Showing papers by "Michael Kremer published in 2017"

Should the WHO withdraw support for mass deworming

Abstract: 1 World Bank, Washington DC, United States of America, 2 Harvard T. H Chan School of Public Health, Boston, Massachusetts, United States of America, 3 Center for Effective Global Action, University of California, Berkeley, Berkeley, California, United States of America, 4 Department of Economics, University of California, Berkeley, Berkeley, California, United States of America, 5 Department of Economics, Harvard University, Cambridge, Massachusetts, United States of America

More Evidence on the Effects of Deworming: What Lessons Can We Learn?

Abstract: In this month’s issue, Liu and others present new evidence on the effects of deworming children from a clusterrandomized trial conducted in rural China. This was a useful, well-conducted study. The study found that a program in which school-age children were given deworming pills in school to take at home reduced infection prevalence modestly (end line prevalence of any worm infection was 31.4% in the control group and 27.7% in the treatment group). However, the authors could not reject the hypothesis that the program had no effect on nutrition, cognition, or school performance. As Liu and others emphasized, their results should be interpreted in light of the context they examined. The authors noted that in their setting, baseline prevalence of soil-transmitted helminths (STH) was low (31% for Ascaris, 23–24% for Trichuris, and 1% for hookworm), infections were “light intensity,” and even within the light-intensity group, measured egg counts were low. For example, mean infection intensity among children infected with Ascaris was less than 1,000 eggs per gram (epg; up to 4,999 epg is a light infection) and infection intensity for Trichuris was less than 70 epg (up to 1,000 epg is a light infection). The authors are to be commended for reporting prevalence and intensity for all three STH species, which greatly aids the interpretability of their results. Compliance with the deworming treatment was low, presumably because the study differed from most schoolbased deworming programs, and in particular from the standard World Health Organization (WHO) protocol; in accordance with Chinese regulations preventing children from taking the drugs at school, students were given pills to consume at home. Only 52% of participants reported taking all the recommended pills (75.6% reported taking at least half of the recommended dose of albendazole [200 mg] in both rounds of treatment). Compliance was self-reported, so any social desirability bias by respondents would mean that true compliance was lower than 52%. The fact that the end line difference in prevalence between the treatment and comparison group was only 3.7% points could potentially be explained by a combination of high rates of reinfection and this relatively low rate of compliance. It is useful to first consider the point estimates of effects in the Liu study and then consider issues of hypothesis testing and statistical power. The estimated effects in the Liu study were largely consistent with the results of a metaanalysis recently conducted by Croke and others on the impact of mass drug administration (MDA) on weight gain. The Croke analysis found that, in the environments where WHO recommends deworming, school-based MDA was effective in increasing weight, and indeed that it was very cost-effective relative to school feeding (school feeding was an example of a nutrition program that targeted similar populations for which, based on a review of randomizedcontrolled trials, the authors were able to identify a costeffectiveness analysis). Although too few studies have been conducted to draw strong conclusions about how effects vary with prevalence, point estimates are consistent with the commonsense view that effects are smaller in lower prevalence environments. Among studies in environments with greater than 50% infection prevalence in the Croke study, the average weight gain estimated in a random effects meta-analysis was 0.18 kg (95% confidence interval [CI] 0.07, 0.29). In contrast, the average effect in settings with less than 50% infection prevalence was 0.06 kg (95% CI −0.13, 0.25). A separate meta-analysis, by TaylorRobinson and others, also estimated an average weight gain of 0.06 kg in under 50% infection prevalence settings. In this lower prevalence sample, the Croke study estimated that the standard deviation of true effects was 0.25 kg. Using the method of Higgins and others, this implies that approximately 95% of places are expected to have a true effect in the range −0.59 to 0.71 kg. To put the outcomes in comparable terms, the weight gain estimate from the Liu study expressed in kilograms (rather than weight-for-age z score) was 0.03 (95% CI −0.25, 0.32). This clearly was well within the expected range. It is also unsurprising that the point estimates in the Liu study were lower than the estimated mean effects from this meta-analysis, since reported compliance was only 52% and infection prevalence at end line was only 12% lower in the treatment group. Although not consistently reported, compliance was typically 80–100% in other trials incorporated in the meta-analysis. Assuming that treatment effects were proportional to compliance and compliance was 0.52 implies that with full compliance, the estimated effect in the Liu study would have been (1/0.52) × 0.03 kg = 0.06 kg, close to the average estimated effect found by Croke and others in studies with less than 50% prevalence. Thus, while the results of the Liu study were consistent with the Croke analysis and with the hypothesis that MDA is cost-effective, the confidence intervals were wide enough to also encompass zero effect. The study was larger and better powered than many earlier trials, but as discussed later, power is inherently limited due to low prevalence and low compliance. Moreover, while the cluster-randomized nature of the trial was necessary to pick up potential epidemiological spillovers from treatment and thus to accurately measure impact, it did limit statistical power. Statistical power is limited in low-prevalence settings, since any effect on infected children will be averaged together with effects on uninfected children, who cannot benefit directly from treatment. The authors state that among the 10 outcomes of interest, STH prevalence was the outcome that *Address correspondence to Michael Kremer, Department of Economics, Harvard University, Littauer Center M-20, Cambridge MA, 02138. E-mail: mkremer@fas.harvard.edu