iconstar paper   HIV Articles  
Back grey arrow rt.gif
 
 
Comparison of subjective and objective adherence measures for pre-exposure prophylaxis against HIV infection among serodiscordant couples in East Africa. - Electronic Monitoring
 
 
  Download the PDF here
 
AIDS Jan 16 2016
 
Musinguzi, Nicholas; Muganzi, Collins D.; Boum, Yap II; Ronald, Allan; Marzinke, Mark A.; Hendrix, Craig W.; Celum, Connie; Baeten, Jared; Bangsberg, David R.; Haberer, Jessica E.
 
Abstract
 
Background: Pre-exposure prophylaxis (PrEP) efficacy is highly dependent on adherence. Yet, it is unclear which adherence measures perform best for PrEP.
 
Methods: We compared three types of self-reported (SR) adherence questions (rating of ability to adhere, frequency of doses taken, percent of doses taken) and three forms of objective adherence measurement (unannounced pill counts [UPC], electronic monitoring [EM], plasma tenofovir levels) using data from an ancillary adherence study within a clinical trial of PrEP among East African serodiscordant couples (Partners PrEP Study). Monthly measures were assessed for the first six months of follow-up.
 
Results: 1,147 participants contributed 6,048 person-months of data to this analysis. Median adherence was high: SR rating (90%), SR frequency (93%), and SR percent (97%); UPC (99%); and EM (97%). Prevalence of steady state daily dosing (SSDD; >40 ng/mL) was 74% in a random subset of tenofovir samples obtained from 365 participants. Discrimination of SSDD versus less than SSDD levels was poor for SR rating (area under the receiver operating curve [AROC] 0.54), SR frequency (AROC 0.52), SR percent (AROC 0.56) and UPC (AROC 0.58), but moderate for EM (AROC 0.70). Correlation was moderate among self-reported measures, adherence (0.61-0.66), but low for these self-reported measures compared with UPC (0.32-0.36) and with EM (0.22-0.28).
 
Conclusions: EM was the only adherence measure with meaningful ability to discriminate between SSDD and less than SSDD plasma tenofovir levels. Correlation between subjective and objective measures was poor. Future research should explore novel approaches to adherence measurement as PrEP moves into demonstration projects and programmatic implementation.
 
Real-time adherence monitoring may improve accuracy of EM through rapid identification of technical failures, and also has the added potential for delivery of real-time adherence feedback and/or intervention
 
Adherence
 
The median adherence was 90, 93, and 97% for the self-reported measures (SR rating, SR frequency, and SR percent, respectively), and 99%, and 97% for UPC and EM respectively; the data were generally left-skewed (Figure 3). The median time between EM openings was 24 hours (IQR 23.5-24.5); the median number of >48 hour gaps between EM openings over the six-month analysis period was 2 (IQR: 1-6). Perfect (100%) adherence was most common with SR frequency and UPC (66% of participant-months in each method) and least common with SR rating (50% of participant-months). Less than optimal adherence (<80%)1 was most commonly seen with SR rating (10% of participant-months) and EM (12% of participant-months). Of note, 18% of participant-months for EM were capped because of adherence >100%. Eighty-nine percent of these exceeded the anticipated adherence by ≤2 additional doses (i.e., 100-110% adherence). Tenofovir drug levels were assessed on 486 randomly selected blood samples taken from 365 participants (32%) in the ancillary adherence study. Age, gender and study arm distribution in these 365 participants were similar to the that of the total cohort in the adherence study1. Prevalence of SSDD was 74% (n=358) of the 486 samples.
 
Predictive validity
 
As shown in Figure 1, mean adherence was non-significantly higher in participants with SSDD plasma tenofovir levels compared to those with less than SSDD: 89% versus 85% (p=0.08) for SR rating, 93% versus 91% (p=0.18) for SR frequency, and 96% versus 93% (p=0.05) for SR percent. In examining the objective measures, mean adherence was significantly higher in participants with SSDD versus those with less than SSDD: 97% versus 92% (p=0.02) for UPC, 93% versus 72% (p<0.001) EM adherence for the 28 days prior to sample collection for tenofovir determination, and 94% versus 79% (p<0.001) for 7 days prior to sample collection. Of all adherence measures, EM explained the most variability (5.5%) in tenofovir levels. Inclusion of any or all other adherence measures explained an additional <1%.
 
As shown in Figure 2, the area under the receiver-operating curve (AROC) for all self-reported measures and for UPC was low (0.51-0.54). EM adherence showed the highest AROC of 0.70 (95% CI, 0.64, 0.76) for the 7 days and 0.68 (95% CI, 0.61, 0.74) for the 28 days prior to collection of the tenofovir samples. In the sensitivity analysis considering the threshold consistent with any dosing in the past week (>0.31 ng/mL), we found no differences in the magnitude of discrimination for any adherence measure compared with the SSDD threshold. Discrimination of SSDD with UPC , however, lost statistical significance (p=0.09).
 
Of note, of the 128 participant-months with less than SSDD, 43 participant-months from 40 participants were found to have 100% EM adherence for the seven days prior to sample collection.
 
DISCUSSION
 
In a cohort of East African serodiscordant couples participating in a clinical trial of PrEP, adherence was high by multiple forms of self-report, unannounced pill counts and electronic monitoring; tenofovir drug levels were moderately high. Despite the similarities in overall adherence values (e.g., medians), correlation between subjective and objective measures was poor and only electronic monitoring was able to meaningfully discriminate between steady state and less than steady state daily dosing plasma tenofovir levels. Tenofovir drug levels serve as an important benchmark for comparison with other adherence measures, because they document medication ingestion. Like any adherence measure, however, drug levels have limitations, including assay failure, variable drug metabolism within and among individuals, and a relatively short half-life in plasma26. Additionally, a dose taken just prior to determination of a drug level will mask non-adherence in the prior several days. Our findings underscore the complexity in accurately measuring adherence behavior.
 
The three self-reported measures had similar inability to discriminate between SSDD and less than SSDD tenofovir with AROCs equivalent to a coin flip. The recent analysis within the iPrEx trial also found low predictive validity for self-reported adherence with an AROC of approximately 0.517. Similarly, in the FEM-PrEP trial, the positive predictive value of self-reported adherence compared to tenofovir levels was low, ranging between 28.7% and 42.2% when averaged over time16.
 
The poor correlation between subjective and objective measures (rho: 0.22-0.36) was likely due in part to social desirability and recall bias in self-reported adherence measurements, and is consistent with prior studies of both PrEP and ART adherence12,17,19,27. The statistical significance of these correlations is likely driven by the large number of participant-months, rather than clinically meaningful relationships. Of note, the low level of correlation may be partially explained by the clustering of self-reported adherence values in this analysis (i.e., six categories for SR rating and SR frequency and 11 categories for SR percent) compared to continuous adherence with the objective measures. Importantly, within the TDF2 trial, a correlation analysis yielded a phi coefficient of 0.2818 and moderate correlation has been seen with self-report and drug levels for ART28. It is therefore possible that self-report may have limited validity in some settings.
 
Correlations among the three types of self-reported measures (rho: 0.61-0.67) were higher; however, the distribution of the three self-reported measures differed. SR rating had the widest distribution and identified individuals with low adherence, suggesting that this type of adherence question may be most informative for identifying adherence challenges. Similar findings have been seen in prior analyses of ART adherence measurements21-23. Indeed, the goal with self-reported adherence is often described as “pulling people off the ceiling”-- even if self-reported adherence is overestimated, those reporting less than perfect adherence may benefit from additional adherence support.
 
Interestingly, average SR rating was lower than EM adherence. Self-reported adherence is typically higher than objective measures, suggesting that individuals may have underestimated their adherence when creating this type of adherence estimation. Lower self-reported adherence when the corresponding EM adherence was also low or moderate has been observed previously29,30, suggesting that participants may be overly self-critical whenever their adherence is less than optimal. Cognitive testing has been used to understand the performance of different types of self-report31 and may be beneficial in this setting to understand how this population views adherence behavior. Alternatively, higher EM than self-reported adherence may have been introduced by extra EM openings unassociated with dosing.
 
This analysis also identified noteworthy findings among the objective adherence measures. First, a low rho correlation was found between UPC and EM adherence (0.4). While predictive validity for UPC was statistically significant, discrimination between SSDD and less than SSDD plasma tenofovir was lower than that of EM (AROC 0.58 versus 0.68, respectively). These findings suggest UPC may have been subject to manipulation (e.g., pill dumping despite the unannounced nature of the pill count). One potential limitation of the comparisons of UPC and EM with plasma tenofovir is the difference in monitored times in which UPC and EM reflect one month of adherence behavior, whereas the window for detection of tenofovir in plasma is seven days. The findings for 28-day and 7-day EM adherence, however, are similar. Overall, 43 participant months had 100% 7-day EM adherence, but less than SSDD tenofovir suggesting that in some cases, participants may have manipulated the EM devices (e.g., opening the device without taking the pill), although assay failure and/or atypical drug metabolism could have impacted the tenofovir findings. Importantly, the number of participants with highly discrepant adherence by EM and plasma tenofovir level reflects only 11% of participants randomly selected for tenofovir level determination).
 
The strength of this analysis lies in the large sample with multiple measures of adherence; however, there are also limitations. First, the time frames associated with the subjective and objective adherence measures were mismatched compared to the tenofovir drug quantification (28 versus 7 days) as noted above, and only EM adherence could be adjusted. Additionally, a sample of participants was used for determining plasma tenofovir levels among approximately one-fifth of the cohort. Although the sampling was random, some bias may have been present.
 
In sum, adherence is paramount for PrEP efficacy in reducing HIV transmission, yet characterizing adherence behavior remains challenging. While the low cost and ease of implementation associated with self-reported adherence makes it an ideal candidate for adherence measurement in future studies and clinical implementation of PrEP, our findings suggest that self-reported adherence may be inaccurate, even when geared toward estimation rather than enumeration. Moreover, objective adherence measures have limitations in accuracy as well, including possible manipulation and/or technical failures. Future efforts should explore novel approaches to PrEP adherence measurement. For example, cognitive testing may help identify more informative self-report questions31. SMS may allow for more frequent assessment of self-report, thus reducing recall bias, as well as potentially reducing social desirability through the relative anonymity of the technology. Real-time adherence monitoring may improve accuracy of EM through rapid identification of technical failures, and also has the added potential for delivery of real-time adherence feedback and/or intervention32,33. Pharmacokinetic measures (e.g., drug levels in dried blood spots, hair)19,34,35 that provide an understanding of adherence over longer time periods compared with plasma are desirable36. These or other measurement approaches will be critical for defining the adherence-efficacy relationship as PrEP is rolled out beyond clinical trials. Ongoing assessment of adherence behavior in demonstration projects and programmatic implementation will be important, because motivations, facilitators, and barriers to adherence may differ (e.g., known PrEP efficacy, intensity of counseling).

 
 
 
 
  iconpaperstack View Older Articles   Back to Top   www.natap.org