Saving Bangladeshi Babies

This continues my last post on global health. I want to wander into more methodological territory, too, concerning one terrific, but puzzling, article concerned with infant mortality prevention. I’d be especially interested in comments by folks with specific maternal and child health expertise.

Especially in global health, important research is done every day under difficult conditions that highlight both the strengths and the inherent weaknesses of randomized trials.

The strengths are important and obvious. Large, well-executed intent-to-treat trials lets us know whether the offer of a particular treatment brings statistically significant differences in health outcomes between a treatment and a control group. Because patients are randomly assigned, we can credibly conclude that differences in treatment actually caused the difference in outcome between the two groups. There is just no credible substitute for such well-controlled trials.

I don’t put much stock, for example, in reports that coffee drinking is associated with increased survival. I’m sure that joining the Mormon Church or the Boy Scouts is associated with similar protective effects, too.  There are too many ways that longitudinal cohort studies can generate spurious findings about such things. You control for the obvious confounders, but you inevitably leave things out. (HAP)

Non-experimental studies are obviously valuable. Associational studies nailed the link between cigarettes and lung cancer long before any rigorous experimental trial could be done. Still, even excellent non-experimental studies easily prove misleading when the resulting correlations are interpreted in a causal way.  Elaborate econometric approaches are especially oversold. We need instrumental variables, Heckman correction, and other tools. We should never take the results entirely at face value or fully trust whatever pops out of the latest Stata run. Over my public health career, I can’t think of a single counter-intuitive finding that emerged from an elaborate econometric analysis that (a) seemed really odd to clinicians and public health experts, and (b) turned out to be true.

The weaknesses of a randomized trials are a bit more subtle, and extend beyond the obvious and crucial point that the right randomized trial is often too expensive, too time-consuming, or infeasible. Even when a strong randomized trial is performed, it often leaves important questions unanswered.

This is especially true when one wants to look beyond the bottom line, to explore how and why we observed the outcomes we did. Such information is critical for designing interventions and for policy. It tells us whether and how one might generalize a promising intervention to other settings, how an intervention might be improved, which patient groups are likely to derive the greatest benefit. Information about mediating pathways also helps us understand how much stock to put in a particular finding, and whether the investigators credibly explained what really produced the outcomes they observed.

Suppose, for example, I perform a randomized trial in which I offer academic mentoring to 500 fifth-graders.  I compare their subsequent grades and academic trajectories to those of 500 similar peers who were given usual services. My intervention works. Kids who received the mentoring were less likely to be held back or to be tracked into special education. But why did it work? Did mentoring improve kids’ reading skills? Maybe mentoring improved kids’ grades by improving their classroom behavior. Maybe mentors were effective advocates when teachers and administrators made difficult decisions about individual children. These different possibilities obviously have different implications for educational interventions.

Now back to the topic at hand. In this week’s JAMA, Lars Ake Persson and collaborators in the MINIMat study team describe one effort to reduce infant and child mortality.  4,436 pregnant rural Bangladeshi women were recruited for a rigorous and complicated trial. This paper rewards a careful read, not least because it displays the sheer craftsmanship required to execute such a thing.

Within several treatment arms, women were offered various forms of vitamins, micronutrients, and food supplements at different points in their pregnancies. Some women were offered food supplementation early in their pregnancies. Others were offered similar supports mid-gestation under usual care.

Infant mortality appeared to drop substantially (from 44/1,000 to about 17/1,000) when pregnant women were assigned to a treatment that included micronutrients, iron, folic acid, and the invitation to receive nutrition supplements really early in their pregnancies. Those offered food supplementation mid-gestation actually appeared to experience worse infant and child mortality.

These striking results, if they hold up, underscore the policy challenge of early intervention in pregnancy. It’s difficult to improve birth outcomes until (a) women know they are pregnant and wish to carry their pregnancies to term, (b) decide to seek care, (c) engage the care system, and (d) some intervention occurs. This sequence often doesn’t complete until women’s pregnancies are pretty far along.

Many studies go back further, identifying the importance of preconceptional care. This sometimes feasible, for example in cultures where first pregnancies typically arise among young brides within the first year of marriage. Here in the U.S., about half of all pregnancies are unplanned. The first prenatal care visit often happens a few months into the pregnancy. Given these realities, public health authorities here and elsewhere rely on universal strategies such as folic acid flour fortification rather than specifically targeted preconceptional approaches.

Getting back to the specific MINIMat trial, it’s hard to get under the hood to understand what really happened. The main survival benefit was observed in first six days of infant life. Yet the team found no comparable protective effects in preventing stillbirths or in preventing infant and child mortality after that initial neonatal period. The usual mediating mechanisms such as birthweight, gestational age, and head circumference don’t look different across the different groups.

An accompanying commentary by Parul Christian and Robert Black raises hard questions about whether the MINIMat results are fully plausible. Even in rural Bangladesh, infant deaths are thankfully rare. The research team observed about fourteen fewer early neonatal deaths than was expected within the preferred intervention group. The specific contrasts in mortality across the different groups don’t really match the obvious causal hypotheses one expects from the intervention. Especially when the different treatment groups were not masked, early invitations to nutritional interventions might have produced led to better coordination and management of labor and delivery, which could confound the results.

I’ve talked with several experts. Every one is having a hard time identifying powerful mediating mechanisms to really explain these results.  I’m still trying to get my head around what happened. Sometimes, clinical trials raise as many questions as they answer. Once again, the data don’t speak for themselves. This intriguing trial should be replicated. The stakes could hardly be higher—child survival across the world.

Share on twitter
Share on facebook
Share on linkedin
Share on reddit
Share on email

Hidden information below

Subscribe

* indicates required
Email Format