Optimal adherence appeared highest when the proportion of visits with adherence ≥90%% was used, and unknown intervals were excluded.
In contrast, reporting the proportion of participants with adherence ≥90%%, with unknown intervals assumed to represent poor adherence, gave the most pessimistic picture of optimal adherence.
The overall average ART adherence level was 84.2% with a range from 2.3%to100%0% and the proportion of patients with adherence level of 80% or more was 70% (81/116).
The adherence percentages according to doses taken and to time regimen prescribed, as well as the proportion of patients with adherence of greater than or equal to 95%%, did not differ significantly between the IG and CG.
In HPTN 039, with more frequent (monthly) visits, the proportion of participants with adherence 90 105 % ranged from a low of 64%% at the first visit (month 1) to a high of 73%% at month 14.
In Mwanza, where scheduled visits were quarterly, the proportion of participants with adherence 90 105 % ranged from a low of 46%% at the second visit (month 6) to a high of 57 % at month 18 (Fig. 2).
The proportion of participants with adherence ≥90%% was most sensitive to the value assigned to unknown intervals, with only 45%% of participants in HPTN 039 having optimal adherence if we assumed adherence was 50%% in unknown intervals, versus 54 % if we assumed adherence was 70%% in unknown intervals.
The proportion of patients with adherence greater than or equal to 95%% during the intervention period and the first follow-up period however, showed a tendency toward increased adherence in the IG, and reduced adherence in the CG, although this did not reach statistical significance (Fig. 2).
In the CG, no falls in adherence percentages were found at the follow up periods.> -wrap-foot> * Kruskal wallis test with 1 d.f.; all p > 0.05 No statistically significant difference was found between the proportion of patients with adherence greater than or equal to 95%% of doses taken in the IG and the CG, for the five measurements obtained, as shown in Table 5.
Summaries of optimal adherence, such as the proportion of person years with adherence ≥90%%, may give a better reflection of actual product use, but are more sensitive to assumptions made about intervals with unknown adherence, particularly when the amount of missing data is large.
