Adherence to HIV antiretroviral therapy is the strongest predictor of virologic suppression[1], [2], HIV drug resistance[3], disease progression and death[4], [5]. While treatment with unboosted protease inhibitors (PI) requires near perfect adherence for virologic suppression[1], the introduction of more potent non-nucleoside reverse transcriptase inhibitors (NNRTI) and ritonavir boosted PI therapy has lead to reliable virologic suppression at moderate levels of adherence for most, but not all patients[6]–[9].

Because not all patients with moderate adherence are suppressed, we asked if all missed doses are the same. Specifically, we hypothesized that patterns of adherence in addition to average adherence may be an important determinant of incomplete viral suppression. We focused on NNRTI based regimens because these drugs are the cornerstone for most first line regimens world-wide. Because the NNRTIs are potent and have a very long-half life in vivo, we hypothesized that once viral suppression was achieved, a sustained treatment interruption rather than frequent missed doses would be associated with virologic failure.

To test this hypothesis, we analyzed prospectively collected data from NNRTI treated individuals who achieved an initial virologic response during adherence monitoring with electronic medication monitors. Electronic medication monitors measure patterns of missed doses with a time-date record of pill bottle opening behaviour. The objective of this study was to identify adherence patterns predictive of virologic rebound on NNRTI-based antiretroviral therapy. We examined the temporal association between average adherence, treatment interruptions and subsequent viral rebound in all subjects, but in particular, subjects with <80% adherence because they are at highest risk for virologic rebound[7], [10].

These data suggest that sustained and repeated NNRTI treatment interruptions are associated with viral rebound. Specifically, sustained treatment interruptions more closely predicted viral rebound than interspersed missed doses in patients with low-to-moderate adherence, which represented 24% of our sample. Average adherence and the potential duration of treatment interruptions are, of course, not independent; 100% adherence precludes an interruption in treatment. As adherence rates decline, however, different patterns of missed doses are possible, as shown in Figure 2. Missed doses can either occur as sustained interruptions or more regularly interspersed missed doses. Our data suggest that not all missed doses are the same. In particular, sustained interruptions of NNRTI-based antiretroviral therapy are more closely associated with viral rebound than the same number of regularly interspersed missed doses among individuals with <80% adherence.

Our finding derived from “real life” treatment interruptions in socio-economically and ethnically diverse patients is consistent with both observational and experimental studies of treatment interruptions. For example, Oyugi et al. found that MEMS defined treatment interruptions with a mean of 11.5 days were associated with NNRTI resistance in Uganda[12]. In two trials, Dybul and colleagues found that repeated long cycles of 4-week off 8-week-on of efavirenz-based therapy was associated with resistance failure[13] while repeated short cycles of 1-week off 1-week-on was not associated with failure[14], even though the 4–8 week schedule delivered more medication over time than that 1–1 week schedule (67% vs 50%, respectively). Similarly, Cohen et al. found that virologic rebound was uncommon among an NNRTI-treated patients with repeated short cycles of 5-days on and 2-days off [15]. Our estimate of 50% probability of viral rebound occurring at 15 days is consistent with these data as well as pharmacokinetic data indicating that efavirenz clearance leads to drug levels that allow for viral replication in 5.8 to 14 days after discontinuation [16].

There are several limitations to our study. The number of events (five) is small, mainly because MEMS technology is not routinely used and because the risk of virologic rebound subsequent to viral suppression on NNRTI-based antiretroviral therapy is low[17]. Larger studies will be needed to confirm our results. Our observational design does not demonstrate causality and we did not measure drug-resistance. However, the risk of resistance after rebound on NNRTI-based therapy is predictably high [13], [18]–[20]. In contrast to structured treatment interruption trials, our study represents the distribution of “unstructured” interruptions common in routine practice. Finally, the result may have been different in a fully antiretroviral naïve population with ongoing viral replication.

Gross et al. found that patients treated with efavirenz-based regimens had suboptimal adherence up to 90 days prior to viral rebound[21]. Our data suggest that patterns of incomplete adherence, namely interruptions in treatment, may narrow this window. In particular, a treatment interruption of 15 days conferred a 50% probability of virologic rebound (Figure 1). Moreover, any treatment interruption of >7 days had a sensitivity of 100% and a specificity of 94% to detect subsequent viral rebound (Figure 2). Thus, the window of opportunity to intervene on risky adherence may depend not only on the level but pattern of adherence.

In summary, near perfect adherence leading to sustained virologic suppression remains the goal of HIV therapy. However, patients with moderate adherence to NNRTIs-based regimens can still achieve virologic control. For these patients with incomplete adherence, missing doses over a continuous and sustained interval may pose more risk for virologic rebound than interspersed missed doses. Limiting sustained NNRTI treatment interruptions may improve durable virologic suppression, especially in patients with incomplete adherence.