Miguel 2004 (Cluster)

Methods	Cluster quasi‐randomized stepped‐wedge trial Method to adjust for clustering: CIs adjusted for clustering in regression modelling, robust standard errors presented (confirmed in correspondence with authors) Cluster unit: schools Average cluster size: 400 ICCs: not reported Length of follow‐up: one year for phased quasi‐randomized comparisons for health outcomes. Two years for school attendance
Participants	All children living in endemic area Number analysed for primary outcomes: 75 primary schools randomized containing 778 children analysed for haemoglobin. 9102 children analysed for weight and height, 32% and 34% of eligible population analysed for exam performance and cognitive tests, and 100% of eligible population analysed for school attendance Age range/mean age: school children 12 years or under Inclusion criteria: children from 75 primary schools in the trial area Exclusion criteria: girls > 13 years old
Interventions	Deworming package of interventions vs no treatment Albendazole 600 mg (Zentel, SZB) every 6 months in 1998 intervention, and albendazole 400 mg (Zentel, SZB) in 1999; plus a) worm prevention education b) schools with schistosomiasis prevalence over 30% were mass treated with praziquantel (40 mg/kg Bayer) annually; 6/25 schools treated with praziquantel in 1998, and 16/50 treated with praziquantel in 1990; No treatment.
Outcomes	Weight‐for‐age z‐score ‐ change; Haemoglobin ‐ change; Exam score performance (ICS administered English, Mathematics and Science‐Agriculture exams in pupils in grades 3 to 8); Cognitive tests including picture search, Raven matrix, verbal fluency, digit span, Spanish learning, and a dynamic test using syllogisms; Height‐for‐age z‐score ‐ change; School participation rate based on external NGO assessment at unannounced visit. Not included in review: worm prevalence and intensity, self reported sickness, worm prevention behaviours: proportion "clean" as per health worker observation, proportion wearing shoes as per health worker observation, self‐reported contact with fresh‐water in past week, access to home latrine, malaria/fever.
Notes	Location: Kenya Community category: 1 Source of funding: Sponsored by the World Bank and the Partnership for Child Development.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	High risk	Schools in a deworming project were stratified by zone, their involvement with other NGO programmes, and then listed alphabetically and every third school assigned to start the programme in 1998, to start it in 1999, or to be a control.
Allocation concealment (selection bias)	High risk	Not concealed (see above).
Blinding (performance bias and detection bias) All outcomes	High risk	Pragmatic cluster implementation trial with no blinding.
Incomplete outcome data (attrition bias) All outcomes	Low risk	For haemoglobin, weight and height the outcomes have been measured on a random sub‐sample of the quasi‐randomized population.
Selective reporting (reporting bias)	Low risk	Outcome data not reported for cognitive tests, though authors state: Deworming treatment effects are not significantly different than zero for any component of the cognitive exam (results available on request).
Other bias	High risk	Recruitment bias: low (no asymmetric migration between schools) Baseline imbalance: low Loss of clusters: low (none reported) Incorrect analysis: low (correctly adjusted for clustering). Comparability with RCTs randomizing individuals: low Other sources of bias: high for confounding due to a co‐intervention. The drug intervention is accompanied by intensive health promotion that could account for some of the effects with key outcomes such as school attendance.