Few development interventions have generated as much argument per dollar as deworming. A pill costing well under a dollar per child per year became, for a time, a flagship example of evidence-based aid, championed by economists and charity evaluators alike. Then it became a flagship example of how evidence can mislead.

The puzzle is genuine. Studies of deworming's direct health effects often find little to measure. Yet some long-run studies find large gains in schooling, earnings and consumption decades later. How can a treatment with modest short-term health effects transform adult lives? Either something real is happening that standard health measures miss, or the long-run findings are less solid than they appear.

The so-called worm wars produced plenty of heat. This review separates what the evidence firmly supports from what remains contested. It then asks the question that matters most for practitioners: what should you do when the expected benefit is large but the uncertainty is larger?

Soil-transmitted helminths, a group that includes roundworm, whipworm and hookworm, infect hundreds of millions of people. So does schistosomiasis. Most of those infected are children in poor, tropical settings with inadequate sanitation. Heavy infections cause anaemia, malnutrition and abdominal pain. Nobody seriously disputes that treating a heavily infected child improves that child's health. The dispute is about mass deworming, which means giving everyone in a school or community the drug without testing first.

Here the evidence is sobering. The Cochrane systematic review, updated several times, concludes that mass deworming probably has little or no effect on average weight, height, haemoglobin, cognition or school attendance. A rival meta-analysis by Croke and colleagues drew on a broader set of trials. It found an average weight gain of roughly 0.13 kilograms, which is statistically significant but small. Effects on haemoglobin were also modest in pooled estimates, even though haemoglobin is the measure most directly linked to how hookworm harms the body.

The two camps disagree less about the data than about how to read it. Mass treatment dilutes effects. If only a fraction of children carry heavy worm loads, the average benefit across everyone treated will be small, even if infected children benefit substantially. Many trials were also run in low-prevalence settings or measured outcomes too soon. Pooling such studies can produce a confident-looking null result that mostly reflects limited statistical power.

There is also a structural problem that individual trials struggle to capture. Worms spread through contaminated soil, so treating many children reduces reinfection for everyone, including those who were not treated. When trials randomise individuals within the same school or village, the control group also gains, so the trial can understate the benefit. Studies that randomised whole schools or regions tended to find larger effects. This is part of why researchers disagree about which study designs deserve the most weight.

Takeaway

A small average effect can hide a large effect concentrated in a minority. Before concluding that an intervention 'doesn't work', ask who it was meant to help and whether the study could have detected that.

The case for deworming rests heavily on one programme. In the late 1990s, the Primary School Deworming Project phased treatment into 75 schools in Busia, Kenya. Miguel and Kremer published their analysis in 2004. They found that school participation in treated schools rose by about 7.5 percentage points, which cut absenteeism by roughly a quarter. Untreated children in nearby schools appeared to benefit too.

Follow-up studies made the result more striking. A decade later, men who had received more years of treatment worked about 3.5 more hours a week, and women were more likely to have attended secondary school. A twenty-year follow-up reported consumption gains of around 14 per cent and higher hourly earnings. Historical research by Hoyt Bleakley on the Rockefeller hookworm campaign in the American South found similar patterns in school enrolment and literacy.

Then came the scrutiny. In 2015, epidemiologists at the London School of Hygiene and Tropical Medicine published a replication. They identified coding errors and applied a different analytical approach, and concluded that the attendance effects were weaker than reported. The original authors replied that the core findings survived once the errors were corrected. They argued that the reanalysis discarded valid information. Both sides published extensively, and neither fully convinced the other.

What remains is a narrow but important evidence base. The long-run results come largely from one setting with high worm prevalence. Because treatment was phased in, the design compares children who received more years of treatment with children who received fewer, rather than comparing treatment with no treatment at all. Other evidence points the same direction, including long-run results from Uganda, cognitive gains among younger siblings of Busia pupils, and historical campaigns. But pointing the same direction is not the same as replication. Effect sizes from a single study also tend to shrink when the intervention is tested elsewhere.

Takeaway

Long-run studies can reveal effects that short-run measures miss. But a striking result from one place is a hypothesis about other places, not a guarantee.

Deworming's appeal lies in its price. School-based programmes can deliver treatment for under a dollar per child per year. The drugs are often donated, and teachers can give them out in a single day. At that cost, an intervention does not need large effects to be worthwhile. It only needs a reasonable chance of meaningful effects.

The charity evaluator GiveWell adopted this logic. It did not take the Busia earnings gains at face value. Instead, it applied steep discounts, at times close to 90 per cent, to reflect doubts about whether the results would replicate and apply in other places. Even after such discounting, deworming often looked competitive with leading health interventions. In 2022, however, GiveWell removed deworming organisations from its top charities list because of the greater uncertainty around their impact, though it continued to fund selected programmes.

The deeper question is how to handle uncertainty itself. An expected-value approach says that a cheap intervention with a small chance of large benefits can beat a costly intervention with a certain but modest benefit. Critics counter that this reasoning rewards optimism. If the one strong study is a false positive, the entire case collapses, and resources flow to the programme with the best story rather than the best evidence.

Context complicates any single answer. Worm prevalence has fallen sharply in many regions, partly because of mass treatment itself, and benefits likely rise with the intensity of infection. The World Health Organization recommends periodic mass treatment where at least 20 per cent of the population is infected with soil-transmitted helminths. A programme that was justified in high-burden Busia in 1998 may offer far less in a low-burden district today. Cost-effectiveness is not a property of the drug; it is a property of the drug in a place.

Takeaway

Cheap interventions can justify action under uncertainty, but only if that uncertainty is honestly priced in. Cost-effectiveness belongs to an intervention in a specific context, not to the intervention alone.

The deworming debate is not really about whether worms are bad for children. It is about how much weight a small number of influential studies can bear, and how to act when evidence is promising but thin.

The robust findings are modest. Treating infected children helps them, mass treatment produces small average health gains, and effects depend heavily on how common infection is. The large long-run gains are plausible and supported by several strands of evidence, but they remain contested.

For practitioners, the lesson extends beyond worms. Target interventions where the problem is actually severe. Discount striking results honestly. And keep collecting evidence after a programme becomes popular, because that is often when it is needed most.