Category | Covariate | Rationale | Balancing criterion |
---|---|---|---|
Potential confounders | 1. Age: % of population < 15 yearsa | Dengue risk is age dependent | Each arm within ± 5% of overall population value |
2. 3-year average dengue incidence ratea | Historical dengue incidence may predict future risk | Each arm within ± 5% of overall population value | |
3. Education: % completed high schoola | Proxy for socioeconomic status that may predict dengue risk | Each arm within ± 5% of overall population value | |
Potential sources of bias | 4. Incidence of other febrile illnessf presenting to Puskesmas clinics in 2014–2015b | Prevent chance association between other febrile illness and intervention | Each arm within ± 5% of overall population value |
Sample size | 5. Number of clusters | To maximise precision and power | 12 clusters per study arm |
6. Cluster populationc | To maximise precision and power | Each arm 45–55% of total population | |
Logistics | 7. Total cluster area (km2)d | Releases to be done over approximately half the city | Each arm 45–55% of total area |
8. Non-release area within cluster (km2)e | To prevent an excess of non-residential areas falling in intervention arm | Each arm 45–55% of total non-release area | |
9. Four spatial strata | To prevent a large contiguous intervention area | Within each spatial stratum, three clusters per study arm |