Skip to main content
. Author manuscript; available in PMC: 2021 Mar 20.
Published in final edited form as: Ann Intern Med. 2020 Mar 24;172(8):523–532. doi: 10.7326/M19-2945

Table 3.

Mediation Analysis Estimating the Proportion of the Observed Canakinumab Effect on Anemia Resulting From Changes in hsCRP After Treatment, Adjusted for Confounders

Effect Relative Mean Survival* 95% CI P Value
Mediator: achieved 3-mo hsCRP level <2 mg/L (threshold effect)
 Controlled direct 1.29 1.05-1.57 0.014
 Natural direct 1.27 1.07-1.51 0.006
 Natural indirect 1.11 1.03-1.20 0.007
 Total§ 1.42 1.20-1.67 <0.001
 Proportion mediated 0.35    –  –
Mediator: log(hsCRP) at 3 mo minus log(hsCRP) at baseline (continuous effect)
 Controlled direct 1.18 0.99-1.41 0.071
 Natural direct 1.18 0.998-1.41 0.053
 Natural indirect 1.18 1.10-1.26 <0.001
 Total§ 1.40 1.19-1.65 <0.001
 Proportion mediated 0.53    –  –

hsCRP = high-sensitivity C-reactive protein.

*

Relative mean survival is estimated in a Weibull accelerated failure time model so that estimates above 1 indicate greater mean anemia-free survival. For example, in the threshold analysis (top), the total effect estimate of 1.42 indicates a 42% longer mean anemia-free survival associated with canakinumab vs. placebo.

Direct effects, both controlled and natural, measure the effect of canakinumab that is not mediated through its effect on hsCRP. Controlled direct effects fix the achieved level of hsCRP at 3 mo at a specific value, identical in treated and untreated participants, whereas natural direct effects consider the mediator effect fixed across the range of achieved hsCRP values at 3 mo observed in the placebo group.

Natural indirect effects compare the effect of canakinumab in 2 treated persons, 1 who achieves the hsCRP level expected with active treatment and the other with the expected hsCRP level seen in the placebo group.

§

Total effect estimates the relative mean anemia-free survival associated with canakinumab treatment vs. placebo, including both the direct effect of treatment and its indirect effect through the mediator, hsCRP change.

Proportion mediated estimates the proportion of the total canakinumab effect that acts through its effect on the mediator, change in hsCRP level. For example, 35% of the effect of canakinumab on anemia is estimated to be mediated by achievement of a threshold hsCRP level <2 mg/L at 3 mo (top), whereas 53% of the effect of canakinumab on anemia is estimated to be mediated by the continuous change in hsCRP level from baseline to 3 mo (bottom).