> For the complete documentation index, see [llms.txt](https://research.ambitiousimpact.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://research.ambitiousimpact.com/bemonc/simulation-based-training-for-emergency-obstetric-and-newborn-care/3-evidence-review.md).

# 3 Evidence review

**This section reviews the evidence for the key links in the ToC, summarized in Table 4.**

**Table 4: Summary of evidence review**

| Research question                                                                                    | Magnitude and direction findings                                                                                                                                                                                                                                 | Notes                                                                                                                                                                                                                                                |
| ---------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Can LDHF training improve delivery of BEmONC signal functions in facilities?**                     | <p>Around a one-third increase in signal function performance one year after initial training.<br><br>Moderate confidence: the direction of the evidence is robustly positive, but estimated effect sizes vary.</p>                                              | The strongest individual study [(2018)](https://doi.org/10.1186/s12884-018-1705-5) used an LDHF model, while a head-to-head trial [(2020)](https://doi.org/10.1186/s12913-020-05450-9) found that LDHF marginally outperformed traditional training. |
| *Do other training regimens work?*                                                                   | We are open to a new nonprofit testing and refining different training approaches to improve efficiency and cost-effectiveness. However, we are confident that hands-on simulation should remain a core component.                                               |                                                                                                                                                                                                                                                      |
| **Does improved delivery of BEmONC signal functions improve health outcomes in mothers and babies?** | Around a one-third decrease in deaths and stillbirth. Moderate to high confidence.                                                                                                                                                                               | BEmONC signal functions have been studied both as a package and individually. Their inclusion in the package partly reflects strong evidence that they reduce health risks.                                                                          |
| **Other open questions**                                                                             | <p>• Will the intervention require commodity supply?<br>• Does adding training on post-natal care improve neonatal outcomes without harming learning gains for care around birth?<br>• How quickly would skills decay if ongoing support were not sustained?</p> |                                                                                                                                                                                                                                                      |

## 3.1 Can LDHF training improve delivery of BEmONC signal functions in facilities?

**Several studies suggest that LDHF training programs have improved knowledge and actual practice for maternal and newborn care at delivery.** Given external validity and heterogeneity concerns, we are uncertain of the magnitude of the effect on knowledge and practice.

**Several practitioner groups and experts seem to support the use of LDHF strategies for improving clinical skills** (e.g., [Azad et al., 2020](https://doi.org/10.1136/bmjgh-2019-001981); [Jhpiego, n.d.-b](https://hms.jhpiego.org/wp-content/uploads/2016/08/LDHF_briefer.pdf))**.** We believe that the best package will include hands-on simulation, and some form of continuous quality monitoring and improvement [(e.g., Kamala et al., 2025)](https://doi.org/10.1056/NEJMoa2406295).

**Our preference for LDHF is also supported by its use across reviewed studies.** Put simply, we assume that if the studies find a health impact, it is because the treatment arm gained and—to some extent—retained knowledge about how to improve quality of care.

**Gomez et al.** [(2018)](https://doi.org/10.1186/s12884-018-1705-5) **trained 403 SBAs across 40 hospitals in Ghana using two four-day on-site sessions covering a comprehensive BEmONC package with weekly practice sessions, SMS-based mentoring, and master trainer support between sessions.** This cluster-randomized controlled trial (RCT) reported knowledge and skill gains up to one year after intervention, relative to baseline (scores on objective structured clinical examination were 31% and 28% above pre-test).

**This study is highly relevant** because it used the same LDHF model we propose for the charity, covered a BEmONC package rather than a narrow intervention, trained all SBAs at each facility (creating facility-wide change), and included ongoing mentorship.

**However, we think it is at medium-high risk of bias**, given lack of blinding and outcome measurement concealment, and the likelihood that treated individuals also received other interventions during the study period.

**Ugwa et al.** [(2020)](https://doi.org/10.1186/s12913-020-05450-9) **carried out a head-to-head comparison in an LMIC context through a cluster-RCT in Ebonyi and Kogi States, Nigeria.** They compared LDHF with mobile mentoring against traditional group-based training.

**We think the study provides evidence that LDHF is at worst no different to traditional off-site training and potentially better in facilitating retention of skills.** The study’s LDHF arm performed marginally better on all competencies at three and 12 months post-training, including active management of the third stage of labor, manual removal of placenta, neonatal resuscitation, and management of pre-eclampsia and eclampsia; only neonatal resuscitation achieved a marginally statistically significant difference.

**LDHF was originally developed for simpler clinical protocols, such as cardiopulmonary resuscitation (CPR) training, whereas BEmONC involves complex emergency decision-making under pressure.** Gomez et al. [(2018)](https://doi.org/10.1186/s12884-018-1705-5) provides the strongest evidence that LDHF can work for BEmONC: the training covered labor management, postpartum hemorrhage, eclampsia, newborn resuscitation, and infection prevention, and SBAs retained skills at one year.

**Other studies provide tentative evidence supporting the role of LDHF training in improving clinical knowledge and skills.**

* **A** [**pragmatic**](#user-content-fn-1)[^1] **cluster-RCT found that LDHF with the inclusion of ongoing peer-assisted mentorship managed to retain some skills for longer, compared to treatment arms that did not include full mentorship** [**(Evans et al., 2018)**](https://doi.org/10.1371/journal.pone.0207909)**. We do not rate the quality of this evidence highly, but as the authors suggest, the trial shows the “adequacy” of LDHF approaches.**
* A pre-post study of LDHF training plus quality improvement across 16 facilities in Kinshasa, DRC, found that at least 75% of providers demonstrated competency in key skills at 12 months [(Kikaya et al., 2024)](https://doi.org/10.9745/GHSP-D-23-00236).
* The AMPLI-PPHI program (Jhpiego, PATH, and FIGO; 2022–2026) applies an LDHF approach to onsite providers, confirming that LDHF is being used as a delivery model in this space, though narrowly scoped to integrating three postpartum hemorrhage medicines into clinical practice rather than the full BEmONC package [(*Jhpiego*, n.d.-a)](https://jhpiego.org/areas-of-expertise/ampli-pphi/).

### Do other training regimens work?

**We remain very open and interested in exploring alternative training regimens that may prove more effective or cost-effective, especially given that much of the evidence specific to LDHF approaches in LMICs is of poor or medium quality.** However, we think the approach is well suited to the context. We particularly think its ongoing and hands-on nature is important for complex clinical practice improvement such as BEmONC signal functions.

Based on meta-analyses of clinical practice interventions [(Rowe et al., 2018)](https://doi.org/10.1016/S2214-109X%2818%2930398-X), we believe training alone (i.e., without some form of ongoing support and monitoring) will result in skills decay as practitioners go back to standard practice and as high staff turnover affects institutional knowledge. We also believe hands-on simulation to be much more effective than theoretical or descriptive approaches [(e.g., Kamala et al., 2025)](https://doi.org/10.1056/NEJMoa2406295).

## 3.2 Does improved delivery of BEmONC signal functions improve health outcomes in mothers and babies?

**We strongly believe that improvements to quality of care based on BEmONC will have a positive impact on maternal and neonatal outcomes.** The magnitude of the effect likely depends on implementation quality, existing baselines and drivers of mortality, and additional barriers to good quality of care, such as supply chain and infrastructure issues.

This assessment relies on two strands of evidence that we view as highly indicative of a true effect:

* Evidence coming from quality-improvement trials.
  * We detail a review of relevant evidence by GiveWell [(2022)](https://www.givewell.org/international/technical/programs/facility-based-maternal-and-neonatal-health-interventions) and a handful of studies published after 2022 that we believe to be relevant.
* Evidence coming from trials focused on BEmONC signal functions.
  * BEmONC signal functions are strongly backed by high-quality research. Their individual effectiveness cannot be used to assess potential magnitudes for BEmONC packages due to confounding and the fact that signal functions address overlapping risk factors.

### Evidence coming from quality-improvement trials

#### GiveWell’s report

**GiveWell** [(2022)](https://www.givewell.org/international/technical/programs/facility-based-maternal-and-neonatal-health-interventions) **reviewed seven cluster-RCTs on the effects of training facility-based health workers on neonatal outcomes and three cluster-RCTs on maternal outcomes.** They concluded the studies provided moderately strong evidence that facility-based improvements could cut neonatal mortality by a third, and weak evidence that maternal mortality rates could be cut by the same rate[^2].

**We largely defer to GiveWell’s analysis.** We focused our research efforts on checking for studies carried out or published after GiveWell’s review, and lightly reviewing the literature. GiveWell rated the neonatal evidence as moderately strong (seven cluster-RCTs) and the maternal evidence as weak (three cluster-RCTs, high heterogeneity).

GiveWell's weighted pooled estimates suggest that training:

* Reduces neonatal deaths before discharge by approximately 33% (individual trial effects ranging from a non-significant 3% increase to a 70% reduction) and stillbirths by approximately 35% (ranging from a 5% increase to a 52% reduction).
* Reduces maternal mortality by approximately 32% (ranging from an 11% increase to an 89% reduction), based on three cluster RCTs. The wide ranges reflect substantial heterogeneity in training model, content, and setting.

**GiveWell also identified six non-experimental studies (five pre-post and one observational with a non-randomized control group) reporting fairly large reductions in maternal mortality and postpartum hemorrhage incidence.** The authors did not weight these in the pooled estimate but treated them as increasing confidence in the trial effects.

**After applying internal and external validity adjustments—discounting for potential bias and the expectation that real-world programs are implemented with lower quality than RCTs—GiveWell arrived at adjusted estimates of approximately a 20% reduction in neonatal mortality, a 20% reduction in stillbirths, and a 12% reduction in maternal mortality.** The adjustments are large: 38% for early neonatal mortality, 42% for fresh stillbirths, and 64% for maternal mortality.

#### Studies published after GiveWell’s review

The most important new evidence since GiveWell's review is the Safer Births Bundle of Care (SBBC) stepped-wedge cluster-randomized study published in the New England Journal of Medicine in February 2025 [(Kamala et al., 2025)](https://doi.org/10.1056/NEJMoa2406295). **This three-year study across 30 high-burden facilities in five regions of Tanzania assessed approximately 281,000 mothers and approximately 278,000 babies.** The trial combined on-site simulation-based training, continuous quality improvement, innovative clinical tools (fetal and neonatal heart rate monitors), and local facilitators. It found that:

* Perinatal death fell from 15.3 to 12.5 per 1,000 births (adjusted RR = 0.82, 95% CI 0.73–0.92), though with substantial heterogeneity between sites.
* Neonatal deaths within 24 hours dropped from 6.4 to 3.9 per 1,000 (adjusted RR = 0.61, 95% CI 0.49–0.77).
* In-hospital maternal deaths within the first week fell from 240 to 60 per 100,000 births (adjusted RR = 0.25, 95% CI 0.14–0.46).
* However, intrapartum stillbirths did not change significantly, increasing from 8.6 deaths per 1000 births at baseline to 8.7 deaths per 1000 births after implementation (adjusted RR = 1.01; 95% CI 0.87–1.17)

**This study matters because it is the largest and most rigorous trial of a bundled simulation-based training program in this space, conducted at scale rather than as a small pilot, with striking maternal mortality results.**

**There are important differences from the proposed model, however**: SBBC operates at comprehensive emergency obstetric and neonatal care (CEmONC) hospital facilities rather than primary facilities; it includes clinical tools beyond training alone; it was implemented by Laerdal Global Health with substantial institutional backing.

**Other post-2022 observational evidence suggests ongoing benefits in several locations** [(e.g., Kikaya et al., 2024)](https://doi.org/10.9745/GHSP-D-23-00236):

* A Kinshasa pre-post study found a 9% reduction in early neonatal death rates ([Jhpiego/GHSP, 2024](https://www.ghspjournal.org/content/12/1/e2300236));
* 2025 India study found maintained competency gains [(Solanky et al., 2025)](https://doi.org/10.1186/s12982-025-01023-5);
* A three-arm RCT in Mali will directly test the proposed model at primary facility level when results are published ([JMIR Research Protocols, 2021](https://www.researchprotocols.org/2021/12/e28644)).

**Well-executed pre-post observational studies also contribute to the evidence base for BEmONC packages** (e.g., [Tiruneh et al., 2018](https://doi.org/10.1186/s12884-018-1751-z)). While the magnitude of effects is heterogeneous—like with the experimental studies reviewed by GiveWell—they are largely supportive of packages of interventions. However, since these studies frequently involved multifaceted interventions, we cannot be confident about the contribution of the training components.

### Evidence coming from trials focused on BEmONC components

**BEmONC signal functions are themselves backed by clinical evidence and standards, and are part of standard care in healthcare centers with low mortality rates.** By way of example:

* **Neonatal resuscitation**: A Cochrane review of 27 studies (randomised trials and quasi-RCTs) found that newborn resuscitation training programs may decrease neonatal mortality, with RRs ranging from 0.47 to 0.88 across different target populations [(Dempsey et al., 2026)](https://doi.org/10.1002/14651858.CD009106.pub3).
* **Oxytocin for postpartum hemorrhage prevention**: Prophylactic oxytocin roughly halves the risk of postpartum hemorrhage (RR 0.53, 95% CI 0.38–0.74), based on a Cochrane meta-analysis of six trials comparing oxytocin with placebo [(Westhoff et al., 2013)](https://doi.org/10.1002/14651858.CD001808.pub2)
* **Chlorhexidine for cord care**: Applying chlorhexidine to umbilical cord stumps reduces neonatal mortality by approximately 15%, based on a meta-analysis of five cluster-RCTs ([GiveWell, 2021](https://www.givewell.org/international/technical/programs/chlorhexidine)).
* **Kangaroo care** is a well established practice, which Ambitious Impact has had some early signals of success with (e.g., [Ansh](https://ansh.health/)), and is backed by Cochrane reviews [(Conde-Agudelo & Díaz-Rossello, 2016)](https://doi.org/10.1002/14651858.CD002771.pub4).
* **Community-based packages**: A Cochrane review of 26 randomized and quasi-randomized trials of community-based packages that overlap substantially with facility-based BEmONC found a 25% reduction in neonatal mortality, a 19% reduction in stillbirths, and a 20% reduction in maternal mortality ([*Community-Based Intervention Packages to Reduce Maternal and Neonatal Mortality*, n.d.](https://www.givewell.org/international/technical/programs/community-intervention-package-programs); [Lassi & Bhutta, 2015](https://doi.org/10.1002/14651858.CD007754.pub3)).

[^1]: A pragmatic trial evaluates how well a medical intervention works in real-world, routine clinical practice rather than under strict, ideal conditions.

[^2]: The effect size GiveWell arrived at is very sensitive to their subjective weighting of the evidence, which favors Gomez et al. [(2018)](https://doi.org/10.1186/s12884-018-1705-5). For a breakdown of the studies’ methodologies and findings, see [this sheet](https://docs.google.com/spreadsheets/d/1zlHPdWDUP9F3UYrU03bkQLoI_4BvnomcyQ02OrMohbw/edit?gid=1868074268#gid=1868074268) by GiveWell.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://research.ambitiousimpact.com/bemonc/simulation-based-training-for-emergency-obstetric-and-newborn-care/3-evidence-review.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.