Predictor | Effect measure | Model | ||
Unadjusted | Adjusted (parsimonious) | Adjusted (full) | ||
RDT subsidy | Risk Differences (98% CI) | 2.9% (−0.4% to 6.1%) | 2.5% (−0.5% to 5.5%) | 2.5% (0.2% to 4.8%) |
Risk Ratios (98% CI) | 1.030 (0.995 to 1.065) | 1.026 (0.995 to 1.058) | 1.025 (1.002 to 1.049) | |
Conditional ACT subsidy | Risk Differences (98% CI) | −0.9% (−2.4% to 0.6%) | −0.8% (−2.6% to 1.0%) | −0.6% (−2.2% to 0.9%) |
Risk Ratios (98% CI) | 0.991 (0.976 to 1.007) | 0.992 (0.974 to 1.010) | 0.994 (0.979 to 1.009) | |
Interaction between RDT and ACT subsidies | Risk Difference (99% CI) | −0.7% (−3.6% to 2.3%) | – | −0.6% (−2.8% to 1.6%) |
Risk Ratios (99% CI) | 0.993 (0.964 to 1.024) | – | 0.994 (0.972 to 1.017) |
The reported average main and interaction effects of RDT and conditional ACT subsidies are approximations of risk ratios due to the nature of log link. Unadjusted model included the main effects of the RDT and conditional ACT subsidies and their interaction to match the 2×2 factorial design. Effect coding was used so that main effects of each subsidy level can be interpreted averaged over the levels of the other subsidy. Fully adjusted model includes age (of patient), gender (of patient), education level (of patient or guardian if patient <18 years), occupation (of patient or guardian if patient <18 years), household size, wealth, and main and interaction effects of RDT and conditional ACT subsidies. Only the main effects and wealth was included in the parsimonious model identified by Beaulieu and O’Meara.26 For main effects, 98% CIs were used and 99% CI for the interaction effect to match the prespecified alpha allocation of 0.02 each for main effects and 0.01 for the interaction effect.
RD, risk difference; RR, risk ratio.