University College London — Research on Child Mortality and Long-Term Health Effects of Participatory Learning and Action (January 2023)

Note: This page summarizes the rationale behind a GiveWell grant to Ed Fottrell at the University College London Institute for Global Health. Professor Fottrell reviewed this page prior to publication.

Published: June 2023

The intervention

Participatory learning and action groups for maternal and neonatal health (hereafter PLA) are a potentially promising intervention that addresses a large, important problem: neonatal mortality. Progress at reducing neonatal mortality hasn’t been as rapid as reducing child mortality overall.1

PLA consists of facilitated community group meetings, especially targeting women of reproductive age and pregnant women, that help them develop localized strategies to increase appropriate care-seeking and improve uptake of prevention practices aiming to improve maternal and newborn health.2

Does PLA work?

There is strong evidence that PLA reduces neonatal and maternal mortality. The randomized controlled trials (RCTs) that we use to form our best guess on the cost-effectiveness of the program find that PLA reduces neonatal mortality by 33% and maternal mortality by 49%.3

One of our key uncertainties is the effect on post-neonatal child mortality, that is, child mortality after the age of 28 days. There is weak evidence that PLA reduces post-neonatal mortality, but it is drawn from a single study (Heys et al. 2018) with some concerns about bias.4

See our PLA intervention report for more information.

Grant activities and budget

This $344,175.07 grant to Professor Ed Fottrell at the UCL Institute for Global Health will fund a study that will follow up on an RCT (Fottrell et al. 2013) of PLA in rural Bangladesh. The primary study outcome will be child mortality between 28 days and 14 years of age.5 The researchers will attempt to ascertain the survival status of approximately 20,000 children whose live births were recorded in the intervention and control areas during the original study. They will also collect data on physical and cognitive measures for a subsample of this group.6

See the grant budget details here.

The case for the grant

• Results from this study could cause us to update our estimate of PLA's effect on post-neonatal child mortality, which could open up funding opportunities. If we update our estimate of PLA’s effect on post-neonatal child mortality upward, PLA may meet our cost-effectiveness threshold in more geographies, including those with lower mortality burdens or higher costs. If we update our estimate downward, we may allocate more funding and staff capacity to interventions that are more cost-effective than PLA.
• By funding this follow-up study, we’ll double the number of studies we can use to assess PLA's effects on child mortality. Our current PLA cost-effectiveness analysis assumes a 6% reduction in post-neonatal child mortality, which is based on heavily discounting the 27% (non-statistically significant) reduction in child mortality found in the only follow up study of a PLA RCT.7

Given the relatively small size of this grant, our assessment of the case for the grant relies primarily on qualitative considerations; we have not estimated the cost-effectiveness of this grant.

Risks and reservations

• Limited statistical power and the long delay between intervention and outcomes could lead us to incorrectly update our beliefs based on the study's results. Due to limited sample size, the study may not have enough statistical power to detect a reduction in post-neonatal child mortality.8 Additionally, causal attribution of any effect is made more difficult given the substantial time between randomization and intervention delivery and the outcomes the study seeks to measure. Cluster randomization took place more than 14 years ago. It is possible that factors that would affect mortality have diverged in these clusters since randomization. This could introduce either noise or bias (if for unknown reasons trends in treatment and control areas diverged) in estimates.9 These factors raise the risk that we will be incorrectly updated by the results (e.g., the study does not find evidence of an effect, even if one exists, causing us to wrongly allocate funding away from PLA). To mitigate that risk, we will likely combine information from this study with that from the one existing study measuring PLA’s impact on post-neonatal child mortality, either informally or in a pooled analysis.
• The grant will fund collection of data on physical and cognitive measures for a subsample of children, and we do not foresee a direct use of this data for GiveWell’s purposes.10 However, it’s a small portion of the budget (approximately $45,000) and the researchers believe that this portion of the study is important in terms of increasing the value of the study for study participants.11
• GiveWell has identified limited potential giving opportunities supporting PLA. GiveWell recently recommended a grant to Spark Microgrants for the preparation of a PLA pilot proposal, but if we decide not to fund Spark’s pilot activities, we would have no immediate implementation opportunity that learnings from this study would affect. However, in that case it's likely that the results of this study would still help us decide whether to prioritize finding another PLA implementer.

Plans for follow up

We plan to request informal biannual written updates about progress toward study milestones. We may have calls with the research team if these updates raise questions or concerns.

Based on the timeline Professor Fottrell shared, and as of the recommendation of this grant, we expect to receive preliminary results in May 2024.12 (As of the publication of this grant page, we now expect to receive preliminary results in October 2024.)13

Internal forecasts

For this grant, we are recording the following forecasts:

Confidence	Prediction	By time​​
60%	We receive results on child mortality from the study.	End of May 2024
90%	We receive results on child mortality from the study.	End of December 2024
85%	Results from the study cause us to update or increase our confidence in our estimate of the effect of PLA on post-neonatal child mortality.14	End of December 2024
50%	Results from the study cause us to update our estimate of the effect of PLA on post-neonatal child mortality upward, i.e., we model PLA as reducing post-neonatal child mortality by more than 6% (after adjustments) in our cost-effectiveness analysis.	End of December 2024
30%	We make an additional grant between May 2024 and December 2025 to fund PLA implementation using information from this study.	End of December 2025

Our process

• Women and Children First, a non-governmental organization that provides technical assistance to PLA program implementers,15 connected us with the investigators of several past RCTs of PLA. We reached out to inquire about their interest in conducting a follow-up study; Professor Ed Fottrell was the only researcher who expressed interest and provided a preliminary budget.
• We requested and received a short proposal from Professor Fottrell and his team.
• We reviewed the proposal, the original study, and the anthropometric follow up study (Fottrell et al. 2018); we looked at basic research design but did not scrutinize methodological choices.

Sources

Document	Source
Fottrell et al. 2013	Source
Fottrell et al. 2018	Source
GiveWell cost-effectiveness analysis of PLA-MNH (Participatory Learning & Action - Maternal and Neonatal Health), 2022	Source
GiveWell, "Participatory Learning and Action – Maternal and Neonatal Health," 2022	Source
Heys et al. 2018	Source (archive)
Prost et al. 2013	Source (archive)
UCL Institute for Global Health, "Child survival, health and development 14 years post-randomisation of successful Participatory Learning and Action community mobilisation intervention in rural Bangladesh," 2022	Source
WHO, "Newborn Mortality," January 2022	Source (archive)

• 1

  "Globally 2.4 million children died in the first month of life in 2020. There are approximately 6,700 newborn deaths every day, amounting to 47% of all child deaths under the age of 5 years, up from 40% in 1990. The world has made substantial progress in child survival since 1990.
  Globally, the number of neonatal deaths declined from 5 million in 1990 to 2.4 million in 2020. However, the decline in neonatal mortality from 1990 to 2020 has been slower than that of post-neonatal under-5 mortality." WHO, "Newborn Mortality," January 2022

• 2

  "A third approach involved women's groups in a four-phase participatory learning and action cycle. Phase 1 was to identify and prioritise problems during pregnancy, delivery, and post partum; phase 2 was to plan and phase 3 implement locally feasible strategies to address the priority problems; phase 4 was to assess their activities. Women's groups aimed to increase appropriate care-seeking (including antenatal care and institutional delivery) and appropriate home prevention and care practices for mothers and newborns." Prost et al. 2013, p. 1736.
  

• 3

  "Since the proportion of pregnant women participating in groups was a key predictor of mortality reduction, for our subgroup analyses we separated the trials into categories of high (≥30% of pregnant women participating in women's groups) and low coverage (30% participating). Figure 4 shows that in high-coverage studies (48 333 livebirths), exposure to women's groups was associated with a 49% reduction in maternal mortality (figure 4A) and a 33% reduction in neonatal mortality (figure 4B)." Prost et al. 2013, p. 1740.

• 4
  • See Heys et al. 2018, pg. 8, table 3, "Relative risks (RRs) weighted according to population size within clusters."
  • Heys et al. 2018, p. 8 finds a high intraclass correlation coefficient: "In fact, the ICCs for survival and disability outcomes were considerably higher than the original study and higher than we predicted, suggesting substantial intercluster variability." This can lead to a "relative lack of statistical power."
  • The study also has a risk of confounding since it is a follow-up to a randomized controlled trial that took place over a decade prior.
    • "In 2001–2003, in Makwanpur, Nepal, a cluster randomised controlled trial (RCT) of community-based women’s groups practising participatory learning and action (PLA) reported improvements in newborn and maternal survival." Heys et al. 2018, p. 2.
    • "...it is possible that residual confounding is a factor although randomisation should have reduced this likelihood; for example, if there were an exposure that was related to the outcome such as quality of water supply, for which data were not collected and which by chance was not equally distributed between the intervention and control arm." Heys et al. 2018, p. 10.

• 5

  "Primary outcome
  Post-neonatal child mortality defined as death after 28 days and before survey date (i.e., between 1 month and up to 14 years of age). Date of death will be recorded to calculate age at death, which in turn can be used to estimate age-specific (e.g., 1-59 month) mortality rates. Mortality rates will be reported as deaths per 1,000 livebirths within this cohort and hazard ratios will be estimated." UCL Institute for Global Health, "Child survival, health and development 14 years post-randomisation of successful Participatory Learning and Action community mobilisation intervention in rural Bangladesh," 2022, p. 3.

• 6

  "Study design
  Cohort follow-up of registered births 12-14 years after birth and 14 years after cluster randomisation within the PWG community mobilisation trial in Faridpur, Bogra and Moulvibazar districts, Bangladesh.
  Approximately 20,000 livebirths were recorded during the PWG trial between 2009 and 2011. We will seek to follow-up all registered births to assess survival status and, if deceased, measure date of death and conduct a verbal autopsy assessment to ascertain probable cause of death.
  Within the sub-sample of approximately 2,500 children who participated in the 2013 follow- up assessment of growth, plus an additional random sub-sample of approximately 1,250 children from intervention clusters, we will seek to measure physical and cognitive development indicators and biomarkers of health status and non-communicable disease risk." UCL Institute for Global Health, "Child survival, health and development 14 years post-randomisation of successful Participatory Learning and Action community mobilisation intervention in rural Bangladesh," 2022, p. 3.

• 7
  • See Heys et al. 2018, Table 3, which reports the risk ratio of reliable child deaths as 0.73, indicating a reduction in child mortality due to the intervention of 27%; however, the p-value for this finding is 0.24.
  • Heys et al. 2018 includes follow up for about 4,400 children across 12 clusters (see Table 2 here). It appears to have been underpowered, though it’s difficult to understand what its power was for mortality outcomes from the description in the paper:
    • “The major limitation of the study was a relative lack of statistical power with which to determine long-term survival outcomes. The size of the study was determined by the original trial, and the sample size calculations for the follow-up study indicated that it would be possible to detect a difference of 5% in disability rates with 80% power assuming a baseline rate in the controls of 27% with a sample size of 3999. In these a priori sample size calculations, α was set at 0.00833 to adjust for two-tailed comparisons (ie, allowing for the possibility of either group—intervention or control—having higher prevalence of childhood disability) and for multiple testing of 3 primary outcomes between intervention versus control arms. We assumed a conservative estimate of coefficient of variation (k) of 0.16 which was estimated using unpublished data on ICC from a study of maternal disability in Nepal from our group and from assumptions based on characteristics of clusters that would account for some of the variation in disability scoring between clusters. In fact, the ICCs for survival and disability outcomes were considerably higher than the original study and higher than we predicted, suggesting substantial intercluster variability.” Heys et al. 2018

• 8

  “Power
  Nine union (clusters) per arm and approx. 20,000 births registered in 2009-2011 (i.e. approx. 1111 births per cluster), an estimated coefficient of variation between clusters within strata (district) of 0.15, baseline 1-59 month mortality rate 1.1% (11 per 1000 live births), 65% follow-up rate will give approximately 60-65% power to detect a 40% relative reduction in 1- 59 month mortality (i.e. an absolute reduction of 440 per 1000 live births in intervention clusters compared to control clusters), at a 95% significance level.” UCL Institute for Global Health, "Child survival, health and development 14 years post-randomisation of successful Participatory Learning and Action community mobilisation intervention in rural Bangladesh," 2022, p. 4.

• 9

  Since there were only 18 clusters in the original study, it is possible that factors affecting mortality in treatment and control areas could have diverged by chance. “We used a cluster randomized controlled trial to evaluate the effect of the participatory learning and action cycle with women’s groups when delivered at higher population coverage in 18 “unions” in 3 districts (Bogra, Molavibazar, and Faridpur) of rural Bangladesh. Unions are the lowest administrative unit in Bangladesh, representing a geographically adjacent collection of villages.” Fottrell et al. 2013

• 10

  UCL Institute for Global Health, "Child survival, health and development 14 years post-randomisation of successful Participatory Learning and Action community mobilisation intervention in rural Bangladesh," 2022, "Secondary outcomes" section, p. 3.

• 11
  • Ed Fottrell, professor of Epidemiology & Global Health, University College London, email to GiveWell, October 25, 2022 (unpublished).
  • Ed Fottrell, professor of Epidemiology & Global Health, University College London, email to GiveWell, November 2, 2022 (unpublished).

• 12

  "Proposed start: 1st April 2023

  • Detailed protocol development – April-May 2023
  • Ethical approvals – May-June 2023
  • Community engagement & outreach – June 2023
  • Initiation of survey work – July 2023
  • Completion of survey work – April 2024
  • Initiation of analysis – April 2024
  • Completion of analysis – May 2024
  • Sharing preliminary analysis with GiveWell – May 2024
  • Submission for publication – June 2024

  Total Duration: 15 months" UCL Institute for Global Health, "Child survival, health and development 14 years post-randomisation of successful Participatory Learning and Action community mobilisation intervention in rural Bangladesh," 2022, p. 1.

• 13

  Ed Fottrell, email to GiveWell, March 29, 2023 (unpublished).

• 14
  • i.e.: we get results from the trial and they are high quality enough to be incorporated into our estimate. We would be likely to consider the results high quality if the characteristics of the groups researchers are able to re-sample do not differ on baseline characteristics.
  • If the results cause us to reaffirm our current estimate of 6% with less uncertainty, that would still count as an update for these purposes.

• 15

  "We spoke with a non-governmental organization that provides technical assistance to implement PLA-MNH programs, Women and Children First (WCF)." GiveWell, "Participatory Learning and Action – Maternal and Neonatal Health," 2022.