A mediation analysis was conducted to examine the mechanisms of change in relation to those hypothesised in an a priori logic model.

The mean adherence between 6 to 12 months was primarily mediated by awareness of medication usage (37% of the total effect mediated, 95% CI 24% to 51%), with habit formation (9%, 95% CI 3% to 16%) the second most important factor (Table 18). Other mediating effects were reduced concerns, increased self-efficacy and ease of effort, resulting in a total mediated effect of 51%. There was some evidence that the intervention increased awareness among patients who used several nebulisers at baseline.

TABLE 18

Mediator–outcome and mediator–exposure association results

As shown in Figure 14, better awareness seems to be associated to better normative medication adherence.

FIGURE 14

Normative medication adherence vs. baseline prescription by randomisation group: (a) usual care; and (b) intervention.

As shown in Figure 15, increase in baseline prescription seems to reduce awareness in the usual-care arm but not in the intervention group.

FIGURE 15

Awareness vs. baseline prescription by randomisation group interaction graph: (a) usual care; and (b) intervention.

As shown in Figure 16, increase in effort seems to improve normative medication adherence but this effect varies for males and females.

FIGURE 16

Effort vs. gender by randomisation group interaction graph: (a) usual care, female; (b) usual care, male; (c) intervention, female; and (d) intervention, male.

Figure 17 and Tables 19–21 and provide additional information and data from this analysis.

FIGURE 17

Mediation pathway DAG. Model fit of the model with concerns affecting self-efficacy was similar to self-efficacy affecting concerns but both models produced similar estimates.

TABLE 19

Association between mediators

TABLE 21

Final mediation model result

TABLE 20

Final mediation model result