the ADHERE intervention did not significantly improve adherence to the discharge SCC among the intervention group compared with the control group. The lack of significant improvement in adherence to discharge SCC may be due to providers’ less attention to discharge evaluation because of work overload, or some clients may self-discharge against medical advice. However, this finding warrants further exploration to understand the real cause of the observed result.

The current study had several strengths. The study had a comparator group comprised of health facilities with nearly similar baseline characteristics, which were verified through formative evaluation. We also assessed additional DiD assumptions, such as consistency. The study tried to ensure consistency in preparing for the implementation of ADHERE innovation across health facilities, including training of HCPs, providing basic amenities for the ADHERE system, and conducting iterative discussions. No differential interventions were deployed at baseline among the groups, and there was also no evidence of concurrent interventions (shocks) that could affect the comparability between the intervention and control arms during project implementation.

Despite its strengths, the study also had limitations. The first potential limitation is related to the data collection method. We deployed record review instead of more robust data collection methods, such as direct observation during service delivery. In a record review, the data collectors cannot verify whether the recorded parameters were actually performed, and conversely, a health care provider may deliver care without documenting it. If the necessary care is provided but not recorded, adherence will be underestimated. On the contrary, if care is not provided but recorded as if provided, adherence will be overestimated. However, we chose the record review because we wanted to have baseline data for the control and intervention arms, a comparator with post-intervention data. The most feasible data source for the baseline evaluation was a retrospective labor and delivery chart review. To minimize the challenges associated with record review, we deployed onsite data collectors in the control sites to ensure daily data capture and reduce retrospective manipulation of the client records for reporting purposes. The second limitation is related to the lack of data on provider characteristics, which is a potential confounder that needs to be controlled. This limitation is also tied to the shortcomings of paper-based recording, which may omit provider information. Hence, the observed intervention effect may be confounded by differences in provider characteristics. The third limitation concerns the small number of health facilities and a design effect of 1.5 used in the study. We included only nine facilities, which falls short of the recommended number for the standard CRVE. The inclusion of small numbers of health facilities may overly estimate the intervention effect. It underestimates the standard error and inflates type I error with a narrower confidence interval and small p-value, leading to a false conclusion that the intervention had a significant effect. The use of a 1.5 design effect may also underestimate the ICC and, then, the sample size. To address this, we used the wild bootstrap method, a more robust alternative to the standard CRVE. Additionally, we employed the linear mixed-effects model alongside the standard linear-based DiD to check the sensitivity of the estimates. The fourth limitation is related to the parallel trend assumption. The study did not directly test the parallel trend assumption because it only had pre- and post-intervention data. However, we provided previous studies demonstrating no significant pre-intervention differences in adherence to labor and delivery care protocol or quality of care in Ethiopia or similar settings [62–66].

The study findings have potential implications for policy, practice, and research. The finding provides evidence of practitioners, policy-makers, and stakeholders for the importance of ADHERE interventions to improve adherence to QI tools during labor and delivery care. This can encourage efforts to scale up the ADHERE interventions in health facilities as a quality improvement initiative. However, as highlighted in our published article, the ADHERE system scale-up should consider the strategies used to overcome the challenges of successfully implementing eHealth innovations [34], such as ICT infrastructure development, power backup, on-site technical support team, adaptive system design, and regular supportive supervision. The use of implementation research models or frameworks is also equally important. Some issues warrant further research, including direct observations to capture robust data on provider adherence to intrapartum care quality improvement tools such as the partograph and WHO SCC. Future research may also focus on the feasibility and acceptability of scale-up of the ADHERE system, addressing interoperability issues and multiplicity of electronic platforms used at the point-of-care in Ethiopia.

Conclusions

The ADHERE eHealth solution significantly improved adherence to quality improvement tools during labor and delivery care. Explicitly, the ADHERE-assisted delivery of intrapartum care significantly increased healthcare providers’ adherence to partograph parameters. Similarly, the ADHERE-assisted implementation of the WHO safe childbirth checklist significantly increased adherence to each subtype of WHO SCC. However, the ADHERE intervention did not significantly improve adherence to the discharge checklist. Based on the success observed in enhancing adherence to quality improvement tools during intrapartum care, the ADHERE system could be considered for broader implementation among health facilities in Ethiopia. The scale-up should consider the strategies proposed to overcome the challenges of successfully implementing eHealth innovations in Ethiopia, such as ICT infrastructure development, power backup, on-site technical support team, adaptive system design, and regular supportive supervision. In addition, some issues warrant further research, including direct observations to capture robust data on provider adherence and provider characteristics. Future research may also focus on the feasibility and acceptability of scale-up of the ADHERE system, addressing interoperability issues to existing systems in health facilities in Ethiopia.

Abbreviations

AIC

Akaike’s information criterion

BIC

Bayesian information criterion

BP

Blood pressure

CFIR

Consolidated Framework for Implementation Research

CRVE

Standard cluster-robust variance estimator

DiD

Difference-in-differences

eCDS

electronic clinical decision support

EBPs

Essential birth practices

FHR

Fetal heart rate

HCPs

Healthcare providers

ICC

Intraclass correlation coefficient

IRB

Institutional Review Board

IRLM

Implementation Research Logic Model

LAN

Local area network

LR

Likelihood ration

LL

Log-likelihood

LMICs

Low- and middle-income countries

MCH

Maternal and child health

POC

Point-of-care

QI

quality improvement

SCC

safe childbirth checklists

WCB

Wild cluster bootstrap

WHO

World Health Organization

Authors’ contributions

DN conceived the study, coordinated the study implementation, performed the data analysis and interpretation, and drafted and critically reviewed the manuscript. MAY, EM, TK, EA, and EA involved in refining the study concept and study implementation. TG and MW oversaw the project implementation. MAY, EM, TG, and DAE have guided the data analysis and interpretation and critically reviewed the manuscript. All the authors read and approved the final manuscript.

Funding

We received the research funds from the Fenot Project, Bahir Dar University, and Amhara Public Health Institute, Ethiopia. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data availability

All data generated or analyzed during this study are included in this published article [and its supplementary information files].

Declarations

Ethics approval and consent to participate

Ethics approval was obtained from the Institutional Review Board (IRB) of Bahir Dar University College of Medicine and Health Sciences (IRB protocol number 377/2022). The Amhara Public Health Institute gave support letters to the relevant public health facilities and administrations. This quality improvement research uses anonymized data derived from medical record. The eHealth intervention was deployed at health facilities serving maternity care to enhance healthcare providers’ adherence to recommended quality improvement tools in labor and delivery care. The study participants were health facilities and their staffs working in the MCH clinics, however, women attending the MCH clinics did not meet the recommendations of the Ottawa Statement for being individual study participants [67, 68]. Hence, they were not required to give informed consent to use the anonymized data for this study. After getting ethical approval, we sought permission from each participating health facility to access the medical records and extract the required data accordingly. We extracted anonymized data from eMCH registry and paper-based labor and delivery chart. In addition to institutional permission, a written consent was obtained from staffs working in the MCH clinics. The study team ensured the anonymity and confidentiality of information throughout the study process.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.