Table 2 Assumptions for sample size required to measure the primary outcome “incidence of births before girls’ 18th birthday”
Parameter | Assumed level | Comment |
|---|---|---|
Incidence of births before girls’ 18th birthday | 0.08 | We assume that 27% of girls in the control arm will have given birth before their 18th birthday. This corresponds to an average incidence rate of (27%-3%) = 8% per year over the average 3-year period (from the time the average age is 15). |
Effectiveness of combined intervention vs control | −40% | i.e. the incidence in combined intervention arm assumed to be 0.048 |
Effectiveness (i.e. (1-RR) × 100 of economic intervention vs control | −25% | i.e. the incidence in economic intervention arm assumed to be 0.06 |
Effectiveness of combined intervention vs economic intervention | −20% | The combined intervention will offer [1-(0.0.048/0.06)], i.e. 20% more relative protection than the economic intervention alone. |
Cluster size | 28 | Average number of girls in grade 7 in the selected schools is 31. If we assume that up to 10% may be lost to follow-up by the time of measuring the outcome, the average cluster size will be 28 |
Person years per cluster | 84 | If 28 participants are followed up for 3 years on average, person years per cluster are 84. |
K | 0.15 | The ICC was 0.00737 for “ever pregnant” after the intervention period in the cash transfer trial in Malawi (estimate obtained from Sarah Baird). This corresponds to k = 0.15 when the total proportion who have given birth by this time is 0.27. |
Z1 | 1.96 | |
Power for comparison of economic intervention vs combined intervention | 70% | We need 63 clusters in each of the intervention arms to have 70% power to detect the assumed difference |
Power for comparison of economic intervention vs control | 80% | We need 39 clusters in each arm to have 80% power to detect the assumed difference. The PASS power calculator for incidence rates does not allow for unequal trial arms, but the PASS function for proportions indicates that 63 economic and 31 control will give slightly higher power |
Power for comparison of combined intervention vs control | >95% | We need 23 clusters in each arm to have 95% power to detect the assumed difference. |