In July 2012, the US Food and Drug Administration approved coformulated tenofovir disoproxil fumarate/emtricitabine (TDF/FTC) for use as pre-exposure prophylaxis (PrEP) to prevent HIV infection.1 Current PrEP guidelines recommend daily use of TDF/FTC while an individual remains at risk of HIV infection, a period that may last many months or years.2 The effectiveness of PrEP in preventing HIV infection is strongly correlated with medication adherence during periods of risk.3
Adherence with TDF/FTC was variable in the clinical trials performed to demonstrate PrEP efficacy but was in most cases above thresholds necessary to prevent HIV infection (ie, >60%).3–5 Participants in these efficacy trials received substantial adherence counseling and monitoring. It is possible that adherence with PrEP will be lower in routine health care settings, where resources to support adherence may be limited. However, little is known about PrEP medication adherence and risk factors for low adherence in routine clinical practice.
The Veterans Health Affairs (VHA) health care system provides an opportunity to examine PrEP medication adherence in routine practice on a national scale. VHA is the largest integrated health care delivery system in the United States, with nearly 6 million Veterans in care in over 900 clinics across the United States.6 VHA is an equal-access health care system with minimal barriers to obtaining care related to insurance status or medication co-pays.7 Moreover, VHA has issued guidance supporting PrEP prescribing in both primary care and specialty care settings,8 and maintains national pharmacy and other administrative databases that allow for tracking of PrEP prescribing and refill patterns at the patient level.
We used VHA pharmacy and other administrative data to create a national cohort of Veterans initiating PrEP in VHA and examined TDF/FTC refill patterns in the year after PrEP initiation. We aimed to estimate TDF/FTC adherence based on pharmacy refill patterns and identify patient characteristics (eg, demographics, comorbidities, and socioeconomic status) and geographic factors (eg, geographic region and rural vs. urban residence) associated with high adherence. We also estimated how often Veterans discontinued PrEP in the first year.
Data Sources and Patient Cohort
We used national data from VHA's Corporate Data Warehouse (CDW) and a previously validated algorithm to create a cohort of Veterans initiating PrEP between July 1, 2012, and June 30, 2016.9 We ended the cohort in June 2016 to ensure a full year of follow-up pharmacy refill data for all Veterans (ie, through June 30, 2017). CDW contains data elements extracted from VHA's integrated electronic health record and administrative files, including patient demographics and residential Zone Improvement Plan (ZIP) codes, inpatient and outpatient visits, and pharmacy records. We also obtained 2012 data from the American Community Survey (ACS) on ZIP-code-level household poverty rates as an area-level measure of socioeconomic status for each patient.10
The algorithm used to identify Veterans initiating PrEP required: (1) at least 1 TDF/FTC fill of more than 30 days in the observation period; (2) no other fills for antiretroviral medications within 180 days of the date of first TDF/FTC fill; (3) no International Classification of Diseases (ICD) version 9 or 10 diagnosis codes for HIV (ie, ICD 9 codes 042 and V08; ICD 10 B20-B24) or hepatitis B virus (HBV) infection (ie, ICD 9 070.3, 070.3, 070.42, 070.52, V02.61, or ICD 10 B18.1) at any time; and (4) no ICD 9 or 10 codes for needle-stick exposure (ie, ICD 9 E920.5 or ICD 10 W46.0XXA) within 60 days of the date of first TDF/FTC fill. We reviewed charts for 864 individuals identified as PrEP users by this algorithm from July 2012 through April 2016. There was medical record documentation that TDF/FTC was prescribed specifically for PrEP among 825 of these 864, yielding a positive predictive for the algorithm of 0.95 (unpublished data). Inclusion of laboratory results for HIV and HBV infection did not improve performance of the algorithm. These laboratory results did not identify individuals receiving TDF/FTC for treatment of HIV or HBV infection who were not already identified by ICD codes and medication use.
Variables Representing Patient Characteristics
We created a series of variables representing patient characteristics that we hypothesized may be associated with PrEP medication adherence, based on existing PrEP literature and our clinical experience providing PrEP.5 These included demographics (ie, age, sex, and race/ethnicity), selected comorbidities (ie, substance use, depression, hypertension, diabetes, and chronic kidney disease), rural vs. urban residence, and geographic region of the United States (ie, Northeast, South, Midwest, or West).
VHA administrative data have high agreement with self-reported race for black and white patients, but lower agreement for other groups.11 We therefore classified race/ethnicity as black, white, other, or missing. We classified comorbidities using diagnosis codes in the year of the first TDF/FTC fill and grouped these based on Elixhauser method and the algorithm proposed by Quan et al.12 Measures for alcohol use and illicit substance use problems were highly correlated (r > 0.6), leading us to combine these to create a single-indicator variable for a diagnosis of a substance use problem.
We classified rural vs. urban patient residence using rural–urban commuting area codes for each patient's residential ZIP code, and geographic region by US Census criteria.13 Rural–urban commuting area codes use data on population density and commuting patterns to classify ZIP codes as rural or urban. We also used ACS data to create an area-based measure of socioeconomic status, based on the percent of households below the federal poverty level in each patient's residential ZIP code. We used the ZIP-code level distribution of household poverty rates to classify ZIP codes as low poverty (ie, below mean poverty, <11%), moderate poverty (ie, between mean and plus 1 SD of the distribution, poverty rates 11%–21%), high poverty (plus 1–2 SDs, poverty rates 21%–31%), and extreme poverty (>2 SDs, >31% poverty).
Medication Refills and Measures of TDF/FTC Adherence and Discontinuation
We assigned an index date for PrEP initiation for each patient based on the date of the first TDF/FTC fill and characterized the longitudinal course of PrEP use during the year after initiation using a cabinet supply approach.14 Briefly, this method uses the pattern of refill dates and days supply values to estimate the medication supply available to a patient on each day during the year. We used these values to calculate the proportion of days covered (PDC) metric as a summary measure reflecting adherence in the entire first year after starting PrEP. We defined PDC as the number of nonzero cabinet supply days during the 1-year follow-up period divided by 365.15 This measure included all 365 days in the first year in the denominator for all individuals in the cohort, based on an assumption that all individuals in the cohort had need for PrEP throughout the first year. Consistent with a previous study of PrEP adherence, we also created a binary measure of high adherence, defined as a PDC >0.80.16 We chose this threshold because it has been previously applied in PrEP studies and indicates adherence substantially over the threshold required for effective HIV prevention (ie, 4 doses/wk or ∼60% adherence), whereas values below 0.8 may indicate need for adherence interventions.3,16
To describe adherence patterns in the first year more completely, we used the method of Steiner et al to create an alternative measure of TDF/FTC adherence that examined only the period between the first and last TDF/FTC fill in the year among patients with more than 1 fill (N = 977, 90% of cohort).15 This method is analogous to the method used by Marcus et al16 to estimate TDF/FTC adherence in a study of PrEP users in the Kaiser Permanente Northern California (KPNC) Health Care System, allowing for direct comparison of adherence estimates between cohorts. Although the PDC measure described above provides an overall measure of TDF/FTC adherence in the first year, it includes all 365 days in the first year in the denominator for all patients and counts days after the last TDF/FTC fill as nonadherence. By contrast, the Steiner method does not include days after the last TDF/FTC fill in the denominator and estimates adherence during periods of TDF/FTC refills, ignoring periods after discontinuation of refills. We calculated the Steiner adherence measure by: (1) determining the number of days with TDF/FTC in possession between the first day of the first TDF/FTC fill and the first day of the last fill in the year and (2) dividing this by the number of days in this interval. As for the PDC measure, we also created a binary indicator variable for high adherence based on the Steiner method (ie, Steiner adherence >0.8).
Finally, following methods previously reported by Marcus et al16 in the study of PrEP users in the KPNC Health Care System, we created a measure of time to PrEP discontinuation during the first year of use. We defined discontinuation as a continuous period of 120 or more days without TDF/FTC in possession beginning in the year after PrEP initiation and time to discontinuation as the number of days between initiation and the first day of the first 120-day gap in supply. We also created a binary variable indicating discontinuation during the first year. Calculation of this measure for the entire year after initiating PrEP required refill data through 485 days after initiation, which were available for the subset of the cohort beginning PrEP before March 1, 2015 (N = 825).
We began by determining the median and interquartile range (IQR) for the PDC in the first year, overall and by each patient characteristic. We examined bivariable associations between each patient characteristic and adherence as a continuous variable, using Wilcoxon rank-sum and Kruskal-Wallis tests to evaluate statistical significance.
Next, we used logistic regression to evaluate bivariable and multivariable associations between patient characteristics and the binary measure of high adherence (ie, PDC >0.80). Models included a random effect for the intercept for each VHA facility to account for clustering of patients in facilities. Apart from the indicator variable for chronic kidney disease, all variables were included in the multivariable model, as these were hypothesized to be associated with adherence before analysis. We did not include the chronic kidney disease indicator variable in models because the causal direction of its association with PDC could not be assumed; it was possible TDF/FTC use led to a CKD diagnosis, leading to PrEP discontinuation. We examined variance inflation factors for variables in the multivariable model to exclude collinearity (ie, variance inflation factor <5). Missing values for race and ZIP-code level poverty were coded separately in regression models. We then repeated these analyses replacing the PDC adherence measure with the alternative Steiner adherence measure and compared results for sensitivity to use of varying methods.
We identified patients who discontinued PrEP in the first year based on a 120 period without TDF/FTC in possession, and generated Kaplan–Maier curves for PrEP discontinuation based on number of days to the beginning of the first 120-day period without TDF/FTC in possession. We also examined bivariable associations between patient characteristics and the indicator variable for discontinuation, using χ2 tests to evaluate statistical significance. All analyses were approved by the institutional review board at the University of Iowa.
We identified 1086 Veterans initiating PrEP in VHA through June 30, 2016. More than 1 in 4 PrEP users (26.7%) in VHA were older than 50 years when initiating PrEP (Table 1). Most were men (96.3%) and white (66.9%); 21.6% were black and 11.6% had other or missing race recorded in administrative records. Over half (53.4%) had a depression diagnosis in the year surrounding PrEP imitation and 21.6% had a diagnosis of a substance use problem. Nearly half (47.6%) lived in ZIP codes with low household poverty, whereas 15.4% lived in a ZIP code with high or extreme household poverty rates. All census regions were represented in this national cohort (35.8% in the South, 40.4% in the West, 12.2% in the Midwest, and 11.6% in the Northeast). Only 4.3% lived in rural areas.
The median PDC for TDF/FTC in the first year was 0.74 (IQR 0.40–0.92, Table 2). PDC as a continuous measure was associated with greater age, male sex, white race, absence of a substance use diagnosis, and presence of diabetes (Table 2). There was little variation in the PDC by geography (ie, census region or rural/urban residence), area socioeconomic status based on ZIP-code level poverty, or coexisting diagnoses of depression or hypertension.
High adherence was common; 40% had a PDC >0.8 in the first year after starting PrEP. In multivariable analysis (Table 3), high adherence based on the PDC was associated with older age (odds ratio 1.97; 1.41–2.74 for age 50–64 compared with <35), white compared with black race (odds ratio 2.12; 1.53–2.93), male sex (odds ratio 3.39; 1.37–8.42), and a coexisting diagnosis of diabetes (odds ratio 2.02; 1.25–3.28). High adherence did not vary significantly by presence of a substance use diagnosis, rural residence, census region, ZIP-code level poverty, or presence of coexisting diagnoses of substance use, depression, or hypertension.
Associations between patient characteristics and adherence were similar using both the PDC and the Steiner method to estimate adherence (results for Steiner method included in Supplemental Digital Content, http://links.lww.com/QAI/B93). As expected, the Steiner method yielded a higher overall adherence (median 0.91; IQR 0.75–0.99) than the PDC because days after the last TDF/FTC fill in the year were not included in the denominator. Most patients in the cohort (60%) had high adherence based on adherence of over 0.8 based on the Steiner method. As in analyses using the PDC, patient characteristics associated with high adherence based on the Steiner method included older age, white compared with black race, and male sex (full results of multivariable modes included in Supplemental Digital Content, http://links.lww.com/QAI/B93).
Forty-four percent of PrEP initiators discontinued PrEP in the first year, based on a 120-day gap in TDF/FTC possession (Table 4). The median number of days to discontinuation in this group was 124. Discontinuation was more common among individuals who were younger (51.5% among those age <35 compared with 39.2% among those 50–64, P = 0.008), black compared with white race (52.4% vs. 41.6%, P = 0.053), and female sex (80.0% vs. 42.8%, P < 0.001) (Table 4). The Kaplan–Meir curve for time to discontinuation (Fig. 1) showed ongoing discontinuation throughout the first year, with prominent steps in discontinuation at 30 and 90 days related to frequent use of 30-day and 90-day medication fills. Overall, 75.5% of TDF/FTC fills were for 30 days, 21.3% for 90 days, 1.7% for 60 days, and 1.5% for other durations. Only 10% had a single fill of TDF/FTC during the year with no refills.
First-year adherence with TDF/FTC was overall high in this nationwide cohort of Veterans initiating PrEP in routine practice settings in VHA. Reassuringly, the median PDC by PrEP medication in the first year (0.74) was above the threshold for adherence associated with PrEP efficacy in clinical trials (ie, >0.60).3 However, there was substantial variation in adherence by patient demographics. Patient characteristics associated with higher PrEP adherence in our analyses (eg, older age, white race, and male sex) have also been associated with higher antiretroviral adherence and virologic suppression among persons in care for HIV infection.17 This suggests potential for race, age, and sex-related disparities in PrEP effectiveness in routine practice similar to those observed for virologic suppression among HIV-infected persons in care for HIV infection.
A previous study in the KPNC health care system also used medication refills to estimate TDF/FTC adherence and discontinuation rates.16 The KPNC study described a lower frequency of discontinuation (23%) compared with our study in VHA (44%), using an identical definition for PrEP discontinuation based on a 120-day gap in medication possession. This may be due to differences in patient characteristics in the KPNC system compared with VHA or in observation time for patients in the 2 studies. The KPNC study reported TDF/FTC adherence of 92% using a pharmacy refill-based method that examined periods between the first and last medication fill for each patient. We observed a remarkably similar first-year adherence (91%) among Veterans initiating PrEP in VHA using the Steiner method that examined only periods between medication fills.
We estimated adherence based on the PDC for TDF/FTC in the entire first year after initiating PrEP because this created a uniform observation period for all patients. This assumed that PrEP was indicated on each day of the first year based on ongoing risk of exposure to HIV infection. In reality, some individuals undoubtedly chose to discontinue PrEP before the end of the first year in response to decreasing perceived risk of HIV infection, medication intolerance, or some other reassessment of the risks and benefits of PrEP. In other cases, apparent discontinuation based on cessation of refills before the end of the year was likely related to an extended period of nonadherence, despite ongoing risk.
Our definition of PrEP discontinuation based on a 120-day gap in medication possession was ultimately arbitrary, and no refill-based measure can perfectly distinguish intentional discontinuation of PrEP from extended nonadherence. We did not have access to behavioral measures of HIV risk of each patient throughout the first year or reasons for discontinuation discussed with providers, which would have been necessary to calculate medication adherence during intervals when PrEP was clearly indicated. We therefore also examined an alternative measure for adherence that examined only the period between the first and last fill in the year, ignoring periods after the last fill. We observed generally high adherence and similar associations between patient characteristics and high adherence using both measures. There is no single best method for estimating adherence based on medication refill patterns and this is a general issue for monitoring PrEP adherence, both for quality assessment and epidemiologic purposes.
A previous, multisite chart review study in VHA found the overwhelming majority (ie, 90%) of Veterans receiving PrEP had initiated a conversation with a health care provider about PrEP, with most specifically requesting the medication.18 Furthermore, a quarter of these individuals were denied PrEP initially and made requests on multiple occasions to obtain it. These individuals were presumably highly motivated and activated to begin PrEP. Viewed in this context, it is not surprising that adherence with TDF/FTC would be high in a self-motivated group of relatively early PrEP adopters. As the PrEP innovation diffuses to larger groups of persons at risk of HIV infection—including populations with lower preexisting motivation to start PrEP—it will be important to monitor rates of medication adherence.
This study has limitations. There were few women in the cohort, consistent with the demographics of patients in VHA care in general. Estimates of differences in adherence and discontinuation by sex were therefore based on relatively small numbers of women using PrEP. The VHA care context may differ in important ways from non-VHA care systems. For example, VHA specialty and primary care clinics often have resources to support medication adherence, such as onsite pharmacists performing medication counseling, which may be lacking in other settings outside VHA. We used administrative data and an algorithm to identify Veterans initiating PrEP, and there was potential for misclassification. Refill-based measures of medication adherence capture only whether the prescription was filled, not whether the patient took the medicine. Refill measures also presume that Veterans do not receive TDF/FTC outside the VHA system, but this is unlikely as most payers would be reluctant to assume high medication costs for VHA enrollees.
We observed high levels of medication adherence in a national cohort of Veterans initiating PrEP in routine clinical practice. Important variation existed in adherence based on age, sex, and race, suggesting potential for disparities in PrEP effectiveness in practice. It will be important to monitor PrEP adherence, as uptake expands from highly motivated early adopters to larger populations.
3. Anderson PL, Glidden DV, Liu A, et al. Emtricitabine-tenofovir concentrations and pre-exposure prophylaxis
efficacy in men who have sex with men. Sci Transl Med. 2012;4:151ra125.
4. Koss CA, Bacchetti P, Hillier SL, et al. Differences in cumulative exposure and adherence
to tenofovir in the VOICE, iPrEx OLE, and PrEP Demo studies as determined via hair concentrations. AIDS Res Hum Retroviruses. 2017. doi: .
5. Yun K, Xu JJ, Zhang J, et al. Female and younger subjects have lower adherence
in PrEP trials: a meta-analysis with implications for the uptake of PrEP service to prevent HIV
. Sex Transm Infect. 2017. doi: .
7. Saha S, Freeman M, Toure J, et al. Racial and ethnic disparities in the VA health care system: a systematic review. J Gen Intern Med. 2008;23:654–671.
, Hepatitis, and Related Pathogens Program, Office of Patient Care Services, Veterans Health Administration. Pre-exposure Prophylaxis
(PrEP) to Prevent HIV
Infection: Clinical Considerations from the Department of Veterans Affairs National HIV
Program. 2016. Available at: https://www.hiv.va.gov/pdf/PrEP-considerations.pdf
. Accessed August 27, 2017.
9. Flash C, Landovitz R, Giler RM, et al. Two years of Truvada for pre-exposure prophylaxis
utilization in the US. J Int AIDS Soc. 2014;17(4 suppl 3):19730.
11. Stroupe KT, Tarlov E, Zhang Q, et al. Use of Medicare and DOD data for improving VA race data quality. J Rehabil Res Dev. 2010;47:781–795.
12. Quan H, Sundararajan V, Halfon P, et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care. 2005;43:1130–1139.
13. Hart LG, Larson EH, Lishner DM. Rural definitions for health policy and research. Am J Public Health. 2005;95:1149–1155.
14. Mosher HJ, Richardson KK, Lund BC. The 1-year treatment course of new opioid recipients in Veterans Health Administration. Pain Med. 2016;17:1282–1291.
15. Raebel MA, Schmittdiel J, Karter AJ, et al. Standardizing terminology and definitions of medication adherence
and persistence in research employing electronic databases. Med Care. 2013;51(8 suppl 3):S11–S21.
16. Marcus JL, Hurley LB, Hare CB, et al. Preexposure prophylaxis for HIV
prevention in a large integrated health care system: adherence
, renal safety, and discontinuation. J Acquir Immune Defic Syndr. 2016;73:540–546.
17. Beer L, Mattson CL, Bradley H, et al; Medical Monitoring P. Understanding cross-sectional racial, ethnic, and gender disparities in antiretroviral use and viral suppression among HIV
patients in the United States. Medicine (Baltimore). 2016;95:e3171.
18. Skolnik A, Giford A, Bokhour B, et al. Pathways to PrEP: what medical records reveal about access to HIV pre-exposure prophylaxis
in VA. Paper presented at: 2017 HSR&D/QUERI National Conference; July 20, 2017; Washington, DC.
HIV; pre-exposure prophylaxis; adherence; Veteran
Supplemental Digital Content
Copyright © 2018 Wolters Kluwer Health, Inc. All rights reserved.